arxiv_id
stringlengths 11
13
| markdown
stringlengths 2.09k
423k
| paper_doi
stringlengths 13
47
⌀ | paper_authors
sequencelengths 1
1.37k
| paper_published_date
stringdate 2014-06-03 19:36:41
2024-08-02 10:23:00
| paper_updated_date
stringdate 2014-06-27 01:00:59
2025-04-23 08:12:32
| categories
sequencelengths 1
7
| title
stringlengths 16
236
| summary
stringlengths 57
2.54k
|
|---|---|---|---|---|---|---|---|---|
1408.0238v1 | ## On the Second Approximate Matsumoto Metric
A. Tayebi, T. Tabatabaeifar and E. Peyghan
August 4, 2014
## Abstract
In this paper, we study the second approximate Matsumoto metric F = α + β + β /α 2 + β /α 3 2 on a manifold M . We prove that F is of scalar flag curvature and isotropic S-curvature if and only if it is isotropic Berwald metric with almost isotropic flag curvature.
Keywords : Isotropic Berwald curvature, S-curvature, almost isotropic flag curvature. 1
## 1 Introduction
The flag curvature in Finsler geometry is a natural extension of the sectional curvature in Riemannian geometry, which is first introduced by L. Berwald. For a Finsler manifold ( M,F ), the flag curvature is a function K ( P, y ) of tangent planes P ⊂ T M x and directions y ∈ P . F is said to be of scalar flag curvature if the flag curvature K ( P, y ) = K ( x, y ) is independent of flags P associated with any fixed flagpole y . F is called of almost isotropic flag curvature if
$$\mathbf K = \frac { 3 c _ { x ^ { m } } y ^ { m } } { F } + \sigma,$$
where c = c x ( ) and σ = σ x ( ) are scalar functions on M . One of the important problems in Finsler geometry is to characterize Finsler manifolds of almost isotropic flag curvature [11].
To study the geometric properties of a Finsler metric, one also considers non-Riemannian quantities. In Finsler geometry, there are several important non-Riemannian quantities: the Cartan torsion C , the Berwald curvature B , the mean Landsberg curvature J and S-curvature S , etc [6][9][11][18]. These are geometric quantities which vanish for Riemnnian metrics.
Among the non-Riemannian quantities, the S-curvature S = S ( x, y ) is closely related to the flag curvature which constructed by Shen for given comparison theorems on Finsler manifolds. A Finsler metric F is called of isotropic S -curvature if
$$\mathbf S = ( n + 1 ) c F,$$
1 2010 Mathematics subject Classification: 53C60, 53C25.
for some scalar function c = c x ( ) on M . In [11], it is proved that if a Finsler metric F of scalar flag curvature is of isotropic S-curvature (2), then it has almost isotropic flag curvature (1).
The geodesic curves of a Finsler metric F = F x, y ( ) on a smooth manifold M , are determined by ¨ c i +2 G c i ( ˙) = 0, where the local functions G i = G x,y i ( ) are called the spray coefficients. A Finsler metric F is called a Berwald metric, if G i are quadratic in y ∈ T M x for any x ∈ M . A Finsler metric F is said to be isotropic Berwald metric if its Berwald curvature is in the following form
$$B ^ { i } _ { \ j k l } = c \Big \{ F _ { y ^ { j } y ^ { k } } \delta ^ { i } _ { \ l } + F _ { y ^ { k } y ^ { l } } \delta ^ { i } _ { \ j } + F _ { y ^ { l } y ^ { j } } \delta ^ { i } _ { \ k } + F _ { y ^ { j } y ^ { k } y ^ { l } } y ^ { i } \Big \}, \quad \quad ( 3 ) \\ \text{where } c = c ( x ) \text{ is a scalar function on } M \ [ 6 ].$$
where c = ( ) is a scalar function on c x M [6].
As a generalization of Berwald curvature, B´cs´-Matsumoto proposed the a o notion of Douglas curvature [1]. A Finsler metric is called a Douglas metric if G i = 1 2 Γ i jk ( x y y ) j k + P x, y y ( ) i .
In order to find explicit examples of Douglas metrics, we consider ( α, β )-metrics. An ( α, β )-metric is a Finsler metric of the form F := αφ ( β α ), where φ = φ s ( ) is a C ∞ on ( -b 0 , b 0 ) with certain regularity, α = √ a ij ( x y y ) i j is a Riemannian metric and β = b i ( x y ) i is a 1-form on M . This class of metrics is were first introduced by Matsumoto [10]. Among the ( α, β )-metrics, the Matsumoto metric is special and significant metric which constitute a majority of actual research. The Matsumoto metric is expressed as
$$F = \alpha & \left [ 1 + \frac { \beta } { \alpha } + ( \frac { \beta } { \alpha } ) ^ { 2 } + ( \frac { \beta } { \alpha } ) ^ { 3 } + \cdots \right ]. \\ \text{as introduced by Matsumoto as a realization of}$$
This metric was introduced by Matsumoto as a realization of Finsler's idea 'a slope measure of a mountain with respect to a time measure' [19]. In the Matsumoto metric, the 1-form β = b y i i was originally to be induced by earth gravity. Hence, we could regard b i ( x ) as the the infinitesimals and neglect the infinitesimals of degree of b i ( x ) more than two [12][13][14][15][16]. An approximate Matsumoto metric is a Finsler metric in the following form
$$F = \alpha \left [ \sum _ { k = 0 } ^ { r } ( \frac { \beta } { \alpha } ) ^ { k } \right ], \\ \text{more information see [13]} \text{ \ This metric was introduced}$$
where | β | < α | | (for more information, see [13]). This metric was introduced by Park-Choi in [13]. By definition, the Matsumoto metric is expressed as lim r →∞ L α, β ( ) = α 2 α -β .
In this paper, we consider second approximate Matsumoto metric F = α + β + β 2 α + β 3 α 2 with some non-Riemannian curvature properties and prove the following.
Theorem 1.1. Let F = α + β + β 2 α + β 3 α 2 be a non-Riemannian second approximate Matsumoto metric on a manifold M of dimension n . Then F is of scaler flag curvature with isotropic S -curvature (2), if and only if it has isotropic Berwald curvature (3) with almost isotropic flag curvature (1). In this case, F must be locally Minkowskian.
## 2 Preliminaries
Let M be a n-dimensional C ∞ manifold. Denote by T M x the tangent space at x ∈ M , by TM = ∪ x ∈ M x T M the tangent bundle of M , and by TM 0 = TM \ { 0 } the slit tangent bundle on M . A Finsler metric on M is a function F : TM → [0 , ∞ ) which has the following properties: (i) F is C ∞ on TM 0 ;
- (ii) F is positively 1-homogeneous on the fibers of tangent bundle TM ;
- (iii) for each y ∈ T M x , the following quadratic form g y on T M x is positive definite,
$$\mathbf g _ { y } ( u, v ) \coloneqq \frac { 1 } { 2 } \frac { \partial ^ { 2 } } { \partial s \partial t } \left [ F ^ { 2 } ( y + s u + t v ) \right ] | _ { s, t = 0 }, \ u, v \in T _ { x } M. \\ \text{et } x \in M \text{ and } F _ { x } \coloneqq F | _ { T _ { x } M }. \text{ To measure the non-Euclidean feature of } F _ { x }$$
Let x ∈ M and F x := F | T M x . To measure the non-Euclidean feature of F x , define C y : T M x ⊗ T M x ⊗ T M x → R by
$$\mathbf C _ { y } ( u, v, w ) \coloneqq \frac { 1 } { 2 } \frac { d } { d t } \left [ \mathbf g _ { y + t w } ( u, v ) \right ] | _ { t = 0 }, \ u, v, w \in T _ { x } M. \\ \intertext { a l m v \, } \mathbf C \coloneqq \{ \mathbf C _ { x } \} _ { \sim \mathbf T _ { M } } \text{ is called the Cartan torsion. It is well known } t$$
The family C := { C y } y ∈ TM 0 is called the Cartan torsion. It is well known that C = 0 if and only if F is Riemannian [17]. For y ∈ T M x 0 , define mean Cartan torsion I y by I y ( u ) := I i ( y u ) i , where I i := g jk C ijk . By Diecke Theorem, F is Riemannian if and only if I y = 0.
The horizontal covariant derivatives of I along geodesics give rise to the mean Landsberg curvature J y ( u ) := J i ( y u ) i , where J i := I i s | y s . A Finsler metric is said to be weakly Landsbergian if J = 0.
Given a Finsler manifold ( M,F ), then a global vector field G is induced by F on TM 0 , which in a standard coordinate ( x , y i i ) for TM 0 is given by G = y i ∂ ∂x i -2 G x,y i ( ) ∂ ∂y i , where
$$G ^ { i } \coloneqq \frac { 1 } { 4 } g ^ { i l } \left [ \frac { \partial ^ { 2 } ( F ^ { 2 } ) } { \partial x ^ { k } \partial y ^ { l } } y ^ { k } - \frac { \partial ( F ^ { 2 } ) } { \partial x ^ { l } } \right ], \quad y \in T _ { x } M. \\ \text{-} \text{-} \text{allad to an even a-occuritized $t_{\mathbb{ }T}_{M}_{F}\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb { }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{t_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathBB{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T$_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathfrak{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T _\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb
}T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{ \mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{
}T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T _$_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\tilde{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T._\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{\ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{ \mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb`T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{ }T_{\mathbb{$$
The G is called the spray associated to ( M,F ). In local coordinates, a curve c t ( ) is a geodesic if and only if its coordinates ( c i ( )) satisfy ¨ t c i +2 G c i ( ˙) = 0.
For a tangent vector y ∈ T M x 0 , define B y : T M x ⊗ T M x ⊗ T M x → T M x and E y : T M x ⊗ T M x → R by B y ( u, v, w ) := B i jkl ( y u v ) j k w l ∂ ∂x i | x and E y ( u, v ) := E jk ( y u v ) j k where
$$B ^ { i } _ { \ j k l } \coloneqq \frac { \partial ^ { 3 } G ^ { i } } { \partial y ^ { j } \partial y ^ { k } \partial y ^ { l } }, \ \ E _ { j k } \coloneqq \frac { 1 } { 2 } B ^ { m } _ { \ j k m }.$$
The B and E are called the Berwald curvature and mean Berwald curvature, respectively. Then F is called a Berwald metric and weakly Berwald metric if B = 0 and E = 0 , respectively.
A Finsler metric F is said to be isotropic mean Berwald metric if its mean Berwald curvature is in the following form
$$E _ { i j } = \frac { n + 1 } { 2 F } c h _ { i j }, \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
\quad. \quad.$$
where c = ( ) is a scalar function on c x M and h ij is the angular metric [6].
Define D y : T M x ⊗ T M x ⊗ T M x → T M x by D y ( u, v, w ) := D i jkl ( y u v ) i j w k ∂ ∂x i | x where
$$\ v n e r e \quad & D ^ { i } _ { \ j k l } \coloneqq B ^ { i } _ { \ j k l } - \frac { 2 } { n + 1 } \{ E _ { j k } \delta ^ { i } _ { l } + E _ { j l } \delta ^ { i } _ { k } + E _ { k l } \delta ^ { i } _ { j } + E _ { j k, l } y ^ { i } \}. \\ \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad.$$
We call D := { D y } y ∈ TM 0 the Douglas curvature. A Finsler metric with D = 0 is called a Douglas metric. The notion of Douglas metrics was proposed by B´cs´-Matsumoto as a generalization of Berwald metrics [1]. a o
For a Finsler metric F on an n -dimensional manifold M , the BusemannHausdorff volume form dV F = σ F ( x dx ) 1 · · · dx n is defined by
∣ In general, the local scalar function σ F ( x ) can not be expressed in terms of elementary functions, even F is locally expressed by elementary functions. Let G i denote the geodesic coefficients of F in the same local coordinate system. The S-curvature can be defined by
$$\cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdot \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \colon \\ \sigma _ { F } ( x ) \coloneqq \frac { \text{Vol} ( \mathbb { B } ^ { n } ( 1 ) ) } { \text{Vol} \left \{ ( y ^ { i } ) \in R ^ { n } \, \Big | \ F \left ( y ^ { i } \frac { \partial } { \partial x ^ { i } } | _ { x } \right ) < 1 \right \} \\ \text{al, the local scalar function $\sigma_{F} ( x ) \, \text{ can not be expressed in}$} \\ \text{ry functions, even $F$ is locally expressed by elementary funct}$$
$$\mathbf S ( \mathbf y ) \coloneqq \frac { \partial G ^ { i } } { \partial y ^ { i } } ( x, y ) - y ^ { i } \frac { \partial } { \partial x ^ { i } } \left [ \ln \sigma _ { F } ( x ) \right ], \\ \frac { \partial } { \partial x } | _ { x } \in T _ { r } M. \text{ It is proved that } \mathbf S = 0 \text{ if } F \text{ is a Be}$$
where y = y i ∂ ∂x i | x ∈ T M x . It is proved that S = 0 if F is a Berwald metric. There are many non-Berwald metrics satisfying S = 0. S said to be isotropic if there is a scalar functions c x ( ) on M such that S = ( n +1) ( c x F ) .
The Riemann curvature R y = R i k dx k ⊗ ∂ ∂x i | x : T M x → T M x is a family of linear maps on tangent spaces, defined by
$$R ^ { i } _ { \ k } = 2 \frac { \partial G ^ { i } } { \partial x ^ { k } } - y ^ { j } \frac { \partial ^ { 2 } G ^ { i } } { \partial x ^ { j } \partial y ^ { k } } + 2 G ^ { j } \frac { \partial ^ { 2 } G ^ { i } } { \partial y ^ { j } \partial y ^ { k } } - \frac { \partial G ^ { i } } { \partial y ^ { j } } \frac { \partial G ^ { j } } { \partial y ^ { k } }.$$
For a flag P = span { y, u } ⊂ T M x with flagpole y , the flag curvature K = K ( P, y ) is defined by
$$\mathbf K ( P, y ) \coloneqq \frac { \mathbf g _ { y } ( u, \mathbf R _ { y } ( u ) ) } { \mathbf g _ { y } ( y, y ) \mathbf g _ { y } ( u, u ) - \mathbf g _ { y } ( y, u ) ^ { 2 } }. \\ \text{a Finsler metric $F$ is of scalar curvature if for any $u$}$$
We say that a Finsler metric F is of scalar curvature if for any y ∈ T M x , the flag curvature K = K ( x, y ) is a scalar function on the slit tangent bundle TM 0 . In this case, for some scalar function K on TM 0 the Riemann curvature is in the following form
$$R ^ { i } _ { \ k } = \mathbf K F ^ { 2 } \{ \delta ^ { i } _ { k } - F ^ { - 1 } F _ { y ^ { k } } y ^ { i } \}. \\ \text{hon } F \text{ is said to be of content } \text{a} \, \text{a} \, \text{$c}$$
If K = constant , then F is said to be of constant flag curvature. A Finsler metric F is called isotropic flag curvature , if K = K ( x ).
## 3 Proof of Theorem 1.1
Let F = αφ s ( ), s = β α be an ( α, β )-metric, where φ = φ s ( ) is a C ∞ on ( -b 0 , b 0 ) with certain regularity, α = √ a ij ( x y y ) i j is a Riemannian metric and β = b i ( x y ) i is a 1-form on a manifold M . Let
$$r _ { i j } \coloneqq \frac { 1 } { 2 } \left [ b _ { i | j } + b _ { j | i } \right ], \quad s _ { i j } \coloneqq \frac { 1 } { 2 } \left [ b _ { i | j } - b _ { j | i } \right ]. \\ r _ { j } \coloneqq b ^ { i } r _ { i j }, \quad s _ { j } \coloneqq b ^ { i } s _ { i j }. \\ \quad \,.$$
where b i j | denote the coefficients of the covariant derivative of β with respect to α . Let
$$r _ { i 0 } \coloneqq r _ { i j } y ^ { j }, \ s _ { i 0 } \coloneqq s _ { i j } y ^ { j }, \ r _ { 0 } \coloneqq r _ { j } y ^ { j }, \ s _ { 0 } \coloneqq s _ { j } y ^ { j }.$$
Put
$$Q & = \frac { \phi ^ { \prime } } { \phi - s \phi }, \\ \Theta & = \frac { \phi \phi ^ { \prime } - s ( \phi \phi ^ { \prime \prime } + \phi ^ { \prime 2 } ) } { 2 \phi \left [ ( \phi - s \phi ^ { \prime } ) + ( b ^ { 2 } - s ^ { 2 } ) \phi ^ { \prime \prime } \right ] } \\ \Psi & = \frac { \phi ^ { \prime \prime } } { 2 \left [ ( \phi - s \phi ^ { \prime } ) + ( b ^ { 2 } - s ^ { 2 } ) \phi ^ { \prime \prime } \right ] }. \\ \text{urvature is given by} \\ \prime \quad \underset { \theta } { \sim } \cap \quad \ \overline { \nu } \cap \overline { \nu } \cap \ \overline { \nu } \cap \ \overline { \nu } \end{pmatrix}$$
Then the S -curvature is given by
$$\mathbf S = \left [ Q ^ { \prime } - 2 \Psi Q s & - \ 2 ( \Psi Q ) ^ { \prime } ( b ^ { 2 } - s ^ { 2 } ) - 2 ( n + 1 ) Q \Theta + 2 \lambda \right ] s _ { 0 } \\ & + \ 2 ( \Psi + \lambda ) s _ { 0 } + \alpha ^ { - 1 } \left [ ( b ^ { 2 } - s ^ { 2 } ) \Psi ^ { \prime } + ( n + 1 ) \Theta \right ] r _ { 0 0 }. \\ \text{Let us put}$$
Let us put
$$\Delta & \coloneqq \ 1 + s Q + ( b ^ { 2 } - s ^ { 2 } ) Q ^ { \prime }, \\ \Phi & \coloneqq \ - ( n \Delta + 1 + s Q ) ( Q - s Q ^ { \prime } ) - ( b ^ { 2 } - s ^ { 2 } ) ( 1 + s Q ) Q ^ { \prime \prime }. \\ \vdots \sim \sim \ \sim \ \sim \ \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot$$
In [5], Cheng-Shen characterize ( α, β )-metrics with isotropic S-curvature.
Lemma 3.1. ([5]) Let F = αφ β/α ( ) be an ( α, β )-metric on an n -manifold. Then, F is of isotropic S-curvature S = ( n + 1) cF , if and only if one of the following holds
## (i) β satisfies
$$r _ { i j } = \varepsilon \left \{ b ^ { 2 } a _ { i j } - b _ { i } b _ { j } \right \}, \quad s _ { j } = 0, & & ( 8 ) \\ x ) \text{ is a scalar function, and } \phi = \phi ( s ) \text{ satisfies}$$
where ε = ( ) is a scalar function, and ε x φ = φ s ( ) satisfies
$$\Phi = - 2 ( n + 1 ) k \frac { \phi \Delta ^ { 2 } } { b ^ { 2 } - s ^ { 2 } }, \\ \prod _ { \text{out in to each } } \varepsilon = \iota _ { \varepsilon }$$
where k is a constant. In this case, c = k/epsilon1 .
(ii) β satisfies
In this case, c = 0.
Let
$$\Psi _ { 1 } \, \coloneqq \, & \sqrt { b ^ { 2 } - s ^ { 2 } } \Delta ^ { \frac { 1 } { 2 } } \left [ \frac { \sqrt { b ^ { 2 } - s ^ { 2 } } \Phi } { \Delta ^ { \frac { 3 } { 2 } } } \right ] ^ { \prime }, \\ \Psi _ { 2 } \, \coloneqq \, & 2 ( n + 1 ) ( Q - s Q ^ { \prime } ) + 3 \frac { \Phi } { \Delta }, \\ \theta \, \coloneqq \, & \frac { Q - s Q ^ { \prime } } { 2 \Delta }. \\ \text{rmula for the mean Cartan torsion of an } ( \alpha, \beta ) \text{-metric is given by}$$
Then the formula for the mean Cartan torsion of an ( α, β )-metric is given by following
$$I _ { i } & = \frac { 1 } { 2 } \frac { \partial } { \partial y ^ { i } } \left [ ( n + 1 ) \frac { \phi ^ { \prime } } { \phi } - ( n - 2 ) \frac { s \phi ^ { \prime \prime } } { \phi - s \phi ^ { \prime } } - \frac { 3 s \phi ^ { \prime \prime } - ( b ^ { 2 } - s ^ { 2 } ) \phi ^ { \prime \prime \prime } } { ( \phi - s \phi ^ { \prime } ) + ( b ^ { 2 } - s ^ { 2 } ) \phi ^ { \prime \prime } } \right ] \\ & = - \frac { \Phi ( \phi - s \phi ^ { \prime } ) } { 2 \Delta \phi \alpha ^ { 2 } } ( \alpha b _ { i } - s y _ { i } ). \\ \vdots \, \vdots \, \vdots \, \dots \, \dots \, \vdots \, \vdots \, \vdots \, \dots \, \dots \, \dots \, \dots \, \dots \, \vdots \, \vdots \, \dots \, \dots \, \dots \, \dots \, \vdots \, \vdots \, \dots \, \dots \, \dots \, \dots \, \vdots \, \dots \, \dots \, \dots \, \dots \, \dots \, \vdots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \end{array}$$
In [7], it is proved that the condition Φ = 0 characterizes the Riemannian metrics among ( α, β )-metrics. Hence, in the continue, we suppose that Φ = 0.
/negationslash
Let G i = G x,y i ( ) and ¯ G i α = ¯ G i α ( x, y ) denote the coefficients of F and α respectively in the same coordinate system. By definition, we have
$$G ^ { i } = \bar { G } _ { \alpha } ^ { i } + P y ^ { i } + Q ^ { i },$$
where
$$P \coloneqq \ \alpha ^ { - 1 } \Theta \Big [ - 2 Q \alpha s _ { 0 } + r _ { 0 0 } \Big ] \\ Q ^ { i } \coloneqq \ \alpha Q s ^ { i } _ { 0 } + \Psi \Big [ - 2 Q \alpha s _ { 0 } + r _ { 0 0 } \Big ] b ^ { i }. \\ \text{ yields the following}$$
Simplifying (13) yields the following
$$G ^ { i } = \bar { G } _ { \alpha } ^ { i } + \alpha Q s _ { 0 } ^ { i } + \theta ( - 2 \alpha Q s _ { 0 } + r _ { 0 0 } ) \left [ \frac { y ^ { i } } { \alpha } + \frac { Q ^ { \prime } } { Q - s Q ^ { \prime } } b ^ { i } \right ]. \quad \ ( 1 4 ) \\ \text{Clearly, if } \beta \text{ is parallel with respect to } \alpha \ ( r _ { i j } = 0 \text{ and } s _ { i j } = 0 ), \text{ then } P = 0 \text{ and }$$
Clearly, if β is parallel with respect to α ( r ij = 0 and s ij = 0), then P = 0 and Q i = 0. In this case, G i = ¯ G i α are quadratic in y , and F is a Berwald metric.
For an ( α, β )-metric F = αφ s ( ), the mean Landsberg curvature is given by
$$& \text{For an } ( \alpha, \beta ) \text{-metric } F = \alpha \phi ( s ), \text{ the mean Landsberg curvature is given by} \\ J _ { i } = & \ - \frac { 1 } { 2 \Delta \alpha ^ { 4 } } \left [ \frac { 2 \alpha ^ { 2 } } { b ^ { 2 } - s ^ { 2 } } [ \frac { \Phi } { \Delta } + ( n + 1 ) ( Q - s Q ^ { \prime } ) ] ( r _ { 0 } + s _ { 0 } ) h _ { i } \\ & \ + \frac { \alpha } { b ^ { 2 } - s ^ { 2 } } ( \Psi _ { 1 } + s \frac { \Phi } { \Delta } ) ( r _ { 0 0 } - 2 \alpha Q s _ { 0 } ) h _ { i } + \alpha \left [ - \alpha Q ^ { \prime } s _ { 0 } h _ { i } + \alpha Q ( \alpha ^ { 2 } s _ { i } - y _ { i } s _ { 0 } ) \\ & \ + \alpha ^ { 2 } \Delta s _ { i 0 } + \alpha ^ { 2 } ( r _ { i 0 } - 2 \alpha Q s _ { i } ) - ( r _ { 0 0 } - 2 \alpha Q s _ { 0 } ) y _ { i } \right ] \frac { \Phi } { \Delta } \right ].$$
$$r _ { i j } = 0, \quad s _ { j } = 0.$$
Contracting (15) with b i = a im b m yields
$$\bar { J } \coloneqq J _ { i } b ^ { i } = - \frac { 1 } { 2 \Delta \alpha ^ { 2 } } \left [ \Psi _ { 1 } ( r _ { 0 0 } - 2 \alpha Q s _ { 0 } ) + \alpha \Psi _ { 2 } ( r _ { 0 } + s _ { 0 } ) \right ]. \quad \ ( 1 6 ) \\ \text{The horizontal covariant derivatives } J _ { i ; m } \text{ and } J _ { i | m } \text{ of } J _ { i } \text{ with respect to } F \text{ and}$$
The horizontal covariant derivatives J i m ; and J i m | of J i with respect to F and α , respectively, are given by
$$J _ { i ; m } = \frac { \partial J _ { i } } { \partial x ^ { m } } - J _ { l } \Gamma ^ { l } _ { i m } - \frac { \partial J _ { i } } { \partial y ^ { l } } N ^ { l } _ { m },$$
$$J _ { i | m } = \frac { \partial J _ { i } } { \partial x ^ { m } } - J _ { l } \bar { \Gamma } ^ { l } _ { i m } - \frac { \bar { \partial J _ { i } } } { \partial y ^ { l } } \bar { N } ^ { l } _ { m }.$$
Then we have
$$J _ { i ; m } y ^ { m } = J _ { i | m } y ^ { m } - J _ { l } ( N _ { i } ^ { l } - \bar { N } _ { i } ^ { l } ) - 2 \frac { \partial J _ { i } } { \partial y ^ { l } } ( G ^ { l } - \bar { G } ^ { l } ). \quad \quad ( 1 7 )$$
Let F be a Finsler metric of scalar flag curvature K . By Akbar-Zadeh's theorem it satisfies following
$$A _ { i j k ; s ; m } y ^ { s } y ^ { m } + \mathbf K F ^ { 2 } A _ { i j k } + \frac { F ^ { 2 } } { 3 } \left [ h _ { i j } \mathbf K _ { k } + h _ { j k } \mathbf K _ { j } + h _ { k i } \mathbf K _ { j } \right ] = 0, \quad ( 1 8 ) \\ \text{where } A _ { i j k } = F C _ { i j k } \text{ is the Cartan torsion and } \mathbf K _ { i } = \frac { \partial \mathbf K } { A. i } \, [ 2 ]. \text{ Contracting } ( 1 8 )$$
$$\Phi _ { \Phi _ { i } ; m } y ^ { m } + \mathbf K F ^ { 2 } I _ { i } + \frac { n + 1 } { 3 } F ^ { 2 } \mathbf K _ { i } = 0. \quad \ \quad \ \ ( 1 9 ) \\ \Phi _ { \Phi _ { i } ; m } \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nolimits \Phi _ { \Phi _ { i } } }$$
where A ijk = FC ijk is the Cartan torsion and K i = ∂ K ∂y i [2]. Contracting (18) with g ij yields
By (17) and (19), for an ( α, β )-metric F = αφ s ( ) of constant flag curvature K , the following holds
$$J _ { i | m } - J _ { l } \frac { \partial ( G ^ { l } - \bar { G } ^ { l } ) } { \partial y ^ { i } } b ^ { i } - 2 \frac { \partial \bar { J } } { \partial y ^ { l } } ( G ^ { l } - \bar { G } ^ { l } ) \mathbf K \alpha ^ { 2 } \phi ^ { 2 } I _ { i } = 0. \quad \ ( 2 0 )$$
Contracting (20) with b i implies that
$$\bar { J } _ { | m } y ^ { m } - J _ { i } a ^ { i k } b _ { k | m } y ^ { m } - J _ { l } \frac { \partial ( G ^ { l } - \bar { G } ^ { l } ) } { \partial y ^ { i } } b ^ { i } - 2 \frac { \partial \bar { J } } { \partial y ^ { l } } ( G ^ { l } - \bar { G } ^ { l } ) + \mathbf K \alpha ^ { 2 } \phi ^ { 2 } I _ { i } b ^ { i } = 0. \ ( 2 1 )$$
There exists a relation between mean Berwald curvature E and the Scurvature S . Indeed, taking twice vertical covariant derivatives of the S-curvature gives rise the E -curvature. It is easy to see that, every Finsler metric of isotropic S-curvature (2) is of isotropic mean Berwald curvature (5). Now, is the equation S = ( n +1) cF equivalent to the equation E = n +1 2 cF -1 h ?
/negationslash
Recently, Cheng-Shen prove that a Randers metric F = α + β is of isotropic S -curvature if and only if it is of isotropic E -curvature [4]. Then, Chun-HuanCheng extend this equivalency to the Finsler metric F = α -m ( α + β ) m +1 for every real constant m , including Randers metric [3]. In [8], Cui extend their result and show that for the Matsumoto metric F = α 2 α -β and the special ( α, β )-metric F = α + /epsilon1β + κ β /α ( 2 ) ( κ = 0), these notions are equivalent.
To prove Theorem 1.1, we need the following.
Proposition 3.2. Let F = α + β + β 2 α + β 3 α 2 be a second approximate Matsumoto metric on a manifold M of dimension n . Then the following are equivalent
- (i) F has isotropic S -curvature, S = ( n +1) ( c x F ) ;
- (ii) F has isotropic mean Berwald curvature, E = n +1 2 c x F ( ) -1 h ;
where c = ( ) c x is a scalar function on the manifold M . In this case, S = 0 . Then β is a Killing 1 -form with constant length with respect to α , that is, r 00 = 0 .
Proof. ( ) i → ( ii ) is obvious. Conversely, suppose that F has isotropic mean Berwald curvature, E = ( n +1) 2 c x F ( ) -1 h . Then we have
$$\mathbf S = ( n + 1 ) [ c F + \eta ],$$
where η = η i ( x y ) i is a 1-form on M . For the second approximate Matsumoto metric, (6) reduces to following
$$\text{etric, (0) requires to following} \\ Q = & - \frac { 1 + 2 s + 3 s ^ { 2 } } { - 1 + s ^ { 2 } + 2 s ^ { 3 } }, \\ \Theta = & \frac { 1 } { 2 } \frac { 1 - 6 s ^ { 2 } - 1 2 s ^ { 3 } - 1 5 s ^ { 4 } - 1 2 s ^ { 5 } } { ( 1 + s + s ^ { 2 } + s ^ { 3 } ) ( 1 - 3 s ^ { 2 } - 8 s ^ { 3 } + 2 b ^ { 2 } + 6 b ^ { 2 } s ) }, \\ \Psi = & \frac { 1 + 3 s } { ( 1 - 3 s ^ { 2 } - 8 s ^ { 3 } + 2 b ^ { 2 } + 6 b ^ { 2 } s ) }. \\ \mu = & \mu \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \colon$$
By substituting (22) and (23) in (7), we have
$$\Psi = & \, \frac { \left ( 1 - 3 s ^ { 2 } - 8 s ^ { 3 } + 2 b ^ { 2 } + 6 b ^ { 2 } s \right ) }. \\ \dots & \, \left ( \infty \, \cdots \, \cdots \, \left ( \infty \, \cdots \, \right ) \, \left ( \infty \, \cdots \, \cdots \, \cdots \, \cdots \,.$$
$$& \gamma - ( 1 - 3 s ^ { 2 } - 8 s ^ { 3 } + 2 b ^ { 2 } + 6 b ^ { 2 } s ) ^ { \cdot } \quad \langle \rangle ^ { \cdot \rangle } \\ \text{By substituting} \, ( 2 2 ) \, \text{and} \, ( 2 3 ) \, \text{in} \, ( 7 ), \, \text{we have} \\ \text{S} & = \, \left [ \frac { 2 ( 1 + 3 s ) ( 1 + s + s ^ { 2 } + s ^ { 3 } ) } { ( - 1 + s ^ { 2 } + 2 s ^ { 3 } ) ^ { 2 } } \, + \, ( 1 - 3 s ^ { 2 } - 8 s ^ { 3 } + 2 b ^ { 2 } + 6 b ^ { 2 } s ) ( - 1 + s ^ { 2 } + 2 s ^ { 3 } ) } \\ & \quad - \frac { 2 ( 5 + 2 6 s + 7 7 s ^ { 2 } + 8 8 s ^ { 3 } - 6 1 s ^ { 4 } - 4 3 0 s ^ { 5 } - 8 0 5 s ^ { 6 } + 4 b ^ { 2 } + 4 0 b ^ { 2 } s + 1 4 8 b ^ { 2 } s ) ( b ^ { 2 } - s ^ { 2 } ) } { ( 1 - 3 s ^ { 2 } - 8 s ^ { 3 } + 2 b ^ { 2 } + 6 b ^ { 2 } s ) ^ { 2 } ( - 1 + s ^ { 2 } + 2 s ^ { 3 } ) ^ { 2 } } \\ & \quad + \frac { ( n + 1 ) ( 1 + 2 s + 3 s ^ { 2 } ) ( 1 - 6 s ^ { 2 } - 1 2 s ^ { 3 } - 1 5 s ^ { 4 } - 1 2 s ^ { 5 } ) } { ( - 1 + s ^ { 2 } + 2 s ^ { 3 } ) ( 1 + s + s ^ { 2 } + s ^ { 3 } ) ( 1 - 3 s ^ { 2 } - 8 s ^ { 3 } + 2 b ^ { 2 } + 6 b ^ { 2 } s ) } + 2 \lambda \right ] s _ { 0 } \\ & \quad + 2 \left [ \frac { ( 1 + 3 s ) } { 1 - 3 s ^ { 2 } - 8 s ^ { 3 } + 2 b ^ { 2 } + 6 b ^ { 2 } s } \right ] + \frac { ( 1 ) s ( b ^ { 2 } - s ^ { 2 } ) ( 1 + 1 s ^ { 2 } + 1 6 s ^ { 3 } + 2 s ) } { \alpha ( 1 - 3 s ^ { 2 } - 8 s ^ { 3 } + 2 b ^ { 2 } + 6 b ^ { 2 } s ) ^ { 2 } } \right ] r _ { 0 0 } \\ & \quad + \left [ \frac { ( n + 1 ) ( 1 - 6 s ^ { 2 } - 1 2 s ^ { 3 } - 1 5 s ^ { 4 } - 1 2 s ^ { 5 } ) } { 2 \alpha ( 1 + s + s ^ { 2 } + s ^ { 3 } ) ( 1 - 3 s ^ { 2 } - 8 s ^ { 3 } + 2 b ^ { 2 } + 6 b ^ { 2 } s ) } \right ] r _ { 0 0 } \\ & = \, ( n + 1 ) \left [ \alpha ( 1 + s + s ^ { 2 } + s ^ { 3 } ) + \eta \right ]. \quad \, ( 2 4 ) \\ \text{Multiplying} \, ( 2 4 ) \, \text{with} \, ( - 1 + s ^ { 2 } + 2 s ^ { 3 } ) ( 1 + s + s ^ { 2 } + s ^ { 3 } ) ( 1 - 3 s ^ { 2 } - 8 s ^ { 3 } + 2 b ^ { 2 } + 6 b ^ { 2 } s ) ^ { 2 } \alpha ^ { 1 4 } \\ \text{ implies that}$$
$$\text{mases} \text{max} \\ M _ { 1 } + M _ { 2 } \alpha ^ { 2 } + M _ { 3 } \alpha ^ { 4 } + M _ { 4 } \alpha ^ { 6 } + M _ { 5 } \alpha ^ { 8 } + M _ { 6 } \alpha ^ { 1 0 } + M _ { 7 } \alpha ^ { 1 2 } + M _ { 8 } \alpha ^ { 1 4 } \\ + \alpha \left [ M _ { 9 } + M _ { 1 0 } \alpha ^ { 2 } + M _ { 1 1 } \alpha ^ { 4 } + M _ { 1 2 } \alpha ^ { 6 } + M _ { 1 3 } \alpha ^ { 8 } + M _ { 1 4 } \alpha ^ { 1 0 } \\ + M _ { 1 5 } \alpha ^ { 1 2 } + M _ { 1 6 } \alpha ^ { 1 4 } \right ] = 0, \quad ( 2 5 ) \\ 8$$
[ ] Multiplying (24) with ( -1+ s 2 +2 s 3 )(1+ s + s 2 + s 3 )(1 -3 s 2 -8 s 3 +2 b 2 +6 b 2 s ) 2 α 14 implies that (25)
## where
- M 1 := -128( n +1) cβ 15 , M 2 := 2 ( [ n +1) ( [ -385 + 96 b 2 ) cβ 2 -64 ηβ ] +128 ( λ r 0 + s 0 ) β +48 nr 00 ] β 11 , M 3 := -[ ( n +1)[4( -283 b 2 +243 + 18 b 4 ) cβ 2 -6(32 b 2 -59) ηβ ] + [12( -59 + 32 b 2 ) λ r ( 0 + s 0 ) -96((3 n -1) s 0 -r 0 ] β + 3( -24 b 2 +36 -219 n +72 nb 2 ) r 00 ] β , 9 M 4 := [ ( n +1) [ -2(29 -738 b 2 +208 b 4 ) cβ 2 -3( -172 b 2 +21 + 24 b 4 ) ηβ ] + ( -159 nb 2 -132 -39 n +96 b 2 ) r 00 -3( -48 b 2 +123 + 72 nb 2 -277 n s β ) 0 + 3(24 b 2 +86) r β 0 +6 ( λ -172 b 2 +21+24 b 4 )( s 0 + r 0 ) β β , ] 7 M 5 := [ ( n +1)[ -8( -27 b 2 +70 b 4 -41) cβ 2 -4(47 b 4 -40 -32 b 2 ) ηβ ] + (15 nb 2 +20 -55 n +108 b 2 ) r 00 +8 (47 λ b 4 -40 -32 b 2 )( r 0 + s 0 ) β -2 ( [ -262 b 2 +278 + 303 nb 2 -147 n s ) 0 -(94 b 2 +32) r 0 ] β β ] 5 M 6 := 2 ( [ n +1) [ -2(4 b +1)(4 b -1)(4 b 2 +13) cβ 2 -10(1 + 20 b 2 +6 b 4 ) ηβ ] + (11 n -40 b 2 +35 nb 2 +20) r 00 +20 (1 + 20 λ b 2 +6 b 4 )( r 0 + s 0 ) β -2(32 n +51 + 129 nb 2 -294 b 2 ) s β 0 -(30 b 2 -50) r β 0 ] β , 3 M 7 := -[ ( n +1) [ -4( -17 b 2 -7 + 30 b 4 ) cβ 2 -6( -1 + 10 b 4 ) β ] +(3 n +12) b 2 r 00 -12 λ (1 -10 b 4 )( s 0 + r 0 ) -6 ( [ nb 2 -n -2 + 14 b 2 ) s 0 -10 b 2 r 0 ] β β, ] M 8 := 4 ( [ n +1) 2(1 + 2 [ b 2 )(8 b 2 +1) cβ +(1 + 2 b 2 ) 2 η ] -2 λ (1 + 2 b 2 ) 2 ( s 0 + r 0 ) + [ -57 nb 2 -64( n +1) b 4 -8 n -55 b 2 +6 ] r 00 -(2 -4 b 2 ) r 0 + [ n +(4 + 2 n b ) 2 -1 ] s 0 ] , M 9 := -416( n +1) cβ 14 , M 10 := [ ( n +1)[( -1037 + 616 b 2 ) cβ 2 -288 ηβ ] + 576 λ r ( 0 + s 0 ) β +(204 n -6) r 00 ] β 12 , M 11 := -1 2 [ ( n +1)[8(57 b 4 -385 b 2 +143) cβ 2 -2(424 b 2 -267) ηβ ] + 4 (424 [ λ b 2 -267)( r 0 + s 0 ) -4( -115 + 330 n s ) 0 +1320 r 0 ] β + (300 nb 2 +249 -189 n -120 b 2 ) r 00 ] β , 8 M 12 := 4 ( [ n +1)[ -(572 b 4 -932 b 2 -275) cβ 2 -4(39 b 4 -28 -102 b 2 ) ηβ ] -75 nb 2 -62 -89 n +144 b 2 r 00 +8 (39 λ b 4 -28 -102 b 2 )( r 0 + s 0 ) β -6[( -58 b 2 +100 + 81 nb 2 -109 n s ) 0 -(26 b 2 +34) r 0 ] β β , ] 6
$$M _ { 1 3 } \coloneqq \ \left [ ( n + 1 ) [ - 8 ( - 2 4 + 3 5 b ^ { 2 } + 4 7 b ^ { 4 } ) c \beta ^ { 2 } - 6 ( 2 6 b ^ { 4 } - 1 1 + 2 0 b ^ { 2 } ) \eta \beta \right ] \\ & + ( 5 1 n b ^ { 2 } + 3 3 - 6 n + 2 4 b ^ { 2 } ) r _ { 0 0 } + 1 2 \lambda ( 2 6 b ^ { 4 } - 1 1 + 2 0 b ^ { 2 } ) ( s _ { 0 } + r _ { 0 } ) \\ & - 6 ( - 1 1 8 b ^ { 2 } + 5 4 + 8 3 n b ^ { 2 } - 3 n ) s _ { 0 } \beta + 6 ( 2 6 b ^ { 2 } - 1 0 ) r _ { 0 } \beta \right ] \beta ^ { 4 }, \\ M _ { 1 4 } \coloneqq \ \left [ ( n + 1 ) [ - ( 5 6 b ^ { 4 } - 2 7 2 b ^ { 2 } - 3 9 ) c \beta ^ { 2 } - 4 ( 7 ( n + 1 ) b ^ { 4 } - 6 ( 1 + n ) - 2 2 n b ^ { 2 } ) \eta \beta \right ] \\ & + ( - 7 n b ^ { 2 } - 8 - 5 n + 3 2 b ^ { 2 } ) r _ { 0 0 } + 8 \lambda ( 7 b ^ { 4 } - 6 - 2 2 b ^ { 2 } ) ( r _ { 0 } + s _ { 0 } ) \beta \\ & + 2 ( - 1 5 4 b ^ { 2 } + 1 2 + 3 3 n b ^ { 2 } + 1 9 n ) s _ { 0 } \beta + 2 ( 1 4 b ^ { 2 } - 2 2 ) s _ { 0 } \beta \right ] \beta ^ { 2 } \\ M _ { 1 5 } \coloneqq \ \left [ ( n + 1 ) [ 4 ( 2 3 b ^ { 4 } + 5 b ^ { 2 } - 1 ) c \beta ^ { 2 } - ( 1 4 b ^ { 2 } + 1 ) ( 2 b ^ { 2 } + 1 ) \eta \beta \right ] \\ & - \frac { n + 1 } { 2 } ( 8 b ^ { 2 } + 1 ) r _ { 0 0 } - 2 \lambda ( 1 4 b ^ { 2 } + 1 ) ( 2 b ^ { 2 } + 1 ) ( s _ { 0 } + r _ { 0 } ) \beta \\ & - 2 ( - 6 b ^ { 2 } + 3 - 5 n b ^ { 2 } ) s _ { 0 } \beta - 2 ( 4 + 1 4 b ^ { 2 } ) r _ { 0 } \beta \right ], \\ M _ { 1 6 } \coloneqq \ ( n + 1 ) ( 1 + 2 b ^ { 2 } ) ^ { 2 } c \\ The term of ( 2 5 ) which is seemingly does not contain \alpha ^ { 2 } \, is \, M _ { 1 }. \, Since \, \beta ^ { 1 5 } \, is \, not \\ divisible by \alpha ^ { 2 }, \, then \, c = 0 \, \text{which implies that}$$
The term of (25) which is seemingly does not contain α 2 is M 1 . Since β 15 is not divisible by α 2 , then c = 0 which implies that
$$M _ { 1 } = M _ { 9 } = 0.$$
Therefore (25) reduces to following
$$M _ { 2 } + M _ { 3 } \alpha ^ { 2 } + M _ { 4 } \alpha ^ { 4 } + M _ { 5 } \alpha ^ { 6 } + M _ { 6 } \alpha ^ { 8 } + M _ { 7 } \alpha ^ { 1 0 } + M _ { 8 } \alpha ^ { 1 2 } = 0, \text{ \quad \ } ( 2 6 )$$
$$M _ { 1 0 } + M _ { 1 1 } \alpha ^ { 2 } + M _ { 1 2 } \alpha ^ { 4 } + M _ { 1 3 } \alpha ^ { 6 } + M _ { 1 4 } \alpha ^ { 8 } + M _ { 1 5 } \alpha ^ { 1 0 } + M _ { 1 6 } \alpha ^ { 1 2 } = 0. \quad ( 2 7 )$$
By plugging c = 0 in M 2 and M 10 , the only equations that don't contain α 2 are the following
$$8 \left [ 8 ( 2 \lambda ( r _ { 0 } + s _ { 0 } ) - ( n + 1 ) \eta ) + 6 n r _ { 0 0 } \right ] & = \tau _ { 1 } \alpha ^ { 2 }, \quad \quad ( 2 8 ) \\ 6 \left [ 4 8 ( 2 \lambda ( r _ { 0 } + s _ { 0 } ) - ( n + 1 ) \eta ) + ( 3 4 n - 1 ) r _ { 0 0 } \right ] & = \tau _ { 2 } \alpha ^ { 2 }, \quad \quad ( 2 9 )$$
where τ 1 = τ 1 ( x ) and τ 2 = τ 2 ( x ) are scalar functions on M . By eliminating [2 λ r ( 0 + s 0 ) -( n +1) ] from (28) and (29), we get η
$$\check { 6 \left [ 4 8 ( 2 \lambda ( r _ { 0 } + s _ { 0 } ) - ( n + 1 ) \eta ) + ( 3 4 n - 1 ) r _ { 0 0 } \check { \right ] } = \tau _ { 2 } \alpha ^ { 2 }, \quad \ ( 2 9 ) \\ \intertext { here } \tau _ { 1 } = \tau _ { 1 } ( x ) \text{ and } \tau _ { 2 } = \tau _ { 2 } ( x ) \text{ are scalar functions on } M. \text{ By eliminating}$$
$$r _ { 0 0 } = \tau \alpha ^ { 2 },$$
where τ = τ 2 -τ 1 -(18 n +1) . By (28) or (29), it follows that
$$2 \lambda ( r _ { 0 } + s _ { 0 } ) - ( n + 1 ) \eta = 0.$$
By (30), we have r 0 = τβ . Putting (30) and (31) in M 10 and M 11 yield
$$M _ { 1 0 } = & ( 2 0 4 n - 6 ) \tau \alpha ^ { 2 } \beta ^ { 1 2 }, & & ( 3 2 ) \\ & r & & ( 3 0 0 n - 1 0 0 \lambda ^ { 2 } \pm 2 0 n - 1 8 0 n } \quad. \,.$$
$$\stackrel { \cdots \nu } { \cdots } \stackrel { \cdots \cdots } { \cdots } \stackrel { \cdots \cdots } { \cdots } \stackrel { \cdots \cdots } { \cdots } \stackrel { \cdots } { \cdots } \,, \\ M _ { 1 1 } = & \left [ [ ( 6 6 0 n - 2 3 0 ) s _ { 0 } - 6 6 0 r _ { 0 } ] \beta - \frac { ( 3 0 0 n - 1 2 0 ) b ^ { 2 } + 2 4 9 - 1 8 9 n } { 2 } r _ { 0 0 } \tau \alpha ^ { 2 } \right ] \beta ^ { 2 } _ { 0 3 } )$$
By putting (32) and (33) into (27), we have
$$& [ ( 6 6 0 n - 2 3 0 ) s _ { 0 } - 6 6 0 r _ { 0 } ] \beta ^ { 1 0 } - \frac { 3 0 0 n b ^ { 2 } + 2 4 9 - 1 8 9 n - 1 2 0 b ^ { 2 } } { 2 } r _ { 0 0 } \tau \alpha ^ { 2 } \beta ^ { 9 } \\ + & \, ( 2 0 4 n - 6 ) \tau \beta ^ { 1 2 } - M _ { 1 2 } \alpha ^ { 2 } + M _ { 1 3 } \alpha ^ { 4 } + M _ { 1 4 } \alpha ^ { 6 } + M _ { 1 5 } \alpha ^ { 8 } + M _ { 1 6 } \alpha ^ { 1 0 } = 0. \ ( 3 4 )$$
$$n - 6 ) \tau \beta ^ { 1 2 } - M$$
The only equations of (34) that do not contain α 2 is [(204 n -6) τβ 2 +(660 n -230) s 0 -660 r 0 ] β 10 . Since β 10 is not divisible by α 2 , then we have
$$[ ( 2 0 4 n - 6 ) \tau \beta ^ { 2 } + ( 6 6 0 n - 2 3 0 ) s _ { 0 } - 6 6 0 r _ { 0 } ] = 0. \quad \quad ( 3 5 )$$
By Lemma 3.1, we always have s j = 0. Then (35), reduces to following
$$( 2 0 4 n - 6 ) \tau \beta ^ { 2 } - 6 6 0 r _ { 0 } = 0.$$
Thus
$$2 ( 2 0 4 n - 6 ) \tau b _ { i } \beta - 6 6 0 \tau b _ { i } = 0. \quad \quad \quad ( 3 7 )$$
By multiplying (37) with b i , we have
$$\tau = 0.$$
Thus by (31), we get η = 0 and then S = ( n +1) cF . By (30), we get r ij = 0. Therefore Lemma 3.1, implies that S = 0. This completes the proof.
Proof of Theorem 1.1: Let F be an isotropic Berwald metric (3) with almost isotropic flag curvature (1). In [20], it is proved that every isotropic Berwald metric (3) has isotropic S-curvature (2).
Conversely, suppose that F is of isotropic S-curvature (2) with scalar flag curvature K . In [11], it is showed that every Finsler metric of isotropic Scurvature (2) has almost isotropic flag curvature (1). Now, we are going to prove that F is a isotropic Berwald metric. In [6], it is proved that F is an isotropic Berwald metric (3) if and only if it is a Douglas metric with isotropic mean Berwald curvature (5). On the other hand, every Finsler metric of isotropic S-curvature (2) has isotropic mean Berwald curvature (5). Thus for completing the proof, we must show that F is a Douglas metric. By Proposition 3.2, we have S = 0. Therefore by Theorem 1.1 in [11], F must be of isotropic flag curvature K = σ x ( ). By Proposition 3.2, β is a Killing 1-form with constant length with respect to α , that is, r ij = s j = 0. Then (14), (15) and (16) reduce to
$$G ^ { i } - \bar { G } ^ { i } = \alpha Q s ^ { i } _ { \ 0 }, \ \ J _ { i } = - \frac { \Phi s _ { i 0 } } { 2 \alpha \Delta }, \ \ \bar { J } = 0.$$
By (12), we get
$$\lambda I _ { i } b ^ { i } = \frac { - \Phi } { 2 \Delta F } ( \phi - s \phi ^ { \prime } ) ( b ^ { 2 } - s ^ { 2 } ).$$
We consider two case:
Case 1. Let dimM ≥ 3. In this case, by Schur Lemma F has constant flag curvature and (21) holds. Thus by (38) and (39), the equation (21) reduces to following
$$\frac { \Phi s _ { i 0 } } { 2 \alpha \Delta } a ^ { i k } s _ { k 0 } + \frac { \Phi s _ { l 0 } } { 2 \alpha \Delta } \Big ( \alpha Q s ^ { l } _ { 0 } \Big ) _ {. i } b ^ { i } - K F \frac { \Phi } { 2 \Delta } ( \phi - s \phi ^ { \prime } ) ( b ^ { 2 } - s ^ { 2 } ) = 0. \quad ( 4 0 ) \\ \text{Bv assumption } \Phi \neq 0. \text{ Thus } b v \ ( 4 0 ). \text{ we get}$$
By assumption Φ = 0. Thus by (40), we get
/negationslash
$$s _ { i 0 } s _ { 0 } ^ { i } + s _ { l 0 } \left ( \alpha Q s _ { 0 } ^ { l } \right ) _ {. i } b ^ { i } - K F \alpha ( \phi - s \phi ^ { \prime } ) ( b ^ { 2 } - s ^ { 2 } ) = 0. \quad \ ( 4 1 ) \\ \text{The following holds}$$
The following holds
$$\left ( \alpha Q s ^ { l } _ { \, 0 } \right ) _ {. i } b ^ { i } = s Q s ^ { i } _ { \, 0 } + Q ^ { \prime } s ^ { i } _ { \, 0 } ( b ^ { 2 } - s ^ { 2 } ). \\ \text{n be rewritten as follows}$$
Then (41) can be rewritten as follows
$$s _ { i 0 } s ^ { i } _ { 0 } \Delta - K \alpha ^ { 2 } \phi ( \phi - s \phi ^ { \prime } ) ( b ^ { 2 } - s ^ { 2 } ) = 0.$$
By (11), (23) and (42), we obtain
$$\begin{array} { l l l l } \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \\ & \left [ 1 - \frac { s ( 1 + 2 s + 3 s ^ { 2 } ) } { ( - 1 + s ^ { 2 } + 2 s ^ { 3 } ) } + \frac { 2 ( b ^ { 2 } - s ^ { 2 } ) ( 1 + 3 s ) ( 1 + s + s ^ { 2 } + s ^ { 3 } ) } { ( - 1 + s ^ { 2 } + 2 s ^ { 3 } ) ^ { 2 } } \right ] s _ { i 0 } s ^ { i } _ { 0 } \\ & - K ( 1 + s + s ^ { 2 } + s ^ { 3 } ) \alpha ^ { 2 } \left [ 1 + s + s ^ { 2 } + s ^ { 3 } - s ( 1 + 2 s + 3 s ^ { 2 } ) \right ] ( b ^ { 2 } - s ^ { 2 } ) = 0 ( 4 3 ) \\ & \text{Multiplying (43) with (-1 + s ^ { 2 } + 2 s ^ { 3 } ) ^ { 2 } \alpha ^ { 1 2 } \text{ yields} \end{array}$$
Multiplying (43) with ( -1 + s 2 +2 s 3 ) 2 α 12 yields
$$A + \alpha B = 0,$$
where
$$\text{where} \quad \\ A & = - K b ^ { 2 } \alpha ^ { 1 4 } + ( 2 b ^ { 2 } + 1 ) ( K \beta ^ { 2 } + s _ { i 0 } s ^ { i } _ { 0 } ) \alpha ^ { 1 2 } \\ & + 2 ( 3 K b ^ { 2 } \beta ^ { 2 } + 4 s _ { i 0 } s ^ { i } _ { 0 } b ^ { 2 } - K \beta ^ { 2 } - s _ { i 0 } s ^ { i } _ { 0 } ) \beta ^ { 2 } \alpha ^ { 1 0 } \\ & - ( 6 K \beta ^ { 2 } + 1 1 s _ { i 0 } s ^ { i } _ { 0 } + 2 0 K \beta ^ { 2 } b ^ { 2 } - 6 s _ { i 0 } s ^ { i } _ { 0 } ) \beta ^ { 4 } \alpha ^ { 8 } \\ & - ( - 2 0 K \beta ^ { 2 } + 5 K \beta ^ { 2 } b ^ { 2 } + 8 s _ { i 0 } s ^ { i } _ { 0 } ) \beta ^ { 6 } \alpha ^ { 6 } + ( K \beta ^ { 1 0 } ) ( 2 6 b ^ { 2 } + 5 ) \alpha ^ { 4 } \\ & - 2 K \beta ^ { 1 2 } ( 1 3 - 4 b ^ { 2 } ) \alpha ^ { 2 } - 8 K \beta ^ { 1 4 }, & ( 4 4 ) \\ B & = - ( K b ^ { 2 } \beta ) \alpha ^ { 1 2 } + ( 1 + 8 b ^ { 2 } ) ( K b ^ { 2 } \beta ^ { 2 } + s _ { i 0 } s ^ { i } _ { 0 } ) \beta \alpha ^ { 1 0 } \\ & - 2 ( 3 K b ^ { 2 } \beta ^ { 2 } - 4 s _ { i 0 } s ^ { i } _ { 0 } b ^ { 2 } + 4 K \beta ^ { 2 } + 5 s _ { i 0 } s ^ { i } _ { 0 } ) \beta ^ { 3 } \alpha ^ { 8 } \\ & + ( 6 K \beta ^ { 2 } - 1 1 s _ { i 0 } s ^ { i } _ { 0 } - 2 0 K \beta ^ { 2 } b ^ { 2 } ) \beta ^ { 5 } \alpha ^ { 6 } \\ & + ( 5 K \beta ^ { 9 } ) ( 3 b ^ { 2 } + 4 ) \alpha ^ { 4 } + ( 5 K \beta ^ { 1 1 } ) ( - 3 + 4 b ^ { 2 } ) \alpha ^ { 2 } - 2 0 \alpha K \beta ^ { 1 3 }. \\ \text{ Obviously, we have } A = 0 \text{ and } B = 0. \text{ } B \text{ } A = 0 \text{ and the fact that } \beta ^ { 1 4 } \text{ is not divisible by } \alpha ^ { 2 }, \text{ we get } K = 0.$$
Obviously, we have A = 0 and B = 0.
By A = 0 and the fact that β 14 is not divisible by α 2 , we get K = 0. Therefore (43) reduces to following
$$s _ { i 0 } s _ { \, 0 } ^ { i } = a _ { i j } s _ { \, 0 } ^ { j } s _ { \, 0 } ^ { i } = 0.$$
Because of positive-definiteness of the Riemannian metric α , we have s i 0 = 0, i.e., β is closed. By r 00 = 0 and s 0 = 0, it follows that β is parallel with respect to α . Then F = α + β + β 2 α + β 3 α 2 is a Berwald metric. Hence F must be locally Minkowskian.
Case 2. Let dim M = 2. Suppose that F has isotropic Berwald curvature (3). In [20], it is proved that every isotropic Berwald metric (3) has isotropic Scurvature, S = ( n +1) cF . By Proposition 3.2, c = 0. Then by (3), F reduces to a Berwald metric. Since F is non-Riemannian, then by Szab´'s rigidity Theorem o for Berwald surface (see [2] page 278), F must be locally Minkowskian.
Acknowledgment. The authors would like to thank the referee's valuable suggestions.
## References
- [1] S. B´cs´ and M. Matsumoto, a o On Finsler spaces of Douglas type, A generalization of notion of Berwald space , Publ. Math. Debrecen. 51 (1997), 385-406.
- [2] D. Bao, S. S. Chern and Z. Shen, An Introduction to Riemann-Finsler Geometry , Springer-Verlag, 2000.
- [3] X. Chun-Huan and X. Cheng, On a class of weakly-Berwald ( α, β ) -metrics , J. Math. Res. Expos. 29 (2009), 227-236.
- [4] X. Cheng and Z. Shen, Randers metric with special curvature properties , Osaka. J. Math. 40 (2003), 87-101.
- [5] X. Cheng and Z. Shen, A class of Finsler metrics with isotropic Scurvature , Israel. J. Math. 169 (2009), 317-340.
- [6] X. Chen and Z. Shen, On Douglas metrics , Publ. Math. Debrecen. 66 (2005), 503-512.
- [7] X. Cheng, H. Wang and M. Wang, ( α, β ) -metrics with relatively isotropic mean Landsberg curvature , Publ. Math. Debrecen. 72 (2008), 475-485.
- [8] N. Cui, On the S-curvature of some ( α, β ) -metrics , Acta. Math. Scientia, Series: A. 26 (7) (2006), 1047-1056.
- [9] I.Y. Lee and M.H. Lee, On weakly-Berwald spaces of special ( α, β ) -metrics , Bull. Korean Math. Soc. 43 (2) (2006), 425-441.
- [10] M. Matsumoto, Theory of Finsler spaces with ( α, β ) -metric , Rep. Math. Phys. 31 (1992), 43-84.
- [11] B. Najafi, Z. Shen and A. Tayebi , Finsler metrics of scalar flag curvature with special non-Riemannian curvature properties , Geom. Dedicata. 131 (2008), 87-97.
- [12] H. S. Park and E. S. Choi, On a Finsler spaces with a special ( α, β ) -metric , Tensor, N. S. 56 (1995), 142-148.
- [13] H. S. Park and E. S. Choi, Finsler spaces with an approximate Matsumoto metric of Douglas type , Comm. Korean. Math. Soc. 14 (1999), 535-544.
- [14] H. S. Park and E. S. Choi, Finsler spaces with the second approximate Matsumoto metric , Bull. Korean. Math. Soc. 39 (1) (2002), 153-163.
- [15] H. S. Park, I. Y. Lee and C. K. Park, Finsler space with the general approximate Matsumoto metric , Indian J. Pure. Appl. Math. 34 (1) (2002), 59-77.
- [16] H. S. Park, I.Y. Lee, H. Y. Park and B. D. Kim, Projectively flat Finsler space with an approximate Matsumoto metric , Comm. Korean. Math. Soc. 18 (2003), 501-513.
- [17] Z. Shen, Differential Geometry of Spray and Finsler Spaces , Kluwer Academic Publishers, Dordrecht, 2001.
- [18] A. Tayebi and B. Najafi, On isotropic Berwald metrics , Ann. Polon. Math. 103 (2012), 109-121.
- [19] A. Tayebi, E. Peyghan and H. Sadeghi, On Matsumoto-type Finsler metrics , Nonlinear Analysis: RWA. 13 (2012), 2556-2561.
- [20] A. Tayebi and M. Rafie Rad, S-curvature of isotropic Berwald metrics , Sci. China. Series A: Math. 51 (2008), 2198-2204.
Akbar Tayebi and Tayebeh Tabatabaeifar Faculty of Science, Department of Mathematics University of Qom Qom. Iran Email: [email protected] Email: [email protected]
Esmaeil Peyghan
Faculty of Science, Department of Mathematics Arak University 38156-8-8349 Arak. Iran
Email: [email protected] | null | [
"A. Tayebi",
"T. Tabatabaeifar",
"E. Peyghan"
] | 2014-07-31T08:55:09+00:00 | 2014-07-31T08:55:09+00:00 | [
"math.DG",
"53B40, 53C60, 53C25"
] | On the Second Approximate Matsumoto Metric | In this paper, we study the second approximate Matsumoto metric on a manifold
M. We prove that F is of scalar flag curvature and isotropic S-curvature if and
only if it is isotropic Berwald metric with almost isotropic flag curvature. |
1408.0239v1 | ## WIYN OPEN CLUSTER STUDY. LX. SPECTROSCOPIC BINARY ORBITS IN NGC 6819
Katelyn E. Milliman 1,6 , Robert D. Mathieu 1,6 , Aaron M. Geller 2,3 , Natalie M. Gosnell 1 , Søren Meibom 4 , and Imants Platais 5
Draft version August 4, 2014
## ABSTRACT
We present the current state of the WOCS radial-velocity (RV) survey for the rich open cluster NGC 6819 (2.5 Gyr) including 93 spectroscopic binary orbits with periods ranging from 1.5 to 8,000 days. These results are the product of our ongoing RV survey of NGC 6819 using the Hydra Multi-Object Spectrograph on the WIYN 3.5 m telescope. We also include a detailed analysis of multiple prior sets of optical photometry for NGC 6819. Within a 1 ◦ field of view, our stellar sample includes the giant branch, the red clump, and blue straggler candidates, and extends to almost 2 mag below the main-sequence (MS) turnoff. For each star observed in our survey we present all RV measurements, the average RV and velocity variability information. Additionally, we discuss notable binaries from our sample, including eclipsing binaries (WOCS 23009, WOCS 24009, and WOCS 40007), stars noted in Kepler asteroseismology studies (WOCS 4008, WOCS 7009, and WOCS 8007), and potential descendants of past blue stragglers (WOCS 1006 and WOCS 6002). We find the incompletenesscorrected binary fraction for all MS binaries with periods less than 10 4 days to be 22% ± 3% and a tidal circularization period of 6 2 . +1 1 . -1 1 . days for NGC 6819.
Subject headings:
## 1. INTRODUCTION
The rich intermediate-age open cluster NGC 6819 (2.5 Gyr; α = 19 41 h m 17 s . 5 (J2000), δ =+40 11 47 ; ◦ ′ ′′ l =74 ◦ . 0, b =+8 ◦ . 5) is one of four open clusters located within the field of view of the Kepler Mission and as such has experienced an increase in interest in recent years. Research on this cluster includes deep photometric studies (Rosvick & Vandenberg 1998, Kalirai et al. 2001, Yang et al. 2013), a time-series study (Street et al. 2002, Street et al. 2005), a moderately high-precision radial-velocity (RV) survey (Hole et al. 2009), studies of metal abundance (Bragaglia et al. 2001, Anthony-Twarog et al. 2013), and proper-motion measurements (Sanders 1972, Platais et al. 2013). Other research has studied the X-ray properties of NGC 6819 (Gosnell et al. 2012) resulting in the discovery of a dozen sources including a candidate quiescent low-mass Xray binary. Kepler data have enabled a number of asteroseismology studies of red giants in NGC 6819 that have investigated cluster membership (Stello et al. 2010, Stello et al. 2011), provided model-independent estimates of cluster parameters such as distance and
Electronic address: email: [email protected]
- 2 Center for Interdisciplinary Exploration and Research in Astrophysics (CIERA) and Department of Physics and Astronomy, Northwestern University, 2145 Sheridan Rd, Evanston, IL 60208, USA
- 1 Department of Astronomy, University of Wisconsin-Madison, 475 North Charter St, Madison, WI 53706, USA
- 3 Department of Astronomy and Astrophysics, University of Chicago, 5640 S. Ellis Avenue, Chicago, IL 60637, USA
- 5 Department of Physics and Astronomy, Johns Hopkins University, 3400 North Charles Street, Baltimore, MD 21218, USA
- 4 Harvard-Smithsonian Center for Astrophysics, 60 Garden Street, Cambridge, MA 02138, USA
6 Visiting Astronomer, Kitt Peak National Observatory, National Optical Astronomy Observatory, which is operated by the Association of Universities for Research in Astronomy (AURA) under cooperative agreement with the National Science Foundation.
age (Basu et al. 2011), and integrated red giant branch (RGB) mass-loss (Miglio et al. 2012). Kepler data have also been used to detect eclipsing binaries in NGC 6819 that have yielded stellar mass, radius, and age measurements (Sandquist et al. 2013, Jeffries et al. 2013).
As part of the WIYN Open Cluster Study (WOCS; Mathieu 2000) we have continued the moderately highprecision RV survey of NGC 6819 discussed in Hole et al. (2009). Our ongoing RV survey covers a 1 ◦ diameter region on the sky and includes giant branch, red clump, upper main-sequence (MS), and blue straggler candidates. In this paper we present 93 spectroscopic binary orbits, as well as variability information for all stars observed as part of our RV survey. In Section 2, we discuss the photometry used in this study. Section 3 details the proper-motion membership information presented in Platais et al. (2013) that we have incorporated into our RV survey and analysis of NGC 6819. We describe the details and results of our RV survey in Section 4 including our observations, completeness, and RV membership probabilities. In Section 5, we provide 93 single-lined (SB1) and double-lined (SB2) orbital solutions for binary cluster members of NGC 6819. Binaries identified as interesting stars from XMM Newton -or Kepler observations and stars that are potential triple systems are discussed in Section 6. We analyze the circularization period (CP) of NGC 6819 in Section 7 and the binary frequency for stars with periods under 10 4 days in Section 8. Section 9 summarizes our current results for the WOCS RV survey of NGC 6819.
## 2. CLUSTER PHOTOMETRY
The first WOCS NGC 6819 RV paper, Hole et al. (2009), utilized four sources of CCD BV photometry: Rosvick & Vandenberg (1998), Kalirai et al. (2001), and two sets of WOCS photometry taken at the WIYN 0.9 m telescope in 1998 and 2003 (Sarrazine et al. 2003).
Figure 1. Comparison of the photometry for overlapping sources between Y13 and Phot03. The top row of figures shows V Y13 minus V Phot03 as a function of V Phot03 (left) and as a function of radial distance from the center of the cluster (right). The bottom row shows the same comparisons for I Y13 and I Phot03 . The dashed line in all the figures marks zero difference between the photometry sets.
We will use RV98, K01, Phot98, and Phot03 to refer to each photometry set, respectively. Hole et al. (2009) calibrated each set of photometry to the same zero point as Phot03 and derived the final photometry by averaging all V and ( B -V ) information available for each star from RV98, Phot98, Phot03, and K01 (more details available in Hole et al. 2009, and references therein). This provided BV photometry for 2220 stars in our sample, yet left a large number of stars, especially in the outer part of the field, with no BV information. Recent photometry of Yang et al. (2013) and Everett, Howell & Kinemuchi (2012), hereafter Y13 and EHK12 respectively, include almost all the stars in our NGC 6819 sample. We have used these photometry sets to revise and expand the photometry for the stars in our RV survey.
## 2.1. Yang et al. (2013)
Y13 provides CCD V I photometry for 3831 stars in our RV sample and covers the entire radial extent of our survey. The data were taken in 2000 July on the WIYN 0.9 m telescope with the NOAO MOSAIC imager at the f /7.5 secondary and provided a field of view of approximately 1 ◦ × 1 . ◦ Yang et al. (2013) used Phot03 to calibrate their instrumental magnitudes and therefore Y13 and Phot03 have no significant offsets. This is seen in Figure 1 and in the average differences, 〈 V Y13 -V Phot03 〉 = 0.005 ± 0.001 and 〈 I Y13 -I Phot03 〉 = 0.000 ± 0.001. The I band shows a larger range in the scatter between Phot03 (root-mean-square error of 0.039) and Y13 than the V band (rms error of 0.036), but neither band has a dependence on magnitude except in I for the very brightest stars.
## 2.2. Everett, Howell & Kinemuchi (2012)
EHK12 provides UBV photometry of 4 414 002 , , sources in the Kepler field, including NGC 6819. They observed the Kepler field using the NOAO Mosaic-1.1 Wide Field Imager and the WIYN 0.9 m telescope with a set of 209 pointings over five nights in 2011 June and achieved typical completeness limits of U ∼ 18.7, B ∼ 19.3, and V ∼ 19.1, well below the magnitude limits of our target sample.
Figure 2 shows a large offset between Phot03 and EHK12 in both V and B as well as a large amount of scatter at all magnitudes and stellar positions. 〈 V EHK12 -V Phot03 〉 and 〈 B EHK12 -B Phot03 〉 are 0.115 ± 0.001 and 0.073 ± 0.002, respectively. Also, V EHK12 -V Phot03 has a rms error of 0.127 and B EHK12 -B Phot03 has an rms error of 0.0935, almost triple the rms errors found with the Y13 photometry comparison.
## 2.3. Photometry Summary
We currently use only the V I photometry from Y13 for our RV survey and in this paper. We made this decision due to the limited radial extent of the previous BV photometry sets and the large offsets and lack of precision in EHK12. Y13 V I photometry provides photometry for 98% of our RV survey and covers the entire radial extent of our survey.
As seen clearly in Figure 1 our photometry comparisons have revealed spatial trends in V Y13 -V Phot03 and I Y13 -I Phot03 with distance from the center of the cluster. Clem & Landolt (2013) note a similar trend in the difference between their calibrated magnitudes and those of Landolt standards which vary as a function of x and y
Figure 2. Comparison of the photometry for overlapping sources between Everett, Howell & Kinemuchi (2012) and Phot03. The top row of figures shows V EHK12 minus V Phot03 as a function of V Phot03 (left) and as a function of radial distance from the center of the cluster (right). The bottom row shows the same comparisons for B EHK12 and B Phot03 .
position relative to the center of the two different detectors used in their study. They attribute this systematic trend to the failure of the flat fields to completely correct for large-scale illumination patterns in the detector systems. This detector-independent flat fielding effect may be the cause of the trends seen in the Y13 data as well. Regardless, the typical magnitude and color errors are less than 0.05 and will not affect any of the kinematic results discussed in the rest of this paper.
P µ , are calculated with local-sample techniques (Kozhurina-Platais et al. 1995) which utilize separate two-dimensional Gaussian frequency distributions of the field and the cluster for a limited sample of stars. The limited samples are chosen to contain stars with similar magnitudes and distance from the cluster center as the specific star being analyzed.
Finally, based on Y13 V I photometry of stars within 10 ′ of the cluster center that are proper-motion members Platais et al. (2013) find differential reddening in NGC 6819. As Platais et al. (2013) only analyzed the inner part of our total sample region and we expect the specific ∆ E V ( -I ) quantities may need to be revisited based on the radial trend in Y13 discussed above, we do not incorporate these findings in the photometry in Table 2 or the color -magnitude diagram (CMD) presented in this paper. However, the maximum differential reddening found by Platais et al. (2013) was ∆ E V ( -I ) = 0.09 and we expect associated scatter in the CMD.
## 3. PROPER MOTIONS
Platais et al. (2013) presents proper-motion measurements for 15 , 408 stars in the NGC 6819 field with a precision that ranges from ∼ 0.2 mas yr -1 within 10 ′ of the cluster center to 1.1 mas yr -1 outside of 10 . ′ The study uses CCD observations taken with the MegaCam imager at the CFHT as well as 23 scanned photographic plates from four different telescopes, all with different epochs, depth, scales, and fields of view.
The proper-motion membership probabilities,
This survey provides proper-motion information for 78% of the targets in our RV survey of NGC 6819. The large majority of stars missing proper-motion measurements are outside the area covered by the photographic plates. Proper-motion measurements are also missing for stars that fall near the MegaCam interchip gaps, suffer from image confusion on the photographic plates, or have proper-motions exceeding the upper limit imposed by Platais et al. (2013) of 20 mas yr -1 in either the x or y coordinate.
In Figure 3 we compare the membership probabilities from proper-motion and our RV (Section 4.4) measurements, splitting the sample at a radius of 10 ′ due to the large difference in the precisions of the proper-motion data at that radius. Overall we have good agreement between the two studies. A large number of stars have both P µ and P RV = 0% and within 10 ′ we see a large concentration of targets with high RV and proper-motion probabilities. In both plots we note a large population of stars around the perimeter (e.g., the vertical lines at P RV = 0%). These over-densities are due to the vast majority of the stars in both studies falling at either end of the distribution, either 0% or 99%, providing a large population to fill up the edges and leaving very few stars to populate the centers of the plots of Figure 3.
Figure 3. Comparison of cluster membership probabilities from radial-velocity and proper-motion measurements. Proper-motion membership probabilities, P µ , are from Platais et al. (2013). Radial-velocity membership probabilities are determined from Equation (3). Sources within 10 ′ of the cluster center are presented on the left and sources outside of 10 ′ are on the right. Proper-motion membership probabilities outside of 10 ′ have lower precisions, peak at P µ of 63%, and have more inconsistencies with our radial-velocity membership determination.
Platais et al. (2013) choose a membership threshold at a P µ of 4%, but note a substantial overlap between the proper-motion distributions of the cluster and field stars in the direction of NGC 6819. They find that a CMD of stars with P µ ≥ 4% within 10 ′ shows a clean MS, subgiant branch, and RGB. A CMD that includes stars with P µ < 4% shows no distinguishable cluster features.
We also adopt the proper-motion membership threshold of 4%. We do caution that proper-motion information outside of 10 ′ from the cluster center have larger errors and disagree with our membership determinations from RVs more frequently (Figure 3) than for stars within 10 . ′
## 4. RADIAL-VELOCITY STUDY
The details of our RV survey of NGC 6819 including the observing procedure, data reduction, and membership classification are discussed in depth in Hole et al. (2009) and Geller et al. (2008). The following sections summarize the details relevant to the results presented in this paper and provide updated information where appropriate.
sample was based on RV98 and PHOT98 photometry. Cutoffs based on V and ( B -V ) were chosen in order to select the upper MS, giant branch, red clump, and potential blue straggler candidates in NGC 6819. Later, the sample was updated and extended in radius with the addition of K01 and Phot03 photometric studies. Stars without optical photometry (primarily at large radii) were included based on V magnitudes converted from 2MASS J and K information. In this paper, as discussed in Section 2, we have incorporated Y13 V I photometry that provides photometry for 98% of our RV survey. In 2010, we incorporated the proper-motion information from Platais et al. (2013) into our RV survey. Stars identified by Platais et al. (2013) as proper-motion non-members, P µ < 4%, were moved to the lowest priority for observations.
## 4.1. Observations, Data Reduction, and Precision
The WOCS RV target sample for NGC 6819 has 3895 stars that span 1 ◦ on the sky centered at α = 19 41 h m 17 s . 5, δ = +40 11 47 ◦ ′ ′′ (J2000). Hole et al. (2009) discuss the target selection and evolution of the RV target sample in great detail. In brief, the initial target
Observations of NGC 6819 with the Hydra MultiObject Spectrograph (MOS; Barden et al. 1994) on the WIYN 7 3.5 m telescope began in 1998 June and are still ongoing. We have almost 14,000 spectra for over 2600 stars. These observations are augmented with 733 RV measurements for 170 stars taken at the HarvardSmithsonian Center for Astrophysics (CfA) facilities between 1988 May and 1995 by R. D. Mathieu & D. W Latham (Hole et al. 2009).
The WIYN data include for each configuration: three
7 The WIYN Observatory is a joint facility of the University of Wisconsin-Madison, Indiana University, Yale University, and the National Optical Astronomy Observatory.
Table 1 Radial-velocity Measurements
| ID | HJD - 2400000 | Telescope a | RV 1 | Correlation Height 1 | O - C 1 | RV 2 | Correlation Height 2 | O - C 2 | Phase |
|------|-----------------|---------------|----------|------------------------|-----------|----------|------------------------|-----------|---------|
| 9030 | | | | | | | | | |
| | 55374.6633 | W | 1.874 | 0.89 | 4.00 | - 19.272 | 0.63 | 3.44 | 0.234 |
| | 55374.7375 | W | - 13.202 | 0.89 | 0.01 | · · · | · · · | · · · | 0.257 |
| | 55375.6625 | W | - 88.546 | 0.63 | - 3.10 | 60.357 | 0.57 | 5.29 | 0.545 |
| | 55375.7354 | W | - 80.345 | 0.72 | 2.46 | 54.662 | 0.65 | 2.06 | 0.568 |
| | 55376.6627 | W | 32.298 | 0.68 | 3.05 | - 50.376 | 0.56 | 1.62 | 0.856 |
| | 55376.7367 | W | 42.208 | 0.73 | 3.55 | - 59.876 | 0.54 | 0.90 | 0.879 |
| | 55412.6959 | W | 58.410 | 0.73 | - 2.79 | - 80.087 | 0.61 | 1.73 | 0.066 |
| | 55413.6906 | W | - 62.518 | 0.73 | 0.39 | 37.516 | 0.62 | 3.49 | 0.375 |
| | 55414.6935 | W | - 45.001 | 0.76 | 3.83 | 23.552 | 0.63 | 2.67 | 0.687 |
| | 55440.8442 | W | 11.868 | 0.81 | - 1.99 | - 34.090 | 0.64 | 3.54 | 0.822 |
a The observatory at which the observations were taken, using 'C' for CfA facilities and 'W' for the WIYN 3.5 m.
(This table is available in its entirety in machine-readable and Virtual Observatory (VO) forms in the online journal. A portion is shown here for guidance regarding its form and content.)
science exposures, one 100 s dome flat, and two 300 s thorium -argon emission lamp spectra, one taken before and one taken after the science exposures. The science observations are split into three integrations of equal time for cosmic ray rejection. All the image processing is done within IRAF where the data are bias subtracted and dispersion corrected, and the extracted spectra are flat-fielded, throughput corrected, and sky subtracted.
as single stars. We quantify our detection completeness for binaries using the Monte Carlo analysis described in Section 8.
RVs for single-lined stars are derived from the centroid of a one-dimensional cross-correlation function (CCF) with an observed solar template, converted to a heliocentric velocity, and corrected for the unique fiber offsets of the Hydra MOS. RVs for double-lined stars are derived using two -dimensional correlation (TODCOR) and we provide more details in Section 5.2. The χ 2 analysis of Hole et al. (2009) finds the precision for NGC 6819 data to be σ WIYN = 0.4 km s -1 and σ CfA = 0.7 km s -1 , which we have confirmed remains true for the additional RVs presented here.
We present all of our RV measurements for each star in Table 1, along with the Heliocentric Julian Date (HJD) of the observation and the height of the CCF. For binary stars with completed orbital solutions we also include the RV residual and the orbital phase of the observation.
## 4.2. Velocity Variables
We define velocity variables stars as having RV standard deviations (the external error, e ) greater than four times the precision (the internal error, i ) or e/i > 4 (Geller et al. 2008). We take into account the different data precisions if a star has RV measurements from both observatories using the following:
$$\left ( \frac { e } { i } \right ) ^ { 2 } = \frac { N ^ { 2 } } { N - 1 } \frac { \sum _ { i } ^ { N } \left ( \mathrm R V _ { i } - \overline { \mathrm R V } \right ) ^ { 2 } / \sigma _ { i } ^ { 2 } } { \sum _ { i } ^ { N } \sigma _ { i } ^ { 2 } \sum _ { i } ^ { N } 1 / \sigma _ { i } ^ { 2 } }, \quad ( 1 ) \quad \text{the} \\ \text{where} \, \mathrm R V \text{ is the mean RV weighted by the precision val-} \quad \text{with} \quad \text{$two$}$$
where RV is the mean RV weighted by the precision values of each observatory:
$$\overline { R V } = \frac { \sum _ { i } ^ { N } R V _ { i } / \sigma _ { i } ^ { 2 } } { \sum _ { i } ^ { N } 1 / \sigma _ { i } ^ { 2 } }. \quad ( 2 ) \quad \begin{pmatrix} t t \\ \text{al} \\ \text{t} \\ \text{ju} \\ \text{S}! } \end{pmatrix}$$
For a star with three or more observations we calculate the RV and e/i and classify it as single or velocity variable. Given our RV measurement precisions binaries that are long-period and/or low-amplitude may be classified
In Table 2, we present the coordinates, V I photometry, number of observations at each observatory, RV, RV standard error, e/i , membership probabilities, and membership classification for all of the stars we have observed in NGC 6819 as of 2014 February. For each binary star with a completed orbital solution the center-of-mass velocity, γ , its error, and whether it is a single- or doublelined binary are also listed in Table 2.
## 4.3. Completeness
The entire target sample is comprised of 3895 stars with RV measurements from the CfA dating back to 1988 and WIYN observations beginning in 1998. Over the entire course of our survey we constantly reevaluate the observing priority of every star. Based on Monte Carlo analyses and observing efficiency arguments (Mathieu 1983, Geller & Mathieu 2012) a star is classified as single or velocity variable after three observations. In general, stars classified as single are given a low priority for further observation. Stars that are classified as velocity variable are given a high observing priority and are observed until an orbital solution is determined. However, stars identified as proper-motion non-members, P µ < 4%, are given the lowest observing priority.
The percentage of stars with either no P µ or P µ ≥ 4% that we are able to classify as single or binary as a function of radius and V magnitude are shown in Figure 4. Lower classification percentages at the center of cluster are due to fiber separation requirements limiting the number of central stars we are able to observe in each pointing. The percentages decrease at larger radii in part because early in the survey we prioritized stars with color information and stars more than ∼ 18 ′ from the cluster center did not have color information until the inclusion of Y13 V I information in 2013. We are able to classify more than 75% of stars with V magnitude less than ∼ 15.5 as single or binary and this number jumps to 96% for stars within 18 ′ of the cluster center. Stars dimmer than this are observed less frequently because they require better observing conditions (clearer skies, dimmer moon, etc.) and longer exposures to meet our signal-to-noise requirements.
Table 2 Radial Velocity Summary Table
| ID W a | α (J2000) | δ (J2000) | V | V - I | N W | N C | RV | Std. Err. | e/i | P RV | P PM | Class b | γ RV | γ RVe | Comment |
|----------|-------------|-------------|-------|---------|-------|-------|----------|-------------|-------|--------|--------|-----------|--------|---------|-----------|
| 1001 | 19 41 18.71 | 40 11 42.9 | 12.69 | 1.41 | 3 | 20 | 6.045 | 3.4 | 23.88 | 90 | 19 | BM | 1.42 | 0.17 | SB1 |
| 1002 | 19 41 17.05 | 40 10 51.8 | 11.65 | 1.73 | 2 | 11 | 0.943 | 0.18 | 0.92 | 84 | 99 | SM | · · · | · · · | · · · |
| 1004 | 19 41 26.59 | 40 11 41.8 | 12.22 | 1.50 | 5 | 5 | 1.547 | 0.22 | 1.15 | 91 | 99 | SM | · · · | · · · | · · · |
| 1005 | 19 41 11.85 | 40 13 30.1 | · · · | · · · | 5 | 6 | - 38.297 | 0.19 | 1.02 | 0 | 87 | SN | · · · | · · · | · · · |
| 1006 | 19 41 17.76 | 40 9 15.9 | 12.83 | 1.23 | 5 | 13 | 1.473 | 0.96 | 6.3 | 94 | 99 | BM | 2.72 | 0.13 | SB1 |
| 1007 | 19 41 13.20 | 40 14 56.7 | · · · | · · · | 3 | 6 | 3.415 | 0.28 | 1.31 | 91 | 0 | SN | · · · | · · · | · · · |
| 1010 | 19 41 42.64 | 40 11 40.2 | 12.82 | 0.29 | 17 | 15 | 2.330 | 0.87 | 8.66 | 86 | 87 | BM | 1.07 | 0.35 | SB1 |
| 1011 | 19 41 44.98 | 40 12 50.3 | 11.80 | 0.59 | 2 | 6 | - 22.497 | 0.27 | 1.13 | 0 | 0 | SN | · · · | · · · | · · · |
| 1012 | 19 41 46.76 | 40 10 2.3 | 12.40 | 0.73 | 11 | 2 | 14.813 | 9.7 | 73.19 | 0 | 0 | BLN | · · · | · · · | · · · |
| 1013 | 19 41 44.81 | 40 8 1.9 | 11.97 | 0.62 | 6 | 9 | - 27.611 | 0.37 | 2.34 | 0 | 0 | SN | · · · | · · · | · · · |
a WOCS ID number. This number is based on V magnitude and radial distance from the cluster center. Full details can be found in Hole et al. (2009).
b See Section 4.5 for explanation of class codes.
(This table is available in its entirety in machine-readable and Virtual Observatory (VO) forms in the online journal. A portion is shown here for guidance regarding its form and content.)
Figure 4. Percentage of stars in our sample that have three or more RV observations and either no P µ information or a P µ ≥ 4% with respect to distance from the cluster center (left) and V magnitude (right).
## 4.4. RV Membership Probabilities
The RV membership probability of a given star, P RV , listed in Table 2 is calculated using the equation:
Table 3 Gaussian Fit Parameters for Cluster and Field RV Distributions
$$P _ { \text{RV} } ( v ) = \frac { F _ { \text{cluster} } ( v ) } { F _ { \text{field} } ( v ) + F _ { \text{cluster} } ( v ) }, \quad \quad ( 3 )$$
where F cluster and F field are separate Gaussian functions fit to the cluster and field-star populations using our sample of single stars with three or more RV observations (Figure 5). The parameters for these Gaussian fits are shown in Table 3. With the observation and completion of more single stars these values have shifted only slightly from the numbers published in Hole et al. (2009) and we continue to use the membership threshold of P RV ≥ 50% adopted by Hole et al. (2009) to identify cluster members. We estimate from the cluster and field Gaussian functions a field star contamination of 13% at this membership threshold (Figure 5). Interestingly, using
| Parameter | Cluster | Field |
|----------------|-----------|---------|
| Ampl. (number) | 112.99 | 8.81 |
| RV (km s - 1 ) | 2.45 | - 14.59 |
| σ (km s - 1 ) | 1.02 | 25.79 |
P µ < 4% as a criterion for non-membership, we find a field-star contamination of 12% ± 2% among stars with P RV ≥ 50%.
## 4.5. Membership Classification of Stars
Stars in our RV survey of NGC 6819 can fall into one of eight different membership classifications determined by membership probability, number of observations, and variability. The number of stars in each membership clas-
Figure 5. Histogram of the RV distribution of single stars, e/i < 4, with three or more RV observations. Also plotted are the Gaussian distributions simultaneously fit to the cluster, the large peak at a mean velocity of 2.4 km s -1 (blue dotted line), and the field (orange dashed line).
sification is listed in Table 4. For single stars and velocity variable stars without completed orbital solutions we calculate P RV from the mean velocity, RV, and for binary stars with completed orbital solutions we use the centerof-mass velocity, γ , from the orbital solution. For stars without P µ information, membership is determined by P RV alone.
Single member (SM) . Stars that have e/i < 4, P RV ≥ 50%, and P µ ≥ 4%.
Single non-member (SN) . Stars that have e/i < 4, and P RV < 50% or P µ < 4%.
Binary member (BM) . Velocity-variable stars that have a completed orbital solution. P µ is ≥ 4% and P RV ≥ 50%.
Binary non-member (BN) . Velocity-variable stars that have a completed orbital solution. Either P µ < 4% or P RV < 50%.
Binary likely member (BLM) . Velocity-variable stars that do not have a completed orbital solution, but P RV ≥ 50% and P µ ≥ 4%. Without a completed orbital solution P RV is calculated from RV and this classification is subject to change because observations are ongoing.
Binary likely non-member (BLN) . Velocity-variable stars that do not have a completed orbital solution. P µ < 4% or the P RV is < 50%. If P µ ≥ 4% but P RV < 50%, the individual RVs do not cross the cluster mean, making it unlikely that the orbital solution will place the star within the cluster distribution. Without a completed orbital solution P RV is calculated from RV and this classification is subject to change because observations are ongoing.
Binary unknown (BU) . Velocity-variable stars that do not have a completed orbital solution. P µ ≥ 4% and P RV is < 50%, plus the range of individual RVs includes the cluster mean, making it more likely that the binary could be a member. Without a completed orbital solution P RV is calculated from RV and this classification is subject to change because observations are ongoing.
Unknown (U) . Any star that does not have three or
| Table 4 Number of Stars Within Each Classification | Table 4 Number of Stars Within Each Classification |
|------------------------------------------------------|------------------------------------------------------|
| Classification | N Stars |
| SM | 566 |
| SN | 1381 |
| BM | 93 |
| BN | 79 |
| BLM | 20 |
| BLN | 172 |
| BU | 22 |
| U | 1562 |
more RV measurements, along with stars such as rapid rotators for which we are unable to derive accurate RVs from our spectra.
## 4.6. Color -Magnitude Diagram
We incorporate all our membership classifications, the proper-motion information of Platais et al. (2013) and the photometry of Yang et al. (2013) to produce a cleaned CMD of NGC 6819 in Figure 6 that includes BM, BLM, and SM stars. Binary stars of note (BoNs) are marked in orange on this CMD and will be discussed in more detail in Section 6. The CMD shows a clear MS turnoff, blue hook, and red clump. To guide the eye we plot a 2.5 Gyr PARSEC isochrone (Bressan et al. 2012) using the E V ( -I ) = 0.20 and ( m -M ) o = 11.93 from Yang et al. (2013).
Candidate blue stragglers, defined as being to the blue of the dashed line in Figure 6, are also apparent. The dashed line indicates the upper magnitude and color limit for binaries that could be compromised of two normal stars from the cluster turnoff region or blue hook. A large fraction of these potential blue stragglers are RV variable stars (7/17, ∼ 40%) and three have completed orbital solutions (WOCS 1010, WOCS 4004, and WOCS 14012; see Section 6 for a more detailed discussion of these stars).
The RGB shows a large amount of scatter that is not caused by the blended light from binary stars and is not improved by incorporating differential reddening (Platais et al. 2013). This scatter on the RGB is very different from the very narrow RGB in M67 (Mathieu et al. 2003), but it is similar to the spread in the RGB of NGC 188. The cause of this scatter in NGC 6819 remains unknown.
## 5. SPECTROSCOPIC BINARY ORBITS
## 5.1. Single-lined Orbital Solutions
For each single-lined spectroscopic binary we use the data given in Table 1 to solve for the orbital solution. In Figure 7 we have plotted the orbital solutions; for each binary the top panel contains the orbit solution and the bottom panel contains the residuals, in the sense of observed minus computed RVs. In Table 5 we provide the orbital elements for each binary in two rows: the first row contains the binary ID, the orbital period ( P ), the number of orbital cycles observed, the center-of-mass RV ( γ ), the orbital amplitude ( K ), the eccentricity ( e ), the longitude of periastron ( ω ), a Julian Date of periastron passage ( T ◦ ), the projected
Figure 6. CMDof all radial-velocity and proper-motion members of NGC 6819. Black dots represent single members. The binary members (BMs, with completed orbital solutions) are indicated by circles, the binary likely members (BLMs, without completed orbital solutions) are indicated by squares. The binaries of note (BoNs) discussed in Section 6 are indicated in orange. A 2.5 Gyr PARSEC isochrone is shown in gray. The dashed line indicates the upper magnitude and color limit for binaries that could be compromised of two normal stars from the cluster turnoff region. Stars bluer than the dashed line are candidate blue stragglers.
semi-major axis ( a sin i ), the mass function ( f ( m )), the rms residual velocity from the orbital solution ( σ ), and the number of RV measurements ( N ). The second row contains the respective 1 σ errors on each of these values where appropriate.
## 5.2. Double-lined Orbital Solutions
We used the TODCOR technique formulated by Zucker & Mazeh (1994) to find the RVs for the primary and secondary stars of a given double-lined spectroscopic binary. TODCOR uses two template spectra to derive the two RVs of an SB2 binary simultaneously, greatly increasing our ability to recover reliable RVs even for observations that appear highly blended in a one-dimensional CCF. We used the same solar template that we use to derive RVs for all single stars and SB1 binaries for both template spectra used in TODCOR. (We found that for most of the stars studied here, substituting a library of synthetic spectra that varied in temperature and gravities for the template spectra only has a minor impact on the RV output from TODCOR relative to our other measurement errors.)
We provide the orbital solutions in Figure 8. The top panel of each orbital solution shows the measured RVs for the primary (filled circles) and secondary (open circles), fitted orbital solution for the primary (solid line) and secondary (dashed line), and γ velocity (dotted line). The bottom panel in each orbital solution shows the observed minus computed values for the primary RVs (filled circles) and secondary RVs (open circles). The orbital elements are listed in Table 6. The first row contains the binary ID, the orbital period ( P ), the number of orbital cycles observed, the center-of-mass RV ( γ ), the orbital amplitude of the primary ( K ), the eccentricity ( e ), the longitude of periastron ( ω ), a Julian Date of periastron passage ( T ◦ ), the projected primary semi-major axis ( a sin i ), m sin 3 i , the mass ratio ( q ), the rms residual velocity for the primary from the orbital solution ( σ ), and the number of RV measurements ( N ). The second row contains the respective errors on each of these values where appropriate. The third and fourth rows are the amplitude ( K ), projected semi-major axis ( a sin i ), m sin 3 i , the rms residual velocity from the orbital solution ( σ ), and the number of RV measurements ( N ) for the secondary star and the respective errors.
Table 5 Orbital Parameters For NGC 6819 Single-lined Binaries
| ID | P (days) | Orbital Cycles | γ (km s - 1 ) | K (km s - 1 ) | e | ω (deg) | T ◦ (HJD-2400000 d) | a sin i (10 6 km) | f ( m ) ( M /circledot ) | σ (km s - 1 | N |
|------|------------------|------------------|-----------------|-----------------|----------------------|------------|-----------------------|---------------------|----------------------------|---------------|-----|
| 1001 | 653.4 ± 0.5 | 6.1 | 1.43 ± 0.17 | 21.49 ± 0.23 | 0.226 ± 0.011 | 88 ± 3 | 48784 ± 4 | 188.1 ± 2.0 | 6.21 e - 1 ± 2.0 e - 2 | 0.65 | 23 |
| 1006 | 1524 ± 5 | 5.5 | 2.73 ± 0.13 | 5.37 ± 0.18 | 0.24 ± 0.04 | 131 ± 10 | 49770 ± 40 | 109 ± 4 | 2.25 e - 2 ± 2.2 e - 3 | 0.49 | 18 |
| 1010 | 1144 ± 3 | 7.7 | 1.1 ± 0.4 | 7.2 ± 0.7 | 0.55 ± 0.04 | 107 ± 11 | 53065 ± 22 | 95 ± 9 | 2.6 e - 2 ± 0.7 e - 2 | 1.56 | 32 |
| 2008 | 2379 ± 9 | 3.3 | 1.79 ± 0.14 | 5.76 ± 0.16 | 0.24 ± 0.04 | 143 ± 5 | 52040 ± 30 | 183 ± 5 | 4.3 e - 2 ± 0.4 e - 2 | 0.51 | 41 |
| 2012 | 2920 ± 30 | 1.8 | 2.64 ± 0.16 | 6.01 ± 0.20 | 0.39 | 19 ± | 51860 ± 30 | 223 ± 8 | 5.1 e - 2 ± 0.5 e - 2 | 0.41 | 19 |
| 3002 | 17.6978 ± 0.0003 | 414.5 | 2.2 ± 0.3 | 42.7 ± 0.5 | ± 0.03 0.022 ± 0.012 | 4 130 ± 30 | 48961.8 ± 1.4 | 10.38 ± 0.12 | 1.42 e - 1 ± 0.5 e - 2 | 2.02 | 37 |
| 3011 | 7369 ± 19 | 1.1 | 3.02 ± 0.11 | 4.28 ± 0.11 | 0.608 ± 0.012 | 345 ± 4 | 54980 ± 30 | 344 ± 10 | 2.99 e - 2 ± 2.3 e - 3 | 0.32 | 26 |
| 4008 | 1449 ± 4 | 4.5 | 1.22 ± 0.09 | 5.13 ± 0.13 | 0.06 ± 0.03 | 180 ± 30 | 50710 ± 130 | 102 ± 3 | 2.01 e - 2 ± 1.5 e - 3 | 0.50 | 33 |
| 6001 | 323.1 | 6.0 | 2.5 | 7.0 | 0.43 | 319 | 55442 | 28 | 9.0 e - 3 | 0.79 | 16 |
| | ± 2.1 | | ± 0.3 | ± 0.7 | ± 0.05 | ± 9 | ± 7 | ± 3 | ± 0.3 e - 2 | | |
| 6002 | 3360 | 1.6 | 2.19 | 4.7 | 0.72 | 113 | 48249 | 151 | 1.2 e - 2 | 0.49 | 21 |
Table 5
## Orbital Parameters For NGC 6819 Single-lined Binaries
| | ± 50 | | ± 0.20 | ± 0.6 | ± 0.06 | ± 4 | ± 24 | ± 23 | | |
|-------------|--------------------------------|-----------|--------------------|-----------------|---------------------|-----------------------|-----------------------|----------------------------------------------|-----------|-------|
| 6004 | 113.615 | 35.8 | 1.36 | 17.2 | 0.162 | 166 49453.3 | 26.5 | ± 0.5 e - 2 5.8 e - 2 | 0.89 | 26 |
| 6006 | ± 0.022 5200 | 1.6 | ± 0.21 0.64 | ± 0.3 3.44 | ± 0.019 0.10 | ± 6 344 | ± 1.9 ± 0.4 56200 245 | ± 0.3 e - 2 2.2 e - 2 | 0.40 | 35 |
| 7009 | ± 120 ± 21 209.89 216.5 | 23.1 | ± 0.12 0.22 | ± 0.14 20.5 | ± 0.04 0.585 | ± 300 | ± 10 | ± 0.3 e - 2 1.00 e - 1 | 0.61 | 38 |
| 8007 | ± 0.04 ± 1.1 66.837 79 | 82.6 3.74 | ± 0.11 | ± 0.3 | ± 0.007 | 49988.9 ± 0.6 51623.2 | 48.0 ± 0.8 3.62 | ± 0.5 e - 2 4.2 e - 4 | 0.21 | 17 |
| 8012 | ± 0.016 ± 6 213.9 | 50.0 | ± 0.10 1.21 | 4.04 ± 0.12 | 0.22 ± 0.03 | ± 1.2 | ± 0.11 | ± 0.4 e - 4 - | 0.45 | 18 |
| 9001 | 40.744 ± 0.008 ± 1.4 | 4.1 | ± 0.12 | 14.21 ± 0.18 | 0.586 ± 0.009 | 52417.46 ± 0.13 | 6.45 ± 0.10 | 6.5 e 3 ± 0.3 e - 3 | 0.57 | 12 |
| | 458.4 249 ± 1.1 ± 5 | | 1.58 ± 0.20 | 8.0 ± 0.4 | 0.54 ± 0.03 | 55368 ± 3 | 43 ± 3 | 1.5 e - 2 ± 0.3 e - 2 | | |
| 9002 | 1584 74 ± 10 ± 7 | 2.7 | 1.99 ± 0.12 | 3.94 | 0.39 | 50031 | 79 ± 5 | 7.9 e - 3 ± 1.5 e - 3 | 0.43 | 16 |
| 9026 | 1240 99 | 1.8 | 3.64 | ± 0.24 4.0 | ± 0.06 0.35 | ± 22 51700 | 63 ± | 6.6 e - 3 ± 1.8 e - 3 | 0.39 | 19 |
| 11006 | ± 30 ± 10 7810 172 | 1.1 | ± 0.12 2.35 | ± 0.4 3.8 | ± 0.06 | ± 30 48470 | 6 310 | 1.9 e - 2 | 0.70 | |
| | ± 60 ± 7 | | ± 0.22 | ± 0.3 | 0.67 ± 0.04 | ± 70 | ± 30 | ± 0.4 e - 2 | | |
| 13001 ± | 4.24079 205 ± 20 | 871.0 | 2.7 ± 0.6 | 55.1 ± 1.0 | 0.047 ± 0.020 | 54002.05 ± 0.24 | 3.21 ± 0.06 | 7.3 e - 2 ± 0.4 e - 2 | 2.43 | 16 |
| 13007 | 0.00004 511 353 ± 21 | 5.5 | 1.92 ± | 1.7 | ± | 52927 ± 20 | 10 ± 3 | 1.7 e - 4 ± 1.2 e - 4 | 0.45 | 15 |
| | ± 3 300 | 860.0 | 0.13 | ± 0.4 | 0.47 0.12 0.02 | 52959.5 | 1.07 ± 0.03 27.2 | 1.66 e - 2 ± 1.5 e - 3 | 2.93 | |
| 14008 ± | 1.719530 0.000024 ± 80 | | 2.6 ± 0.9 | 45.3 ± 1.3 | ± 0.03 | ± 0.4 | | | | 14 |
| 14009 | 174.464 84 ± ± 4 | 23.4 | 2.63 ± 0.23 | 15.7 | 0.691 ± | 53620.3 ± | | 2.6 e - 2 | 0.72 | 32 |
| 14012 | 0.020 762 193 | 5.7 | 3.6 ± 0.3 | ± 0.8 | 0.017 0.13 | 0.5 55610 | ± 1.6 73 | ± 0.4 e - 2 2.7 e - 2 | 1.37 | 42 |
| | ± 4 ± 24 620.5 211.9 | 8.8 | 2.12 | 7.0 ± 0.5 | ± 0.05 | ± 50 53378.9 | ± 5 63 | ± 0.6 e - 2 2.5 e - 2 | 0.77 | 33 |
| 16006 | ± 0.6 ± 2.2 | 6.9 | ± 0.17 | 10.5 ± | 0.714 | ± 2.3 | ± 3 20 | ± 0.3 e - 2 7.7 e - 3 | 0.26 | |
| 16011 | 209.446 273 | | 3.1 | 0.4 18 | ± 0.019 0.92 | 55232.0 | | | | 15 21 |
| 17005 | ± 0.019 ± 21 543.1 249.3 ± 0.7 | 6.7 | ± 0.5 1.08 | ± 10 9.83 | ± 0.07 0.498 | ± 0.7 52892 | ± 14 63.7 | ± 1.4 e - 2 3.49 e - 2 | 0.33 0.46 | 16 |
| 17008 | ± 1.5 2230 97 ± 80 | 1.3 | ± 0.11 2.99 ± 0.15 | ± 0.17 ± | ± 0.015 ± | ± 3 ± | ± 1.3 107 | ± 1.9 e - 3 9.8 e - 3 | | |
| | ± 16 | | 0.7 | 3.54 0.19 | 0.16 0.07 | 54710 90 | ± 6 | ± 1.6 e - 3 | | 19 |
| 17025 | 135.0 300 ± 0.3 ± 8 142.3 | 8.8 | ± 0.4 | 10.3 ± 0.3 | 0.28 ± 0.03 | 52094 ± 4 | 18.3 ± 0.5 | 1.35 e - 2 ± 1.2 e - 3 | 0.68 | 19 |
| 18019 20002 | 304 ± 0.3 ± 5 141.74 197 | 13.3 | 1.84 ± 0.12 | 4.78 ± 0.17 | 0.32 ± 0.03 0.52 | 52045.1 ± 1.8 | 8.9 ± 0.3 | 1.38 e - 3 ± 1.5 e - 4 | 0.34 | 17 |
| | ± 0.22 ± 8 3540 | 30.4 | 2.8 ± 0.4 | 7.1 ± ± | ± 0.09 ± | 52665 ± 4 53300 ± 900 | 11.8 ± 2.3 154 | 3.3 e - 3 ± 1.9 e - 3 1.1 e - 2 | 1.00 0.64 | |
| 20003 | 40 ± 80 | 1.3 | 2.3 | 1.3 3.17 0.25 | 0.08 0.12 0.37 | | ± 12 | ± 0.3 e - 2 | | 16 15 |
| 20008 | ± 140 2540 283 ± 160 ± 14 | 1.6 | ± 0.3 3.3 ± 0.5 | 7.9 ± | ± | 55630 ± 80 53102.8 | 258 ± 19 9.4 | 1.06 e - 1 ± 2.2 e - 2 | 1.26 | |
| 20010 | 135 | 33.1 | 3.1 | 0.5 15.8 | 0.07 0.53 ± 0.06 | ± 1.2 | ± 1.5 | 1.3 e - 2 | 1.50 | |
| 20013 | 51.00 ± 0.07 ± 9 255 | | ± 0.5 2.23 ± 0.20 | ± 2.4 | | 52296.6 ± | 23.2 | ± 0.6 e - 2 | 0.26 | 15 15 |
| 21007 | 121.570 0.017 ± 3 | 15.1 49.6 | 2.3 | 15.6 ± 0.3 16.2 | 0.454 0.008 0.40 | 0.6 52123.3 | ± 0.4 4.87 | 3.36 e - 2 ± 1.7 e - 3 | | 15 |
| | ± 23.921 248 ± 0.017 ± 6 | | ± 0.5 | ± | ± ± 0.03 | ± 0.4 51960 | ± 0.13 147 | 8.0 e - 3 ± 0.6 e - 3 | 0.88 0.82 | 16 |
| 22006 | 2390 311 | 1.5 | 2.55 ± 0.24 | 0.4 5.5 | 0.59 ± | ± 50 | ± 12 | 2.2 e - 2 ± 0.5 e - 2 | | |
| 22020 | ± 40 ± 9 81.68 352 ± 0.13 | 13.7 | 0.46 ± | ± 0.4 | 0.04 ± | 52125 ± 3 | 11.4 ± 0.3 | 8.8 e - 3 ± - | 0.69 | |
| 23009 | ± 14 771 291 ± 3 | 7.2 | 0.23 2.22 ± 0.14 | 10.2 ± 0.3 | 0.14 0.03 | 53411 ± 12 | 71.1 | 0.8 e 3 2.41 e - 2 | 0.65 | 12 32 |
| | ± 5 83.33 ± 0.11 | 13.2 | 1.08 | 7.04 ± | 0.302 ± | 55281.1 ± 0.5 | ± 2.1 20.2 ± 0.4 | ± 2.1 e - 3 4.8 e - 2 ± 0.3 e - 2 | 0.62 | 12 |
| 24004 | 172 ± 3 | | ± 0.22 | 0.20 20.4 ± 0.4 | 0.024 0.501 ± 0.010 | 52137 | 6.69 | | | 18 |
| 24012 ± | 25.5266 120 0.0013 | 114.6 | 1.69 ± 0.15 | 19.1 | ± | ± 12 54884.0 | ± 0.09 20.1 | 1.83 e - 2 | 0.53 | |
| 24031 | ± 170 91.20 14 ± 0.04 ± 3 | 16.2 | 2.15 ± 0.19 | ± 0.3 16.9 | 0.004 0.011 0.323 ± | ± 0.7 52029.50 | ± 0.4 4.66 | ± 0.7 e - 3 3.90 e - 2 | 0.69 | 18 |
| 25004 | 21.284 297.4 ± 0.003 | 138.3 | 1.2 ± 0.3 | ± 0.3 17.53 ± | 0.016 ± | ± 0.21 54467.8 | ± 0.07 0.677 | ± 2.4 e - 3 8.9 e - 3 ± 0.4 e - 3 4.99 e - 3 | 0.48 | |
| | ± 2.1 1.574695 | 1947.6 | 1.47 | 0.23 31.3 | | | | | 1.36 | 21 33 |
| 25011 | | | ± 0.24 | ± 0.4 | 0.417 0.016 0.009 | | ± 0.008 | ± 1.7 e - 4 | | |
| ± 26007 | 140 0.000005 ± 90 | | 2.91 | 24.9 | ± | ± 0.4 53070.2 | 15.35 | - | | 13 |
| | 46.693 297 ± 0.007 | 87.7 | ± 0.24 | ± 0.4 | 0.010 ± | ± 0.5 | ± 0.23 | 6.6 e 2 ± 0.3 e - 2 | 0.74 | |
| 27011 | ± 4 1340 | 2.8 | 2.0 | 4.4 | 0.277 0.011 | 52240 | 52 | 3.0 e - 3 | 0.67 | 20 |
| | 83 ± 30 | | ± 0.3 | ± 1.1 | ± | ± | ± 17 | ± 0.3 e - 2 | | |
| 27020 | ± 13 | 4.2 | 2.44 | 5.3 | 0.77 0.11 0.26 | 30 52267 | 41.1 | 8.2 e - 3 | 0.52 | 16 |
| | 583 | | | ± 0.3 | ± | | | ± 1.2 e - 3 | | |
| | 268 ± 3 | | ± 0.15 | | 0.05 | ± 13 | ± 2.1 | 3.4 e - 2 | 0.61 | 13 |
| 28004 | 414 | 3.7 | 2.2 | 13.0 | | | 52 | | | |
| | | | ± | ± | | | | | | |
| | | | 0.4 | 1.8 | ± | | ± | ± - | | |
| | | | | | | 55639 ± 4 | 9 | | | |
| | | | | | | | | 1.5 e 2 | | |
| | ± 3 | | | | | | | | | |
| | | | | | 0.71 0.08 | | | | | |
| | | | | | ± 9 119 ± 3 | | | | | |
Table 5 Orbital Parameters For NGC 6819 Single-lined Binaries
| Orbital Parameters For NGC 6819 Single-lined Binaries | Orbital Parameters For NGC 6819 Single-lined Binaries | Orbital Parameters For NGC 6819 Single-lined Binaries | Orbital Parameters For NGC 6819 Single-lined Binaries | Orbital Parameters For NGC 6819 Single-lined Binaries | Orbital Parameters For NGC 6819 Single-lined Binaries | Orbital Parameters For NGC 6819 Single-lined Binaries | Orbital Parameters For NGC 6819 Single-lined Binaries | Orbital Parameters For NGC 6819 Single-lined Binaries | Orbital Parameters For NGC 6819 Single-lined Binaries | Orbital Parameters For NGC 6819 Single-lined Binaries |
|-------------------------------------------------------------------------------|--------------------------------------------------------------------------------|---------------------------------------------------------|---------------------------------------------------------|---------------------------------------------------------|-------------------------------------------------------------------|---------------------------------------------------------|------------------------------------------------------------------|--------------------------------------------------------------|---------------------------------------------------------|---------------------------------------------------------|
| 29021 | 370.3 | 4.6 | 1.2 | 12.8 | 0.11 | 118 | 54967 | 65 - | 0.92 | 16 |
| | ± 2.1 | | ± 0.4 | ± 0.7 | ± 0.04 | ± 16 | ± 16 ± | 7.9 e 2 3 ± 1.3 e - 2 | | |
| 30008 | 93.16 ± 0.13 | 41.8 ± | 1.14 ± | 4.43 | 0.46 55174 ± 3 | 297 | 5.0 | 5.9 e - 4 | 0.44 | 14 |
| 31004 | 22.857 | 84.3 | 0.20 2.8 | 0.23 14 | ± 0.04 0.39 55742.2 | ± 8 307 | ± 0.3 4.0 | ± 1.0 e - 4 5.0 e - 3 | 0.42 | 14 |
| 32006 | ± 0.003 1.84716 | 2623.5 | ± 1.3 2.2 ± 0.3 | ± 3 18.2 ± 0.3 | ± 0.07 ± 1.1 0.05 52305.65 | ± 8 330 | ± 0.8 0.462 | ± 0.3 e - 2 1.15 e - 3 | | 20 |
| 33002 0.016 ± 30 0.26 165 | ± 0.00004 10.9025 ± 0.0006 839 4.4 | | 3.1 ± 0.5 | 40.5 ± 0.5 ± | ± 0.03 0.14 0.022 54194.1 ± 1.0 51882 | ± 30 ± 10 | ± 0.009 6.07 | ± 0.7 e - 4 7.5 e - 2 ± 0.3 e - 2 6.6 e - 1 | 1.08 1.56 | |
| 33005 ± 0.03 ± 34006 | ± 5 60.48 ± 0.04 24.737 ± 0.005 4.64123 0.00013 176.4 ± 0.6 71.70 ± 0.04 288.5 | 232.3 | 2.8 ± 0.5 1.3 ± 0.3 1.40 | 20.3 ± 0.8 14.2 ± 0.4 | ± 19 0.24 55353.8 ± 0.03 ± 1.2 0.36 52042.16 ± 0.19 51999.66 0.03 | 8 226 ± 7 136 ± 4 ± 2.4 ± | ± 0.08 226 ± ± ± | 1.63 e - 2 ± 1.4 e - 3 4.0 e - 3 ± 0.4 e - 3 2.71 e - 2 | 1.73 0.89 | 18 19 |
| 35020 | 20.6 ± 1.7 36.708 ± 0.010 2070 | 20.2 | | 12.4 ± 0.5 39.2 | 0.03 0.208 0.012 0.65 ± 0.04 5 0.09 | 8 | 10 11.4 0.3 3.93 0.15 | ± 0.8 e - 1 | 0.58 0.74 0.69 | 15 |
| | ± 60 370.5 ± 0.7 14.2111 ± 0.0015 1.615916 ± 0.000010 646 ± 3 | 45.4 | ± 0.17 2.45 | ± 0.5 5.8 ± 0.7 | ± 0.03 0.31 ± 0.06 | 117.2 ± 299 ± 348 ± 15 250 ± 12 240 | 2.45 ± 0.03 10.6 ± 1.4 | ± 1.0 e - 3 | | 15 |
| 35025 36012 | ± | 199.4 | ± 0.22 3.3 ± 0.4 | 14.2 ± 0.3 | ± 3 0.06 | | 13.97 ± 0.25 | 1.5 e - 3 ± 0.6 e - 3 - | 0.47 | 17 |
| | | 13.3 11.5 | 3.3 ± 0.4 1.32 ± 0.17 2.7 | 4.05 0.21 17.1 ± 0.8 4.4 | 9 ± 0.05 0.38 | ± 40 111 | 15.3 ± ± ± | 2.11 e 2 1.1 e - 3 1.7 e - 3 0.3 e - 3 1.9 e - 2 0.3 e - 2 | | 12 19 |
| 36043 ± 256 | 247.34 ± 0.18 42.54 ± 0.10 73.027 ± | | | ± ± 0.4 6.1 | ± 20 | 54410 ± 56244 ± 54638 ± 54427 ± 51560 | ± 2.0 1.79 ± 0.10 1.19 ± 0.03 100 ± 50 47.6 ± 0.8 2.9 ± 0.6 17.9 | ± ± ± 1.5 e - 2 ± 0.4 e - 2 8.3 e - 3 ± 1.6 e - 3 1.13 e - 3 | 0.56 1.41 | 23 |
| 37028 39013 | | 99.2 1.8 | ± 0.6 2.09 | | 4 ± 0.08 | ± 13 332 20 | 0.8 8.6 0.4 116 | | 2.00 | 18 |
| 40006 | | | ± 0.22 2.5 | ± 0.4 10.8 | ± 90 0.16 53502 ± 0.05 0.53 | 51878.89 | 10 30.5 | | 0.81 | 17 |
| ± 240 | | 11.7 | ± 0.3 2.3 ± 0.3 | ± 0.5 53.7 ± 1.2 | ± 0.020 0.68 ± 0.12 | 5 ± 53291 | | 1.8 e - 4 2.59 e - 2 | 0.85 | 24 26 |
| 40017 46013 | | 105.4 1573.2 | 3.8 ± 0.8 | | ± 0.04 0.022 | ± 0.11 53897.1 0.3 53207 ± 5 | | ± 1.7 e - 3 | | |
| 49002 | | | 0.5 ± 1.0 | ± | | | | ± 1.09 e - 1 | 3.18 | 20 |
| | | 7.9 | | 16 7 | | ± 60 129 ± 11 326 | | 1.1 e - 2 | | 23 |
| 49004 | | | 1.36 ± 0.21 | 14.03 0.25 | 0.053 ± 0.018 | ± 21 | | ± | 0.75 | |
| 52015 | | 12.2 | 1.9 | ± | | 77 | | 7.1 e - 2 ± 0.4 e - 2 | 0.69 | 18 |
| 53003 | | 23.7 | ± 0.5 | ± | 0.39 ± 0.14 | ± 14 | ± 15 55107.7 ± 2.0 | 5.0 e - ± 0.3 e - | 0.73 | 16 |
| | | | | 5.4 1.0 | | 41.3 | | 4 3 4.25 e - 2 | | |
| 54027 ± | | 31.1 | 3.09 ± 0.22 | 21.2 ± 0.3 | 0.55 ± 0.03 | 2.2 90 | 55589.2 ± 0.7 ± 0.4 | ± 2.2 e - 3 2.0 e - 1 | 0.49 | 13 |
| 56010 ± 0.006 ± | 0.019 2600 ± 40 9.1537 ± 0.0004 | 1.5 128.4 | 2.6 ± 1.9 3.40 ± 0.18 ± | 9 ± 3 40.03 0.22 ± | 0.23 ± 0.19 ± 0.123 56052.14 0.06 | 30 59.3 2.3 ± | 54510 320 150 ± 110 5.00 ± 0.03 | ± 1.9 e - 1 | 0.77 0.47 | 12 |
| 57004 59003 0.275 187.1 0.016 ± 2.3 60006 0.019 319 0.011 ± 21 62010 0.70 338 | 21.368 99.7 ± 0.003 7.8431 153.2 ± 0.0003 1332 3.3 ± 5 1560 ± | ± ± 49.9 | 2.5 0.3 1.9 0.3 ± 2.12 2.2 ± 0.3 3.9 | 31.6 ± 0.4 ± 33.14 0.22 ± 7.0 ± 0.3 38.6 ± 0.7 19.8 | 52336.13 0.16 ± ± 0.017 0.190 0.023 0.64 | ± 15 3 | 8.93 ± 0.11 51728.8 3.574 0.5 ± 0.024 54382 17 53178.6 ± | 5.95 e - 2 ± 1.0 e - 3 6.22 e - 2 ± 2.2 e - 3 2.96 e - 2 | 0.80 0.66 0.79 | 12 17 |
| 66016 | 30 | 4.9 | ± 0.9 1.56 | ± 0.3 15 ± 6 | ± 0.09 | | 53500 | 1.1 e - ± 1.2 e - 6.6 e - | | |
| | | | | | | | ± ± | ± 2.5 e - 3 1.35 e - 2 ± 0.8 e - 3 4.51 e - 2 | | |
| 66004 | | | | | ± ± | 5 | ± 0.8 54836 ± 4 | ± 0.6 e - 3 1.72 e - 2 | | |
| | | | | | | | 15.9 0.3 100 | ± 2.2 e - 3 | | |
| | 640 | | ± 0.5 | | | 70 | ± 0.5 54456.9 | | | |
| | ± 0.018 | | | | | | 1.208 0.023 | | | |
| | 0.000015 59.383 | | | | 0.016 | 150 ± 329 | | | 0.87 | |
| | | | | | | | | | | 17 |
| | 2.277936 | | 2.9 | | | | ± 6 | | 1.80 | |
| | ± 7 | | ± 0.20 | | ± 0.03 | ± 5 | 91 | | | 23 |
| | | 1324.8 | | | | | | | | |
| | ± | | | | | ± | | | | 16 |
| 69005 | | | | | | | | 1 | | |
| | | | | | | | 40 | | 0.77 | 18 |
| 77007 | | 2.8 | | 3.9 | | ± | 74 | 1 3 | | |
| | | | 0.22 | ± 0.4 | 0.48 ± 0.10 | 298 11 | ± 9 | ± 2.2 e - 3 | 0.83 | 21 |
| | | | ± | | | | | | | |
| | | | | | | ± | | | | |
| | | | | | | | ± 50 | | | |
+
+
+
Figure 7. NGC 6819 SB1 orbit plots. For each binary, we plot RV against orbital phase, showing the data points with filled circles and the orbital fit to the data with the solid line. The dotted line marks the γ -velocity. Beneath each orbital plot, we show the residuals from the fit. Above each plot, we give the binary WOCS ID and orbital period.
Figure 7. (Continued)
Figure 7. (Continued)
Figure 8. NGC 6819 SB2 orbit plots. For each binary, we plot RV against orbital phase, showing the primary-star data points with filled circles and the secondary-star data points with open circles. The orbital fits to the data are plotted in the solid and dashed lines for the primary and secondary stars, respectively. The dotted line marks the γ -velocity. Beneath each orbital plot, we show the residuals from the fit. Above each plot, we give the binary WOCS ID and orbital period.
Table 6 Orbital Parameters For NGC 6819 Double-lined Binaries
| ID | P (days) | Orbital Cycles | γ (km s - 1 ) | K (km s - 1 ) | e | ω (deg) | T ◦ (HJD-2400000 d) | a sin i (10 6 km) | m sin 3 i ( M /circledot ) | q | σ (km s - 1 | N |
|--------------|-----------------------------|------------------|-----------------|-----------------------------|---------------------|-------------|-------------------------|---------------------------------------|-----------------------------------------|---------------------|----------------|----------|
| 4004 | 297.17 ± 0.09 | 27.4 | 2.5 ± 0.3 | 18.1 ± 0.5 22.9 | 0.227 ± 0.017 | 89 ± 5 | 51907 ± 4 | 72.2 ± 2.0 91.3 | 1.10 ± 0.06 0.87 | 0.79 ± 0.03 | 2.10 1.56 | 38 21 |
| 11004 | 91.336 ± 0.012 | 20.2 | 2.48 ± 0.22 | ± 0.4 31.8 ± 0.3 40.1 | 0.447 ± 0.007 | 115.6 ± 1.5 | 51635.69 ± 0.20 | ± 1.9 35.7 ± 0.4 45.1 | ± 0.05 1.41 ± 0.06 1.11 | 0.792 ± 0.017 | 0.72 2.14 | 19 17 |
| 12006 | 33.2393 ± 0.0022 | 111.3 | 1.75 ± 0.14 | ± 0.8 39.97 ± 0.21 54.3 | 0.366 ± 0.006 | 286.5 ± 0.8 | 52256.78 ± 0.08 | ± 0.9 17.00 ± 0.09 23.11 | ± 0.04 1.34 ± 0.03 0.986 | 0.735 ± 0.008 | 0.54 1.31 | 18 16 |
| 14005 | 29.7184 ± 0.0008 | 145.0 | 3.27 ± 0.20 | ± 0.5 46.3 ± 0.4 52.2 | 0.308 ± 0.007 | 171.9 ± 1.4 | 53425.03 ± 0.13 | ± 0.21 18.02 ± 0.15 20.3 | ± 0.014 1.34 ± 0.05 1.19 | 0.888 ± 0.016 | 0.73 1.89 | 22 20 |
| 17003 | 20.8904 ± 0.0011 | 106.5 | 2.9 ± 0.3 | ± 0.8 41.1 ± 1.3 43.2 | 0.059 ± 0.015 | 351 ± 10 | 52489.2 ± 0.6 | ± 0.3 11.8 ± 0.4 12.39 | ± 0.03 0.66 ± 0.03 0.63 | 0.95 ± 0.04 | 2.70 1.29 | 15 12 |
| 23018 24009b | 3.574911 ± 0.000021 3.64921 | 609.4 983.5 | 2.9 ± 0.3 0.4 | ± 0.6 76.3 ± 0.5 97.3 ± 0.5 | 0.010 ± 0.005 | 326 ± 25 69 | 55064.78 ± 0.24 54836.1 | ± 0.21 3.75 ± 0.03 4.783 ± 0.024 4.52 | ± 0.05 1.086 ± 0.014 0.852 ± 0.013 1.05 | 0.785 ± 0.007 1.023 | 2.39 2.04 4.86 | 49 37 26 |
| 25010 | ± 0.00004 18.78991 | | ± 0.7 3.72 | 90.0 ± 1.3 88.0 ± 1.3 56.7 | 0.005 ± 0.011 0.397 | ± 140 305.8 | ± 1.4 53192.52 | ± 0.07 4.42 ± 0.07 13.45 | ± 0.04 1.08 ± 0.04 1.29 | ± 0.023 | 5.00 | 22 |
| | ± 0.00020 | 136.0 | ± 0.21 | ± 0.6 61.4 ± 0.4 | ± 0.006 | ± 1.4 | ± 0.05 | ± 0.16 14.56 ± 0.13 | ± 0.03 1.19 ± 0.03 | 0.924 ± 0.013 | 1.16 0.88 | 12 11 |
| 37013 | 167.19 ± 0.04 | 24.0 | 1.7 ± 0.3 | 24.3 ± 0.7 23.9 | 0.414 ± 0.019 | 17 ± 3 | 54072.7 ± 0.9 | 50.8 ± 1.6 50.0 | 0.73 ± 0.06 0.74 | 1.02 ± 0.04 | 1.76 2.08 | 23 19 |
| 38008 | 5.49633 ± 0.00004 | 269.3 | 3.47 ± 0.11 | ± 0.8 57.75 ± 0.15 67.5 | 0.002 ± 0.002 | 190 ± 70 | 51994.5 ± 1.1 | ± 1.8 4.365 ± 0.013 5.10 | ± 0.06 0.604 ± 0.017 0.516 | 0.855 ± 0.012 | 0.45 2.62 | 19 19 |
| 40007 | 3.185087 ± 0.000012 | 1023.6 | 3.7 ± 0.5 | ± 0.8 87.9 ± 1.0 103.3 | 0.002 ± 0.008 | 90 ± 300 | 54329.3 ± 2.4 | ± 0.07 3.85 ± 0.05 4.53 | ± 0.008 1.25 ± 0.03 1.06 | 0.851 ± 0.014 | 4.25 4.87 | 40 37 |
| 42002 | 278.98 ± 0.23 | 14.5 | 3.1 ± 0.3 | ± 1.1 19.4 ± 0.5 20.0 | 0.375 ± 0.018 | 305 ± 6 | 53522 ± 4 | ± 0.05 69.2 ± 2.1 71 | ± 0.03 0.72 ± 0.13 0.70 | 0.97 ± 0.09 | 1.09 3.36 | 14 11 |
| 42007 | 145.342 ± 0.018 | 30.7 | 2.05 ± 0.25 | ± 1.5 26.4 ± 0.8 26.4 | 0.541 ± 0.019 | 170.8 ± 1.7 | 52910.5 ± 0.5 | ± 6 44.4 ± 1.0 44.3 | ± 0.08 0.66 ± 0.06 0.66 | 1.00 ± 0.05 | 1.37 2.80 | 29 26 |
| 55007 | | | 3.17 0.25 | ± 1.2 54.7 ± 0.4 | 0.259 | 160.0 ± | 54360.34 | ± 1.9 9.97 ± 0.09 | ± 0.04 1.017 0.023 | | 1.18 | 23 |
| | 13.72355 ± 0.00018 | 232.8 | ± | 60.1 | ± 0.006 | 1.2 | ± 0.04 | 10.96 | ± 0.925 | 0.910 ± 0.011 | 1.68 | 20 38 |
| 76004 | 4.601485 ± 0.000021 | 724.1 | 0.79 ± 0.23 | ± 0.5 66.7 ± 0.4 91.5 | 0.009 ± 0.005 | 12 ± 25 | 54704.7 ± 0.3 | ± 0.11 4.222 ± 0.024 5.79 | ± 0.020 1.092 ± 0.023 0.796 | 0.729 ± 0.008 | 1.44 3.48 | 32 |
## 6.
## BINARIES OF NOTE 6.1. WOCS 3002
WOCS 3002 is an SB1 binary located near the red clump, with V = 12.73 and ( V -I ) = 1.29. This system has a circular orbit and a short period of 17.7 days and is most likely a cluster member with P µ = 99% and P RV = 94%. As mentioned in Hole et al. (2009), the current orbital separation is too small for the primary to have evolved to the tip of the RGB without overfilling its Roche lobe. Gosnell et al. (2007) explores the likelihood that the current state of this system is a result of a dynamical exchange between a horizontal branch star and a binary system. They found that this scenario is possible but that only 0.1% of the parameter space could reproduce this system. Based on optical, UV, and Xray information Gosnell et al. (2012) classify this system as currently an RS Canum Venaticorum type object, a
tightly orbiting binary system with two chromospherically active late-type stars.
## 6.2. WOCS 13001
WOCS 13001 has V = 15.87 and ( V -I ) = 0.85 with a P µ = 99% and is a likely triple system. The light contribution from two stars is clearly seen in the spectra. The RVs for the more luminous of the two stars have an average velocity of 1.870 km s -1 and standard deviation of 1.386 km s -1 , resulting in an SM classification with a P RV of 93%. The velocity values of the second crosscorrelation peak have much greater variability. Fitting an orbital solution to just the velocities of the second peak, as shown in Figure 7, gives a 4.24 day orbit with an eccentricity of 0.048 and P RV = 94% based on the center-of-mass velocity of the solution. It is unlikely to have a chance alignment between a binary system and an SM of this cluster within our 3 ′′ fiber diameter. We suggest that the 4.24 day binary orbit is the inner binary of a long period hierarchical triple system, where the more luminous star is the tertiary companion.
## 6.3. WOCS 1006, WOCS 4008, and WOCS 6002
WOCS1006, WOCS 4008, and WOCS 6002 are all SB1 binaries. They are identified by Corsaro et al. (2012) as seismic outliers from the period spacing of mixed modes, lying away from the 17 other NGC 6819 stars in a ∆ P obs -∆ ν diagram.
WOCS 4008 has a P µ of 98%, a P RV of 88%, and is located slightly fainter and to the blue of the red clump. Corsaro et al. (2012) conclude that WOCS 4008 is a binary composed of a RGB member and a fainter less evolved companion, with the anomalous ∆ P obs potentially originating from a binary interaction that altered the core structure of the primary star. WOCS 4008 has a very low eccentricity ( e = 0.06 ± 0.03 ) at a period of almost four years, potentially indicating a history of mass transfer in this system. Taking the primary mass of 1.75 M /circledot from Corsaro et al. (2012), we find the lower limit for the secondary mass to be 0.46 M /circledot .
WOCS 6002 has a P µ of 99% and P RV of 94% and WOCS 1006 has a P µ of 99% and a P RV of 93%. Both are located near the red clump and have ∆ P obs that place them in the sequence of He-core burning stars, but with much higher masses (WOCS 6002: 2.3 M /circledot and WOCS 1006: 2.45 M /circledot ; Corsaro et al. 2012) than the other red clump stars in NGC 6819 ( ∼ 1.65 M /circledot ; Miglio et al. 2012). This supports the suggestion by Rosvick & Vandenberg (1998) that stars in that region of the CMD, brighter and bluer than the zero-age horizontal branch, may be the more evolved descendants of past blue stragglers. We find WOCS 6002 and WOCS 1006 to have periods of 3360 days and 1524 days, respectively. This is consistent with the trend found for the blue straggler population in NGC 188 (Geller & Mathieu 2012) where 13 of the 16 blue stragglers in binaries have periods of order 1000 days. From the primary mass estimates by Corsaro et al. (2012), we put the lower limit for the secondary mass for WOCS 6002 at 0.45 M /circledot and for WOCS 1006 at 0.59 M /circledot . Again this is consistent with the companion-mass distribution for the NGC 188 blue straggler binaries with periods of order 1000 days, which is narrow and peaked near 0.5 M /circledot (Geller & Mathieu
2011). Geller & Mathieu (2011) suggest this distribution indicates white dwarf companions and indeed white dwarf companions were found for three NGC 188 blue stragglers by Gosnell et al. (2014).
## 6.4. WOCS 7009 and WOCS 8007
Basu et al. (2011) used Kepler observations taken from 2009 May to 2009 December to study the oscillations of 21 red giant stars in NGC 6819 and derive estimates of mass, radius, and log( g ). Two of the giants listed in Basu et al. (2011), WOCS 7009 and WOCS 8007, have completed SB1 orbital solutions (Figure 7). 7009 has a P RV of 58% placing it near our member cutoff threshold, but it has P µ of 99% and Stello et al. (2011) identify it as a member from seismic data. Basu et al. (2011) estimate the primary mass to be 1.74 M /circledot . This mass combined with the mass function produced by the orbital solution gives a lower limit on the secondary mass of 0.88 M /circledot . WOCS 8007 has a P µ of 94%, P RV of 88%, and is also determined by Stello et al. (2011) to be a seismic member of NGC 6819. Basu et al. (2011) mass estimate for 1.62 M /circledot , putting the lower limit on the secondary mass at 0.108 M /circledot .
## 6.5. WOCS 40007
WOCS 40007 is a detached eclipsing SB2 located on the MS with V = 15.65, ( V -I ) = 0.84, and an orbital period of 3.185 days. It has a P µ = 97% and a P RV = 89%. Examining the observed minus expected RVs of the primary and secondary components reveals the influence of a tertiary companion in this system with an orbital period greater than ∼ 3000 days (Figure 9).
Jeffries et al. (2013) also find evidence for a triple companion in the Kepler eclipse timing of primary and secondary stars. Jeffries et al. (2013) used extensive BVR c I c data (obtained on the 1 m telescope on at the Mount Laguna Observatory), WOCS RV data, and a simultaneous three-body fit to derive properties for the primary (p) and secondary (s) components. They find M p = 1.236 ± 0.020 M /circledot , R p = 1.399 ± 0.007 R /circledot , M s of 1.086 ± 0.018 M /circledot , and R s = 1.098 ± 0.004 R /circledot .
## 6.6. WOCS 23009
WOCS 23009 is a long-period ( P = 773 days) SB1 with a P µ of 99% and P RV of 94%, located near the MS turnoff with ( V -I ) = 0.77 and V = 15.11. A total eclipse was detected in quarter 4 of the Kepler satellite. Sandquist et al. (2013) do a complete analysis of this system using the WOCS RV and Kepler information. They find a secondary mass, M s , of ∼ 0.45 M /circledot and use this system to estimate an age of 2.62 ± 0.25 Gyr for NGC 6819.
## 6.7. WOCS 24009
WOCS 24009 is near the MS turnoff with a V = 15.18 and ( V -I ) = 0.80 and a P µ of 99%. The spectra show contributions from three sources. The brightest source, WOCS 24009a, is a long-period velocity-variable star with a e/i = 7.60 and RV = 4.911 km s -1 . Examining the RVs of WOCS 24009a versus HJD (Figure 10) we find a slow change in RV over the course of more than 4000 days. The RVs peak at 12.3 km s -1 and have an observed minimum of ∼ 0 km s -1 , but it appears we have
Figure 9. Observed RVs minus computed RVs for the primary (filled circles) and secondary (open circles) stars in WOCS 40007. These residuals reveal the signature of a tertiary companion fit with an orbital period of ∼ 3100 days (dashed line). Evidence for a tertiary is also present in the photometry and eclipse timing of this system (Jeffries et al. 2013).
not observed this binary for a full orbit and any orbital solution fit to the RVs has very large errors on all the parameters. With the current RV information this star is classified as a binary with unknown membership (BU).
Figure 10. WOCS 24009 spectra show the contributions from three sources, all with variable RVs. Radial velocities of the brightest source in the spectra, 24009a, vs. Heliocentric Julian Date (filled circles) show a binary star with a period greater than 4500 days.
The two dimmer sources in the spectra, WOCS 24009b, are very similar. The CCF consistently shows almost identical peak heights and FWHM for these sources. We are able to obtain velocities for these two sources using TODCOR when the CCFs are not too blended with the brightest source mentioned above. Using the date of periastron and period obtained from Kepler observation of the eclipses in this inner binary (L. Brewer, private communication) we are able to obtain an orbital solu- tion for this binary (Figure 7). The γ -velocity for this short-period binary is 0.4 ± 0.7 km s -1 and results in a P RV of 66%. This velocity is lower than the cluster average of 2.45, but the P RV meets our threshold for cluster membership.
With a fiber diameter of 3 ′′ it is unlikely that we have a chance alignment of a field and cluster binary or even two binary cluster members. This system has a high potential of being a hierarchical triple system in NGC 6819.
## 6.8. WOCS 1010, WOCS 4004, and WOCS 14012
In NGC 6819 we identify 17 potential blue stragglers as being brighter or bluer than the MS turnoff and not on the blue hook. Seven of these blue stragglers are velocity variable and for WOCS 1010, WOCS 4004, and WOCS 14012 we have obtained orbital solutions.
WOCS 1010 has a V = 12.82 and ( V -I ) = 0.29, a P RV 86% and P µ of 87%. It is the bluest and one of the most luminous blue stragglers we have identified in NGC 6819. Its orbital period of 1,144 days and eccentricity of 0.55 are similar to those found for most of the blue stragglers in NGC 188 (Geller & Mathieu 2012). WOCS 1010 falls on the evolutionary track for a 2.4 M /circledot star, which is less than twice the ∼ 1.4 M /circledot turnoff mass of NGC 6819 (Bressan et al. 2012). However, we expect 2.4 M /circledot to be a lower mass limit for WOCS 1010 based on evidence that blue stragglers are less luminous than normal stars of the same mass (Sandquist et al. 2003, Geller & Mathieu 2012).
WOCS 4004 is a double-lined spectroscopic binary with a P RV of 94%, P µ of 99%, V = 13.98, and ( V -I ) = 0.61. It has a period of 297.17 days, eccentricity of 0.23, and mass ratio, q = 0.79. In an effort to determine more information about the companion star, we cross-correlate the spectra of WOCS 4004 against a grid of synthetic spectra 8 (Meibom et al. 2009) with solar-metallicity, log( g ) = 4.0, and spanning appropriate ranges in temperature and projected rotation velocity. From the best matches (the highest cross-correlation peak height) we derive an effective temperature of 6900 K ± 125 K for the primary and 6650 K ± 160 K for the secondary. Based on a 2.5 Gyr solar-metallicity PARSEC isochrone (Bressan et al. 2012) the hottest MS star in NGC 6819 has a T eff of 6382 K. Unfortunately since this temperature is within 2 σ of the temperature derived for the secondary star we are unable to conclude if the secondary star is another blue straggler or an MS star. But clearly the secondary is not a white dwarf.
WOCS14012has a P RV of 90%, P µ of 99%, V =14.85, and ( V -I ) = 0.58. It has the broadest absorption lines of any of the blue straggler candidates in NGC 6819. Using the same grid of synthetic spectra described above, we derive an effective temperature of 6465 K ± 130 K and a projected rotation velocity, v sin , of 59 km s i -1 ± 6 km s -1 for WOCS 14012. This projected rotation velocity is much faster than the 10 -20 km s -1 of typical MS stars in NGC 6819, but is similar to the v sin i values Mathieu & Geller (2009) found for some blue stragglers in NGC 188. We obtained more precise RVs, resulting
8 The library is based on ATLAS9 (http://kurucz.harvard.edu) and the companion program SYNTHE used to compute the synthetic spectrum from the model atmosphere and line list.
in a more precise orbital solution, by cross-correlating the spectra of WOCS 14012 against a synthetic spectrum with T eff of 6500 K and v sin i of 60 km s -1 , instead of the typical solar template. The RVs from using the synthetic spectrum are presented in Table 1 and used for Figure 7. The resulting orbital solution has a period of 762 days and an eccentricity of 0.13. Again, these are similar to the properties found for most of the blue straggler binaries in NGC 188 (Geller & Mathieu 2012).
## 7. TIDAL CIRCULARIZATION IN NGC 6819
We use the binary orbits of 70 MS stars to construct the distribution of orbital eccentricity versus orbital period ( e -logP ). The distribution in Figure 11 shows the usual transition from a wide range in eccentricities at longer periods to circular orbits at shorter periods, thought to result from tidal circularization over the age of the cluster. In particular, we find a tidal CP of 6 2 . +1 1 . -1 1 . days for NGC 6819. We find the CP and the associated uncertainty using the information and fitting method of Meibom & Mathieu (2005). Specifically, we fit
Figure 11. Distribution of orbital eccentricity, e , as a function of the log of orbital period in days for 62 MS binary members of NGC 6819. The dashed line marks an eccentricity of 0.01, the thin solid line marks an eccentricity of zero, and the thick solid line represents the best-fit circularization function. The circularization period of NGC 6819, 6 2 . +1 1 . -1 1 . days, is marked by a thick dash near the log( P ) axis.
$$e ( P ) = \begin{cases} 0. 0 & \text{if $P\leq P^{\prime}$}, \\ \alpha \ ( 1 - e ^ { \beta ( P ^ { \prime } - P ) } ) ^ { \Gamma } & \text{if $P>P^{\prime}$}, \end{cases} ( 4 ) \quad \text{the} \quad \text{a la}$$
to the observed period -eccentricity distribution. Based on Monte Carlo experiments and in order to minimize the sensitivity to the choice of eccentricity threshold, Γ is set to 1.0 and β is set to 0.14. The value of α is set to 0.35, the mean eccentricity of all observed binary orbits with periods longer than 50 days in the Pleiades, M35, Hyades, M67, and NGC 188. The location of the circularization function and P ′ are determined by minimizing the total absolute deviation between the observed eccentricity and the model. The tidal CP is defined as the period at which the best-fit circularization function equals 0.01, e (CP) = 0.01.
NGC 6819 is an intermediate-age open cluster, log(Age(Gyr))= 0.4, and falls roughly in the middle of the age range of the eight MS binary populations studied by Meibom & Mathieu (2005). Figure 12 shows tidal CP as a function of age for all of these systems. Also plotted are three theoretical predictions. The flat grey line is from Zahn & Bouchet (1989), in which tidal circularization is only significant during the pre-mainsequence (PMS) phase. The other predictions plotted are from the Binary-Star Evolution algorithm used in Hurley, Tout & Pols (2002) and the ad hoc tidal energy dissipation rate used by Geller, Hurley & Mathieu (2013). The ad hoc rate from Geller, Hurley & Mathieu (2013) is a combination of the PMS circularization algorithm from Kroupa (1995) and the Binary-Star evolution algorithm of Hurley, Tout & Pols 2002 with the convective damping term increased ad hoc by a factor of 100 over the default value in order to match the distribution of tidal CP with age (e.g., Figure 12).
A χ 2 -analysis shows that the CPs for the five youngest binary populations -PMS binaries, Pleiades, M35, Hyades/Praesepe, and NGC 6819 -are consistent with the theoretical prediction of Zahn & Bouchet (1989) that MS tidal circularization will be negligible and the CP will remain constant as the system ages. On the other hand, Figure 12 indicates that for binary populations older than NGC 6819 tidal circularization acts to increase their tidal CPs. As yet no self-consistent theory is able to explain the distribution of cutoff periods in Figure 12 (Meibom & Mathieu 2005).
## 8. BINARY FREQUENCY
## 8.1. Completeness in Binary Detection
In order to characterize the completeness of binary detections we use a Monte Carlo approach to produce simulated observations of a set of artificial binaries, following the procedure outlined in Geller & Mathieu (2012). Specifically, we generated artificial binaries with period and eccentricity distributions of the Galactic field binary population found by Raghavan et al. (2010). We used 1 M /circledot for the primary mass and selected the mass ratio, q, from a flat distribution. Orbital inclination, longitude of periastron, and phases were chosen randomly. Appropriate to NGC 6819, we set the eccentricity to zero for periods under 6.2 days to reflect the CP of NGC 6819 discussed in Section 7, we used the actual observation dates in the Monte Carlo, and the precision values of 0.7 km s -1 and 0.4 km s -1 for observations made at the CfA and WIYN, respectively, were incorporated. Given these assumptions, RV measurements were produced for a large number of synthetic binaries and their detectability assessed.
The resulting binary detection success rate varies most notably with distance from the cluster center due to using actual observation dates in the simulation and the less frequent observation near the outskirts of the cluster, and with increasing orbital period as long period binaries generally have lower RV variations. Our overall binary detection percentages are 89% for binaries with periods under 1,000 days, 81% for periods under 3,000 days, and 67% for periods under 10,000 days. These numbers are in good agreement to those for NGC 188 found by Geller & Mathieu (2012).
Figure 12. Tidal circularization periods determined by Meibom & Mathieu (2005) for eight different stellar populations (filled circles). The CP of NGC 6819 is plotted for comparison (large open circle around filled circle). The theoretical prediction of Zahn & Bouchet (1989) in which tidal circularization is only significant during the PMS phase is indicated by the large gray line and the prediction of the Binary-Star Evolution algorithm of Hurley, Tout & Pols (2002) is shown as the dark gray curve. The dashed line shows the ad hoc relationship introduced by Geller, Hurley & Mathieu (2013) that adds the pre-main-sequence circularization from Kroupa (1995) to the Binary-Star Evolution algorithm with convective damping increased by a factor of 100. This relationship is ad hoc, but fits the increase in tidal circularization observed in the older populations.
## 8.2. Binary Frequency
We have identified 679 members and likely members of NGC 6819; 113 of these are binaries (BMs and BLMs). On the MS, we find 460 single and 85 binary member and likely member stars. To calculate our MS binary frequency we first remove four binary stars from the sample that, by deconvolving the combined light with possible companions, contain a primary star fainter than our magnitude cutoff of V = 16.5. Next, we incorporate our 67% binary-detection percentage for binaries with periods less than 10 4 days. This results in an incompletenesscorrected MS binary frequency of 22% ± 3% for binaries with periods less than 10 4 days. This is consistent with the MS binary frequency of 24% ± 3% found in M35 (150 Myr; Geller et al. 2010) and 29% ± 3% found for NGC 188 (7 Gyr; Geller & Mathieu 2012) for binaries with periods under 10 4 days. This is also consistent with the observable MS binary frequency predicted for the age of NGC 6819 by N -body modeling of the older cluster NGC 188 Geller, Hurley & Mathieu (2013).
Using a fiducial fitting method Kalirai et al. (2001) quantified the location of the equal-mass binaries in NGC 6819 at ∼ 0.75 mag above the MS track and found an equal-mass binary percentage of ∼ 11% ± 2%. Using MS
SB2 binaries as a proxy for equal-mass binary systems, we find the incompleteness-corrected equal-mass binary frequency for P < 10 4 days to be much lower at 3% ± 1% (11/545). Investigating the MS equal-mass binaries in NGC 188 we find a similar incompleteness-corrected frequency of 5% ± 2% for binaries with period under 10 4 days.
## 9. SUMMARY
In this paper we present the current state of the WOCS RV survey of the rich open cluster NGC 6819 (2.5 Gyr) including the orbital solutions for 93 cluster binaries. This paper builds on the work of Hole et al. (2009), and presents the entire current WOCS database of RV measurements. We use our RV results, the proper-motion information of Platais et al. (2013), and the photometry of Yang et al. (2013) to produce a cleaned CMD in Figure 6 that includes BMs, BLMs, and SM stars. The CMD shows a clear MS turnoff, blue hook, red clump and potential blue stragglers, a large fraction of which are RV variable stars. We call special attention to a handful of interesting binaries in NGC 6819 including eclipsing binaries, Kepler asteroseismology targets, potential higher order multiple star systems and possible descendants of past blue stragglers. We find the incompleteness-corrected binary fraction for all MS binaries with periods less than 10 4 days to be 22% ± 3%. This binary fraction is similar to the results for the young open cluster M35 and the old open cluster NGC 188 (Section 8). With the MS binary orbits and the techniques of Meibom & Mathieu (2005) we determine a CP of 6 2 . +1 1 . -1 1 . days for NGC 6819. The addition of this CP indicates that tidal circularization of MS binary orbits does not present itself until after 2-3 Gyr. As part of WOCS we plan to continue to observe NGC 6819 and in the future we hope to address the mass segregation, blue straggler population, and tidal circularization of the cluster in more detail.
The authors express their gratitude to the WIYN Observatory staff for the many long nights and excellent spectra. Also, we would like to thank the many undergraduate and graduate students who helped obtain and reduce the many hours WIYN data that underlie this paper, including E. Braden, T. Hole, E. Leiner and B. Tofflemire. This work was funded by the National Science Foundation grant AST-0908082 to the University of Wisconsin-Madison. A.M.G. is funded by a National Science Foundation Astronomy and Astrophysics Postdoctoral Fellowship under Award No. AST1302765. K.E.M. is supported in part by the National Space Grant College and Fellowship Program and the Wisconsin Space Grant Consortium.
## REFERENCES
Anthony-Twarog B. J., Deliyannis C. P., Rich E., Twarog B. A., 2013, ApJL, 767, L19
Barden S. C., Armandroff T., Muller G., et al., 1994, in Proc. SPIE, Vol. 2198, Instrumentation in Astronomy VIII, Crawford D. L., Craine E. R., eds., pp. 87-97
- Basu S., Grundahl F., Stello D., et al., 2011, ApJL, 729, L10 Bragaglia A., Carretta E., Gratton R. G., et al., 2001, AJ, 121, 327
Bressan A., Marigo P., Girardi L., et al., 2012, MNRAS, 427, 127 Clem J. L., Landolt A. U., 2013, AJ, 146, 88
Corsaro E., Stello D., Huber D., et al., 2012, ApJ, 757, 190 Everett M. E., Howell S. B., Kinemuchi K., 2012, PASP, 124, 316 Geller A. M., Hurley J. R., Mathieu R. D., 2013, AJ, 145, 8 Geller A. M., Mathieu R. D., 2011, Nature, 478, 356 Geller A. M., Mathieu R. D., 2012, AJ, 144, 54 Geller A. M., Mathieu R. D., Braden E. K., et al., 2010, AJ, 139, 1383
- Geller A. M., Mathieu R. D., Harris H. C., McClure R. D., 2008, AJ, 135, 2264
- Gosnell N., DiPompeo M. A., Braden E. K., et al., 2007, in BAAS, Vol. 39, American Astronomical Society Meeting Abstracts, p. 839
- Gosnell N. M., Mathieu R. D., Geller A. M., et al., 2014, ApJL, 783, L8
- Gosnell N. M., Pooley D., Geller A. M., et al., 2012, ApJ, 745, 57 Hole K. T., Geller A. M., Mathieu R. D., et al., 2009, AJ, 138, 159 Hurley J. R., Tout C. A., Pols O. R., 2002, MNRAS, 329, 897 Jeffries, Jr. M. W., Sandquist E. L., Mathieu R. D., et al., 2013, AJ, 146, 58
- Kalirai J. S., Richer H. B., Fahlman G. G., et al., 2001, AJ, 122, 266
- Kozhurina-Platais V., Girard T. M., Platais I., et al., 1995, AJ, 109, 672
- Kroupa P., 1995, MNRAS, 277, 1507
- Mathieu R. D., 1983, ApJL, 267, L97
- Mathieu R. D., 2000, in ASP Conf. Ser., Vol. 198, Stellar Clusters and Associations: Convection, Rotation, and Dynamos,
Pallavicini R., Micela G., Sciortino S., eds., p. 517
- Mathieu R. D., Geller A. M., 2009, Nature, 462, 1032
- Mathieu R. D., van den Berg M., Torres G., et al., 2003, AJ, 125, 246
- Meibom S., Grundahl F., Clausen J. V., et al., 2009, AJ, 137, 5086
- Meibom S., Mathieu R. D., 2005, ApJ, 620, 970
- Miglio A., Brogaard K., Stello D., et al., 2012, MNRAS, 419, 2077 Platais I., Gosnell N. M., Meibom S., et al., 2013, AJ, 146, 43 Raghavan D., McAlister H. A., Henry T. J., et al., 2010, ApJS,
- 190, 1
- Rosvick J. M., Vandenberg D. A., 1998, AJ, 115, 1516 Sanders W. L., 1972, A&A, 19, 155
- Sandquist E. L., Latham D. W., Shetrone M. D., Milone A. A. E., 2003, AJ, 125, 810
- Sandquist E. L., Mathieu R. D., Brogaard K., et al., 2013, ApJ, 762, 58
- Sarrazine A. R., Deliyannis C. P., Sarajedini A., Platais I., 2003, in BASS, Vol. 35, American Astronomical Society Meeting Abstracts, p. 1230
- Stello D., Basu S., Bruntt H., et al., 2010, ApJL, 713, L182
Stello D., Meibom S., Gilliland R. L., et al., 2011, ApJ, 739, 13
- Street R. A., Horne K., Lister T. A., et al., 2002, MNRAS, 330, 737
- Street R. A., Horne K., Lister T. A., et al., 2005, MNRAS, 358, 795
- Yang S.-C., Sarajedini A., Deliyannis C. P., et al., 2013, ApJ, 762, 3
- Zahn J.-P., Bouchet L., 1989, A&A, 223, 112
- Zucker S., Mazeh T., 1994, ApJ, 420, 806 | 10.1088/0004-6256/148/2/38 | [
"Katelyn E. Milliman",
"Robert D. Mathieu",
"Aaron M. Geller",
"Natalie M. Gosnell",
"Søren Meibom",
"Imants Platais"
] | 2014-08-01T17:13:13+00:00 | 2014-08-01T17:13:13+00:00 | [
"astro-ph.SR"
] | WIYN Open Cluster Study. LX. Spectroscopic Binary Orbits in NGC 6819 | We present the current state of the WOCS radial-velocity (RV) survey for the
rich open cluster NGC 6819 (2.5 Gyr) including 93 spectroscopic binary orbits
with periods ranging from 1.5 to 8,000 days. These results are the product of
our ongoing RV survey of NGC 6819 using the Hydra Multi-Object Spectrograph on
the WIYN 3.5 m telescope. We also include a detailed analysis of multiple prior
sets of optical photometry for NGC 6819. Within a 1-degree field of view, our
stellar sample includes the giant branch, the red clump, and blue straggler
candidates, and extends to almost 2 mag below the main-sequence (MS) turnoff.
For each star observed in our survey we present all RV measurements, the
average RV and velocity variability information. Additionally, we discuss
notable binaries from our sample, including eclipsing binaries (WOCS 23009,
WOCS 24009, and WOCS 40007), stars noted in Kepler asteroseismology studies
(WOCS 4008, WOCS 7009, and WOCS 8007), and potential descendants of past blue
stragglers (WOCS 1006 and WOCS 6002). We find the incompleteness-corrected
binary fraction for all MS binaries with periods less than 10,000 days to be
22% +/- 3% and a tidal circularization period of 6.2 +/- 1.1 days for NGC 6819. |
1408.0242v1 | ## Symmetry Reduction of the Two-Dimensional Ricci Flow Equation
Mehdi Nadjafikhah
∗ Mehdi Jafari †
August 4, 2014
## Abstract
This paper is devoted to obtain the one-dimensional group invariant solutions of the two-dimensional Ricci flow ((2 D ) Rf) equation. By classifying the orbits of the adjoint representation, the optimal system of one-dimensional subalgebras of the ((2 D ) Rf) equation is obtained. For each class, we will find the reduced equation by method of similarity reduction. By solving these reduced equations we will obtain new sets of group invariant solutions for the ((2 D ) Rf) equation.
Keywords: Lie symmetry group, two-dimensional Ricci flow equation, optimal system, group invariant solution
2010 Mathematics Subject Classification : 70G65, 34C14, 53C44
## 1 Introduction
The Ricci flow was introduced by Hamilton in his seminal paper, 'Three-manifolds with positive Ricci curvature' in 1982 [7]. Since then, Ricci flow has been a very useful tool for studying the special geometries which a manifold admits. Ricci flow is an evolution equation for a Riemannian metric which sometimes can be used in order to deform an arbitrary metric into a suitable metric that can specify the topology of the underlying manifold. If ( M,g t ( )) be a smooth Riemannian manifold, Ricci flow is defined by the equation
$$\frac { \partial } { \partial t } g ( t ) = - 2 R i c,$$
where Ric denotes the Ricci tensor of the metric g . By using the concept of Ricci flow, Grisha Perelman completely proved the Poincar´ e conjecture around 2003 [12-14]. The Ricci flow also is used as an approximation to the renormalization group flow for the twodimensional nonlinear σ -model, in quantum field theory. see [6] and references therein.
∗ Corresponding author. School of Mathematics, Iran University of Science and Technology, Narmak, Tehran 1684613114, Iran. E-mail: m [email protected]
† Department of Complementary Education, Payame Noor University, PO BOX 19395-3697, Tehran, Iran. E-mail: [email protected]
In this paper we want to obtain new solutions of the ((2 D ) Rf) equation by method of Lie symmetry group. As it is well known, the Lie symmetry group method has an important role in the analysis of differential equations. The theory of Lie symmetry groups of differential equations was developed by Sophus Lie [8]. By this method we can reduce the order of ODEs and investigate the invariant solutions. Also we can construct new solutions from known ones (for more details about the applications of Lie symmetries, see [10, 1, 3]). Lei's method led to an algorithmic approach to find special solution of differential equation by its symmetry group. These solutions are called group invariant solutions and obtain by solving the reduced system of differential equation having fewer independent variables than the original system. Bluman and Cole generalized the Lie's symmetry method for finding the group-invariant solutions [2]. In this paper we apply this method to obtain the invariant solutions of ((2 D ) Rf) equation and classify them.
This paper is organized as follows. In section 2, by using the mechanical model of Ricci flow, Lie symmetries of ((2 D ) Rf) equation will be state and some results yield from the structure of the Lie algebra of the Lie symmetry group. In section 3, we will construct an optimal system of one-dimensional subalgebras of the ((2 D ) Rf) equation which is useful for classifying of group invariant solutions. In section 4, the reduced equation for each element of optimal system is obtained. In section 5, we will solve the reduced equations by method of Lie symmetry group and obtain the group invariant solutions of ((2 D ) Rf) equation.
## 2 Lie Symmetries of ( (2 D ) Rf) Equation
As we know, transformations which map solutions of a differential equation to other solutions are called symmetries of the equation. In [10] Olver has introduced the procedure of finding Lie symmetry group of a differential equation. Cimpoiaus and Constantinescu have introduced the mechanical model for the Ricci flow as follows:
$$u ^ { 2 } u _ { t } + u _ { y } u _ { x } - u u _ { x y } = 0$$
and obtained the Lie symmetry group of this equation [5]. They proved that this equation admits a 6-parameter Lie group, G , with the following infinitesimal generators for its Lie algebra, g .
$$\begin{array} { c c c } X _ { 1 } = \partial _ { x }, & X _ { 2 } = \partial _ { y }, & X _ { 3 } = \partial _ { t }, \\ X _ { 4 } = t \partial _ { t } + u \partial _ { u }, & X _ { 5 } = x \partial _ { x } - u \partial _ { u }, & X _ { 6 } = y \partial _ { y } - u \partial _ { u }. \end{array}$$
The commutator table of Lie algebra for g is given below, where the entry in the i th row and j th column is [ X , X i j ] = X X i j -X X j i , i, j = 1 , ..., 6.
Exponentiating the infinitesimal symmetries (3) we obtain the one-parameter groups g k ( s ) generated by X k , k = 1 , ..., 6:
↦
↦
$$\cdots \cdots \cdots \cdots _ { J } \cdots _ { \kappa }, \cdots \cdots \\ g _ { 1 } ( s ) & \colon \, ( x, y, t, u ) \, \longmapsto \, ( x + s, y, t, u ), \\ g _ { 2 } ( s ) & \colon \, ( x, y, t, u ) \, \longmapsto \, ( x, y + s, t, u ), \\ g _ { 3 } ( s ) & \colon \, ( x, y, t, u ) \, \longmapsto \, ( x, y, t + s, u ), \\ g _ { 4 } ( s ) & \colon \, ( x, y, t, u ) \, \longmapsto \, ( x, y, t e ^ { s }, u e ^ { s } ), \\ g _ { 5 } ( s ) & \colon \, ( x, y, t, u ) \, \longmapsto \, ( x e ^ { s }, y, t, u e ^ { - s } ), \\ g _ { 6 } ( s ) & \colon \, ( x, y, t, u ) \, \longmapsto \, ( x, y e ^ { s }, t, u e ^ { - s } ),$$
↦
↦
↦
↦
Table 1: The commutator table of g .
| [ , ] | X 1 | X 2 | X 3 | X 4 | X 5 | X 6 |
|---------|-------|-------|-------|-------|-------|-------|
| X 1 | 0 | 0 | 0 | 0 | X 1 | 0 |
| X 2 | 0 | 0 | 0 | 0 | 0 | X 2 |
| X 3 | 0 | 0 | 0 | X 3 | 0 | 0 |
| X 4 | 0 | 0 | - X 3 | 0 | 0 | 0 |
| X 5 | - X 1 | 0 | 0 | 0 | 0 | 0 |
| X 6 | 0 | - X 2 | 0 | 0 | 0 | 0 |
Consequently, we can state the following theorem:
Theorem 2.1. If f = f ( x, y, t ) is a solution of (2), so are functions
$$g _ { 1 } ( s ). f & = f ( x - s, y, t ), \quad g _ { 4 } ( s ). f & = f ( x, y, t e ^ { - s } ) e ^ { s }, \\ g _ { 2 } ( s ). f & = f ( x, y - s, t ), \quad g _ { 5 } ( s ). f & = f ( x e ^ { - s }, y, t ) e ^ { - s }, \\ g _ { 3 } ( s ). f & = f ( x, y, t - s ), \quad g _ { 6 } ( s ). f & = f ( x, y e ^ { - s }, t ) e ^ { - s }.$$
## 3 One-dimensional optimal system of subalgebras for the (( 2 D) Rf) equation
In this section, we obtain the one-dimensional optimal system of ((2D) Rf) equation by using symmetry group. Since every linear combination of infinitesimal symmetries is an infinitesimal symmetry, there is an infinite number of one-dimensional subgroups for G . Therefore, it's important to determine which subgroups give different types of solutions. For this, we must find invariant solutions which can not be transformed to each other by symmetry transformations in the full symmetry group. This led to concept of an optimal system of subalgebra. For one-dimensional subalgebras, this classification problem is the same as the problem of classifying the orbits of the adjoint representation [10]. Optimal set of subalgebras is obtaining from taking only one representative from each class of equivalent subalgebras. The problem of classifying the orbits is solved by taking a general element in the lie algebra and simplify it as much as possible by imposing various adjoint transformation on it [11, 9]. Adjoint representation of each X i , i = 1 , ..., 6 is defined as follow:
$$\text{of} \, \text{allow} \, \colon \, & \text{Ad} ( \exp ( s. X _ { i } ). X _ { j } ) = X _ { j } - s. [ X _ { i }, X _ { j } ] + \frac { s ^ { 2 } } { 2 }. [ X _ { i }, [ X _ { i }, X _ { j } ] ] - \cdots, \quad & \quad & ( 6 ) \\ \,. \quad \, & \quad \cdots \, \quad \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \end{$$
where s is a parameter and [ X , X i j ] is the commutator of the Lie algebra for i, j = 1 , · · · , 6 ([10],page 199). Taking into account the table of commutator, we can compute all the adjoint representations corresponding to the Lie group of the ((2D) Rf) equation. They are presented in Table 2. Not that, the ( i, j ) entry indicate Ad(exp( s.X i ) .X j ).
Now we can state the following theorem:
Theorem 3.1. A one-dimensional optimal system for Lie algebra of (( 2 D) Rf) equation is given by
$$\begin{smallmatrix} 1 \right ) X _ { 4 } + a X _ { 5 } + b X _ { 6 }, & & 4 \right ) \varepsilon X _ { 1 } + \varepsilon ^ { \prime } X _ { 2 } + X _ { 4 }, & & 7 \right ) \varepsilon X _ { 1 } + \varepsilon ^ { \prime } X _ { 3 } + X _ { 6 }, \\ 2 \right ) \varepsilon X _ { 2 } + X _ { 4 } + a X _ { 5 }, & & 5 \right ) \varepsilon X _ { 3 } + X _ { 5 } + a X _ { 6 }, & & 8 \right ) \varepsilon X _ { 1 } + c X _ { 2 } + \varepsilon ^ { \prime } X _ { 3 }, \\ 3 \right ) \varepsilon X _ { 1 } + X _ { 4 } + a X _ { 6 }, & & 6 \right ) \varepsilon X _ { 2 } + \varepsilon ^ { \prime } X _ { 3 } + X _ { 5 }, \end{smallmatrix}$$
1
Table 2: The adjoint representation table of the infinitesimal generators X i .
| Ad | X 1 | X 2 | X 3 | X 4 | X 5 | X 6 |
|------|---------|---------|---------|------------|------------|------------|
| X 1 | X 1 | X 2 | X 3 | X 4 | X 5 - sX 1 | X 6 |
| X 2 | X 1 | X 2 | X 3 | X 4 | X 5 | X 6 - sX 2 |
| X 3 | X 1 | X 2 | X 3 | X 4 - sX 3 | X 5 | X 6 |
| X 4 | X 1 | X 2 | e s X 3 | X 4 | X 5 | X 6 |
| X 5 | e s X 1 | X 2 | X 3 | X 4 | X 5 | X 6 |
| X 6 | X 1 | e s X 2 | X 3 | X 4 | X 5 | X 6 |
/negationslash where ε and ε ′ are ± 1 or zero. Also a, b, c ∈ R and a = 0 , b = 0 .
/negationslash
↦
Proof. Let F s i : g → g be the adjoint transformation defined by X → Ad(exp( sX i ) .X ), for i = 1 , · · · , 6. The matrix of F s i , i = 1 , · · · , 6, with respect to basis { X , 1 · · · , X 6 } is:
M
M
s
s
4
=
1
0
0
0
-
s
0
1
0
0
0
0
0
1
0
0
0
0
0
=
0
0
0
0
0
1
0
0
0
1
0
0
0
0
0
1
0
0
0
1
0
0
0
0
0
e
s
0
0
0
1
0
0
0
0
0
1
0
0
0
1
0
0
0
0
0
0
0
0
0
1
0
0
0
0
,
M
s
2
M
5
=
=
0
0
0
0
0
e
s
0
0
0
0
1
0
0
0
0
0
0
1
0
0
0
0
0
0
0
0
1
0
0
0
0
1
0
0
0
0
0
0
1
0
0
0
0
0
0
1
0
-
s
0
0
0
1
0
0
0
0
0
1
=
=
0
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
0
e
s
0
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
-
s
1
0
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
0
1
respectively. If X = ∑ 6 i =1 a X i i , then we have
↦
$$F _ { 6 } ^ { s _ { 6 } } \circ F _ { 5 } ^ { s _ { 5 } } \circ \cdots \circ F _ { 1 } ^ { s _ { 1 } } \quad \colon \, & \mapsto \quad e ^ { s _ { 5 } } a _ { 1 } X _ { 1 } + e ^ { s _ { 6 } } a _ { 2 } X _ { 2 } + e ^ { s _ { 4 } } a _ { 3 } X _ { 3 } + ( a _ { 4 } - e ^ { s _ { 4 } } s _ { 3 } a _ { 3 } ) X _ { 4 } + \\ & ( a _ { 5 } - e ^ { s _ { 5 } } s _ { 1 } a _ { 1 } ) X _ { 5 } + ( a _ { 6 } - e ^ { s _ { 6 } } s _ { 2 } a _ { 2 } ) X _ { 6 }.$$
Now, we try to vanish the coefficients of X by acting the adjoint representations M s i i on X , by choosing suitable parameters s i in each step. Therefor we can simplify X as follows:
/negationslash
If a 4 = 0, a 5 = 0 and a 6 = 0, we can make the coefficients of X 3 , X 1 and X 2 vanish by F S 3 3 , F S 1 1 and F S 2 2 . By setting s 3 = -a 3 a 4 , s 1 = -a 1 a 5 and s 2 = -a 2 a 6 respectively. Scaling X if necessary, we can assume that a 4 = 1. So X is reduced to the case (1).
/negationslash
/negationslash
/negationslash
If a 4 = 0, a 5 = 0 and a 6 = 0, we can make the coefficients of X 3 and X 1 vanish by F S 3 3 and F S 1 1 ; By setting s 3 = -a 3 a 4 and s 1 = -a 1 a 5 , respectively. Also by setting s 6 = ln | a 2 | in F S 6 6 , we can make the coefficient of X 2 vanish or ± 1. Scaling X if necessary, we can assume that a 4 = 1. So X is reduced to the case (2).
/negationslash
/negationslash
/negationslash
If a 4 = 0, a 6 = 0 and a 5 = 0, we can make the coefficients of X 3 and X 2 vanish by F S 3 3 and F S 2 2 ; By setting s 3 = -a 3 a 4 and s 2 = -a 2 a 6 , respectively. Also by setting s 5 = ln | a 1 | in F S 5 5 , we can make the coefficient of X 1 vanish or ± 1. Scaling X if necessary, we can assume that a 4 = 1. So X is reduced to the case (3).
,
1
0
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
0
1
0
.
,
,
M
M
s
3
s
6
s
,
/negationslash
If a 4 = 0 and a 5 = a 6 = 0, we can make the coefficient of X 3 vanish by F S 3 3 ; By setting s 3 = -a 3 a 4 . Also by setting s 5 = ln | a 1 | and s 6 = ln | a 2 | in F S 5 5 and F S 6 6 , we can make the coefficients of X 1 and X 2 vanish or ± 1, respectively. Scaling X if necessary, we can assume that a 4 = 1. So X is reduced to the case (4).
/negationslash
If a 5 = 0, a 6 = 0 and a 4 = 0, we can make the coefficients of X 1 and X 2 vanish by F S 1 1 and F S 2 2 ; By setting s 1 = -a 1 a 5 and s 2 = -a 2 a 6 , respectively. Also by setting s 4 = ln | a 3 | in F S 4 4 , we can make the coefficient of X 1 vanish or ± 1. Scaling X if necessary, we can assume that a 5 = 1. So X is reduced to the case (5).
/negationslash
/negationslash
If a 5 = 0 and a 4 = a 6 = 0, we can make the coefficient of X 1 vanish by F S 1 1 ; By setting s 1 = -a 1 a 5 . Also by setting s 6 = ln | a 2 | and s 4 = ln | a 3 | in F S 6 6 and F S 4 4 , we can make the coefficients of X 2 and X 3 vanish or ± 1, respectively. Scaling X if necessary, we can assume that a 5 = 1. So X is reduced to the case (6).
/negationslash
If a 4 = a 5 = 0 and a 6 = 0, we can make the coefficient of X 2 vanish by F S 2 2 ; By setting s 2 = -a 2 a 6 . Also by setting s 5 = ln | a 1 | and s 4 = ln | a 3 | in F S 5 5 and F S 4 4 , we can make the coefficients of X 1 and X 3 vanish or ± 1, respectively. Scaling X if necessary, we can assume that a 6 = 1. So X is reduced to the case (7).
If a 4 = a 5 = a 6 = 0, we can make the coefficients of X 1 and X 3 vanish or ± 1 by F S 5 5 and F S 4 4 ; By setting s 5 = ln | a 1 | and s 4 = ln | a 3 | , respectively. So X is reduced to the case (8).
## 4 Similarity Reduction of (( 2 D) Rf) Equation
In this section, the two-dimensional Ricci flow equation will be reduced by expressing it in the new coordinates. The ((2D) Rf) equation is expressed in the coordinates ( x, y, t, u ), we must search for this equation's form in the suitable coordinates for reducing it. These new coordinates will be obtained by looking for independent invariants ( z, w, f ) corresponding to the generators of the symmetry group. Hence, By using the new coordinates and applying the chain rule, we obtain the reduced equation. We express this procedure for one of the infinitesimal generators in the optimal system (7) and list the result for some other cases.
For example, consider the case (7) in theorem (3 1) when . ε = 1 and ε ′ = 0, therefor we have X := X 1 + X 6 . For determining independent invariants I , we ought to solve the PDEs X I ( ) = 0, that is
$$( X _ { 1 } + X _ { 6 } ) I = ( \partial _ { x } + y \partial _ { y } - u \partial _ { u } ) I = \frac { \partial I } { \partial x } + y \frac { \partial I } { \partial y } + 0 \frac { \partial I } { \partial t } - u \frac { \partial I } { \partial u } = 0.$$
For solving this PDE, the following associated characteristic ODE must be solved:
$$\frac { d x } { 1 } = \frac { d y } { y } = \frac { d t } { 0 } = \frac { d u } { - u }.$$
Hence, three functionally independent invariants z = ye -x , w = t and f = uy are obtained. If we treat f as a function of z and w , we can compute formulae for the derivatives of u
with respect to x , y and t in terms of z , w f , and the derivatives of f with respect to z and w . By using the chain rule and the fact that u = f ( z, w y ) -1 we have
$$\begin{array} { c } u _ { t } = ( f _ { z } z _ { t } + f _ { w } w _ { t } ) y ^ { - 1 } = f _ { w } y ^ { - 1 }, \quad & u _ { x } = - f _ { z } e ^ { - x }, \\ u _ { y } = f _ { z } e ^ { - x } y ^ { - 1 } - f y ^ { - 2 }, \quad & u _ { x y } = - e ^ { - 2 x } f _ { z z }. \end{array}$$
After substituting the above relations into the equation (2), we obtain:
$$u ^ { 2 } u _ { t } + u _ { y } u _ { x } - u u _ { x y } = y ^ { - 3 } ( f ^ { 2 } f _ { w } - f _ { z } ^ { 2 } z ^ { 2 } + f f _ { z } z + f f _ { z z } z ^ { 2 } ) = 0.$$
So the reduced equation is
$$f ^ { 2 } f _ { w } - z ^ { 2 } f _ { z } ^ { 2 } + z f f _ { z } + z ^ { 2 } f f _ { z z } = 0.$$
This equation has two independent variables z and w and one dependent variable f . In a similar way, we can compute all of the similarity reduction equations corresponding to the infinitesimal symmetries that mentioned in theorem (3 1). . some of them are listed in Table 3.
Table 3: Lie invariants, similarity solutions and reduced equations.
| i | h i | { z i , w i , v i } | u i | Similarity reduced equations |
|-----|------------------|--------------------------------|------------------|--------------------------------------------------------------|
| 1 | X 1 + X 6 | { ye - x , t, uy } | f ( z,w ) y | f 2 f w - z 2 f 2 z + zff z + z 2 ff zz = 0 |
| 2 | X 2 + X 4 | { x, te - y , ue - y } | f ( z, w ) e y | f 2 f w - wf z f w + wff zw = 0 |
| 3 | X 3 + X 5 + aX 6 | { y x a , ln e t x , ux a +1 } | f ( z,w ) x a +1 | f w ( f 2 - f z ) - azf 2 z + f ( af z + azf zz + f zw ) = 0 |
| 4 | X 5 + aX 6 | { y x a , t, ux a +1 } | f ( z,w ) x a +1 | f 2 f w + a ( ff z - zf 2 z + zff zz ) = 0 |
| 5 | X 2 + X 3 + X 5 | { ln e y x , ln e t x , ux } | f ( z,w ) x | f 2 f w - f 2 z - f w f z + ff zz + ff zw = 0 |
| 6 | X 2 + X 5 | { ln e y x , t, ux } | f ( z,w ) x | f 2 f w - f 2 z + ff zz = 0 |
| 7 | - X 3 + X 6 | { x, t +ln y, uy } | f ( z,w ) y | f 2 f w + f z f w - ff zw = 0 |
| 8 | X 1 + aX 2 | { y - ax, t, u } | f ( z, w ) | f 2 f w - af 2 z + aff zz = 0 |
| 9 | aX 2 + X 3 | { x, ta - y a , u } | f ( z, w ) | af 2 f w - f z f w + ff zw = 0 |
## 5 Group Invariant Solutions of (( D) Rf) Equation 2
In this section we reduce the equations obtained in last section to ODEs and solve them. For example, the equation (12) admits a 4-parameters family of Lie operators with following infinitesimal generators
$$\begin{smallmatrix} V _ { 1 } = \frac { 1 } { 2 } z \ln z \partial _ { z } + w \partial _ { w }, & & V _ { 3 } = - \frac { 1 } { 2 } z \ln z \partial _ { z } + f \partial _ { f }, \\ V _ { 2 } = \partial _ { w }, & & V _ { 4 } = z \partial _ { z }. \end{smallmatrix}$$
The invariants associated to the infinitesimal generator V 2 , are s = z and g = f . By substituting these invariants into the equation (12) and using chain rule, the reduced equation is obtained as follows:
$$s g ^ { \prime 2 } - g g ^ { \prime } - s g g ^ { \prime \prime } = 0$$
the solution of this equation is g s ( ) = c s 2 c 1 = c 2 z c 1 . therefore we have f ( z ) = c 2 z c 1 = c 2 ( ye -x c ) 1 = c 2 y c 1 e -c 1 x . So u = fy -1 = c 2 y c 1 -1 e -c 1 x is a solution of ((2D) Rf) equation.
By similar arguments, we can obtain other invariant solutions of the equation (12). Also by reducing other equations in Table 3, we can find other solutions of ((2D) Rf) equation. Some of the similarity reduced equations and their invariant solutions are listed in Table 4 and Table 5 respectively.
## Conclusion
In this paper, by using the adjoint representation of the symmetry group on its Lie algebra, we have constructed an optimal system of one-dimensional subalgebras of the two-dimensional Ricci flow equation. Moreover, we have obtained the similarity reduced equations for each element of optimal system as well as some group invariant solutions of two-dimensional Ricci flow equation.
## References
- [1] Bluman, G. W. and Cole, J. D., Similarity Methods for Differential equations, Appl. Math. Sci., No. 13, Springer-Verlag, New York, 1974.
- [2] Bluman, G. W. and Cole, J. D., The general similarity solutions of the heat equation, Journal of Mathematics and Mechanics, 18 , 1969, 1025-1042.
- [3] Bluman, G. W. and Kumei, S., Symmetries and Differential Equations, Springer, New York, 1989.
- [4] Chow, B. and Knopf, D., The Ricci flow: an introduction , volume 110 of Mathematical Surveys and Monographs. AMS, Providence, RI, 2004.
- [5] Cimpoiasu, R. and Constantinescu, R., Symmetries and invariants for the 2D-Ricci flow model, Journal of Nonlinear Mathematical Physics, Volume13, Number 2, 2006, 285-292.
- [6] Gawedzki, K., Lectures on conformal field theory. Quantum fields and strings: a course for mathematicians (Princeton, 1996-97), 727-805.
- [7] Hamilton, R.S., Three-manifolds with positive Ricci curvature, J. Diff. Geom, 17 (2): 1982, 255-306.
- [8] Lie, S., On integration of a class of linear partial differential equations by means of definite integrals, Arch. for Math., 6 , (328) 1881. translation by Ibragimov, N. H.
- [9] Nadjafikhah, M., Lie symmetries of inviscid Burgers' equation, Advances in Applied Clifford Algebras, 19 , no. 1, 2009, 101-112.
- [10] Olver, P.J., Applications of Lie Groups to Differential Equations , Springer, New York, 1986.
- [11] Ovsiannikov, L.V., Group Analysis of Differential Equations , Academic Press, New York, 1982.
- [12] Perelman, G., Finite extinction time for the solutions to the Ricci flow on certain Threemanifolds, math. DG/0307245, 2003.
- [13] Perelman, G., Ricci flow with surgery on three-manifolds, math. DG/0303109, 2003.
- [14] Perelman, G., The entropy formula for the Ricci flow and its geometric applications, math. DG/0211159, 2002.
Table 4: ODEs that obtain from the reduced equations of table 3.
| i | Simmetry group generators | Optimal system | Invariants { s, g } | Reduced equation |
|-----|------------------------------------------------------------------------------|-------------------------------------------------------------|--------------------------------------------------------|-----------------------------------------------------------------------------------------------------------------------------|
| 1 | V 1 = 1 2 z ln z∂ z + w∂ w V 2 = ∂ w V 3 = - 1 2 z ln z∂ z + f∂ f V 4 = z∂ z | A 1 1 : V 2 A 2 1 : V 3 A 3 1 : V 1 + V 3 A 4 1 : V 2 + V 4 | { z, f } { w, f (ln z ) 2 } { z, f w } { w - ln z, f } | sg ′ 2 - gg ′ - sgg ′′ = 0 g 2 ( g ′ +2) = 0 g 3 - s 2 g ′ 2 + sgg ′ + s 2 gg ′′ = 0 g 2 g ′ - g ′ 2 + gg ′′ = 0 |
| 2 | V 1 = w∂ w + f∂ f V 2 = z∂ z - f∂ f V 3 = ∂ z | A 1 2 : V 2 A 2 2 : V 3 | { w, fz } { w, f } | g 2 g ′ = 0 g 2 g ′ = 0 |
| 3 | V 1 = z ln z∂ z + w∂ w - f (1 +ln z ) ∂ f V 2 = ∂ w V 3 = z∂ z - f∂ f | A 1 3 : V 1 A 2 3 : V 2 A 3 3 : V 3 | { w ln z , fz ln z } { z, f } { w, fz } | gg ′ ( g +2 as - 1) + ag 2 + s ( as - 1)( gg ′′ - g ′ 2 ) = 0 sg ′ 2 - gg ′ - sgg ′′ = 0 g 2 g ′ = 0 |
| 4 | V 1 = z∂ z + w∂ w V 2 = ∂ w V 3 = - z∂ z + f∂ f V 4 = z (2 - ln z ) ∂ z + | A 1 4 : V 4 A 2 4 : V 1 + V 3 | { w, fz (4 - ln z (4 - ln z )) } { z, f w } | g 2 ( g ′ +2 a ) = 0 g 3 - a ( sg ′ 2 + gg ′ + sgg ′′ ) = 0 |
| 5 | V 1 = z∂ z + w∂ w - f∂ f V 2 = ∂ w V 3 = ∂ z | A 1 5 : V 1 A 2 5 : V 2 A 3 5 : V 2 - V 3 | { w z , fz } { z, f } { z + w, f } | gg ′ (1 - g - 2 s ) - g 2 + s ( s - 1)( g ′ 2 - gg ′′ ) = 0 - g ′ 2 + gg ′′ = 0 g 2 g ′ - 2 g ′ 2 +2 sgg ′′ = 0 |
| 6 | V 1 = z 2 ∂ z + w∂ w V 2 = ∂ w V 3 = - z 2 ∂ z + f∂ f V 4 = ∂ z | A 1 6 : V 3 A 2 6 : V 1 + V 3 A 3 6 : V 2 + V 4 | { w, fz 2 } { z, f w } { w - z, f } | g ′ +2 = 0 g 3 - g ′ 2 + gg ′′ = 0 g 2 g ′ - g ′ 2 + gg ′′ = 0 |
| 7 | V 1 = w∂ w V 2 = ∂ w V 3 = z∂ z - f∂ f V 4 = ∂ z | A 1 7 : V 4 A 2 7 : V 2 + V 3 | { w, f } { w - ln z, zf } | g ′ = 0 g 2 g ′ - g ′ 2 + gg ′′ = 0 |
| 8 | V 1 = z 2 ∂ z + w∂ w V 2 = ∂ w V 3 = - z 2 ∂ z + f∂ f V 4 = ∂ z | A 1 8 : V 2 A 2 8 : V 3 A 3 8 : V 2 + V 4 A 4 8 : V 1 - V 3 | { z, f } { w, fz 2 } { w - z, f } { w z , fz } | g ′ 2 - gg ′′ = 0 g 2 ( g ′ +2 a ) = 0 g 2 g ′ - ag ′ 2 + agg ′′ = 0 g 2 g ′ - as 2 g ′ 2 +2 asgg ′ + ag 2 + as 2 gg ′′ = 0 |
| 9 | V 1 = w∂ w V 3 = z∂ z - f∂ f V 2 = ∂ w V 4 = ∂ z | A 1 9 : V 2 + V 4 A 2 9 : V 2 + V 3 | { w - z, f } { w - ln z, zf } | ag 2 g ′ + g ′ 2 - gg ′′ = 0 ag 2 g ′ + g ′ 2 - gg ′′ = 0 |
Table 5: Group invariant solutions of the ((2D) Rf) equation.
| A i j | Invariant solution | Solution after substitiuting |
|---------|--------------------------------------------------------------------------------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------|
| A 1 1 | c 2 s c 1 | c 2 y c 1 - 1 e - c 1 x |
| A 2 1 | - 2 s + c 1 | - 2 t + c 1 y ( ln y - x ) 2 |
| A 3 1 | 1 2 c 2 1 (1 - tanh( ln s - c 2 2 c 1 ) 2 ) | t 2 c 2 1 y (1 - tanh( ln y - x - c 2 2 c 1 ) 2 ) |
| A 4 1 | c 1 e c 1 ( s + c 2 ) - 1+ e c 1 ( s + c 2 ) | c 1 e c 1 ( x + t - ln y + c 2 ) - y + ye c 1 ( x + t - ln y + c 2 ) |
| A 1 2 | c 1 | c 1 x e y |
| A 2 2 | c 1 | c 1 e y |
| A 1 3 | c 1 (1+ c 1 ) s c 1 - s c 1 (1+ c 1 - as )+ ac 1 c 2 (1+ c 1 )( as - 1) c 1 +1 | c 1 (1+ c 1 )( t - ln x ln y - ln x ) c 1 xy ln y x ( - ( t - ln x ln ln ) c 1 (1+ c 1 - at - a ln x ln ln )+ ac 1 c 2 (1+ c 1 )( at - a ln x ln ln - 1) c 1 +1 |
| A 2 3 | c 2 s c 1 | c 2 x 2 ( y x ) c 1 |
| A 3 3 | c 1 | c 1 yx |
| A 1 4 | - 2 as + c 1 | - 2 at + c 1 xy (4 - 4 ln y x +( ln y x ) 2 ) |
| A 2 4 | 1 2 c 1 s (1 - tanh( ln s + c 2 2 √ ac 1 ) 2 ) | t 2 c 1 xy (1 - tanh( ln y - ln x + c 2 2 √ ac 1 ) 2 ) |
| A 1 5 | c 1 (1+ c 1 ) s c 1 - s c 1 (1+ c 1 - s )+ c 1 c 2 (1+ c 1 )( s - 1) c 1 +1 | c 1 (1+ c 1 )( y - ln x ) x ( y - ln x )( - ( t - ln x y - ln x ) c 1 (1+ c 1 - t - ln x y - ln x )+ c 1 c 2 (1+ c 1 )( t - y y - ln x ) c 1 +1 ) |
| A 2 5 | c 2 e c 1 s | c 2 x e c 1 ( y - ln x ) |
| A 3 5 | 2 c 1 e c 1 ( s + c 2 ) - 1+ e c 1 ( s + c 2 ) | 2 c 1 e c 1 ( t + y - 2 ln x + c 2 ) - x + xe c 1 ( t + y - 2 ln x + c 2 ) |
| A 1 6 | - 2 s + c 1 | - 2 t + c 1 x ( y - ln x ) 2 |
| A 2 6 | 1 2 c 2 1 (1 - tanh( s + c 2 2 c 1 ) 2 ) | t 2 c 2 1 x (1 - tanh( y - ln x + c 2 2 c 1 ) 2 ) |
| A 3 6 | c 1 e c 1 ( s + c 2 ) - 1+ e c 1 ( s + c 2 ) | c 1 e c 1 ( t - y + ln x + c 2 ) - x + xe c 1 ( t - y + ln x + c 2 ) |
| A 1 7 | c 1 | c 1 y |
| A 2 7 | c 1 e c 1 ( s + c 2 ) - 1+ e c 1 ( s + c 2 ) | c 1 e c 1 ( ln y - ln x + t + c 2 ) xy ( - 1+ e c 1 ( ln y - ln x + t + c 2 ) ) |
| A 1 8 | c 2 e c 1 s | c 2 e c 1 ( y - ax ) |
| A 2 8 | - 2 as + c 1 | - 2 at + c 1 ( y - ax ) 2 |
| A 3 8 | c 1 ae c 1 ( s + c 2 ) - 1+ e c 1 ( s + c 2 ) | c 1 ae c 1 ( ax - y + t + c 2 ) - 1+ e c 1 ( ax - y + t + c 2 ) |
| A 4 8 | - c 2 1 ae c 1 s c 1 | - c 2 1 ae c 1 ( y - ax ) t c ( y - ax ) |
| A 1 9 | c 1 e c 1 ( s + c 2 ) 1 - ae c 1 ( s + c 2 ) | - - - c 1 e c 1 ( t - x - y a + c 2 ) 1 - ae c 1 ( t - x - y a + c 2 ) |
| A 2 | c 1 e c 1 ( s + c 2 ) c ( s + c ) | c 1 e c 1 ( t - ln x - y a + c 2 ) y 9 | | 10.1155/2013/373701 | [
"Mehdi Nadjafikhah",
"Mehdi Jafari"
] | 2014-07-31T08:36:06+00:00 | 2014-07-31T08:36:06+00:00 | [
"math.DG"
] | Symmetry Reduction of the Two-Dimensional Ricci Flow Equation | This paper is devoted to obtain the one-dimensional group invariant solutions
of the two-dimensional Ricci flow ((2D) Rf) equation. By classifying the orbits
of the adjoint representation of the symmetry group on its Lie algebra, the
optimal system of one-dimensional subalgebras of the ((2D) Rf) equation is
obtained. For each class, we will find the reduced equation by the method of
similarity reduction. By solving these reduced equations, we will obtain new
sets of group invariant solutions for the ((2D) Rf) equation. |
1408.0243v1 | ## Computation of Partially Invariant Solutions for the Einstein Walker Manifolds' Identifying Equations
Mehdi Nadjafikhah a, ∗ , Mehdi Jafari b
a b
- School of Mathematics, Iran University of Science and Technology, Narmak, Tehran 1684613114, Iran
- Department of Complementary Education, Payame Noor University, PO BOX 19395-3697, Tehran, Iran
## Abstract
In this paper, partially invariant solutions (PISs) method is applied in order to obtain new four-dimensional Einstein Walker manifolds. This method is based on subgroup classification for the symmetry group of partial differential equations (PDEs) and can be regarded as the generalization of the similarity reduction method. For this purpose, those cases of PISs which have the defect structure δ = 1 and are resulted from two-dimensional subalgebras are considered in the present paper. Also it is shown that the obtained PISs are distinct from the invariant solutions that obtained by similarity reduction method.
Keywords: Einstein Walker manifolds, Lie symmetry group, Optimal system of Lie subalgebras, Partially invariant solutions (PISs)
2000 MSC:
70G65, 34C14, 53C50
## 1. Introduction
The idea of analyzing the differential equations by applying the transformation groups implied a new theory: the symmetry group theory, also called Lie group analysis. This method was originated at the end of nineteenth century from the pioneering work of Sophus Lie [1]. Since that time, several books have been dedicated to this concept and its generalizations [2-5]. Classification of the group invariant solutions and reduction of the original system can be regarded as significant applications of the Lie group method in the theory of differential equations. The fact that symmetry reductions for many PDEs can not be obtained via the classical symmetry method, motivated the creation of several generalizations of the classical Lie group method for symmetry reductions. Consequently, several alternative reduction methods have been proposed, going beyond Lie's classical procedure and providing further solutions. Partially invariant solutions (PISs) method is one of these procedures. This algorithmic method is a powerful tool for the reduction of PDEs and is based on classifying the subgroups of the symmetry group. The notion of PISs was first developed by Ovsiannikov [5] and can be considered as the extension of invariant solutions. The algorithm of constructing PISs is similar to that of invariant solutions. Indeed, obtaining the invariant solutions by applying the PISs method is easier than similarity reduction method whenever we deal with low-dimensional groups. One significant concept which appears while constructing PISs is the defect structure. This quantity is determined by the dimension of orbits and is denoted by δ .
In this paper, the PISs method will be applied in order to construct some new classes of four-dimensional Einstein Walker manifolds. Those Manifolds which admit null parallel distributions are called Walker manifolds . A Walker manifold is called Einstein Walker manifold if its Ricci tensor is a scaler multiple of the metric at each point. Four-dimensional Einstein Walker manifolds form the underling structure of many geometric and physical models such as; hh-space in general relativity, pp-wave model and other areas [6-12].
∗ Corresponding author. Tel.:+98 21 73225426; Fax: +98 21 73228426 Email addresses: m [email protected] (Mehdi Nadjafikhah), [email protected] (Mehdi Jafari)
Preprint submitted to Commun Nonlinear Sci Numer Simulat
The general form of the metric tensor of four-dimensional walker manifolds is expressed as follows [13]:
$$g _ { a, b, c } & \coloneqq 2 ( d x \circ d y$$
where a , b and c are smooth functions with respect to ( x, t, y, z ). Let M a,b,c := ( O , g a,b,c ), where O be an open subset of R 4 . We can see that M a,b,c is Einstein if and only if the functions a , b and c verify the following system of PDEs ([13], page 81).
$$\begin{array} { c } a _ { 1 1 } - b _ { 2 2 } =$$
where the index 1, 2, 3 and 4 for functions a , b and c represent the derivatives of these functions with respect to x , t , y and z , respectively. The system (2) is hard to handle, so we consider a spacial case in this paper; where a , b and c only depend on x and t . Therefore the following system must be solved.
$$\begin{array} {$$
In system (3), x and t are independent and a , b and c are dependent variables. It is worthwhile to say that some other special cases have been considered in many references which yield some results about the structures admitted by these manifolds [14-16].
In [17], we have comprehensively analyzed the problem of symmetries of the system (3). By applying the basic Lie symmetry method, we have obtained the classical Lie point symmetry operators of the system (3) and proved the following result (refer to [17] for more details):
Corollary 1.1. The Lie group of point symmetries of the PDE system (3) has a seven-dimensional Lie subalgebra generated by the following vector fields:
$$\begin{array} { c c c } X _ { 1 } = \partial _ { x }$$
$$( \partial _ { x } \equiv \frac { \partial } { \partial x },$$
Mainly, we have constructed an optimal system of one-dimensional subalgebras in [17] which provides the preliminary classification of group invariant solutions for the system (3). Also, we have obtained the corresponding invariant solutions of this system via the method of similarity reduction.
It is worth considering that some of the partially invariant solutions are not invariant with respect to the subalgebras of lower dimensions. They are known as non-reducible PISs. By determining non-reducible PISs for the system (3), we can obtain some new four-dimensional Einstein Walker manifolds.
This paper is organized as follows: In section 2, we recall the general procedure of determining PISs. In section 3, we classify the two-dimensional subalgebras of the symmetry Lie algebra and construct an optimal system. Section 4 is devoted to the computation of some of the PISs associated to the system (3). In section 5, the relation between PISs and invariant solutions is investigated and the condition for obtaining the non-reducible PISs is stated. Meanwhile, some non-reducible PISs for the system (3) are presented. Some concluding remarks are declared at the end of the paper.
## 2. Partially invariant solutions method
In this section, we recall the general procedure for determining PISs for an arbitrary system of PDEs. First, we present a brief review of the concept of PISs [18, 5]. Consider a system of PDEs of n th order with
p independent variables ( x = ( x i ) ∈ X,i = 1 , ..., p ) and q dependent variables ( u = ( u j ) ∈ U, j = 1 , ..., q ) such as:
$$\Delta = \Delta _ { \mu } ( x, u ^ { ( n ) } ) = 0, \quad \ \mu = 1, \dots, r.$$
Let G be a local symmetry group of the above system which acts on the total space X × U with r -dimensional orbits. If u = f ( x ) is a solution of the system (5) with graph Γ f , then the orbit space of Γ f can be defined as follows:
$$G \Gamma _ { f } = \{ g. ( x, u ) \ | \ ( x, u ) \in \Gamma _ { f }, g \in G \}$$
which is the union of the orbits of the Γ f -elements.
The defect structure of the solution u = f ( x ) with respect to the group G is computed by the matrix of generators' characteristics and is defined as:
$$\delta = \dim ( G \Gamma _ { f } ) - \dim ( \Gamma _ { f } ) = \dim ( G \Gamma _ { f } ) - p.$$
Also, we have 0 ≤ δ ≤ min { r, q } ([5], p 276-277).
If δ = 0, then u = f ( x ) is an invariant solution and if 0 < δ < min { r, q } , then u = f ( x ) is a partially invariant solution.
In order to calculate the PISs, first of all, it is necessary to classify the symmetry group into conjugacy classes. For obtaining the PISs with the defect structure δ , those subgroups H ⊂ G which have the property that if the dimension of the orbits of H on the space X × U is r , then the dimension of the orbits is r -δ on the space X , must be selected [19]. Let H be a subgroup with this property that mentioned above and h be its Lie algebra with infinitesimal generators { v , ..., v 1 s } . Hence, we can obtain a complete set of functionally independent invariants of the form
$$\{ \xi _ { i } ( x ), I _ { j } ( x, u ) \},$$
where i = 1 , ..., p + δ -s and j = 1 , ..., q -δ . Then we have
$$\text{rank} \left ( \frac { \partial I _ { j } ( x, u ) } { \partial u } \right ) = q - \delta = q ^ { \prime }. \quad \text{--} \cdots \text{--} \text{rrc} \quad \text{--} \cdots \cdots \cdots \quad \text{--} \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \ddots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots\cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdets \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \ \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \collections \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \dots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots$$
If u = f ( x ) is a function, then the manifold H Γ f can be expressed in terms of the invariants (8). So, we have:
$$I _ { j } ( x, u ) = f _ { j } ( \xi _ { i } ( x ) )$$
where the functions f j are arbitrary. Now, by applying the implicit function theorem, we conclude that
$$u ^ { i _ { \alpha } } = U ^ { i _ { \alpha } } ( x, u ^ { j \beta }, f _ { j } ( \xi _ { i } ( x ) ) )$$
where α = 1 , ..., q ′ and β = 1 , ..., δ . The remaining dependent variables only depend on the original independent variables:
$$u ^ { j _ { \beta } } = U ^ { j _ { \beta } } ( x _ { 1 }, \dots, x _ { p } ), \quad \ \beta = 1, \dots, \delta.$$
Now, the derivatives of the functions u , ..., u 1 q with respect to the new variables which are obtained from equations (11) and (12) must be calculated. Hence, by substituting these quantities into the original system, a new system is obtained, involving the q ′ functions f j and the invariants ξ i . The resulted equations are not generally consistent, so that the compatibility conditions must be computed. Consequently, a system of PDEs is deduced from these constraints which is denoted by ∆ /H . On the other hand, a system of PDEs is resulted from (12) denoted by ∆ . ′ Now, the system ∆ /H must be solved first. Then corresponding to each of the solution of this system, the system ∆ ′ will be solved. Finally, the partially invariant solutions are obtained by substituting the resulted solutions into equations (11) and (12). For more details about the methods of determining the PISs refer to [20, 5].
## 3. Classification of Subalgebras for the system (3)
In this section, we want to classify the subgroups of symmetry group of the system (3), into conjugacy classes. Searching for the invariant solutions can be regarded as the main motivation of computing the symmetries of a differential equation. As it is well known, the problem of classifying invariant solutions is equivalent to the problem of classifying the subgroups of the full symmetry group under conjugation. Let H and ˜ be two connected, H s -dimensional Lie subgroups of the Lie group G with corresponding Lie subalgebras h and ˜ of the Lie algebra h g of G . Let g ∈ G , then ˜ H = gHg -1 are conjugate subgroups if and only if ˜ = Ad( h g ) · h are conjugate subalgebras, where Ad( g ) is adjoint represen Hence, the problem of determining an optimal system of subgroups is equivalent to that of obtaining an optimal system of subalgebras, and so we focus on it [4, 5]. The latter problem, tends to obtain a list of conjugacy inequivalent subalgebras with the property that any other subalgebra is equivalent to only a unique member of the list under some element of the adjoint representation for some element of the investigated Lie group.
## 3.1. Optimal system of one-dimensional subalgebras for the system (3)
Indeed, for one-dimensional subalgebras, the classification problem is necessarily the same as the problem of classifying the orbits of the adjoint representation. Thus, an optimal set of subalgebras is constructed if we select just one representative from each family of equivalent subalgebras. Consequently, the associated set of invariant solutions is then the minimal list from which we can obtain all other invariant solutions of one-dimensional subalgebras simply via transformations.
In [17], we have presented a comprehensive analysis of this problem and have constructed an optimal system of one-dimensional subalgebras for the system (3) as follows :
Theorem 3.1. An optimal system of one-dimensional Lie subalgebras of the system (3) is provided by the following generators:
$$\text{now many genericours} \colon \\ 1 ) \, X ^ { 1 } & = X _ { 7 }, \quad & \, 8 ) \, X ^ { 8 } \ = X _ { 4 } + a X _ { 5 } + b X _ { 6 } + c X _ { 7 }, \\ 2 ) \, X ^ { 2 } & = X _ { 1 } + a X _ { 7 }, \quad & \, 9 ) \, X ^ { 9 } \ = \varepsilon X _ { 1 } + X _ { 4 } + a X _ { 5 } + b X _ { 6 } + c X _ { 7 }, \\ 3 ) \, X ^ { 3 } & = X _ { 2 } + a X _ { 7 }, \quad & \, 1 0 ) \, X ^ { 1 0 } & = X _ { 3 } + a X _ { 5 } + b X _ { 6 } + c X _ { 7 }, \\ 4 ) \, X ^ { 4 } & = X _ { 6 } + a X _ { 7 } \quad & \, 1 1 ) \, X ^ { 1 1 } & = \varepsilon X _ { 2 } + X _ { 3 } + a X _ { 5 } + b X _ { 6 } + c X _ { 7 }, \quad & \, ( 1 3 ) \\ 5 ) \, X ^ { 5 } & = \varepsilon X _ { 1 } + X _ { 6 } + a X _ { 7 }, \quad & \, 1 2 ) \, X ^ { 1 2 } & = X _ { 3 } + \varepsilon X _ { 4 } + a X _ { 5 } + b X _ { 6 } + c X _ { 7 }, \\ 6 ) \, X ^ { 6 } & = X _ { 5 } + a X _ { 6 } + b X _ { 7 }, \quad & \, 1 3 ) \, X ^ { 1 3 } & = \varepsilon X _ { 2 } + X _ { 3 } + \varepsilon ^ { \prime } X _ { 4 } + a X _ { 5 } + b X _ { 6 } + c X _ { 7 }. \\ 7 ) \, X ^ { 7 } & = \varepsilon X _ { 2 } + X _ { 5 } + a X _ { 6 } + b X _ { 7 },$$
- 7) X = εX 2 + X 5 + aX 6 + bX , 7
where ε and ε ′ are ± 1 and a, b, c ∈ R are arbitrary numbers [17].
## 3.2. Optimal system of two-dimensional subalgebras for the system (3)
In this paper, we need to classify the two-dimensional subalgebras. Because, we want to calculate those PISs which have the defect structure δ = 1 and the reduced system ∆ /H is a system of ordinary differential equations. Since p = 2, δ = 1 and p + δ -s = 1, then we have s = 2. So, we should consider the twodimensional subgroups.
Consequently, constructing the two-dimensional optimal system, i.e., classification of the two-dimensional subalgebras of g is our next step. This process is performed by choosing one of the vector fields as stated in theorem (3.1). Let us consider X 1 (or X i , i = 2 , · · · , 13). Corresponding to it, an optional vector field Y = b 1 X 1 + · · · + b 7 X 7 is selected, so we must have
$$[ \mathbf X ^ { 1 }, Y ] = \lambda \mathbf X ^ { 1 } + \mu Y.$$
Equation (14) leads us to the system
$$C _ { j k } ^ { i } \alpha _ { j } a _ { k } = \lambda a _ { i } + \mu \alpha _ { i } & & ( i = 1, \cdots, 7 ). & & ( 1 5 )$$
where the constant coefficients C i jk are the structure constants. The solutions of the system (15), give one of the two-dimensional generators and the second generator is X 1 (or X i , i = 2 , · · · , 13) if selected.
After the construction of all two-dimensional subalgebras, for every vector fields of theorem (3.1), they need to be simplified by the action of adjoint transformations in the manner analogous to the way of onedimensional optimal system. Hence, we can state the following theorem:
## Theorem 3.2. Any two-dimensional subalgebras of (4) is conjugate to precisely one of the following subalgebras:
| A 1 1 : 〈 X 1 , X 3 + αX 6 + βX 7 〉 , A 4 1 : 〈 X 1 , X 3 + /epsilon1X 5 + X 6 + αX 7 〉 , A 7 1 : 〈 X 1 , X 7 〉 , | A 2 1 : 〈 X 1 , X 2 + αX 5 + βX 7 〉 , A 5 1 : 〈 X 1 , X 2 + αX 3 + βX 7 〉 , | A 3 1 : 〈 X 1 , X 5 + αX 7 〉 , A 6 1 : 〈 X 1 , X 6 + αX 7 〉 , |
|---------------------------------------------------------------------------------------------------------------------------|----------------------------------------------------------------------------------------|---------------------------------------------------------------------|
| A 1 2 : 〈 X 2 , X 3 + αX 6 + βX 7 〉 , A 4 2 : 〈 X 2 , X 3 + /epsilon1X 4 + X 6 + αX 7 〉 , A 7 2 : 〈 X 2 , X 7 〉 , | A 2 2 : 〈 X 2 , X 1 + /epsilon1X 4 + βX 7 〉 , A 5 2 : 〈 X 2 , X 1 + X 6 + αX 7 〉 , | A 3 2 : 〈 X 2 , X 4 + αX 7 〉 , A 6 2 : 〈 X 2 , X 6 + αX 7 〉 , |
| A 1 3 : 〈 X 6 , X 3 + αX 7 〉 , A 4 3 : 〈 X 6 , X 1 + αX 7 〉 , | A 2 3 : 〈 X 6 , X 4 〉 , A 5 3 : 〈 X 6 , X 2 〉 , | A 3 3 : 〈 X 6 , X 5 〉 , A 6 3 : 〈 X 6 , X 7 〉 , |
| A 1 4 : 〈 εX 1 + X 6 , X 2 〉 , | A 2 4 : 〈 εX 1 + X 6 , X 5 〉 , | A 3 4 : 〈 εX 1 + X 6 , X 7 〉 , |
| A 1 5 : 〈 X 5 , X 3 + αX 6 + βX 7 〉 , A 4 5 : 〈 X 5 , X 7 〉 , | A 2 5 : 〈 X 5 , X 1 + αX 6 + βX 7 〉 , | A 3 5 : 〈 X 5 , X 6 + αX 7 〉 , |
| A 1 6 : 〈 εX 2 + X 5 , X 3 + 1 2 X 6 + αX 7 〉 , | A 2 6 : 〈 εX 2 + X 5 , X 1 + αX 7 〉 , | A 3 6 : 〈 εX 2 + X 5 , X 7 〉 , |
| A 1 7 : 〈 X 4 , X 3 + αX 6 + βX 7 〉 , A 4 7 : 〈 X 4 , X 7 〉 , | A 2 7 : 〈 X 4 , X 2 + αX 3 + βX 7 〉 , | A 3 7 : 〈 X 4 , X 6 + αX 7 〉 , |
| A 1 8 : 〈 εX 1 + X 4 , X 3 +2 X 6 + αX 7 〉 , | A 2 8 : 〈 εX 1 + X 4 , X 2 + αX 7 〉 , | A 3 8 : 〈 εX 1 + X 4 , X 7 〉 , |
| A 1 9 : 〈 X 3 , X 2 + αX 7 〉 , A 4 9 : 〈 X 3 , X 6 + αX 7 〉 , | A 2 9 : 〈 X 3 , X 5 〉 , A 5 9 : 〈 X 3 , X 7 〉 , | A 3 9 : 〈 X 3 , X 1 〉 , A 6 9 : 〈 X 3 , X 4 〉 , |
| A 1 10 : 〈 εX 2 + X 3 , X 1 〉 , A 4 10 : 〈 X 2 , X 3 + αX 7 〉 , | A 2 10 : 〈 εX 2 + X 3 , X 4 〉 , | A 3 10 : 〈 εX 2 + X 3 , X 7 〉 , |
| A 1 11 : 〈 X 3 + εX 4 , εX 5 + X 6 - 2 X 7 〉 , A 4 11 : 〈 X 3 + εX 4 , X 3 + X 6 + αX 7 〉 , | A 2 11 : 〈 X 3 + εX 4 , X 2 + αX 7 〉 , A 5 11 : 〈 X 3 + εX 4 , X 1 + εX 2 〉 , | A 3 11 : 〈 X 3 + εX 4 , X 7 〉 , |
| 1 12 : εX 2 + X 3 + ε ′ X 4 , X 1 + ε ′ X 2 , | 2 12 : εX 2 + X 3 + ε ′ X 4 , X 7 , | |
A 1 12 : 〈 εX 2 + X 3 + ε X , X ′ 4 1 + ε X ′ 2 〉 , A 2 12 : 〈 εX 2 + X 3 + ε X , X ′ 4 7 〉 , where α and β are arbitrary constants, ε and ε ′ are ± 1 and /epsilon1 is ± 1 or 0 .
Proof. Each of the two-dimensional subalgebra has two generators. For classifying two-dimensional subalgebras, we must select one of the generators from the list of one-dimensional optimal system (13) and another generator must be taken optionally.
Suppose that h = span { X,Y } is a two-dimensional subalgebra of g where X is a one-dimensional subalgebra which is chosen from the list (13) and Y is an optional vector defined by: Y = b 1 X 1 + · · · + b 7 X 7 . Now, we must simplify h as much as possible by imposing various adjoint transformations on it [4].
Each adjoint transformation is a linear map F s i : g → g defined by X → Ad(exp( sX i ) .X ), for i = 1 , · · · , 7.
↦
Since this procedure is so lengthy, we only explain one of the cases in the following.
Case1 : If X = X 1 + aX 7 then
$$\mathfrak { h } = \langle X, Y \rangle & = \langle X _ { 1 } + a X _ { 7 }, \sum _ { i = 1 } ^ { 7 } b _ { i } X _ { i } \rangle \\ & \quad = \langle X _ { 1 }, b _ { 2 } X _ { 2 } + b _ { 3 } X _ { 3 } + b _ { 4 } X _ { 4 } + b _ { 5 } X _ { 5 } + b _ { 6 } X _ { 6 } + b _ { 7 } X _ { 7 } \rangle \\.$$
So, we have:
- a) If b 2 = b 3 = b 4 = b 5 = b 6 = 0, then we have h = 〈 X ,b X 1 7 7 〉 = 〈 X ,X 1 7 〉 . Since [ X ,X 1 7 ] = 0 = 0 X 1 +0 X 7 , then h is closed under the Lie bracket. So, h is reduced to the case A 7 1 .
/negationslash
- b) If b 2 = b 3 = b 4 = b 5 = 0 and b 6 = 0, then h = 〈 X ,b X 1 6 6 + b 7 X 7 〉 = 〈 X ,X 1 6 + αX 7 〉 . Since [ X ,X 1 6 + αX 7 ] = 0, then h is closed under the Lie bracket. So, h is reduced to the case A 6 1 .
/negationslash
- c) If b 2 = b 3 = b 4 = b 6 = 0 and b 5 = 0, then h = 〈 X ,b X 1 5 5 + b 7 X 7 〉 = 〈 X ,X 1 5 + αX 7 〉 . Since [ X ,X 1 5 + αX 7 ] = 0, then h is closed under the Lie bracket. So, h is reduced to the case A 3 1 .
/negationslash
- d) If b 2 = b 3 = b 4 = 0 and b 6 = 0, then we can make the coefficient of X 5 vanish by F s 5 5 ; By setting s 5 = 1 b 6 . So, we have h = 〈 X ,b X 1 6 6 + b 7 X 7 〉 which is reduced to the case A 6 1 like (b).
/negationslash
- e) If b 2 = b 3 = 0 and b 4 = 0, then h = 〈 X ,b X 1 4 4 + · · · + b 7 X 7 〉 and [ X ,b X 1 4 4 + · · · + b 7 X 7 ] = b 4 X 2 = rX 1 + ( s b 4 X 4 + · · · + b 7 X 7 ) for any r, s ∈ R . So, h is not closed under the Lie bracket and we have not any two-dimensional subalgebras in this case.
/negationslash
/negationslash
- f) If b 2 = b 4 = 0, b 3 = 0 and b 6 = 1, then we can make the coefficient of X 5 vanish by F s 5 5 ; By setting s 5 = b 5 -1+ b 6 . Also, by scaling if necessary, we can assume that b 3 = 1. Then we have h = 〈 X ,X 1 3 + b 6 X 6 + b 7 X 7 〉 and [ X ,X 1 3 + b 6 X 6 + b 7 X 7 ] = X 1 . So, h is closed under the Lie bracket and the case A 1 1 is concluded.
/negationslash
/negationslash
- g) If b 2 = b 4 = 0, b 3 = 0 and b 6 = 1, then by scaling if necessary, we can assume that b 3 = 1. Also, the coefficient of X 5 can be vanished or be ± 1 by F s 6 6 ; By setting s 6 = -ln | b 5 | . Then we have h = 〈 X ,X 1 3 + εX 5 + X 6 + b 7 X 7 〉 and [ X ,X 1 3 + εX 5 + X 6 + b 7 X 7 ] = X 1 . So, h is closed under the Lie bracket and the case A 4 1 is deduced.
/negationslash
- h) If b 3 = b 4 = b 6 = 0 and b 2 = 0, then by scaling if necessary, we can assume that b 2 = 1. Also, we have h = 〈 X ,X 1 2 + b 5 X 5 + b 7 X 7 〉 and [ X ,X 1 2 + b 5 X 5 + b 7 X 7 ] = 0. So, h is closed under the Lie bracket and the case A 2 1 is resulted.
/negationslash
- i) If b 4 = b 6 = 0, b 2 = 0 and b 3 = 0, then by scaling if necessary, we can assume that b 2 = 1. Also we can make the coefficient of X 5 vanish by F s 5 5 ; By setting s 5 = -b 5 b 3 . Then we have h = 〈 X ,X 1 2 + b 3 X 3 + b 7 X 7 〉 and [ X ,X 1 2 + b 3 X 3 + b 7 X 7 ] = b 3 X 1 . So, h is closed under the Lie bracket and the case A 5 1 is resulted .
/negationslash
/negationslash
- j) If b 4 = 0, b 6 = 0 and b 3 = 1, then by scaling if necessary, we can assume that b 6 = 1. We can make the coefficient of X 2 vanish by F s 2 2 ; By setting s 2 = -1 b 6 . Also, the coefficient of X 5 can be vanished or be ± 1 by F s 6 6 ; By setting s 6 = -ln b | 5 | . So, we have h = 〈 X ,X 1 3 + εX 5 + X 6 + b 7 X 7 〉 which is reduced to the case A 4 1 similar to (g).
/negationslash
- k) If b 4 = 0, b 6 = 0 and b 3 = 1, then we can make the coefficient of X 2 and X 5 vanish by F s 2 2 and F s 5 5 ; By setting s 2 = -1 b 6 and s 5 = -b 5 b 3 -1 , respectively. By scaling if necessary, we have h = 〈 X ,X 1 3 + b 6 X 6 + b 7 X 7 〉 that is reduced to the case A 1 1 similar to (f).
/negationslash
In each case, h can not be simplified any more by F s i i , i = 1 , ..., 7. By a similar method, we can find the two-dimensional subalgebras for the other 11 cases.
## 4. Computation of the partially invariant solutions for the system (3)
In this section, we will calculate the PISs of the system (3). For example, consider the Lie subalgebra A 7 1 : 〈 X ,X 1 7 〉 . The set of functionally independent invariants for A 7 1 is a set of functions I with the following property: X 1 ( I ) = X 7 ( I ) = 0. By calculating these functions, a set of functionally independent invariants is obtained as follows:
$$\{ t, \frac { b } { a }, \frac { c } { a } \}$$
So, we have:
$$\text{ive} \colon & & \text{rank} \left ( \frac { \partial ( \frac { b } { a }, \frac { c } { a } ) } { \partial ( a, b, c ) } \right ) = \text{rank} \left ( \begin{array} { c c } - b / a ^ { 2 } & - c / a ^ { 2 } \\ 1 / a & 0 \\ 0 & 1 / a \end{array} \right ) = 2 = q - \delta. & & ( 1 8 ) \\ \colon & & 3, \text{then} \, \delta = 1. \text{ The equations corresponding to the relation } ( 1 0 ) \, \text{ are}$$
Since q = 3, then δ = 1. The equations corresponding to the relation (10) are
$$\frac { b } { a } = f ( t ), \quad \frac { c } { a } = g ( t ),$$
and the solutions corresponding to the equations (11) and (12) are expressed as follows:
$$b = a f ( t ), \quad c = a g ( t ), \quad a = a ( x, t ).$$
Note that a x, t ( ) is an arbitrary function.
Now, we can compute the derivatives of the functions a, b and c from relations (20):
$$\begin{array} { c c c } b _ { 1 } = a _ { 1 } f, & b _ { 2 } = a _ { 2 } f + a f ^ { \prime }$$
Hence, by substituting the above relations into the system (3), we obtain:
$$\cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cd$$
The consistency conditions, obtained from the system (22), conclude the following equations
$$\begin{array} { c c c c c c c } \cdot & \cdot & \cdot & \cdot & \cdot \\ & a ^ { 2 } f ^ { \$$
and these inequations
$$f \neq 0, \quad g \neq 0, \quad a \neq 0, \quad f - g ^ { 2 } \neq 0. \quad \quad \quad$$
/negationslash
/negationslash
/negationslash
/negationslash
Equations (23) form a system of ODEs. By solving this system, we obtain four types of solutions as follows:
$$\text{ons} \, ( 2 3 ) \, \text{form} \, a \$$
Now, by applying the relations (20), the partially invariant solutions associated to the subalgebra A 7 1 can be obtained for the system (3).
$$\begin{array} {$$
These partially invariant solutions seem to be trivial. In a similar way, we can calculate nontrivial PISs by applying other subalgebras listed in theorem (3.2). Some other of these PISs are presented in Table 1.
## 5. Non-reducible partially invariant solutions
In this section, we will deal with those partially invariant solutions which are not invariant with respect to some of the subgroups of G . They are called non-reducible PISs and they are usually rare. For a reducible partially invariant solution u = f ( x ), we can determine a subgroup H ′ ⊂ H which u is a H ′ -invariant solution and dim( H ′ Γ ) f /greaterorequalslant dim( H ) -δ = s -δ ([5], page 290). So, reducible PISs can be obtained via the method of similarity reduction from the reduced system involving p -dim( H ′ Γ ) f /lessorequalslant p + δ -s independent variables, which is indeed easier than obtaining them by PISs method. For example, the PISs that we computed in the previous section are non-reducible PISs. Because, reducible PISs must be invariant with respect to the one-parameter subgroups of its Lie group. These subgroups have Lie subalgebras of A 7 1 generated by an infinitesimal generator of the form αX 1 + βX 7 where α, β ∈ R . In [17], we obtained the invariant solutions of the system (3) with respect to this generator by applying the method of similarity reduction:
$$\text{use system} \, ( \mathcal { J } \, \text{with respect$$
Since, none of these solutions is similar to the PISs in (26), we conclude that PISs in (26) are non-reducible PISs. Several other non-reducible partially invariant solutions of the system (3) are listed in the following table.
Table 1: Non-reducible partially invariant solutions of (3).
| Subalgebras | Invariants | Dependent variables | PISs |
|-----------------------|-------------------------------------------------|-----------------------------------------------------------------------------------|--------------------------------------------------------------------------------------------------------------------------|
| 〈 X 2 , X 7 〉 | { x, b a , c a } | a = a ( x, t ) b = af ( x ) c = ag ( x ) | a = c 1 x + c 2 b = c 1 c 2 3 x +( ∗ ) c = c 3 ( c 1 x + c 2 ) + c 1 c 4 |
| 〈 X 5 , X 7 〉 | { t, c 2 - ba b 2 , - ct + bx b } | a = b ( x - g ( t )) 2 t 2 - bf ( t ) b = b ( x, t ) c = b t ( x - g ( t )) | a = ( c 1 + c 2 t 3 )( x - c 3 ) 2 t 3 - ( ∗∗ ) b = c 1 + c 2 t 3 t c = ( c 1 + c 2 t 3 )( x - c 3 ) t 2 |
| 〈 X 2 , X 6 + X 7 〉 | { x, b a 3 , c a 2 } | a = a ( x, t ) b = a 3 f ( x ) c = a 2 g ( x ) | a = 4( t + c 1 ) ( c 2 x + c 3 ) 2 b = 4 c 2 2 ( t + c 1 ) 3 ( c 2 x + c 3 ) 4 c = 4 c 2 ( t + c 1 ) 2 ( c 2 x + c 3 ) 3 |
| 〈 X 4 , X 7 〉 | { x, - at + cx xa , at 2 - 2 xtc + bx 2 x 2 a } | a = a ( x, t ) b = a ( g ( x ) +2 t x f ( x ) + t 2 x 2 ) c = a ( f ( x ) + t x ) | a = c 1 + c 2 x 3 x b = ( c 1 + c 2 x 3 )( t + c 3 ) 2 x 3 +( ∗ ∗ ∗ ) c = ( c 1 + c 2 x 3 )( t + c 3 ) x 2 |
where c i 's are arbitrary constants and
- ∗ = ( c 5 c 1 -c 2 3 c 2 )ln( c 1 x + c 2 ) + c 6 , ∗∗ = ( t c 4 + c 5 )(ln( t +( c 1 c 2 ) ( 1 3 ) ) 2 -ln( t 2 -t ( c 1 c 2 ) ( 1 3 ) +( c 1 c 2 ) ( 2 3 ) ) + 2 √ 3tan -1 ( 2 c 2 t √ 3 c 1 ( c 1 c 2 ) ( 2 3 ) -1 √ 3 )) , √
$$* * * = x ( c _ { 4 } + c _ { 5 } ) ( \ln ($$
## Conclusion
Partially invariant solutions can be regarded as the natural extension of the invariant ones and the method of obtaining them is algorithmic. Some of the PISs are not invariant with respect to the lower dimensional subalgebras. They are called non-reducible PISs and can be applied in order to obtain some new solutions for an arbitrary system of PDEs. For example, in the present paper, we have obtained a non-reducible PISs for the system (3) as follows:
$$a = \frac { 4 ( t + c _ { 1 } ) } { ( c _ { 2$$
Consequently, according to relation (1), we could determine a general form for the metric associated to a set of four-dimensional Einstein Walker manifolds as follows:
$$g _ { a, b, c } & \coloneqq 2 ( d x \circ d y + d t \circ d z ) + \frac { 4 ( t + c _ { 1 } ) } { ( c _ { 2 } x + c _ { 3 } ) ^ { 2 } } d y \circ d y \\ & \quad + \frac { 4 c _ { 2 } ^ { 2 } ( t + c _ { 1 } ) ^ { 3 } } { ( c _ { 2 } x + c _ { 3 } ) ^ { 4 } } d z \circ d z + \frac { 8 c _ { 2 } ( t + c _ { 1 } ) ^ { 2 } } { ( c _ { 2 } x + c _ { 3 } ) ^ { 3 } } d y \circ d z. \\ \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot$$
This set of solutions can not be obtained from the similarity reduction method. In a similar manner, we can deduce some other four-dimensional Einstein Walker manifolds from Table 1.
## Acknowledgements
It is a pleasure to thank the anonymous referees for their constructive suggestions and helpful comments which have improved the presentation of the paper. The authors wish to express their sincere gratitude to Fatemeh Ahangari for her useful advise and suggestions.
## References
- [1] S. Lie, On integration of a class of linear partial differential equations by means of definite integrals, translation by N.H. Ibragimov, Arch Math. 6 (1881) 328-368.
- [2] G. Bluman, A.F. Cheviakov, S. Anco, Application of symmetry methods to partial differential equations, Springer, New York, 2010.
- [3] A. Kushner, V. Lychagin, V. Rubstov, Contact geometry and nonlinear differential equations, Cambridge University Press, Cambridge, 2007.
- [4] P.J. Olver, Applications of Lie Groups to Differential Equations, Springer, New York, 1986.
- [5] L.V. Ovsiannikov, Group Analysis of Differential Equations, Academic Press. New York, 1982.
- [6] C.P. Boyer, J. D. Finley III, and J. F. Pleba´ski, Complex general relativity, n h and hh spaces - a survey of one approach, Gen. Relativity Gravitation, 2 (1980) 241-281, Plenum, New York-London.
- [7] J.D. Finley III, and J. F. Pleba´ski, The intrinsic spinorial structure of hyperheavens, J. Math. Phys. 17 (1976), 2207-2214. n
- [8] R. Abounasr, A. Belhaj, J. Rasmussen, and E.H. Saidi, Superstring theory on pp-waves with ADE geometries, J. Phys. A 39 (2006), 2797-2841.
- [9] J. Kerimo, AdS pp-waves, J. High Energy Phys. (2005), 025, 18 pp.
- [10] J. Michelson, and X. Wu, Dynamics of antimembranes in the maximally supersymmetric eleven-dimensional pp-wave, J. High Energy Phys. (2006), 028, 37 pp.
- [11] V. Pravda, A. Pravdova, A. Coley, and R. Milson, All spacetimes with vanishing curvature invariants, Classical Quantum Gravity 19 (2002), 6213-6236.
- [12] Th. Leistner, Screen bundles of Lorentzian manifolds and some generalizations of pp-waves, J. Geom. Phys. 56 (2006), 2117-2134.
- [13] M. Brozos-V´ azquez, E. Garcia-Rio, P. Gilkey, S. Nikˇ cevi´, c R. V´ azquez-Lorenzo, The Geometry of Walker Manifolds. Synthesis Lectures on Mathematics and Statistics, Vol. 5, Morgan and Claypool Publ, 2009.
- [14] Y. Matsushita, Walker 4-manifolds with proper almost complex structures, J. Geom. Phys. 55 (2005), 385-398.
- [15] M. Brozos-V´ azquez, E. Garcia-Rio, and R. V´zquez-Lorenzo, Conformally Osserman four-dimensional manifolds whose a conformal Jacobi operators have complex eigenvalues, Proc. R. Soc. A. 462 (2006), 1425-1441.
- [16] P. Gilkey, The geometry of curvature homogeneous pseudo-Riemannian manifolds. ICP Advanced Texts in Mathematics, 2. Imperial College Press, London, 2007.
[17] M. Nadjafikhah, M. Jafari, Some general new Einstein Walker manifolds, arXiv:1206.3730v1 [math.DG] 17 Jun 2012.
- [18] S.V. Meleshko, Methods for Constructing Exact Solutions of Partial Differential Equations, Springer Science+Business Media, Inc, NewYork, 2005.
- [19] A.M. Grundland, L. Lalague, Invariant and partially-invariant solutions of the equations describing a non-stationary and isentropic flow for an ideal and compressible fluid in (3 + 1) dimensions, J. Phys. A: Math. Gen. 29 (1996) 1723-1739.
- [20] A.M. Grundland, P. Tempesta, P. Winternitz, Weak transversality and partially invariant solutions, J. Math. Phys. 44(6) (2003) 2704-2722. | 10.1016/j.cnsns.2013.04.018 | [
"Mehdi Nadjafikhah",
"Mehdi Jafari"
] | 2014-07-31T08:57:25+00:00 | 2014-07-31T08:57:25+00:00 | [
"math.DG"
] | Computation of Partially Invariant Solutions for the Einstein Walker Manifolds' Identifying Equations | In this paper, partially invariant solutions (PISs) method is applied in
order to obtain new four-dimensional Einstein Walker manifolds. This method is
based on subgroup classification for the symmetry group of partial differential
equations (PDEs) and can be regarded as the generalization of the similarity
reduction method. For this purpose, those cases of PISs which have the defect
structure delta=1 and are resulted from two-dimensional subalgebras are
considered in the present paper. Also it is shown that the obtained PISs are
distinct from the invariant solutions that obtained by similarity reduction
method. |
1408.0244v2 | ## Less trouble with orbits: The semi-classical hydrogen problem in parabolic and spherical coordinates
## Manfred Bucher
Physics Department, California State University, Fresno,
Fresno, California 93740-8031
(Dated: June 3, 2014)
## Abstract
Historically, the eccentricity of Sommerfeld orbits from quantization conditions in either parabolic or spherical coordinates was found to differ in almost all cases. To do the orbit comparison correctly, one must use amended instead of traditional Sommerfeld orbits.
## I. INTRODUCTION
There is no question that the orbit-based old quantum theory of the hydrogen atom by Bohr and Sommerfeld is incomplete. While it accounts for the particle character of the electron, it completely ignores its wave character. One consequence is that the orbits depend on the choice of the coordinate system. This was one of the reasons why historically the old quantum theory was dismissed. Another reason concerned wrong values of orbital angular momentum. The subsequently developed quantum mechanics of Heisenberg (matrix mechanics) and Schr¨dinger (wave mechanics) resolved all these problems. In the meantime o it has been found that the difficulty with orbital angular momentum originates with Sommerfeld's discarding of straight-line electron orbits through the nucleus, called 'Coulomb oscillators.' 1 When these are included (and circular Bohr orbits of principal quantum number n consequently omitted) by the use of angular quantum number
$$l = 0, 1, \dots, n - 1,$$
instead of l = 1 2 , , ..., n , along with a correction of both the orbit's semi-focal axis f nl (semilatus rectum) and angular momentum | L nl | in proportion
$$f _ { n l } \in | \mathbf L _ { n l } | ^ { 2 } \in l ( l + 1 ),$$
instead of ∝ l 2 , agreement of those orbit quantities with angular orbital momentum and average size of quantum-mechanical orbitals is achieved. 1
On the other hand, the non-uniqueness of Bohr-Sommerfeld orbits, that is, their dependence on the coordinate system, persists. Historically, it first arose in the old quantum theory of the (linear) Stark effect-the splitting of spectral lines in an external electric field E -independently developed by Epstein and Schwarzschild. 2 A great success at the time, it accounts for the experimental findings and agrees with subsequent quantum-mechanical results. The theory obtains orbits for the electron of a hydrogen atom when quantization conditions in parabolic coordinates are imposed. However, in the absence of an electric field, E = 0, these orbits disagree with the traditional Sommerfeld orbits of the H atom, quantized in spherical coordinates-hence the conflict of non-uniqueness.
## II. FORMALISM
Geometrically, two planar ellipses are equal if they agree in the length of two of their axes, say the length of major axis, 2 a , and of focal axis (latus rectum), 2 f . Equivalently, two ellipses of the same length of major axis are equal if they agree in their eccentricity,
$$\varepsilon = \sqrt { \frac { a - f } { a } }.$$
By the Bohr-Sommerfeld theory of the hydrogen atom the semi-major axis of an elliptical orbit is
$$a _ { n l } = n ^ { 2 } r _ { B }$$
where r B is the Bohr radius and n is the principal quantum number. In terms of spherical quantization conditions, n is the sum of the radial and angular quantum numbers,
$$n = n _ { r } + n _ { \theta } + n _ { \varphi } = n _ { r } + l,$$
aptly called 'quantum sum' by Sommerfeld. Expressed in terms of parabolic quantum numbers n ξ and n η , and the azimuthal, or magnetic, quantum number n ϕ = m , the quantum sum is
$$n = n _ { \xi } + n _ { \eta } + | m |.$$
For a vanishing external electric field, E = 0, the eccentricity of an electron orbit from quantization conditions in parabolic coordinates is given by
$$\varepsilon _ { \xi \eta | m | } = \frac { 1 } { n } \left [ \sqrt { n _ { \xi } ( n _ { \xi } + | m | ) } + \sqrt { n _ { \eta } ( n _ { \eta } + | m | ) } \right ].$$
When this is compared with the eccentricity of a traditional ( n, l, m ) Sommerfeld orbit with quantization conditions in spherical coordinates,
$$\varepsilon _ { n l m } = \frac { 1 } { n } \sqrt { n ^ { 2 } - l ^ { 2 } },$$
disagreement is found in all cases except the circular Bohr orbits ( ε = 0). 2 However, since the traditional Sommerfeld orbits suffer from the defects mentioned above, one must use in this comparison the correct sequence and width of Sommerfeld orbits according to Eqs. (1) and (2). The eccentricity of such amended Sommerfeld orbits with spherical quantization conditions, denoted 〈 nlm 〉 , is then
$$\varepsilon _ { n l m } = \frac { 1 } { n } \sqrt { n ^ { 2 } - l ( l + 1 ) },$$
instead of Eq. (8). This provides agreement of orbit eccentricity, ε ξη m | | = ε nlm , in the cases of extreme | m | and close values in all other cases. It also reveals relationships between orbits from quantization conditions in both coordinate systems for common quantum numbers n and | m | .
## III. RESULTS
Table I gives calculated eccentricities ε ξη m | | and ε nlm for principal quantum numbers n = 1 2 3. , , By Eq. (1), circular Bohr orbits, l = n and ε = 0, are excluded from the amended Sommerfeld model (cases 2, 5, 11). The ground state, { 100 } or 〈 100 , is a straight-〉 line orbit through the nucleus, called Coulomb oscillator, with eccentricity ε ξη m | | = ε nlm = 1, in agreement for quantization in both coordinate systems (case 1). So are higher Coulomb oscillators (cases 3 and 6). Agreement of eccentricity, ε ξη m | | = ε nlm , also holds when, for a given n , the magnetic quantum number takes on the maximum allowed value, | m | = l = n -1 (cases 1, 4, 10). Disagreement is found for the cases 7 and 9 whose eccentricity ε ξη m | | { 111 } = 1 3 √ 8 = 0 94 and . ε ξη m | | { 201 } = 1 3 √ 6 = 0 82 flank the amended Sommerfeld . value ε nlm 〈 31 m 〉 = 1 3 √ 7 = 0 88 by . ± 0 06. .
The same pattern as in Table I can be seen in Table II which lists the eccentricities of orbits with quantum sum n = 4. Agreement between the parabolic and spherical expressions, Eqs. (7) and (9), holds for Coulomb oscillators, | m | = 0 (case 12), and for maximum | m | , here | m | = 3 (case 19). Otherwise the values of the parabolic expression of eccentricity (cases 13 and 15 or 16 and 18) flank the eccentricity values of the related Sommerfeld ellipses, | m | = l (cases 14 or 17, respectively). The inspection of Tables I and II then raises the question, why is there agreement in the extreme cases of minimum and maximum | m | and disagreement otherwise?
## IV. DISCUSSION
By Eq. (9) the eccentricity of an amended Sommerfeld orbit depends only on the principal and angular quantum numbers, n and l , not on the magnetic quantum number m . The latter determines different orientations of a given 〈 nlm 〉 Sommerfeld ellipse (space quantization) for
= 0, 1
Table I. Eccentricities of Sommerfeld orbits for principal quantum number n = 1 2 3 , , with parabolic quantization conditions { n n ξ η | m |} (left side) and spherical quantization conditions 〈 nlm 〉 (right side). The eccentricity ε ξη m | | was calculated with Eq. (7) and ε nlm with Eq. (9). The symbol / ∈ here denotes non-existence.
| Epstein, Schwarzschild parabolic coordinates quantum n = n ξ + n η + | m | | Epstein, Schwarzschild parabolic coordinates quantum n = n ξ + n η + | m | | Epstein, Schwarzschild parabolic coordinates quantum n = n ξ + n η + | m | | amended Sommerfeld model spherical coordinates n = n r + n θ + n ϕ = n r + n ϕ = m | amended Sommerfeld model spherical coordinates n = n r + n θ + n ϕ = n r + n ϕ = m |
|------------------------------------------------------------------------------|------------------------------------------------------------------------------------------|-------------------------------------------------------------------------------------------|----------------------------------------------------------------------------------------------|----------------------------------------------------------------------------------------|
| case | case | case | | |
| ↓ | n { n ξ n η | m |} | ε ξη | m | ______________________________________________________________________________ | < n m > | ε n m |
| 1 | 1 {1 0 0} & {0 1 0} | 1 | < 1 0 0 > | 1 |
| 2 | 1 {0 0 1} ______________________________________________________________________________ | 0 | ∉ | - - - |
| 3 | 2 {1 1 0} 2 {2 0 0} & {0 2 0} | 1 | < 2 0 0 > | 1 |
| 4 | 2 {1 0 1} & {0 1 1} | 1 / 2 √ 2 | < 2 1 m > m = 0, 1 | 1 / 2 √ 2 |
| 5 | 2 {0 0 2} | 0 | ∉ ______________________________________________________________________________ | - - - |
| 6 | 3 {2 1 0} & {2 1 0} | 1 = 1 / 3 √ 9 | < 3 0 0 > | 1 |
| 7 | 3 {1 1 1} | 1 / 3 √ 8 | | 1 / √ 7 |
| 8 9 | 3 3 {2 0 1} & {0 2 1} | 1 / 3 √ 6 | < 3 1 m > m = 0, 1 | 3 |
| 10 | 3 {1 0 2} & {0 1 2} | 1 / 3 √ 3 | < 3 2 m > | 1 / 3 √ 3 |
| 11 | 3 {0 0 3} | 0 | m = 0, 1, 2 ∉ ______________________________________________________________________________ | - - - |
Table II. Eccentricities of Sommerfeld orbits for principal quantum number n = 4 with parabolic quantization conditions { n n ξ η | m |} (left side) and spherical quantization conditions 〈 nlm 〉 (right side). The eccentricity ε ξη m | | was calculated with Eq. (7) and ε nlm with Eq. (9).
| Epstein, Schwarzschild parabolic coordinates quantum n = n + n + | m | | Epstein, Schwarzschild parabolic coordinates quantum n = n + n + | m | | Epstein, Schwarzschild parabolic coordinates quantum n = n + n + | m | | Epstein, Schwarzschild parabolic coordinates quantum n = n + n + | m | | Epstein, Schwarzschild parabolic coordinates quantum n = n + n + | m | | Epstein, Schwarzschild parabolic coordinates quantum n = n + n + | m | | Epstein, Schwarzschild parabolic coordinates quantum n = n + n + | m | | Epstein, Schwarzschild parabolic coordinates quantum n = n + n + | m | | Epstein, Schwarzschild parabolic coordinates quantum n = n + n + | m | | Epstein, Schwarzschild parabolic coordinates quantum n = n + n + | m | | amended Sommerfeld model spherical coordinates n = n r + n θ + n ϕ = n r + n ϕ = m | amended Sommerfeld model spherical coordinates n = n r + n θ + n ϕ = n r + n ϕ = m | amended Sommerfeld model spherical coordinates n = n r + n θ + n ϕ = n r + n ϕ = m | amended Sommerfeld model spherical coordinates n = n r + n θ + n ϕ = n r + n ϕ = m |
|--------------------------------------------------------------------------|--------------------------------------------------------------------------|--------------------------------------------------------------------------|--------------------------------------------------------------------------|--------------------------------------------------------------------------|--------------------------------------------------------------------------|--------------------------------------------------------------------------|--------------------------------------------------------------------------|--------------------------------------------------------------------------|--------------------------------------------------------------------------|----------------------------------------------------------------------------------------|----------------------------------------------------------------------------------------|----------------------------------------------------------------------------------------|----------------------------------------------------------------------------------------|
| sum case | sum case | sum case | sum case | sum case | sum case | sum case | sum case | sum case | ε | | | > | ε n m |
| ↓ | n | { n | n | η | m |} | | | | | ξη | m | | | < n | m | |
| | 4 | {4 | | 0 0} | & | {0 | 4 | 0} | 1 | = 1 / 4 √ 16 | | | |
| 12 | 4 | {3 | | 1 0} | & | {1 | 3 | 0} | 1 | | <4 | 0 0> | 1 |
| | 4 | {2 | 2 | 0} | | | | | 1 | | | | ______________________________________________________________________________ |
| 13 | 4 | {2 | 1 | 1} & | | {1 | 2 | 1} | 1 / 4 ( √ | 6 + √ 2) ≈ 1 | | / 4 √ | 14.93 = 0.97 |
| 14 | 4 | | | | | | | | | <4 | | 1 m > | 1 / 4 √ 14 = 0.94 |
| 15 | 4 | {3 | 0 | 1} | & | {0 | 3 | 1} | | 1 / 4 √ 12 = 0.87 | | ______________________________________________________________________________ | |
| 16 | 4 | {1 | 1 | 2} | | | | | 1 / 4 √ 12 = | 0.87 | | | |
| 17 | 4 | | | | | | | | | <4 | | 2 m > | 1 / 4 √ 10 = 0.79 |
| 18 | 4 | {2 | 0 | 2} | & | {0 | 2 | 2} | 1 | / 4 √ 8 = 0.71 | | ______________________________________________________________________________ | |
| 19 | 4 | {1 | 0 | 3} & | {0 | | 1 | 3} | 1 / 4 √ 4 | <4 | | 3 m > | 1 / 4 √ 4 |
| | | | | | | | | | = 0.50 | ______________________________________________________________________________ | | | = 0.50 |
$$| m | = 0, 1, \dots, l.$$
An apt expression for m would be ' lean quantum number', as it affects the lean angle,
$$\theta = a r c s i n \frac { | m | } { \sqrt { l ( l + 1 ) ] } },$$
by which the minor axis of a Sommerfeld ellipse of given major axis, 2 a = 2 n r 2 B , and focal axis, 2 f = 2 ( l l +1) r B , leans away from the coordinate z axis. This aligns the minor axis of an 〈 nl 0 〉 orbit parallel to the z axis, θ min = 0, and gives the largest lean angle of an 〈 nl m | |〉 orbit,
$$\theta _ { m a x } = a r c s i n \frac { n - 1 } { \sqrt { n ( n - 1 ) } } < \frac { \pi } { 2 },$$
when
$$| m | = l = n - 1.$$
The latter case is the only instance where the angular quantum number l of the orbit is fully known from the magnetic quantum number | m | in the spherical quantization conditions 〈 nlm 〉 under the constraint of Eq. (10).
Different constraints on the quantum numbers hold for quantization in parabolic coordinates. An analysis of them is beyond the scope of this paper. However, four patterns are noticeable: (1) If m = 0 in { n , n ξ η , | m |} , then ε ξη m | | = 0 and the orbit is a Coulomb oscillator (cases 1, 3, 6, 12).
(2) If | m | is maximal in { n , n ξ η , | m |} , that is { 1 0 , , n - } 1 or { 0 1 , , n - } 1 , then one can assign a 'good' (integer) angular quantum number l = n -1 to the orbit, in agreement with its role in an 〈 n, l, m 〉 orbit (cases 1, 4, 10, 19). The eccentricity then is
$$\varepsilon _ { \xi \eta | m | } \{ 1, 0, n - 1 \} = \varepsilon _ { n l m } \langle n, n - 1, m \rangle = \frac { 1 } { n } \sqrt { n },$$
and the angular momentum L of the { n , n ξ η , | m |} orbit is solely determined by | m | . (3) Orbits { 1 1 , , n - } 2 have a fractional value of l , but such that √ l ( l +1) = n -2 = | m | (cases 3, 7, 16). Again, their angular momentum L is solely determined by | m | .
(4) In other cases (9, 15, 13) not only | m | but the parabolic quantum numbers n ξ and n η too, contribute to the angular momentum L via a fractional angular quantum number l with l ( l +1) = 3, = 2, and glyph[similarequal] 1, respectively. The contributing role of parabolic quantum numbers can be seen in case 15, achieving the same orbit eccentricity ε ξη m | | as in case 16.
Epstein, Schwarzschild amended Sommerfeld model parabolic coordinates spherical coordinates
n
=
n
ξ
+
n
η
+ |
m
|
n
=
nr
+
n
θ
+
n
ϕ
=
nr
+
n
ϕ
=
m
| case | { n ξ n η | m |} | ε ξη | m | < n | m > | ε n n | < n | m | > ε n | m | |
|----------|-------------------------------------|--------------------------------|----------------------------------------------------------------------------------------------|---------------------------------|-----------------------------------------------|
| 7 | {1 1 1} | 1 / 3 √ 8 | | < 3 0 1 > | 1 / 3 √ 8 |
| 8 9 | {2 0 1} & {0 2 1} | < 3 1 / 3 √ 6 | 1 m > 1 / 3 √ 7 | < 3 1 1 > < 3 2 1 > | 1 / 3 √ 7 1 / 3 √ 6 |
| 13 14 | {2 1 1} & {1 2 1} | 1 / 4 √ 14.94 < 4 | ______________________________________________________________________________ 1 m > 1 / 4 √ | < 4 0 1 > 14 < 4 | 1 / 4 √ 15 1 1 > 1 / 4 √ 14 |
| 15 - - - | {3 0 1} & {0 3 1} - - - - - - - - - | 1 / 4 √ 12 - - - - - - - - - - | - - - - - | < 4 2 1 > < 4 3 1 > - - - - - - | 1 / 4 √ 13 1 / 4 √ 12 - - - - - - - - - - - - |
| 16 | {1 1 2} | 1 / 4 √ 12 | - - - - - - - | < 4 1 2 > | 1 / 4 √ 12 |
| 17 18 | {2 0 2} & | < 4 1 / 4 √ 8 | 2 m > 1 / 4 √ 10 | < 4 2 2 > < 4 3 2 > | 1 / 4 √ 10 1 / 4 √ 8 |
Table III. Eccentricities of Sommerfeld orbits for the cases from Table I and II where disagreement is found, with parabolic quantization conditions [left side, Eq. (7)] and spherical quantization conditions [center columns, Eq. (9)]. The right side shows eccentricities of projected orbits (denoted by underline) obtained with Eq. (15).
## V. VISUALIZATION
Imagine the projection of a leaning 〈 nl m | |〉 Sommerfeld orbit onto the xy plane. The minor axis of the projected ellipse, denoted by underline as 〈 nl m | |〉 , is shortened in proportion of m/l . 3 Combined with Eq. (9) this renders its eccentricity as
$$\varepsilon _ { \underline { n } l | \underline { m } } | = \frac { 1 } { n } \sqrt { n ^ { 2 } - | m | ( l + 1 ) }.$$
The expression agrees with the eccentricity in parabolic quantization, Eq. (7), for maximum | m | , Eq. (13), as obtained in Eq. (14).
For the minimum magnetic quantum number, | m | = 0, Eq. (15) gives
$$\varepsilon _ { \xi \eta | m | } \{ n _ { \xi }, n _ { \eta }, 0 \} = \varepsilon _ { \underline { n } \underline { l } | \underline { m } \underline { l } } \langle n, l, 0 \rangle = 1.$$
In this instance the 〈 nlm 〉 Sommerfeld ellipses don't lean, so their projection onto the xy plane gives line ellipses of ε = 1, as expressed in Eq. (16), which explains those values in Tables I and II.
In the remaining cases, for orbits with common n and
$$| m | = l,$$
agreement of ε ξη m | | and ε nlm from Eqs. (7) and (9) is close but not exact, as Table III shows in the left and center columns. Instead, the eccentricity of the leaning 〈 nlm 〉 orbits (center column) agrees with that of their projected ellipses, 〈 nl m | |〉 (right column),
$$\varepsilon _ { n l m } = \varepsilon _ { \underline { n } \left | \underline { l } \underline { m } \right | },$$
cases 8, 14, and 17, where the lean angle of a given 〈 nlm 〉 orbit is maximal, Eq. (17).
Next we compare the eccentricity of orbits with parabolic quantization conditions { n n ξ η | m |} (left column) with that of projected ellipses 〈 nl m | |〉 (right column) for common n and | m | . If
$$| m | < \underline { l }$$
(cases 9, 15 and 18), agreement of eccentricity ε ξη m | | from Eq. (7) can be found with one of the ε nl m | | solutions of Eq. (15).
In the opposite situation,
$$| m | > \underline { l }$$
Eq. (15) still provides exact solutions, ε ξη m | | = ε nl m | | , in two cases (7 and 16). The only exception is case 13 which is numerically slightly off, ε ξη m | | glyph[similarequal] ε nl m | | . The three cases are highlighted by underlined and italic notation, 〈 nl m | |〉 , on the right side of Table III. Note that Eq. (20) does not refer to the angular quantum number l from quantization in spherical coordinates, as | m > l | would violate the constraint of Eq. (10). Instead Eq. (15), with (underlined) l from Eq. (20), expresses the eccentricity of orbits { 111 } , { 112 } and { 211 } which obey constraints of quantum numbers in parabolic quantization but not those of spherical quantization.
## VI. CONCLUSION
The historical comparison of orbit eccentricity, based on quantization conditions in parabolic coordinates and, respectively, spherical coordinates for traditional Sommerfeld orbits, gives disagreement in all cases except for the circular Bohr orbits. This suggests a large dissimilarity between the two sets of orbits. When instead the comparison is done with the amended Sommerfeld model, the situation drastically changes: The previous cases of agreement disappear with the rejection of circular orbits whereas the cases of previous disagreement become exact or close. The reason for only approximate agreement of some eccentricities originates with different constraints on quantum numbers in different coordinate systems. All orbits with parabolic quantization conditions can be visualized as projected ellipses from leaning 〈 nlm 〉 Sommerfeld orbits onto the xy plane.
The trouble with orbits continues-but less so than previously thought.
## ACKNOWLEDGMENT
I thank Duane Siemens for valuable discussions and critique.
1 M. Bucher, 'Rise and premature fall of the old quantum theory,' arXiv:0802.1366v1
2 A. Duncan and M. Janssen, 'The trouble with orbits: The Stark effect in the old and the new quantum theory,' arXiv:1404.5333v1
3 The choice of the projection ratio m/l is ad hoc , justified only by its numerical result and the potential of visualization. It differs from the ratio m/ √ l ( l +1) for the lean angle, Eq. (11). | null | [
"Manfred Bucher"
] | 2014-06-03T19:36:41+00:00 | 2014-10-14T21:58:24+00:00 | [
"physics.hist-ph"
] | Less trouble with orbits: The semi-classical hydrogen problem in parabolic and spherical coordinates | Historically, the eccentricity of Sommerfeld orbits from quantization
conditions in either parabolic or spherical coordinates was found to differ in
almost all cases. To do the orbit comparison correctly, one must use amended
instead of traditional Sommerfeld orbits. |
1408.0245v1 | ## Electronic Transport in Graphene with Aggregated Hydrogen Adatoms
Fernando Gargiulo, 1 Gabriel Autès, 1 Naunidh Virk, 1 Stefan Barthel, 2, 3 Malte Rösner, 2, 3 Lisa R. M. Toller, 1 Tim O. Wehling, 2, 3 and Oleg V. Yazyev 1
1 Institute of Theoretical Physics, Ecole Polytechnique Fédérale de Lausanne (EPFL), CH-1015 Lausanne, Switzerland 2 Institut für Theoretische Physik, Universität Bremen, Otto-Hahn-Allee 1, D-28359 Bremen, Germany 3 Bremen Center for Computational Materials Science, Am Fallturm 1a, D-28359 Bremen, Germany (Dated: August 4, 2014)
Hydrogen adatoms and other species covalently bound to graphene act as resonant scattering centers affecting the electronic transport properties and inducing Anderson localization. We show that attractive interactions between adatoms on graphene and their diffusion mobility strongly modify the spatial distribution, thus fully eliminating isolated adatoms and increasing the population of larger size adatom aggregates. Our scaling analysis shows that such aggregation of adatoms increases conductance by up to several orders of magnitude and results in significant extension of the Anderson localization length in the strong localization regime. We introduce a simple definition of the effective adatom concentration x glyph[star] , which describes the transport properties of both random and correlated distributions of hydrogen adatoms on graphene across a broad range of concentrations.
PACS numbers: 72.80.Vp, 73.20.Hb, 71.23.An
Graphene has unveiled a plethora of unconventional transport phenomena [1-3], such as the universal minimal conductivity [4], Klein tunneling [5] and the anomalous quantum Hall effect [6, 7]. On the applied side, graphene is interesting because of its exceptionally high charge-carrier mobility, which is typically limited by the presence of various types of disorder. Resonant scattering impurities, such as chemical functionalization defects [8] and dislocations [9], show the most pronounced effects on charge-carrier transport in graphene. Hydrogen adatoms represent a prototypical resonant scattering impurity, which can be experimentally introduced in a controlled fashion [10] and allows for a simple theoretical description [11]. A hydrogen adatom covalently binds to a single carbon atom of graphene resulting in rehybridization into the sp 3 state, thus effectively removing that site from the honeycomb network of p z orbitals. This gives rise to a zero-energy state localized around the defect, and results in the resonant scattering of charge carriers.
At a fundamental level, the classical scaling theory of Anderson transition predicts complete localization of the electronic spectrum in two dimensions (2D), regardless of the amount of disorder [12]. For hydrogenated graphene, a model based on massless Dirac fermions with δ -function-like potentials confirms this prediction of the unitary class, though in 2D systems localization lengths can be strongly energy-dependent and, eventually, very large [13]. However, no unanimous consensus has been reached since experiments on hydrogenated graphene point towards metal-insulator transition, theoretically justified by the presence of electron-hole puddles (2D percolation class) [14-17].
Early works treating finite concentrations of resonant impurities in graphene assumed that the total scattering cross-section deviates little from the incoherent addition of the individual cross-sections, for example in the Boltzmann equation framework [18]. This picture is valid for low defect concentrations, low charge-carrier densities and random adatom distributions. A better description requires including the effect of coherent superposition of wavefunctions scattered by distinct adatoms [8, 19-21]. This is particularly important when impurities are in proximity to each other, with the limiting case being the formation of compact clusters in which hydrogen adatoms populate neighboring carbon atoms [22-24]. Indeed, the overall short-range attractive interaction between individual hydrogen adatoms on graphene [25, 26], combined with their relatively high diffusion mobility at room temperature [27-29], suggests a high degree of spatial correlation between adatoms.
In this Letter, we address the effects of spatial correlation of resonant impurities on electronic transport in graphene. The equilibrium configurations of hydrogen adatoms on graphene, obtained by means of MonteCarlo simulations, show a strong tendency towards aggregation into small clusters essentially eliminating isolated adatoms. Electronic transport properties investigated using the Landauer-Büttiker approach complemented with kernel polynomial method calculations show that aggregation dramatically increases both the conductivity and the localization length. We propose a unified framework to account for the effects of spatial correlation of resonant scattering centers on electronic transport in graphene.
Upon adsorption, a hydrogen adatom covalently binds to a single carbon atom of graphene changing its hybridization state to sp 3 and its coordination sphere to tetrahedral as shown in Fig. 1(a) [30]. The covalent binding of a second adatom to the nearest neighbor carbon atom partially releases the elastic energy due to the change of coordination sphere, thus resulting in effective attractive interaction between adatoms [25, 26]. This suggests that the interaction between adatoms can be
Figure 1: (a) Atomic structure of a hydrogen adatom covalently bound to graphene. (b) Structures of small clusters of hydrogen adatoms used for fitting the pair potential of Eq. (1) (c) Predicted energy ˜ E of aggregation of hydrogen adatoms as a function of aggregation energy E DFT calculated from first principles for the set of adatom clusters shown in panel (b).
accurately described using a short-ranged pair potential. In our study, we expand an Ising-like interaction energy E up to the second nearest-neighbor term
$$E = \gamma _ { 1 } \sum _ { \langle i, j \rangle } s _ { i } s _ { j } + \gamma _ { 2 } \sum _ { \langle \langle i, j \rangle \rangle } s _ { i } s _ { j }, \quad \ \ ( 1 ) \quad \ s u b _ { \text{tain} }$$
where γ 1 and γ 2 are the corresponding first and second nearest-neighbor parameters. Here, s i = 1 if a carbon atom i is populated by an adatom, otherwise s i = 0 . Parameters γ 1 and γ 2 are obtained by fitting the interaction energies of adatoms calculated from first principles [31] for a set of small adatom aggregates shown in Fig. 1(b). Under the assumption of single-side functionalized graphene, the obtained parameters γ 1 = -1 182 . eV and γ 2 = 0 484 . eV [32] signify a considerable firstnearest-neighbor attraction alongside a weaker secondnearest-neighbor repulsion. The excellent agreement between the interaction energy ˜ E , estimated using the fitted γ 1 and γ 2 , and the first-principles values E DFT , confirms the applicability of the short-range pair potential form (1) for describing small clusters [Fig. 1(c)].
In order to assess the effect of interaction between adatoms on their spatial distribution and transport properties we perform Monte-Carlo simulations using the introduced pair potential (1). The simulations employ a Monte-Carlo move based on the displacement of a randomly selected adatom to a random unoccupied carbon atom in combination with the Metropolis acceptance criterion [33]. We considered models containing up to N C = 10 6 carbon atoms ( 165 × 165 nm ) and adatom con2 centrations x = N /N H C ranging from 0.1% to 10%. All simulations have been performed at T = 300 K . A representative configuration of randomly distributed adatoms [Fig. 2(a)] is compared with a configuration obtained from a Monte-Carlo simulation [Fig. 2(b)] at x = 5% .
Further details are revealed by comparing the cluster size distributions P n ( ) [Fig. 2(c)], with adatoms populating neighboring carbon atoms being assigned to the same cluster. In the case of a random distribution most adatoms are isolated ( n = 1 ), while the occurrence of clusters ( n > 1 ) is merely a probabilistic effect. In contrast, no isolated adatoms are found in the presence of interactions, with the most abundant species being adatom dimers ( n = 2 ). The size distribution for the correlated case shows a longer tail with a significant probability of observing up to n = 6 clusters. The dependence of P n ( ) on adatom concentration x is relatively weak [Fig. 2(d)].
We now focus on electronic and transport properties calculated using the nearest-neighbor tight-binding Hamiltonian for p z orbitals
$$H = - t \sum _ { \langle i, j \rangle } [ c _ { i } ^ { \dagger } c _ { j } + \text{h.c.} ]$$
with the hopping integral constant t = 2 7 . eV [2]. An adsorbed hydrogen atom is modeled by excluding the p z orbital of the carbon atom to which it is bound as a consequence of sp 3 hybridization, making it similar to a vacancy defect [27, 30, 34]. We stress that adatoms do not introduce coupling between the sites owing to the same sublattice of the bipartite lattice of graphene, thus maintaining electron-hole symmetry of the electronic spectrum. Figures 3(a,b) compare the density of states (DOS) of graphene with random and correlated distributions of adatoms at different concentrations. In the case of a ran-
Figure 2: (Color online) Representative configurations of (a) randomly distributed and (b) correlated hydrogen adatoms on graphene at x = 5% concentration. (c) Comparison of random and correlated adatoms cluster size distributions P x ( ) at x = 5% concentration. (d) Cluster size distributions P x ( ) of correlated hydrogen adatoms at different concentrations. All correlated configurations are obtained by means of MonteCarlo simulations carried out at T = 300 K.
Figure 3: (Color online) Density of states of graphene in the presence of (a) randomly distributed and (b) correlated hydrogen adatoms at different concentrations. Local density of states (LDOS) on carbon atoms in the vicinity of (c) an isolated hydrogen adatom and (d) a dimer of hydrogen adatoms. In panels (c,d) LDOS referred to as 'analyt.' have been obtained using the analytical Green's function calculations [36].
dom distribution one observes a strong peak at E = 0 due to the resonant modes originating from isolated adatoms [19, 34]. The corresponding wave function is localized on the sublattice opposite to that of the carbon atom binding the adatom and decays from the defect position [35]. At high adatom concentrations, x > 1% , the E = 0 peak is accompanied by flat density regions at higher energies with a noticeable overall renormalization of the DOS, in agreement with previous calculations [34]. In comparison, the DOS calculated for the correlated impurity configurations shows a less intense peak at E = 0 and an increased weight for -0 9 . t < E < 0 9 . t that is more evident at higher concentrations. This change is a direct consequence of different cluster size distributions. The dominant cluster type in the case of the correlated adatom distribution is the dimer ( n = 2 ), which is known to be non-resonant [11, 24, 36], meaning that no localized states emerge at any energy. The local density of states (LDOS) calculated on neighboring atoms of an isolated single adatom shows a singularity at E = 0 [Fig. 3(c)]. In contrast, an enhancement of the LDOS in a broad energy region -t < E < t is observed on certain carbon atoms in the vicinity of the adatom dimer [Fig. 3(d)]. The numerical LDOS for selected impurities compared to the results of analytical Green's function calculations shows no discrepancies [36].
Based on these observations we conclude that the residual peak at E = 0 in the case of the correlated distribution is due to the n > 2 adatom aggregates, which break the bipartite symmetry of graphene (that is, populate different number of sites in the two sublattices). This gives rise to resonant modes at E = 0 akin to isolated adatoms. All oddn aggregates and certain configurations of evenn clusters lead to resonant modes at E = 0 . Judging on the cluster size distributions [Fig. 2(d)], the largest contribution comes from adatom trimers. This suggests that adatom aggregation has strong effects on the electronic transport properties that we investigate by performing a scaling analysis of conductivity g using the Landauer-Büttiker approach [37]. In this approach, the conductance G E ( ) is given as G E ( ) = G T E 0 ( ) , where G 0 is the conductance quantum and T E ( ) is the transmission probability across the scattering region at energy E . We assume a two-terminal device configuration with a scattering region of width W = 40 nm perpendicular to the current direction, and of variable length 1 nm < L < 60 nm. The scattering region is attached to pristine graphene contacts and populated by adatoms according to concentrations and statistical distributions discussed above. Further details of our methodology are given in the Supplementary Material [31].
The characteristic functional laws for the conductivity g = G × L/W in the ballistic, diffusive and localized transport regimes are g ∝ L , g = const and g ∝ exp( -L/ξ loc ) , respectively, where ξ loc is the localization length. In the localized regime ln g ( ) follows a broad positively-skewed distribution, which means that g can show strong fluctuations depending on the exact configuration of defects, especially in the presence of strong localization [38]. An estimate of the mean value for such a distribution is given by the typical conductivity g typ = exp ln( 〈 g ) 〉 [38, 39]. In our scaling analysis, g typ has been obtained averaging over 9600 disorder realizations. Figure 4(a,b) shows g typ as a function of scattering region length L at different energies for the random and correlated impurity distributions, both at x = 5% concentration. We observe a short transition from ballistic to diffusive and subsequently to localized regime within the first 10 nm. The crossover lengths are expected to be of the order of the elastic mean free path ξ el and localization length ξ loc , respectively. The general trend is that the localized regime is accentuated at low energy, whereas at higher energy the onset of exponential decay occurs at larger L and the absolute slope of the conductance curves is smaller. The scaling of g also depends strongly on the impurity concentration. For low adatom concentrations ( x < ∼ 0 5% . ) the onset of the localized regime is only observable in the vicinity of the Dirac energy (see Supplementary Material [31] for complete results).
We stress that the conductance curves vary smoothly in the whole range of E and x , never showing singularities which would indicate a phase transition such as the metal-insulator transition (MIT). Thus, we ascribe the non-observance of the localized regime to an insufficient scattering region length of our model, which is shorter than ξ loc for some choices of E and x . From Fig-
Figure 4: (Color online) (a,b) Averaged conductivity g typ for the random and correlated adatoms distributions, respectively, as a function of scattering region length L at different energies E and at x = 5% concentration. (c,d) Conductivity g as a function of energy E calculated using the kernel polynomial method for both adatom distributions at concentration x = 0 5% . and x = 5 0% . , respectively. The dashed lines show the minimum conductivity g = 4 /π .
ure 4(a,b) it follows that the presence of spatial correlation between adatoms enhances the conductance by up to five orders of magnitude in the vicinity of the Dirac point ( E = 2 7 . meV). This is a direct consequence of the suppressed weight of low-energy resonant states, as explained above. A closely related effect is a significant increase of the localization length ξ loc at all energies upon adatom aggregation. ξ loc was obtained by fitting conductance curves to the expected law g typ ∝ exp( -L/ξ loc ) [40], and is shown in Fig. 5(a). At x = 5% , the localization length is well defined for the entire range of investigated energies -1 eV < E < 1 eV, whereas at lower concentrations it is well defined only in proximity of the Dirac point (see Supplementary Material [31]). However, as long as a finite positive ξ loc can be determined, it proves to be up to an order of magnitude larger for the correlated adatom distribution compared to the random case.
Our Landauer-Büttiker results are complemented by the Kubo formula calculations based on the kernel polynomial method (KPM) [41]. This approach allows for a direct calculation of the DOS and conductivity independent of sample geometry and contacts. Figure 4(c,d) shows the conductivity g for adatom concentrations x = 0 5% . and x = 5% , and both random and correlated adatom distributions. In comparison to the LandauerBüttiker ansatz, the results are similar at both concentrations, confirming the enhancement of conductivity upon aggregation of adatoms. The only difference appears when the charge-carrier energy E falls into the resonant
Figure 5: (Color online) (a) Localization length ξ loc as a function of charge-carrier energy E for the case of random and correlated adatom distribution at x = 5% . (b) Charge-carrier localization length ξ loc at low energy ( E = 10 -3 t = 2 7 . meV) as a function of simple concentration x and effective concentration x glyph[star] for the random and correlated distributions.
peak region where the KPM conductivity exceeds the Landauer-Büttiker results. This can be proved to be an effect of the increased DOS due to the formation of an impurity band. While the KPM conductivity reflects the excess of states at the Fermi energy, in the LandauerBüttiker ansatz the number of charge carriers is limited by the DOS of the pristine graphene leads, and subsequently the conductivity is determined only by the mean free path and the localization length of the scattering region. We note that upon doping the leads, the LB conductivity at low energies increases while the localization length is unaffected (see Supplementary Material [31]).
On the basis of the identification of resonant adatom clusters we introduce an effective concentration
$$x ^ { * } = \frac { 1 } { N _ { C } } \sum _ { i } N _ { i } | n _ { i } ^ { A } - n _ { i } ^ { B } |,$$
where N i is the number of instances of the cluster configuration i , and n A i ( n B i ) is the number adatoms bound to carbon atoms in sublattice A(B) in this configuration. The main contribution to x glyph[star] comes primarily from adatom trimers as discussed above. We assume that the contribution of non-resonant clusters to the total scattering cross-section can be neglected. It should be stressed that this approach accounts for the diverse scattering effect of different clusters. In order to validate the applicability of effective concentration x glyph[star] , we compare ξ loc ( x ) with ξ loc ( x glyph[star] ) at low charge-carrier energy for the correlated impurity case [Fig. 5(b)]. One can see that replacing x with effective concentration x glyph[star] brings ξ loc ( x glyph[star] ) evaluated for the case of the correlated impurity distribution in good agreement with ξ loc ( x ) calculated for the random distribution. The agreement is particularly good in the low concentration regime where randomly distributed impurities consist almost exclusively of isolated monomers. Deviations at higher adatom concentrations can be ascribed to the inter-cluster interference effects, which become important at the reduced average distance between
the clusters.
To summarize, hydrogen adsorbed on graphene has a strong tendency toward aggregation, resulting in the formation of small clusters and fully eliminating isolated adatoms. Some of the larger clusters, notably trimers, are responsible for the residual resonant scattering, but the overall conductance and localization length dramatically increases upon aggregation. Within the range of parameters investigated in our work, we find no metal-insulator transition, with the graphene spectrum being fully localized. The predicted effects of adatom aggregation can be investigated experimentally by varying the temperature regimes, since the diffusion of hydrogen adatoms occurring at normal conditions can be effectively suppressed at low temperatures. Alternatively, time-resolved transport measurements should evince a rise in conductivity upon formation of adatom clusters.
We would like to thank C. W. J. Beenakker, P. Lugan and V. Savona for discussions. This work was supported by the Swiss National Science Foundation (Grant No. PP00P2\_133552).
- [1] A. K. Geim and K. S. Novoselov, Nature Mater. 6 , 183 (2007).
- [2] A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81 , 109 (2009).
- [3] S. Das Sarma, S. Adam, E. H. Hwang, and E. Rossi, Rev. Mod. Phys. 83 , 407 (2011).
- [4] J. Tworzydło, B. Trauzettel, M. Titov, A. Rycerz, and C. W. J. Beenakker, Phys. Rev. Lett. 96 , 246802 (2006).
- [5] M. I. Katsnelson, K. S. Novoselov, and A. K. Geim, Nature Phys. 2 , 620 (2006).
- [6] Y. Zhang, Y.-W. Tan, H. L. Stormer, and P. Kim, Nature (London) 438 , 201 (2005).
- [7] K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, M. I. Katsnelson, I. V. Grigorieva, S. V. Dubonos, and A. A. Firsov, Nature (London) 438 , 197 (2005).
- [8] T. O. Wehling, S. Yuan, A. I. Lichtenstein, A. K. Geim, and M. I. Katsnelson, Phys. Rev. Lett. 105 , 056802 (2010).
- [9] F. Gargiulo and O. V. Yazyev, Nano Lett. 14 , 250 (2014).
- [10] Z. H. Ni, L. A. Ponomarenko, R. R. Nair, R. Yang, S. Anissimova, I. V. Grigorieva, F. Schedin, P. Blake, Z. X. Shen, E. H. Hill, K. S. Novoselov, and A. K. Geim, Nano Lett. 10 , 3868 (2010).
- [11] T. O. Wehling, A. V. Balatsky, M. I. Katsnelson, A. I. Lichtenstein, K. Scharnberg, and R. Wiesendanger, Phys. Rev. B 75 , 125425 (2007).
- [12] B. Kramer and A. MacKinnon, Rep. Prog. Phys. 56 , 1469 (1993).
- [13] F. Evers and A. D. Mirlin, Rev. Mod. Phys. 80 , 1355 (2008).
- [14] A. Bostwick, J. L. McChesney, K. V. Emtsev, T. Seyller, K. Horn, S. D. Kevan, and E. Rotenberg, Phys. Rev. Lett. 103 , 056404 (2009).
- [15] Y. Song, H. Song, and S. Feng, J. Phys.: Condens. Mat-
ter 23 , 205501 (2011).
- [16] R. Jayasingha, A. Sherehiy, S.-Y. Wu, and G. U. Sumanasekera, Nano Lett. 13 , 5098 (2013).
- [17] S. Adam, S. Cho, M. S. Fuhrer, and S. Das Sarma, Phys. Rev. Lett. 101 , 046404 (2008).
- [18] N. Shon and T. Ando, J. Phys. Soc. Jpn. 67 , 2421 (1998).
- [19] T. O. Wehling, M. I. Katsnelson, and A. I. Lichtenstein, Phys. Rev. B 80 , 085428 (2009).
- [20] A. Ferreira, J. Viana-Gomes, J. Nilsson, E. R. Mucciolo, N. M. R. Peres, and A. H. Castro Neto, Phys. Rev. B 83 , 165402 (2011).
- [21] A. Cresti, F. Ortmann, T. Louvet, D. Van Tuan, and S. Roche, Phys. Rev. Lett. 110 , 196601 (2013).
- [22] J. J. Palacios, J. Fernández-Rossier, and L. Brey, Phys. Rev. B 77 , 195428 (2008).
- [23] N. Leconte, D. Soriano, S. Roche, P. Ordejon, J.-C. Charlier, and J. J. Palacios, ACS Nano 5 , 3987 (2011).
- [24] G. Trambly de Laissardière and D. Mayou, Phys. Rev. Lett. 111 , 146601 (2013).
- [25] D. W. Boukhvalov, M. I. Katsnelson, and A. I. Lichtenstein, Phys. Rev. B 77 , 035427 (2008).
- [26] Y. Lin, F. Ding, and B. I. Yakobson, Phys. Rev. B 78 , 041402 (2008).
- [27] O. V. Yazyev, Phys. Rev. Lett. 101 , 037203 (2008).
- [28] C. P. Herrero and R. Ramírez, Phys. Rev. B 79 , 115429 (2009).
- [29] M. Moaied, J. A. Moreno, M. J. Caturla, and J. J. Palacios, arXiv:1405.3165 (2014).
- [30] O. V. Yazyev and L. Helm, Phys. Rev. B 75 , 125408 (2007).
- [31] See Supplemental Material at [URL will be inserted by publisher] for detailed description of the methodology and additional calculations.
- [32] In the case of hydrogen adsorption on both sides of graphene the fitted parameters γ 1 = -1 461 . eV and γ 2 = 0 342 . eV. The resulting cluster size distributions, however, are very close to those obtained for single-side adsorption [31].
- [33] N. Metropolis, A. W. Rosenbluth, M. N. Rosenbluth, A. H. Teller, and E. Teller, J. Chem Phys 21 , 1087 (1953).
- [34] V. M. Pereira, F. Guinea, J. M. B. Lopes dos Santos, N. M. R. Peres, and A. H. Castro Neto, Phys. Rev. Lett. 96 , 036801 (2006).
- [35] D. M. Basko, Phys. Rev. B 78 , 115432 (2008).
- [36] F. Gargiulo, G. Autès, and O. V. Yazyev, to be published (2014).
- [37] M. Büttiker, Y. Imry, R. Landauer, and S. Pinhas, Phys. Rev. B 31 , 6207 (1985).
- [38] D.-H. Choe and K. J. Chang, Nano Lett. 12 , 5175 (2012).
- [39] A. Uppstu, Z. Fan, and A. Harju, Phys. Rev. B 89 , 075420 (2014).
- [40] P. A. Lee and T. V. Ramakrishnan, Rev. Mod. Phys. 57 , 287 (1985).
- [41] A. Weiße, G. Wellein, A. Alvermann, and H. Fehske, Rev. Mod. Phys. 78 , 275 (2006)
## Supplementary Material for 'Electronic Transport in Graphene with Aggregated Hydrogen Adatoms'
## A. First-principles calculations of the energies of hydrogen adatom clusters
First-principles calculations of the interaction energies of hydrogen adatoms on graphene have been performed within the density functional theory (DFT) framework employing the generalized gradient approximation (GGA) to the exchange-correlation functional [S1]. Ultrasoft pseudopotentials [S2] for carbon and hydrogen atoms have been used in combination with a plane-wave basis set with a kinetic energy cutoff of 30 Ry for the wavefunctions. Models of hydrogen adatom clusters are based on a graphene 6 × 6 supercell with 15 Å of vacuum separating the periodic replicas. We used a 2 × 2 × 1 Monkhorst-Pack k-point mesh for the Brillouin zone integration [S3]. All hydrogen adatom cluster models were relaxed until a maximum force of 0.15 eV/Å on individual atoms was reached. We verified that the chosen parameters provide sufficiently accurate total energies. All calculations have been performed using the pwscf code of the Quantum ESPRESSO package [S4].
The interaction energy of a cluster of hydrogen adatoms calculated from first principles E DFT is defined as
$$E _ { \text{DFT} } = E _ { \text{gr+nH} } - E _ { \text{gr} } - n \left ( E _ { \text{gr+H} } - E _ { \text{gr} } \right ),$$
where E gr+ n H , E gr and E gr+H are the total energies of graphene with a cluster of n hydrogen adatoms, pristine graphene and graphene with a single hydrogen adatom, respectively.
In the main text of our manuscript we focused on the situation where hydrogen adatoms are deposited on a single side of graphene. The values of the fitted parameters in expr. (1) of the main text are γ 1 = -1 182 . eV and γ 2 = 0 484 . eV. This scenario is relevant to the case of graphene on a substrate, however in the situation of suspended graphene both sides of graphene are available for binding adatoms. We investigated this situation by studying the same set of adatom clusters as shown in Fig. 1(b) of the main text, but with adatoms placed on the opposite sides of graphene sheet when functionalized carbon atoms belong to different sublattices. The fitted interaction parameters are γ 1 = -1 461 . eV and γ 2 = 0 342 . eV. The excellent agreement between the estimated interaction energies ˜ E and the first-principles values E DFT is illustrated in Fig. S1(a). This case is characterized by a stronger attractive contribution and a weaker repulsion compared to the single-side adsorption, thus reflecting the known tendency of forming more stable adatom aggregates upon adsorption on both sides [S5, S6]. The cluster size distributions P n ( ) calculated for single-side and both-sides adsorption at adatom concentration x = 5% and T = 300 K are compared in Fig. S1(b). While the distributions are qualitatively very similar, one notes that both-sides adsorption exhibits a somewhat stronger tendency to form larger clusters.
(a)
n
Figure S1: (Color online) (a) Predicted energy ˜ E of aggregation of hydrogen adatoms as a function of aggregation energy E DFT calculated from first principles for the set of small clusters shown in Fig. 1(b) of the main text with adatoms adsorbed on both sides of graphene. (b) Comparison of the cluster size distributions P n ( ) for the cases of single-side and both-sides adatom adsorption at x = 5% concentration and T = 300 K.
## B. Monte-Carlo simulations of the hydrogen adatoms aggregation
We performed Monte-Carlo simulations with an elementary trial move being the displacement of a randomly chosen adatom to a random carbon atom not populated by another adatom. This move insures the fulfillment of detailed balance. The Metropolis algorithm has been employed for the acceptance/rejection criterion. Once a move has been performed, the system is updated from the old configuration S old to the new one S new with a probability
$$P ( S _ { \text{old} } \rightarrow S _ { \text{new} } ) & = \min ( 1, e ^ { - \beta [ \tilde { E } ( S _ { \text{new} } ) - \tilde { E } ( S _ { \text{old} } ) ] } ),$$
where β is the inverse temperature 1 / k ( B T ) . In all our simulations T = 300 K . The number of equilibration steps N eq disregarded from statistical sampling varied between 10 6 and 1 6 . × 10 7 , depending on adatom concentration (larger concentrations need more equilibration steps). The total number of steps in our simulations varied between 10 7 and 3 × 10 7 . The equilibration efficacy of our simulation was tested by comparing certain equilibrium properties such as the total energy and cluster size distribution. The properties obtained from Monte-Carlo simulations performed with and without temperature annealing show negligible differences.
## C. Landauer-Büttiker electronic transport calculations and scaling analysis of conductivity
In order to investigate the transport properties of graphene with resonant scattering impurities we perform Landauer-Büttiker calculations in a two-terminal configuration with a scattering region composed of hydrogenated graphene attached to two semi-infinite leads of pristine graphene, as shown in Fig. S2. The overall configuration is periodic along the transverse direction y . For such a setup, conductance as a function of energy G E ( ) is given by
$$G ( E ) = \frac { W } { 2 \pi } \int _ { - \tilde { w } } ^ { \tilde { w } } T ( E, k _ { \| } ) d k _ { \| },$$
Figure S2: (Color online) Schematic drawing of the two-terminal configuration employed for investigating the transport properties of graphene with hydrogen adatoms. The transport direction is along the x axis while the system is periodic along the y axis. The periodicity along the y axis is W = 40 nm. The unit cells of the left and right leads composed of pristine graphene are indicated by blue lines. The scattering region is populated by adatoms either randomly or according to the configurations produced by Monte-Carlo simulations in the case of correlated adatom distributions. One of the principal layers (PL) in the scattering region is indicated by means of green dashed lines.
where T E,k ( ‖ ) is the transmission probability and k ‖ is the momentum along y [S7]. Due to the large width W = 40 nm of the model employed, transmission is only evaluated at the Γ point ( k ‖ = 0 ).
In order to calculate T E ( ) we decompose the scattering region into principal layers (PLs), that is, the layers in which the atoms are coupled at most to those located in the next layer. Since our tight-binding model is limited to first-nearest-neighbor interactions and the transport direction is oriented along zig-zag direction [see Fig. S2], the minimal width of the PL is d PL = √ 3 2 d CC , where d CC = 1 42 . Å is the carbon-carbon bond length. The Hamiltonian restricted to the i -th principal layer is H i , while t i is the tight-binding hopping matrix connecting i -th and i +1 -th principal layers. We introduce an imaginary cleavage plane between the n -th and n +1 -th principal layers dividing the system into two independent parts. We define g L n and g R n +1 as the surface Green's functions of the two non-interacting semi-infinite systems located on the left and on the right sides of this cleavage plane, respectively.
Following Ref. [S8], the transmission is given by
$$T ( E ) = \text{Tr} [ T _ { n } I m ( g _ { n } ^ { L } ) T _ { n } ^ { \dagger } I m ( g _ { n + 1 } ^ { R } ) ],$$
where operator T n is defined as
$$T _ { n } = t _ { n } ( 1 - g _ { n + 1 } ^ { R } t _ { n } ^ { \dagger } g _ { n } ^ { L } t _ { n } ) ^ { - 1 }.$$
The choice of the position of the cleavage plane is immaterial because of current conservation.
Surface Green's functions g L n and g R n can be related to the preceding (successive) surface Green's functions g L n -1 ( g R n +1 ) by applying the Dyson equations [S9]
$$g _ { n } ^ { L } = ( E - H _ { n } - t _ { n - 1 } ^ { \dagger } g _ { n - 1 } ^ { L } t _ { n - 1 } ) ^ { - 1 }$$
and
$$g _ { n } ^ { \text{R} } & = ( E - H _ { n } - t _ { n } g _ { n + 1 } ^ { \text{R} } t _ { n } ^ { \dagger } ) ^ { - 1 }.$$
Further iterations of Eqns. (S6) and (S7) reduce the problem to the knowledge of the Green's functions at the surfaces separating the scattering region from the left and right leads, g LL and g RL , that we calculated according to the analytic closed form solution described in Ref. [S10].
The time complexity of the Green's function calculation for each lead with respect to the number N lead of orbitals in the lead unit cell is O N ( 3 lead ) [S9]. On the other hand, as follows from Eqns. (S6) and (S7), the complexity of the addition of the layers required to reach the cleavage plane is O M ( × N 3 layer ) , where M and N layer are the number of principal layers and the number of orbitals in each layer, respectively. Consequently, the overall complexity of the method is cubic with respect to the width and linear with respect to the length of the system.
In order to perform our conductance scaling analysis we vary the length of the scattering region in the range L = 1 ... 60 nm by steps of 8 PLs, which corresponds to ∆ L ≈ 1 nm. At each step the right lead is moved rightwards whereas the left lead is kept fixed [Fig. S2]. The values of conductance G E ( ) are averaged over an ensemble of N ens =9600 disorder realizations for proper statistical sampling.
## D. Complete account of the results of calculations of conductivity and localization length
Conductivity g for the entire investigated range of concentrations is presented in Fig. S3(a,b). The overall enhancement of conductivity upon the aggregation of adatoms is a common feature at all investigated concentrations. This is particularly visible in the strong localization regime, that is at low energies E and large scattering region lengths L . Localization length ξ loc can only be determined for the g L ( ) curves which exhibit a well-defined negative slope in the large length region. This is the case for ξ loc < 60 nm, which is the maximum scattering region length considered in our study. This does not imply that the system does not undergo localization, but rather that a longer scattering region is needed in order to estimate ξ loc correctly. For this reason many values of localization length ξ loc at x < 5% , especially in the case of correlated impurities, are missing in Fig. S3(c).
## E. Effect of the doping of leads in the Landauer-Büttiker calculations
Figure S4(a,b) shows the conductivity curves g L ( ) for x = 5% concentration of randomly distributed adatoms obtained by shifting the charge neutrality point of leads by ∆ E L = -0 5 . eV and ∆ E L = -1 0 . eV, respectively.
Figure S3: (Color online) Scaling analysis of conductivity g and localization length ξ loc for concentrations x = 0 25% . ... 10% . (a,b) Conductivity g as a function of scattering region length L calculated for graphene with random and correlated adatom distributions, respectively, at charge-carrier energies 0 eV < E < 1 eV. (c) Localization length ξ loc as a function of charge-carrier energy E for random and correlated adatom distributions.
The effect of the doping of leads is two-fold. Firstly, the DOS of pristine graphene increases away from the charge neutrality point. Hence, upon doping the number of transport channels increases, which may result in larger values of conductance g . This is particularly important when the scattering region has an enhanced DOS at zero energy, such as graphene with resonant impurities. A comparison of Fig. S4(a) and Fig. S3(a) for x = 5% shows that conductivity is indeed enhanced at low energies with a crossing of the g L ( ) curves at L ≈ 10 nm. Secondly, doping results in a mismatch between the Fermi wavelength of the leads and that of the scattering region, which has a detrimental effect on conductance g . This effect is expected to be more pronounced at higher doping and high energy, where localization plays a smaller role. Indeed, at higher doping ( ∆ E L = -1 0 . eV) and high energy ( E = 1 0 . eV) the conductivity is reduced, notably at short distance L < 10 nm [Fig. S4(a,b)]. On the other hand, at large distances a general increase of the conductance is progressively restored since the conductivity becomes predominantly determined by the localization of the wavefunction and the increased number of available states. Finally, as shown in Fig. S4(c), localization length ξ loc is practically unaffected by the doping of the leads, since it is an intrinsic property of the scattering region.
Figure S4: (Color online) (a,b) Conductivity g as a function of scattering region length L for graphene with randomly distributed adatoms at x = 5% . The charge neutrality point of the leads has been shifted by (a) ∆ E L = -0 5 . eV and (b) ∆ E L = -1 0 . eV, respectively. (c) Localization length ξ loc as a function of charge-carrier energy E for the two investigated lead doping levels compared to undoped leads ∆ E L = 0 eV.
## F. Electronic transport calculations within the Kubo-Greenwood formalism using the kernel polynomial method
The diagonal elements of the real part of the frequency dependent conductivity tensor in linear response theory is given by the Kubo-Greenwood formula
$$R e \{ \sigma _ { \alpha \alpha } ( \omega ) \} = \frac { \pi } { V } \int d E \frac { f ( E ) - f ( E + \hbar { \omega } ) } { \omega } \times \text{Tr} \{ \delta ( E - H ) j _ { \alpha } \delta ( E - H + \hbar { \omega } ) j _ { \alpha } \}.$$
Here, f ( E ) is the Fermi function and H the single-particle tight-binding Hamiltonian of the system under consideration, whereas the vectorial component of the current operator j α is defined later. In the thermodynamic limit for zero temperature and zero frequency this equation reduces to
$$R e \{ \sigma _ { \alpha \alpha } ( 0 ) \} = \frac { \pi \hbar } { V } \text{Tr} \{ \delta ( E _ { \text{F} } - H ) j _ { \alpha } \delta ( E _ { \text{F} } - H ) j _ { \alpha } \},$$
which defines the DC-conductivity studied in this work. The evaluation of this expression is carried out by employing the kernel polynomial method (KPM), as described in detail in Ref. [S11]. In this framework, the following matrix element density is defined
$$j ( E, E ^ { \prime } ) = \frac { 1 } { V } \sum _ { n, m } \langle n | j _ { \alpha } | m \rangle \langle m | j _ { \alpha } | n \rangle \delta ( E - \hbar { \omega } _ { n } ) \delta ( E ^ { \prime } - \hbar { \omega } _ { m } ),$$
which is then expanded up to finite order M within the two-dimensional KPM using the Jackson kernel. For the studied supercells containing up to N C ∼ 10 6 carbon atoms, we used M = 1280 . The real-part of the frequency
Figure S5: (Color online) (a,b) Conductivity g as a function of charge-carrier energy E calculated using the kernel polynomial method (KPM) for the random and correlated impurity distributions at x = 0 5% . and x = 5 0% . adatom concentrations, respectively. (c,d) Conductivity g as a function of expansion order M for the random and correlated impurity distributions at x = 0 5% . and x = 5 0% . adatom concentrations, respectively. The dashed lines indicate the minimum conductivity of graphene.
dependent conductivity tensor is then given by a double integration over the matrix density for arbitrary temperature and Fermi level:
$$R e \{ \sigma _ { \alpha \alpha } ( \omega ) \} = \frac { \pi } { \omega } \int _ { - \infty } ^ { \infty } d E \int _ { - \infty } ^ { \infty } d E ^ { \prime } j ( E, E ^ { \prime } ) \times [ f ( E ) - f ( E ^ { \prime } ) ] \, \delta ( \hbar { \omega } - ( E ^ { \prime } - E ) ).$$
As a result, the DC-conductivity can be expressed in terms of the diagonal elements of j ( E,E ′ ) :
$$R e \{ \sigma _ { \alpha \alpha } ( 0 ) \} = \pi \hbar { j } ( E _ { \text{F} }, E _ { \text{F} } ).$$
The calculation of transport properties requires the definition of a current operator appropriate to the employed single-particle tight-binding model. Thus, we approximate the spatial operator r in a diagonal form:
$$r ^ { \gamma } \approx \sum _ { i } r _ { i } ^ { \gamma } c _ { i } ^ { \dagger } c _ { i }.$$
Consequently, one can derive the following current operator used in this work,
$$j _ { \gamma } = - i \frac { e _ { 0 } } { \hbar } \sum _ { i j } t _ { i j } ( r _ { j } ^ { \gamma } - r _ { i } ^ { \gamma } ) c _ { i } ^ { \dagger } c _ { j },$$
where indices i and j run over atomic positions and γ denotes the vectorial component. For details of a similar derivation see e.g. Ref. [S12].
## G. Results for the calculations of conductivity using the kernel polynomial method
Figure S5(a,b) shows conductivity g as a function of charge-carrier energy E calculated using the KPM for random and correlated impurity distributions at two different adatom concentrations x = 0 5% . and x = 5 0% . . A scaling analysis of the conductivity for three charge carrier energies ( E = 1 7 . × 10 -3 eV, E = 0 1 . eV and E = 1 0 . eV) based on the expansion order M is shown in Fig. S5(c,d) for the same values of adatom concentration x . The order M
relates to a timescale τ ( M ) [S13]. For E = 1 0 . eV and x = 5% the conductivity rises with increasing expansion order M , exhibits a maximum around M ∼ 200 and then decreases slowly again for both correlated (dashed line) and random (solid line) impurity distributions [Fig. S5(d)]. We interpret these maxima as the semiclassical values of conductivity g sc corresponding to the diffusive regime, followed by quantum corrections which result in the decrease of the conductivity. For a concentration of x = 0 5% . and E = 1 0 . eV no conductivity maximum is observed, which indicates that the results still correspond to the pre-diffusive/ballistic regime for the largest expansion order investigated. A shift of g sc to larger expansion orders for decreasing x is expected, because fewer scattering centers are present. The situation is, in general, different for small energies close to the Dirac point (here, E = 1 7 . × 10 -3 eV), in both cases of correlated and random adatom distributions as well as concentrations of x = 0 5% . and x = 5% . A distinct maximum for M > 100 is completely absent, and only a reduction of the conductivity is observed for increasing M . Thus, we relate this characteristic to the quantum regime with localization effects resulting in a reduced localization length. If the expansion order is sufficiently large, the curves for E = 1 7 . × 10 -3 eV reach values well below g = 4 /π e /h ( 2 ) , and one may expect the conductivity to converge asymptotically to zero in the limit of M →∞ if the modes are completely localized.
- [S1] J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77 , 3865 (1996).
- [S2] D. Vanderbilt, Phys. Rev. B 41 , 7892 (1990).
- [S3] H. J. Monkhorst and J. D. Pack, Phys. Rev. B 13 , 5188 (1976).
- [S4] P. Giannozzi, S. Baroni, N. Bonini, M. Calandra, R. Car, C. Cavazzoni, D. Ceresoli, G. L. Chiarotti, M. Cococcioni, I. Dabo, A. D. Corso, S. d. Gironcoli, S. Fabris, G. Fratesi, R. Gebauer, U. Gerstmann, C. Gougoussis, A. Kokalj, M. Lazzeri, L. Martin-Samos, N. Marzari, F. Mauri, R. Mazzarello, S. Paolini, A. Pasquarello, L. Paulatto, C. Sbraccia, S. Scandolo, G. Sclauzero, A. P. Seitsonen, A. Smogunov, P. Umari, and R. M. Wentzcovitch, J. Phys.: Condens. Matt. 21 , 395502 (2009).
- [S5] Y. Lin, F. Ding, and B. I. Yakobson, Phys. Rev. B 78 , 041402 (2008).
- [S6] D. W. Boukhvalov, M. I. Katsnelson, and A. I. Lichtenstein, Phys. Rev. B 77 , 035427 (2008).
- [S7] M. Büttiker, Y. Imry, R. Landauer, and S. Pinhas, Phys. Rev. B 31 , 6207 (1985).
- [S8] J. Mathon and A. Umerski, Phys. Rev. B 63 , 220403 (2001).
- [S9] A. Umerski, Phys. Rev. B 55 , 5266 (1997).
- [S10] S. Sanvito, C. J. Lambert, J. H. Jefferson, and A. M. Bratkovsky, Phys. Rev. B 59 , 11936 (1999).
- [S11] A. Weiße, G. Wellein, A. Alvermann, and H. Fehske, Rev. Mod. Phys. 78 , 275 (2006).
- [S12] J. M. Tomczak and S. Biermann, Phys. Rev. B 80 , 085117 (2009).
- [S13] S. Roche, Phys. Rev. B 59 , 2284 (1999) | 10.1103/PhysRevLett.113.246601 | [
"Fernando Gargiulo",
"Gabriel Autès",
"Naunidh Virk",
"Stefan Barthel",
"Malte Rösner",
"Lisa R. M. Toller",
"Tim O. Wehling",
"Oleg V. Yazyev"
] | 2014-08-01T17:44:12+00:00 | 2014-08-01T17:44:12+00:00 | [
"cond-mat.mes-hall",
"cond-mat.dis-nn",
"cond-mat.mtrl-sci"
] | Electronic Transport in Graphene with Aggregated Hydrogen Adatoms | Hydrogen adatoms and other species covalently bound to graphene act as
resonant scattering centers affecting the electronic transport properties and
inducing Anderson localization. We show that attractive interactions between
adatoms on graphene and their diffusion mobility strongly modify the spatial
distribution, thus fully eliminating isolated adatoms and increasing the
population of larger size adatom aggregates. Our scaling analysis shows that
such aggregation of adatoms increases conductance by up to several orders of
magnitude and results in significant extension of the Anderson localization
length in the strong localization regime. We introduce a simple definition of
the effective adatom concentration $x^\star$ , which describes the transport
properties of both random and correlated distributions of hydrogen adatoms on
graphene across a broad range of concentrations. |
1408.0246v1 | ## Unconventional Magnetic Order on the Hyperhoneycomb Kitaev Lattice in β -Li IrO : 2 3 Full Solution via Magnetic Resonant X-ray Diffraction
A. Biffin, 1 R. D. Johnson, 1 Sungkyun Choi, 1 F. Freund, 2 S. Manni, 2
A. Bombardi, 3 P. Manuel, 4 P. Gegenwart 2 and R. Coldea 1
1 Clarendon Laboratory, University of Oxford, Parks Road, Oxford OX1 3PU, U.K.
2 EP VI, Center for Electronic Correlations and Magnetism,
Augsburg University, D-86159 Augsburg, Germany
3 Diamond Light Source Ltd., Harwell Science and Innovation Campus, OX11 0DE, U.K. and
4 ISIS Facility, Rutherford Appleton Laboratory, Harwell Oxford, OX11 0QX, U.K.
The recently-synthesized iridate β -Li 2 IrO 3 has been proposed as a candidate to display novel magnetic behavior stabilized by frustration effects from bond-dependent, anisotropic interactions (Kitaev model) on a three-dimensional 'hyperhoneycomb' lattice. Here we report a combined study using neutron powder diffraction and magnetic resonant x-ray diffraction to solve the complete magnetic structure. We find a complex, incommensurate magnetic order with non-coplanar and counter-rotating Ir moments, which surprisingly shares many of its features with the related structural polytype 'stripyhoneycomb' γ -Li 2 IrO , where dominant Kitaev interactions have been 3 invoked to explain the stability of the observed magnetic structure. The similarities of behavior between those two structural polytypes, which have different global lattice topologies but the same local connectivity, is strongly suggestive that the same magnetic interactions and the same underlying mechanism governs the stability of the magnetic order in both materials, indicating that both β -and γ -Li 2 IrO 3 are strong candidates to realize dominant Kitaev interactions in a solid state material.
PACS numbers: 75.25.-j, 75.10.Jm
## I. INTRODUCTION
Materials containing ions with a strong spin-orbit interaction are attracting much attention as candidates to display novel electronic states ranging from topological insulators with protected gapless surface states to quantum spin liquids with exotic excitations. 1 Complex magnetic behavior can arise when the strong spin-orbit coupling (as found for heavy transition metal 4 d and 5 d ions) and crystal field effects stabilize spin-orbit entangled magnetic moments, which may interact via anisotropic exchange interactions where the anisotropy axis depends on the bond orientation. On certain lattices these interactions may be strongly frustrated, leading potentially to novel forms of cooperative magnetic order and/or excitations, not found for magnets of 3 d ions (where the spin-orbit coupling is much weaker and the orbital moment is in general quenched). The magnetic ground state of Ir 4+ ions in a cubic IrO 6 octahedron is a doublet with a mixed spin-orbital character, 2 J eff = 1 2, and for edge-/ sharing IrO 6 octahedra it was proposed 3,4 that the leading super-exchange is a ferromagnetic Ising coupling between the Ir magnetic moment components perpendicular to the Ir-O -Ir 2 plane. For three-fold coordinated, edge-sharing octahedra this leads to orthogonal Ir-O 2 -Ir planes for the three Ir-Ir bonds and correspondingly orthogonal x y , , z components coupled along the three Ir-Ir bonds. For a honyecomb lattice this realizes the Kitaev model, 5 where the strong frustration effects stabilize an (exactly solvable) quantum spin liquid state with novel excitations (Majorana fermions and fluxes). 6 In search for realizations of such physics, α -Na IrO 2 3 [Refs. 7-11] and α -Li 2 IrO 3 [Refs. 12,13] have been actively explored experimentally, however no clear evidence of novel Kitaev phenomena has been observed.
Generalization of the Kitaev model to threedimensional (3D) lattices have also been shown to have quantum spin liquid ground states. 14-16 Furthermore, non-trivial magnetic behavior has been predicted when (finite) additional interactions suppress the spin liquid and stabilize magnetic order. 15,17 The recently synthesized 'hyperhoneycomb' 18 β -Li 2 IrO 3 [see Fig. 1] and 'stripyhoneycomb' 19 γ -Li 2 IrO , 3 which have the local connectivity of three-fold coordinated IrO 6 octahedra with near-orthogonal Ir-O -Ir planes meeting at each Ir 2 site, are prime candidates to realize 3D Kitaev physics. The availability of single crystal samples of both of those two distinct structural polytypes offers unique opportunities to perform comparative studies between them and gain insight into the underlying physics. In recent experiments 20 on the γ -polytype we have observed an unexpectedly complex, yet highly-symmetric incommensurate magnetic structure with non-coplanar and counterrotating moments, and theoretical calculations showed that this structure could be stabilized by a spin Hamiltonian with dominant Kitaev couplings and some additional interactions. Here we extend our experimental studies to the β -polytype, where we find some striking similarities in the magnetic structure. This shows that the key features of the magnetic order are robust, independent of the changes in the global lattice topology between the two distinct structural polytypes that share the same building blocks. This is strongly suggestive that the same underlying magnetic interactions govern the stability of the magnetic order in both structural polytypes, which would be an important constraint on any theoreti-
cal model of the cooperative magnetism in this family of materials.
The structural polytype β -Li 2 IrO 3 was discovered by Takayama et. al. 18 and the crystal structure is illustrated in Fig. 1(left). It has an orthorhombic unit cell (space group Fddd , for full structural details see Appendix A) and the same mass density and same fundamental building blocks - edge-sharing LiO 6 and IrO 6 octahedra (the latter shaded in red) as in the layered polytype α -Li 2 IrO 3 (Ref. 12), where edge-sharing IrO 6 octahedra form planar honeycomb layers separated by hexagonal Li layers. In the β -structure [see Fig. 1] the Ir lattice is three-fold coordinated as in a planar honeycomb, but now the links of the lattice form a 3D network; this 3D lattice connectivity has been named a 'hyperhoneycomb'. 18 It can be thought of as being made of zig-zag chains [shaded cyan and purple in Fig. 1(right)] stacked along c and alternating in direction between the two basal plane diagonals a ± b (the planar honeycomb structure is obtained when the zig-zag iridium chains are not alternating in orientation, but are all running parallel to either the a + , or b the a -b directions).
FIG. 1: (color online) Crystal structure of β -Li 2 IrO . 3 Two unit cells are shown: (left) full structure with Li (white balls), O (black) and Ir (red) located inside IrO 6 octahedra (shaded polyhedra), (right) Ir ions are arranged in a 'hyperhoneycomb' structure. Light (cyan) and dark (purple) colors indicate zig-zag chains stacked along c and directed alternatingly along a ± b .
Earlier studies 18 have shown that β -Li 2 IrO 3 is an insulator. The temperature-dependence of the magnetic susceptibility parameterized in terms of a Curie-Weiss law gave an effective magnetic moment µ eff = 1.61 µ B , consistent with localized, J eff = 1 2 moments at the /
iridium sites. Both specific heat and susceptibility data showed an anomaly near T N = 38 K attributed to the onset of magnetic order. Here we confirm that the lowtemperature phase has magnetic long-range order and we provide a full experimental magnetic structure solution. Our results give uniquely detailed information about the correlations that govern the cooperative magnetism and the relevant magnetic interactions that stabilize them. We first used neutron diffraction on a powder sample to confirm the presence of magnetic order at low temperatures and to obtain candidate magnetic propagation vectors. We then performed magnetic resonant x-ray diffraction (MRXD) experiments on a 17 µ m diameter single crystal at the L 3 edge of iridium, where experiments on other iridates 2,8,21 have reported a strong enhancement of the magnetic scattering cross-section. We exploit the polarization dependence of the MRXD intensity (probed via azimuth scans) to deduce that the magnetic structure has rotating magnetic moments and we determine the plane of rotation and relative phases between all 16 iridium sites in the structural unit cell.
The paper is organized as follows: Sec. II presents the neutron powder diffraction measurements and the analysis by which we obtain candidate magnetic propagation vectors. Sec. III presents the single-crystal MRXD measurements, which observe magnetic Bragg peaks with an incommensurate propagation vector along the (100) direction. The observed diffraction pattern is analyzed in terms of magnetic basis vectors in Sec. III A and III B. The azimuth dependence of the diffraction intensity is used to determine the polarization of all magnetic basis vectors in the ground state and the relative phase between them (in Sec. III C). Finally the absolute value of the ordered magnetic moment is extracted from the neutron powder data in Sec. III D. The resulting magnetic structure is presented in Sec. III E and similarities with the magnetic structure in the γ -polytype are discussed in Sec. IV. Finally, conclusions are summarized in Sec. V. Further technical details of the analysis are presented in the Appendices: (A) crystal structure refinement at low temperature from single-crystal x-ray data, (B) decomposition of the magnetic structure in terms of its Fourier components, (C) derivation of the selection rules for magnetic scattering for the various basis vectors, (D) the azimuth dependence of the MRXD intensity enabling determination of the relative phase between basis vectors, and (E) an equivalence mapping between the magnetic structures in the β - and γ -polytypes of Li 2 IrO . 3
## II. MAGNETIC NEUTRON POWDER DIFFRACTION
Neutron powder diffraction measurements to obtain information about the magnetic propagation vector were performed using the time-of-flight diffractometer WISH at ISIS. 0.71g of powder β -Li 2 IrO 3 (synthesized as described in Appendix A) was placed in an aluminium can
of annular geometry (to minimize the strong neutron absorption by the Ir nuclei) and was mounted on the cold finger of a closed cycle refrigerator. Diffraction patterns were collected at a selection of temperatures from base (5.6 K) to paramagnetic (70 K), well above the magnetic ordering transition T N = 38 K inferred from thermodynamic measurements. 18 The obtained neutron diffraction pattern did not allow for a full structural refinement (most likely due to neutron absorption effects in the presence of iridium). We therefore performed additional x-ray measurements on a small single crystal piece extracted from the same powder batch to determine precise internal atomic positions in the unit cell at a temperature of 100 K, considered to be cold enough to be representative of the crystal structure in the low-temperature limit. The x-ray diffraction was performed on a Mo-source Oxford Diffraction Supernova x-ray diffractometer under a N 2 gas flow, which gave a temperature of 100 K at the sample position. The data confirmed the expected space group and full structural refinement (for details see Appendix A) gave atomic positions consistent with those previously reported at room temperature. 18
Fig. 2 shows the neutron powder diffraction pattern in the lowest angle bank of detectors, which covers the region of large d -spacings where magnetic diffraction is expected. In the main panel the visible diffraction peaks are of structural origin. β -Li 2 IrO 3 peak positions are denoted by the upper row of vertical marks (below the data) and the two large peaks in the data at d -spacing near 2 ˚ (labelled 'Al') are the lowest-order diffraction peaks A from the Al can and were fitted as such. The pattern also showed two peaks labelled (*), originating from a small impurity phase identified as pure Ir, and those regions in the data were excluded from the analysis. We performed the structural fit (using FullProf 22 ) to the neutron data (both paramagnetic and base temperature) with the internal atomic positions kept fixed, and only the lattice parameters left free to vary. The structural fit was consistent over all detector banks - corresponding to a range of scattering angles and resolutions - and we were also able to consistently fit the instrument parameters (which provide the conversion from time-of-flight to d -spacing) over all the detector banks tested. Fig. 2 illustrates the very good agreement in describing the structural diffraction pattern.
## A. Candidate magnetic propagation vectors
The inset in Fig. 2 highlights the region of large d -spacing, where upon cooling below T N two new Bragg peaks appear, attributed to the onset of magnetic ordering. The peak widths are comparable to those of structural Bragg peaks, implying long-range magnetic order. With only two magnetic peaks the magnetic structure cannot be solved uniquely, however the data provides strong constraints on the magnetic propagation vector, q . Typically q will lie on a point, line or plane of high sym-
FIG. 2: (color online) Neutron powder diffraction at base temperature (5.6 K, red circles) and in the paramagnetic regime (70 K, green circles) in the lowest-angle detector bank. Solid black line shows the fit to the structural and magnetic contributions as discussed in the text. Positions of structural β -Li IrO 2 3 peaks, aluminium peaks and magnetic Bragg peaks are marked below the pattern in the upper, middle and lower rows, respectively and the blue line underneath represents the difference between data and fit. Labels 'Al' indicate aluminium peaks and (*) denote two peaks from an impurity phase identified as pure Ir, removed from the fit. Inset: zoom into the large d -spacing region showing two magnetic Bragg peaks, labels indicate the magnetic peaks index (symmetrically equivalent indexes are omitted).
metry in the Brillouin zone. Using the k-search tool in FullProf 22 we systematically searched all such positions for the Fddd space group in order of decreasing symmetry to find propagation vectors that could reproduce the magnetic peak positions observed in the powder neutron data. All commensurate orderings with wavevectors at the Γ, Z Y , , T and L points (where the Miller-Love convention for labelling high-symmetry Brillouin zone points has been used 23 ) were ruled out by this analysis. Therefore we considered propagation vectors along lines of high symmetry with general points denoted by Σ, ∆, Λ. Only the first two were compatible with the neutron data and the obtained solutions are listed in Table I, along with possible solutions for other high symmetry directions.
TABLE I: Candidate magnetic propagation vectors along high symmetry directions in the Brillouin zone compatible with the magnetic neutron powder diffraction pattern. The ellipses denotes the fact that multiple solutions exist for a general propagation vector in the ab -plane.
| Position | q fit |
|------------|------------------|
| Σ | (0.57,0,0) |
| ∆ | (0,0.81,0) |
| J | (0.27,0, 1 / 2 ) |
| E | (0,0.39, 1 / 2 ) |
| M | (0.43,0.53,0) |
| | . . . |
## III. MAGNETIC RESONANT X-RAY DIFFRACTION
To determine which of the candidate magnetic propagation wavevectors identified by the neutron data actually occurs, we have performed a magnetic resonant x-ray diffraction experiment in reflection geometry at the L 3 edge of Ir using the I16 beamline at Diamond. We used a single crystal of β -Li 2 IrO 3 (characterized via x-ray diffraction as described in Appendix A) with an orthorhombic morphology with its maximum dimension less than 17 µ m. The sample was placed on a Si (111) disk with the reciprocal lattice vector (-6,5,7) being approximately surface normal and it was cooled using a closedcycle refrigerator with Be domes. The x-ray energy was tuned to the L 3 edge of Iridium (11.215 keV) and on cooling below T N new Bragg peaks appeared at satellite positions of reciprocal lattice points at ( h k l , , ) ± q , with h k l , , integers and an incommensurate ordering wavevector q =(0.57(1), 0, 0). Scans through such a peak are shown in Fig. 3a-c). The scans emphasize the incommensurate position along h and centering at integer k and l values. In the language of Table I this is then a propagation vector along the Σ line of symmetry, and the magnitude found from the x-ray data is consistent with the positions of peaks in the neutron data (first line in Table I). Lorentzian squared fits to the peak shape profiles (dashed lines in Figs. 3a-c),e) gave widths comparable to those measured for the nearby structural Bragg peak (-6,6,12), indicating that the magnetic order is longrange in all three directions. To estimate a lower bound on the correlation lengths we use the inverse of the peak half-widths, which gives values in excess of 100, 200 and 300 ˚ along the A a , b and c axes, respectively (more than ≈ 20 unit cells in each direction), and in reality the correlation lengths will be much larger as the above estimate did not include the peak broadening due to the finite instrumental angular resolution.
Several independent tests can be performed to confirm the magnetic origin of the observed diffraction peaks. Firstly, in Fig. 3(d) the integrated intensity of (-6,6,12)q as a function of temperature (filled circles) shows a typical order-parameter behavior. A power-law fit (solid line) gives an onset temperature of T fit = 36.7 K, which is very close to the transition temperature deduced from specific heat measurements. 18 We attribute the apparent small offset of ∆ T glyph[similarequal] 1.3 K to the fact that the sensor where the temperature was measured in the x-ray experiments was at the bottom of the sample support and the actual sample temperature may have been higher by ∆ T due to local heating effects from the very intense synchrotron beam, a common occurrence in resonant x-ray experiments.
Further evidence of the magnetic origin of the incommensurate diffraction peaks can be obtained via polarization analysis of the scattered beam. For a scattering experiment with incident beam polarization ˆ glyph[epsilon1] normal to the ( k k , ′ ) scattering plane ( σ -polarized), mag-
FIG. 3: (color online) Magnetic Bragg peak at (-6,6,12) -q ; (a) Scan shows that the peak occurs at an incommensurate position along h , and is well-centered at integer (b) k , and (c) l . (d) Temperature-dependence of the integrated peak intensity (solid line is a power-law fit).(e) Scans with a polarizer in the scattered beam show that the peak is only present in the rotated polarization channel ( σ π -′ ) (light blue, filled) as expected for magnetic scattering. (f) Intensity as a function of the x-ray energy (filled points) shows that the signal is resonant on the edge of the fluorescence signal onset (dashed line), as expected for magnetic scattering.
netic scattering is expected to rotate the polarization by 90 ◦ compared to the incident beam direction (scattering in the σ π -′ channel, see diagram in Fig. 4a) (inset)), whereas conventional (charge) scattering would leave the polarization direction unchanged (scattering in the σ σ -′ channel). 25 In the experiments polarization analysis was achieved via the 3 rd harmonic reflection from the (1,1,1) planes of an Au crystal, the d -spacing of 0.785 ˚ being A ideally suited for polarization analysis at the L 3 edge of Ir (giving perfect filtering at 11.16 keV, see Ref. 24). Fig. 3(e) shows an h -scan across the incommensurate peak performed with polarizer parallel (open symbols, σ σ -′ ) and perpendicular (filled symbols, σ π -′ ) to the original incident beam polarization. The signal appears only in the rotated polarization channel, as expected for magnetic scattering. Note that the higher background in the σ σ -′ channel is due to incoherent x-ray scattering, which is largely filtered out by the polarization analysis in the σ π -′ channel revealing a pure magnetic signal.
A third test is the dependence on the x-ray energy as the magnetic scattering is only expected at resonance. Indeed, the incommensurate peak intensity has a characteristic resonant response as shown in Fig. 3(f) (filled
blue symbols) with a maximum at the resonance energy of 11.215 keV in agreement with previous resonant studies on other iridates, 8,21 and coincides with the edge of the fluorescence signal from the sample (black, dotted line).
## A. Magnetic Basis Vectors
The observation of incommensurate magnetic Bragg peaks indicates a magnetic structure with rotating moments or a spin-density wave, both with a wavelength that is incommensurate with the underlying crystal lattice periodicity. In order to distinguish between those two scenarios, and to determine fully the orientations of the magnetic moments at all magnetic sites in the unit cell, we consider below symmetry-allowed basis vectors for magnetic structures compatible with the observed propagation vector q . This provides a natural framework to directly link diffraction data with a magnetic structure model and allows one to develop a systematic strategy in experiments to determine the magnetic structure completely.
The magnetic ions, Ir 4+ , occupy a single crystallographic site with ions at (1 / 8, 1 / 8, z ) [ z = 0.70845(7) glyph[similarequal] 5/8+1/12] and symmetry equivalent positions, giving four Ir ions in the primitive unit cell (labelled 1-4 in Fig. 6 and with positions listed explicitly in Table IV in Appendix B). Taking into account that at each site the magnetic moment could have components along the x , y and z -axes (along the orthorhombic a , b , c -axes), gives a 12-component representation of the magnetic structure. Its irreducible representations and associated magnetic basis vectors (obtained using the BasIrreps tool, part of the FullProf suite 22 ) are listed in Table II.
TABLE II: Irreducible representations and basis vectors for a magnetic structure with propagation vector q =( , 0, 0). q
| Irreducible Representation | Basis Vectors |
|------------------------------|-----------------|
| Γ 1 | F x , G y , A z |
| Γ 2 | C x , A y , G z |
| Γ 3 | G x , F y , C z |
| Γ 4 | A x , C y , F z |
The basis vectors contain symmetry-imposed relations between the Fourier components of the magnetic structure between the four sites in the primitive cell. The types of basis vectors that can occur are:
$$F = \begin{bmatrix} 1 \\ 1 \\ \delta \\ \delta \end{bmatrix}, C = \begin{bmatrix} 1 \\ 1 \\ - \delta \\ - \delta \end{bmatrix}, A = \begin{bmatrix} 1 \\ - 1 \\ - \delta \\ \delta \end{bmatrix}, G = \begin{bmatrix} 1 \\ - 1 \\ \delta \\ - \delta \end{bmatrix}, \quad & \text{where} \\ ( 1 ) \quad & \text{where} \\ \text{where for each vector the four values are relative phase} \quad & ( h, k,$$
where for each vector the four values are relative phase factors between the Fourier components at the four sites.
δ = e -i q · ( r 3 -r 1 ) = e -iπq/ 2 is a displacement phase factor that takes into account the fact that sites 3 and 4 are displaced relative to sites 1 and 2 by a/ 4 in the direction of the propagation vector q . In an F basis vector the Fourier components at the 4 sites are related by M q ,1 = M q ,2 = δ -1 M q ,3 = δ -1 M q ,4 , i.e. there are no additional phase factors between the four sites apart from the natural displacement phase factor, and in the limit q → 0 ( δ → 1) one recovers ferromagnetic order. For the basis vectors C , A and G two out of the four sites have a change in sign in addition to the normal displacement phase factor, and in the limit q → 0 one finds an antiferromagnetic alignment between the four sites. To determine the full magnetic structure, we first identify which basis vectors are present, determine their polarization (i.e. along x , y or z ) and then find the relative phase between them, as follows.
## B. Selection Rules for Magnetic Scattering
Each magnetic basis vector contains a strict, symmetry imposed phase relation between moments on all Ir sites in the primitive unit cell. As such, one can derive selection rules for non-zero magnetic diffraction intensity, i.e., each basis vector will only contribute to magnetic satellite reflections of certain reciprocal lattice points and will have zero structure factor for others. It follows that simply the presence or absence of magnetic Bragg peaks at certain positions can already identify which basis vectors are present.
Explicitly, the structure factor for a magnetic superlattice reflection at Q = ( h k l , , ) ± q is
$$\mathcal { F } ( Q ) = \mathcal { F } ( ( h, k, l ) \pm q ) = f _ { F } \, \sum _ { n } M _ { \pm q, n } e ^ { i Q \cdot r _ { n } }, \quad ( 2 )$$
where the pre-factor f F = 1+ e iπ h ( + ) k + e iπ k ( + ) l + e iπ l ( + h ) is due to the F -centering of the orthorhombic structural unit cell. The sum extends over all sites in the primitive cell ( n =1 -4) and M ± q , n are the Fourier components of the magnetic moments at site n with position in the unit cell r n . For an F -basis vector we obtain the structure factor
$$\text{ions} \\ \text{ruc-} \\ \text{The} \quad \mathcal { F } ^ { F } ( Q ) & = \begin{cases} \ 1 6 M _ { \pm q, 1 } \cos \frac { \pi l } { 6 } e ^ { i \xi _ { \pm } }, \ h, k, l \, \text{all even and} \\ h + k + l = 4 p, \ p \, \text{integer} \\ \ 1 6 M _ { \pm q, 1 } \cos \frac { \pi l } { 6 } \cos \frac { \pi ( h + l - k ) } { 4 } e ^ { i \chi _ { \pm } }, \\ h, k, l \, \text{all odd} \\ 0, \quad \text{otherwise,} \end{cases} \\ \Big \rfloor \quad & \text{where} \, \xi _ { \pm } = \pi ( h + k - 3 l \, \pm q ) / 4 \, \text{and} \, \chi _ { \pm } = \pi ( 2 h - 2 l \, \pm }$$
where ξ ± = π h ( + k -3 l ± q / ) 4 and χ ± = π (2 h -2 l ± q / ) 4 and to obtain a closed-form analytic expression we used the 'ideal' iridium position z =5/8+1/12. From (3) we conclude that an F basis vector will contribute to the intensity of ± q satellites of reciprocal lattice points ( h k l , , ) with the selection rule h k l , , all odd, or all even with h + k + l = 4 , p p integer. This is the same selection
rule as for structural Bragg scattering from the iridium ion sublattice at the ( h k l , , ) position, as expected since in the limit q → 0 the F -type magnetic basis vectors recover ferromagnetic order, and this has the same selection rule as structural scattering. Structure factor expressions for the other basis vectors are given in Appendix C. The resulting selection rule for A is same as for F , but with no contribution from A for l = 6 m m , integer. The selection rule for C and G is h k l , , all odd, or all even and h + k + l = 4 p + 2, p integer with no contribution from G if l = 6 m m , integer.
The experimental geometry constrained the search for magnetic satellite reflections to regions of reciprocal space in the vicinity of the surface normal reflection, ≈ (-6,5,7). Using (3) and similar expressions for C G , and A (see Appendix C) we ascertained 3 positions of interest for identifying which basis vectors contribute to the magnetic structure, namely (-6,6,12) ± q (pure F ), (-8,6,12) ± q (pure C ) and (-5,5,9) ± q (mixed A and G ) and found magnetic peaks at all these positions. Therefore we deduce that the structure contains F , C and one (or both) of A and G . To determine the direction of the magnetic moment components for each of those basis vectors we make use of the polarization dependence of the magnetic x-ray diffraction intensity, as follows.
## C. Azimuth Scans to Determine the Moment's Direction and Relative Phases Between Basis Vectors
In the dipolar approximation the magnetic x-ray diffraction intensity at resonance is proportional to
$$L ( \theta ) \mathcal { A } \left | ( \hat { \epsilon } ^ { \prime } \times \hat { \epsilon } ) \cdot \mathcal { F } ( Q ) \right | ^ { 2 },$$
where L θ ( ) is the Lorentz factor at the scattering angle 2 , θ A is an absorption correction dependent upon the experimental geometry, F ( Q ) is the magnetic structure factor vector given in (2), and ˆ glyph[epsilon1] ′ and ˆ glyph[epsilon1] are unit vectors along the polarization of the electric field component of the scattered and incident x-ray beams, respectively. 25 For a σ -polarized incident beam magnetic resonant scattering occurs only in the σ π -′ channel [see diagram in Fig. 4a) inset], meaning that the product of the electric field polarization vectors is along the scattered beam direction, i.e. ˆ glyph[epsilon1] ′ × ˆ glyph[epsilon1] = ˆ k ′ , so only the component of the structure factor vector along the scattered beam direction, F ‖ = F · ˆ k ′ , contributes to the magnetic intensity. By keeping the instrument in the scattering condition and rotating the sample around the scattering wavevector, Q = k ′ -k , the projection F ‖ of the structure factor vector onto the (fixed) direction ˆ k ′ varies depending on the azimuth angle Ψ, with maximum magnetic intensity when the magnetic moments that give rise to the scattering make the smallest angle with ˆ k ′ and zero intensity when they are perpendicular.
FIG. 4: (color online) Intensity as a function of azimuth for three magnetic Bragg peaks a) C y , b) F z and c) A x . Top diagram illustrates the scattering geometry. Data points (filled circles) are integrated peak intensities (in sample rocking curve scans) corrected for absorption and Lorentz factor. Thick (red) lines show fits that include all contributions to the magnetic scattering intensity 26 for the magnetic structure model Γ 4 depicted in Fig. 6. Orange, green and blue lines denote other models (Γ 1,2,3 , respectively), which can be easily ruled out. The azimuth origin Ψ = 0 corresponds to the case when the (010) axis is in the scattering plane making the smallest angle with the incident beam direction ˆ k .
Azimuth scans to test the polarization dependence are shown in Figs. 4a-c) where the data points are obtained by integrating the magnetic peak intensity in sample rocking curve scans for each value of the azimuth, Ψ. As discussed earlier, the selection rules identify the signal in panel a) at (-8,6,12)+ q to be of C -character and the observed maximum intensity at Ψ = 0 uniquely iden-
tifies the polarization to be along the y -axis (red thick line) as at Ψ = 0 the y -axis is in the scattering plane and makes the smallest angle with the ˆ k ′ direction and makes a larger angle with ˆ k ′ when the azimuth is displaced away from 0 in both directions; green and blue curves show the expected intensity dependence for other polarizations that can be clearly ruled out. Similarly, in panel b) the peak at (-6,6,12) -q is identified as being of F character and polarized along z (thick red line). Panel c) shows data at (-5,5,9) -q , clearly of x -polarization (thick red line). The structure factor is compatible with both an A or G character and we will show later that only a pure A character can describe this and other azimuth scans quantitatively, so we identify this peak as coming from an A x basis vector.
FIG. 5: (color online) Intensity as a function of azimuth for 'mixed' magnetic Bragg peaks (black circles) corrected for absorption and Lorentz factor, compared to calculations (red thick line) for the magnetic structure model in Fig. 6, as well as calculations for other possible models (a) (-6,6,8)+ q is a mixed F : A peak with contributions in the ratio 1 : 3, (b) (-5,5,7) -q is a mixed F : C : A peak with contributions in the ratio 3 : 3 : 1. Note that the data clearly distinguishes between different relative phases of basis vectors (cyan and purple).
The next step in solving the structure is to determine the relative phase between the A x , C y and F z basis vectors. A phase of 0 or π between basis vectors with orthogonal components, say along A x and F z , means that the moments vary sinusoidally along a direction in the xz plane, whereas a phase offset of π/ 2 means that the moments rotate in the xz plane. The relative phase between two basis vectors can be determined from an azimuth scan at a magnetic Bragg peak position where both basis vectors contribute as the intensity is the sum of the intensities due to the two separate basis vectors plus an additional cross-term that is sensitive to the phase offset (for more details see Appendix C).
Fig. 5(a) shows the azimuth scan at (-6,6,8)+ q where structure-factor calculations show that both F and A basis vectors contribute in the ratio 1:3. The intensity dependence uniquely identifies it as originating from the particular basis vector combination F z + iA x , i.e. the x components are π/ 2 out-of-phase with respect to the z -components, meaning that the magnetic moments rotate in the xz plane (in a particular sense on each site, note that the opposite sense of rotations encoded in the basis vector F z -iA x is not consistent with the data cyan line). The data clearly rules out the case when the moments are not rotating in the xz plane, but are varying sinusoidally along a line in the xz plane ( F z ± A x , purple line), a model with no A x component is also ruled out (not shown). This analysis shows that an A x basis vector must exist in the ground state and that it is π/ 2 out-of-phase with respect to F z . Similarly, to determine the phase offset between the F z and C y basis vectors we measured the azimuth scan at the magnetic reflection (5,5,7) -q of mixed F , C and A contributions in the ratio 3:3:1, shown in Fig. 5(b), and only a model with the basis vector contribution F z + iC y (thick red line) matches the data, with other combinations (blue and green curves) clearly ruled out (note that those calculations also include the iA x contribution as already determined).
Having established all the basis vectors that are present, their polarization, and their relative phases, the relative magnitudes are obtained by quantitatively fitting the azimuth intensity dependencies. Three free parameters were fit; the relative ratios M /M x z and M /M y z , and a single overall intensity scale factor for all the data. The simultaneous fit of this magnetic structure model to the data in Figs. 4a-c) and Fig. 5a-b) (shown by the thick red solid lines) gave the moment amplitudes in the relative ratio M x : M y : M z = 0.45(1) : 0.65(1) : 1. Note that the basis vector combination ( iA x , iC y , F z ) with the above moment amplitudes quantitatively explains the mixed F z , A x peak in Fig. 5a), and the same magnitude A x basis vector also quantitatively accounts for the azimuth dependence in Fig. 4c), further confirming the identification of the signal in this magnetic Bragg peak as coming from a pure A x basis vector (if a G x basis vector is also present and contributing to the signal in Fig. 4c), its magnitude is very small, below the accuracy of the present experiments). Given this we conclude that the magnetic structure is described by the basis vector combination
( iA x , iC y , F z ), which corresponds to a single irreducible representation, Γ 4 in Table II.
## D. Absolute Value of the Ordered Moment
The only remaining parameter still to be determined is the absolute magnitude of the magnetic moments. This is difficult to extract reliably from the magnetic resonant x-ray diffraction data as it requires accurate determination of scale factors between the magnetic and structural peaks (the latter being of order 10 5 -10 6 more intense). However, in neutron scattering experiments the magnetic and structural peaks are of comparable magnitudes and one can obtain a reliable determination of the relative scale factor. To this aim we return to the powder neutron data in Fig. 2 and simultaneously fit three contributions to the data: structural peaks of the sample calculated using the crystal structure with fixed internal parameters (as deduced from single crystal x-ray diffraction), aluminium structural peaks (from the sample can), and magnetic peaks calculated using the full magnetic structure model deduced from the resonant x-ray experiments. The fit is plotted as a solid black line in Fig. 2 and shows excellent agreement with the data. The resulting magnetic moment is of magnitude 0.47(1) µ B when aligned along the c -axis, and slightly reduced by ∼ 20% when in the ab plane. The propagation vector was also fitted and we find q =(0.5768(3), 0, 0), consistent with the x-ray measurements.
## E. Magnetic Structure
The obtained magnetic structure projected onto the ac plane is shown in Fig. 6a). The magnetic moments are counter-rotating between all nearest-neighbor sites and the plane of rotation alternates between the two sites of every vertical ( c -axis) bond as illustrated by the pattern of light and dark shaded elliptical envelopes. The planes of rotation are obtained from the ac plane by rotation around the c -axis by an angle ± φ , with φ = tan -1 M y M x = 55(1) ◦ . Each zig-zag chain has the top and bottom sites counter-rotating in one and the same plane, which then alternates between consecutive zig-zag chains verticallylinked along the c -axis. The zig-zag chains are directed alternatingly along the a ± b directions and are made up of iridium sites of type (2,4) and (1,3), respectively, with vertical ( c -axis) bonds coupling sites of types 1-2 and 34. The magnetic order pattern is such that for every vertical bond the spin components along the y -axis are ferromagnetically-aligned (see Fig. 7), in accordance with the basis vector C y , which has equal Fourier components at sites 1-2 and 3-4, respectively [see eq. (1)].
We note that the magnetic moment rotation at each site defines an elliptical envelope, distorted from circular with the moment slightly smaller when in the ab plane compared to when along the c -axis, √ M 2 x + M /M 2 y z = 0.80(1). This effect may be due to a larger g -factor along the c -axis compared to the ˆ a cos φ ± ˆ b sin φ directions in the ab plane. We note that single-crystal susceptibility data 19 in the related polytype γ -Li 2 IrO 3 do provide evidence for the presence of g -tensor anisotropy for the iridium moments (in the high-temperature limit the susceptibility, χ , is expected to be proportional to the squared g -factor along the applied field direction, and experiments observe χ c larger than both χ a and χ b , implying an anisotropic g -tensor). The very similar local environment around the iridium sites in β -Li 2 IrO 3 suggests that an anisotropic g -tensor is also likely here. Another possibility is that zero-point quantum fluctuations may prefer a non-fixed length ordered moment between sites, with a larger ordered moment along the c -axis, which is the only direction that is common to all planes of rotation. Evidence for the presence of zero-point quantum fluctuations in the ground state is provided by the fact that the absolute magnitude of the ordered moment (0.47(1) µ B ) is significantly reduced from what is believed to be the available full-moment value (estimated at gµ B J eff glyph[similarequal] 1 µ B assuming g glyph[similarequal] 2), so a structure with small modulations on an already significantly reduced ordered moment could be compatible with the experimental results.
## IV. DISCUSSION
It is interesting to note that the obtained magnetic structure has striking similarities with the magnetic structure in the related polytype γ -Li 2 IrO 3 [Ref. 20], with which β -Li 2 IrO 3 shares the same size orthorhombic unit cell ( a × b × c ) and also many key structural features. In both polytypes the iridium lattice is locally three-fold coordinated and is made up of vertically-linked zig-zag chains that alternate in orientation between the a ± b direction. The only difference between them is that in the hyperhoneycomb β -polytype this alternation occurs between consecutive zig-zag chains, whereas in the stripyhoneycomb γ -polytype the alternation occurs between pairs of parallel zig-zag chains (which form a honeycomb row). The magnetic structures in both polytypes are incommensurate with the same propagation vectors within experimental error, q =(0.57(1), 0, 0), the moments are counter-rotating between the sites of every nearest-neighbor bond, and the plane of rotation alternates between two orientations tilted away (by an angle φ ) from the ac plane between consecutive zig-zag iridium chains vertically-linked along the c -axis. The φ angle was found to be somewhat smaller in γ -Li 2 IrO , 3 but apart from this difference the magnetic structure in γ -Li 2 IrO 3 can be regarded as 'equivalent' to that in β -Li 2 IrO 3 but in a different lattice setting (for a formal mapping see Appendix E). Those similarities are strongly suggestive that the defining features of the magnetic structure, namely
FIG. 6: (color online) a) Projection of the magnetic structure on the ac plane showing the sites 1-4 of the primitive cell. Left curly arrows indicate counter-rotation of moments between consecutive sites along c . In unit cell 2 light and dark shaded elliptical envelopes indicate an alternating tilt of the plane of moments' rotation away from the ac face. In unit cell 3 the color of bonds shows the anisotropy axis of the exchange in a Kitaev model 15 (ferromagnetic Ising exchange for each bond, but with a different Ising axis x y , , z for the blue/green/red bonds). Right-hand labels (b)-(e) indicate where slices through the magnetic structure are taken at different heights in the unit cell and projected onto the ab plane to illustrate the direction of the zig-zag chains and the alternating tilt of the plane of rotation away from the ac plane by ± φ between adjacent zig-zag chains stacked along c .
FIG. 7: (Color online) Projection of the iridium lattice on the ac plane showing the ordering of the magnetic moment components along the b -axis only. These are shown rotated (for ease of visualization) from the b to the a -axis and indicated by horizontal red arrows (length of arrow indicates magnitude of M y ( r ) at each site. Note the ferromagnetic alignment between the two sites of each vertical ( c -axis) bond.
non-coplanarity, the counter-rotation on every nearestneighbor bond, and the direction and even the magnitude of the incommensurate ordering wavevector, are determined by the same type of short-range magnetic interactions in both polytypes, and those features appear to be robust against changes in the global lattice connectivity (i.e. how zig-zag chains are vertically-linked in a long-range pattern) which leave the local connectivity of the zig-zag chains unchanged. In the γ -polytype it was shown 20 that the key features of the magnetic order can be stabilized by a spin Hamiltonian with dominant Kitaev couplings with some additional interactions; the observation of similar magnetic structures in the β -and γ -polytypes suggests the same underlying magnetic interactions occur in both cases. The relevant parent Kitaev model for the β -polytype has been considered in Refs. 15,16 and is illustrated in Fig. 6a) (unit cell 3) where the color of the bonds indicates the anisotropy axis of the exchange, i.e. a ferromagnetic Ising coupling between the components normal to the Ir-O -Ir plane of 2 each Ir-Ir bond, those directions define the cubic axes x , y and z related to the crystallographic axes by ˆ x = ( ˆ a + ) ˆ c / √ 2, ˆ y = ( ˆ a -ˆ c ) / √ 2 and ˆ z = ˆ b (where we have assumed the parent 'idealized' lattice with cubic IrO 6 octahedra and a b c : : =1: √ 2:3). We note that the magnetic interactions may also be very similar in magnitude between the β and γ -polytypes, as both materials show very similar values for the magnetic ordering temperature.
## V. CONCLUSIONS
We have reported a combined study using magnetic neutron powder diffraction and single-crystal magnetic resonant x-ray diffraction experiments at the L 3 edge of Ir to explore the magnetic structure of β -Li 2 IrO , which 3 is a unique realization of a hyperhoneycomb iridium lattice, a candidate to display 3D Kitaev physics of strongly-frustrated, bond-anisotropic interactions. Using the azimuth dependence of the magnetic Bragg peak intensities combined with magnetic symmetry analysis we have obtained a complete solution for the magnetic structure for all 16 iridium sites in the structural unit cell. We find an incommensurate, non-coplanar magnetic structure with moments counter-rotating on
- 1 for a review see W. Witczak-Krempa, G. Chen, Y. B. Kim, L. Balents, Annual Review of Condensed Matter Physics 5 , 57 (2014).
- 2 B. J. Kim, H. Ohsumi, T. Komesu, S. Sakai, T. Morita, H. Takagi, and T. Arima, Science 323 , 1329 (2009).
- 3 G. Jackeli and G. Khaliullin, Phys. Rev. Lett. 102 , 017205 (2009).
- 4 J. Chaloupka, G. Jackeli and G. Khaliullin, Phys. Rev. Lett. 105 , 027204 (2010).
- 5 A. Kitaev, Ann. Phys. (N.Y.) 321 , 2 (2006).
- 6 J. Knolle, D. L. Kovrizhin, J. T. Chalker, and R. Moessner, Phys. Rev. Lett. 112 , 207203 (2014).
- 7 Y. Singh and P. Gegenwart, Phys. Rev. B 82 , 064412 (2010).
- 8 X. Liu, T. Berlijn, W.-G. Yin, W. Ku, A. M. Tsvelik, Young-June Kim, H. Gretarsson, Yogesh Singh, P. Gegenwart, and J. P. Hill, Phys. Rev. B 83 , 220403(R) (2011).
- 9 S. K. Choi, R. Coldea, A. N. Kolmogorov, T. Lancaster, I. I. Mazin, S. J. Blundell, P. G. Radaelli, Yogesh Singh, P. Gegenwart, K. R. Choi, S.-W. Cheong, P. J. Baker, C. Stock, and J. Taylor, Phys. Rev. Lett. 108 , 127204 (2012).
- 10 F. Ye, S. Chi, H. Cao, B. C. Chakoumakos, J. A. Fernandez-Baca, R. Custelcean, T. F. Qi, O. B. Korneta, and G. Cao, Phys. Rev. B 85 , 180403(R) (2012).
- 11 H. Gretarsson, J. P. Clancy, Y. Singh, P. Gegenwart, J. P. Hill, J. Kim, M. H. Upton, A. H. Said, D. Casa, T. Gog, and Y.-J. Kim, Phys. Rev. B 87 , 220407(R) (2013).
- 12 M. J. O'Malley, H. Verweij and P.M. Woodward, J. Solid State Chem. 181 , 1803 (2008).
- 13 Y. Singh, S. Manni, J. Reuther, T. Berlijn, R. Thomale, W. Ku, S. Trebst, and P. Gegenwart, Phys. Rev. Lett. 108 , 127203 (2012).
- 14 S. Mandal and N. Surendran, Phys. Rev. B 79 , 024426 (2009).
- 15 E. K.-H. Lee, R. Schaffer, S. Bhattacharjee, and Y. B. Kim, Phys. Rev. B 89 , 045117 (2014).
- 16 I. Kimchi, J. G. Analytis, A. Vishwanath, arXiv:1309.1171 (2013).
- 17 E. K.-H. Lee and Y. B. Kim, arXiv:1407.4125 (2014).
- 18 T. Takayama, A. Kato, R. Dinnebier, J. Nuss, and H. Takagi, arXiv:1403.3296 (2014).
- 19 K. A. Modic, T. E. Smidt, I. Kimchi, N. P. Breznay, A. Biffin, S. Choi, R. D. Johnson, R. Coldea, P. Watkins-Curry,
every nearest-neighbor bond. The magnetic structure shows striking similarities to the magnetic structure of the related structural polytype γ -Li 2 IrO , 3 suggesting that the same underlying Hamiltonian (with dominant Kitaev interactions) stabilizes the defining features of the magnetic structure in both structural polytypes.
## VI. ACKNOWLEDGEMENTS
Work at Oxford was supported by EPSRC (UK) and at Diamond and ISIS by STFC (UK). Work at Augsburg was supported by the Helmholtz Virtual Institute 521 ('New states of matter and their excitations').
- G. T. McCandless, J. Y. Chan, F. Gandara, Z. Islam, A. Vishwanath, A. Shekhter, R. D. McDonald, and J. G. Analytis, Nature Comm. 5 , 4203 (2014).
- 20 A. Biffin, R. D. Johnson, I. Kimchi, R. Morris, A. Bombardi, J. G. Analytis, A. Vishwanath, and R. Coldea, arXiv:1407.3954 (2014).
- 21 S. Boseggia, R. Springell, H. C. Walker, H. M. Rønnow, Ch. R¨egg, H. Okabe, M. Isobe, R. S. Perry, S. P. Collins, u and D. F. McMorrow, Phys. Rev. Lett. 110 , 117207 (2013).
- 22 J. Rodriguez-Carvajal, Physica B. 192 , 55 (1993).
- 23 S.C. Miller and W.F. Love, Tables of Irreducible Representations of Space Groups and Co-Representations of Magnetic Space Groups (Pruett, Boulder, 1967)
- 24 Resonant Elastic X-ray Scattering , Aussois, France (2011) 25 J. P. Hill and D. F. McMorrow, Acta Cryst. A 52 , 236, (1996).
- 26 http://forge.ill.eu/svn/magnetix
## Appendix A: Sample synthesis, characterization and low-temperature structural refinement
The synthesis route to obtain the samples used in the neutron and x-ray experiments was as follows. First, a powder sample of the layered honeycomb phase α -Li IrO 2 3 was prepared as described in Ref. 13. Then repetitive annealing at 1100 C transformed the sample ◦ into hyperhoneycomb β -Li 2 IrO , 3 confirmed via powder x-ray diffraction. Bulk magnetic susceptibility and heat capacity measurements on the powder sample indicated evidence for magnetic ordering at 38 K, in full agreement with Ref. 18. The polycrystalline powder contained small shiny crystallites. Several such crystallites were extracted from the powder and their diffraction pattern measured using a Mo source single-crystal SuperNova xray diffractometer. Representative diffraction patterns in the ( hk 0) and ( h l 1 ) planes obtained from the sample used in the magnetic resonant x-ray diffraction experiments (diameter 17 µ m) are shown in Fig. 8. The patterns observe sharp Bragg peaks, indicating a highquality single crystal, with selection rules consistent with
the Fddd space group expected for β -Li 2 IrO . 3 The re-
FIG. 8: (color online) X-ray diffraction pattern at 300 K in the ( hk 0) and ( h l 1 ) planes from the crystal used in the resonant x-ray diffraction experiments.
sults of the full structural refinement of a complete xray diffraction data set collected at 100 K from another crystal of comparable dimensions from the same batch are given in Table III where we have used the same convention for the origin of the unit cell as in Ref. 18. In the table A.D.P. stands for Atomic Displacement Parameters, which were assumed to be isotropic. The reliability factors of the Rietveld refinement were R F ( 2 ) = 4.50%, wR F ( 2 ) = 5.24%, R F ( ) = 2.42% and χ 2 = 0.318.
TABLE III: β -Li 2 IrO 3 structural parameters at 100 K
## Cell parameters
Space group: Fddd (#70, origin choice 2) Z = 16
a b c , , ( ˚ ): A 5.8903(2) 8.4261(3) 17.7924(7)
$$\alpha, \beta, \gamma \colon \quad 9 0 ^ { \circ } \quad \ 9 0 ^ { \circ } \quad \ 9 0 ^ { \circ }$$
Volume ( ˚ A ): 883.08(6) 3
Atomic fractional coordinates and isotropic A.D.P.'s
| Atom | x | y | z | U iso (A ˚ 2 ) |
|--------|----------|------------|------------|------------------|
| Ir | 0.125 | 0.125 | 0.70845(7) | 0.0025(3) |
| Li1 | 0.125 | 0.125 | 0.04167 | 0.00633 |
| Li2 | 0.125 | 0.125 | 0.875 | 0.00633 |
| O1 | 0.855(7) | 0.125 | 0.125 | 0.002(5) |
| O2 | 0.621(8) | 0.3669(19) | 0.0384(7) | 0.002(3) |
## Data collection
# measured reflections: 3770
Data reduction R int :
6.99%
(Criterion for observed reflections: I > 2.0 σ I ( ))
# observed independent reflections: 298
# fitted parameters: 9
## Appendix B: Magnetic Structure and Fourier Decomposition
This section lists the coordinates of the four iridium ions in the primitive cell and also gives explicitly the expressions for the magnetic moments at all sites in the unit cell and their Fourier decomposition. The iridium ions occupy a single crystallographic site with generating position (1 / 8, 1 / 8, z ) with z = 0.70445(7) and the four sites per primitive cell (labelled 1-4 in Fig. 6a) (left panel) have the coordinates listed in Table. IV.
TABLE IV: Fractional atomic coordinates of the iridium sites in the primitive cell and corresponding magnetic basis vector components in the determined magnetic structure.
| Site | Coordinates | v x | v y | v z |
|--------|----------------------------|-------|-------|-------|
| 1 | (0.125, 0.125, z ) | + | + | + |
| 2 | (0.125, 0.625, 3 / 4 - z ) | - | + | + |
| 3 | (0.375, 0.375, 1 - z ) | - | - | + |
| 4 | (0.375, 0.875, 1 / 4 + z ) | + | - | + |
The determined magnetic structure is described by the basis vector combination ( iA x , iC y , F z ) with magnitudes M x , M y and M z . The magnetic moment at position r belonging to site index n = 1 -4 is given by
$$M _ { n } ( r ) \ & = \ \hat { x } M _ { x } v _ { x } ( n ) \sin q \cdot r + \hat { y } M _ { y } v _ { y } ( n ) \sin q \cdot r \\ & + \ \hat { z } M _ { z } v _ { z } ( n ) \cos q \cdot r,$$
where ˆ x y , ˆ , ˆ z are unit vectors along the orthorhombic a , b , c axes, respectively. The pre-factors v x y z , , are obtained from the basis vectors along the corresponding axes, but without the displacement phase factors e -i q · ( r n -r 1 ) , i.e. v x = [+ --+] (from A x ), v y = [+ + --] (from C y ) and v z = [+ + ++] (from F z ). Eq. (B1) describes all iridium sites, including those related by F -centering translations, where r is the actual position of the ion and n is the site index at the equivalent position (1 -4) in the primitive unit cell.
The magnetic moments are expressed in terms of their Fourier components as M r n ( ) = ∑ k = ± q M k , n e -i k r · , where
$$\text{where} \\ M _ { q, n } \, & = \, \left \{ i \left [ \hat { x } \frac { M _ { x } } { 2 } v _ { x } ( n ) + \hat { y } \frac { M _ { y } } { 2 } v _ { y } ( n ) \right ] \\ & \quad + \, \hat { z } \frac { M _ { z } } { 2 } v _ { z } ( n ) \right \} e ^ { - i q \cdot ( r _ { n } - r _ { 1 } ) }, \quad \text{(B2)} \\ \dots \, \vec { \epsilon } \quad \vec { \epsilon } \quad \dots \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \,.$$
with M -q , n = M ∗ q , n as the magnetic moment distribution is real.
## Appendix C: Magnetic Structure Factors
This section gives the analytic expressions for the magnetic structure factors for all the magnetic basis vectors.
Starting from the general expression in eq. (2) we obtain the magnetic structure factors for the C A , , and G basis vectors as follows
$$\text{vectors as follows} \\ \mathcal { F } ^ { C } ( Q ) = \begin{cases} \ 1 6 M _ { \pm q, 1 } \cos \frac { \pi l } { 6 } e ^ { i \xi _ { \pm } }, \ h, k, l \text{ all even and} \\ h + k + l = 4 p + 2, p \text{ integer} \\ \ 1 6 M _ { \pm q, 1 } \cos \frac { \pi l } { 6 } \sin \frac { \pi ( h + l - k ) } { 4 } e ^ { i ( \chi _ { \pm } + \pi / 2 ) }, \\ h, k, l \text{ all odd} \\ 0, & \text{ otherwise} \end{cases}$$
$$\mathcal { F } ^ { A } ( Q ) = \begin{cases} \ 1 6 M _ { \pm q, 1 } \sin \frac { \pi l } { 6 } e ^ { i ( \xi _ { \pm } - \pi / 2 ) }, \ h, k, l \, \text{all even and} } & \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \ \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \end{cases} \\ \mathcal { F } ^ { A } ( Q ) = \begin{cases} \ 1 6 M _ { \pm q, 1 } \sin \frac { \pi l } { 6 } e ^ { i ( \xi _ { \pm } - \pi / 2 ) }, \ h, k, l \, \text{all even and} } & \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad\, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad\, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \ \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \ \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \,
\quad \ 1 6 M _ { \pm q, 1 } \sin \frac { \pi l } { 6 } \cos \frac { \pi ( h + l - k ) } { 4 } e ^ { i ( \chi _ { \pm } - \pi / 2 ) }, \\ \ h, k, l \, \text{all odd} } & \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad\, \ \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \ \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \,
\quad \ 0, & \quad \text{otherwise} & \quad \quad \quad$$
$$\mathcal { F } ^ { G } ( Q ) = \begin{cases} 1 6 M _ { \pm q, 1 } \sin \frac { \pi l } { 6 } e ^ { i ( \xi _ { \pm } + \pi / 2 ) }, \ h, k, l \, \text{all even and } } \text{entation} \\ h + k + l = 4 p + 2, p \, \text{integer} \\ 1 6 M _ { \pm q, 1 } \sin \frac { \pi l } { 6 } \sin \frac { \pi ( h + l - k ) } { 4 } e ^ { i ( \chi _ { \pm } + \pi ) }, & \text{the a ve} \\ h, k, l \, \text{all odd} \\ 0, & \text{otherwise}, & \text{the sens} \\ \text{ifed} \,. \\ \text{ing plan} \\ \end{cases}$$
where to obtain an analytic closed-form we used the 'ideal' iridium positions z =5/8+1/12.
## Appendix D: Azimuth Dependence of the Magnetic Resonant X-ray Diffraction Intensity
Here we obtain the azimuth dependence of the magnetic x-ray diffraction intensity and show how it can be used to determine the relative phase between magnetic basis vectors. As discussed in Sec. III C the magnetic diffraction intensity is proportional to |F | ‖ 2 , where F ‖ = F · ˆ k ′ is the projection of the structure factor vector onto the scattered beam direction. To calculate this dot product it is convenient to express both vectors in terms of their components in the Laboratory reference frame 25 illustrated in Fig. 9 and defined by the orthonormal set of vectors ˆ U 2 = ˆ k × ˆ k ′ , ˆ U 3 = -ˆ Q and ˆ U 1 = ˆ U 2 × ˆ U 3 , i.e. ˆ U 2 is normal to the scattering plane and ˆ U 1 is in the scattering plane making an angle θ with both k and k ′ , where 2 θ is the total scattering angle. The scattered beam direction then can be written as ˆ k ′ = (cos θ , 0, -sin θ ) L , where the L subscript indicates that the components are with respect to the Laboratory basis of ˆ U 1,2,3 vectors, as opposed to the Crystal basis, defined by the orthonormal set of vectors ˆ x y , ˆ , ˆ z (along the orthorhombic axes a b , , c ).
The magnetic structure factor vector F defined in (2) is most naturally expressed in terms of the Crystal basis as ( F x , F y , F z ) C . The components in the Laboratory basis ( F 1 , F 2 , F 3 ) L are obtained via a linear transformation
FIG. 9: Schematic diagram of the x-ray diffraction experiment at zero azimuth (Ψ = 0). k and k ′ are the incident and scattered x-ray beam wavevectors, and Q = k ′ -k is the scattering wavevector. Main (bottom) panel shows the orientation of the Laboratory frame unit vectors ˆ U 1,2,3 and the orthorhombic crystal axes ( a b c , , ). Top left inset: view along b showing the angle ζ subtended from the scattering plane by the a vector. Right inset: view along Q . During an azimuth scan the crystal axes are rotated around Q by an angle Ψ (in the sense show in the figure), the reference position Ψ = 0 is defined as the azimuth angle for which b lies in the scattering plane making the smallest angle with the incident beam direction ˆ k .
encoded in a matrix T (Ψ) via
$$\left [ \begin{array} { c } \mathcal { F } _ { 1 } \\ \mathcal { F } _ { 2 } \\ \mathcal { F } _ { 3 } \end{array} \right ] = T ( \Psi ) \left [ \begin{array} { c } \mathcal { F } _ { x } \\ \mathcal { F } _ { y } \\ \mathcal { F } _ { z } \end{array} \right ], \quad \quad \text{(D1)}$$
where we have explicitly introduced the dependence of the transformation matrix on the azimuth angle Ψ, which defines the rotation angle of the crystal axes around the scattering wavevector Q . To find the transformation matrix we first consider the reference position of zero azimuth, Ψ = 0, depicted in Fig. 9. In this case the orientation of the crystal axes with respect to the Laboratory axes is specified by two angles η and ζ . η is the angle between the b -axis and the scattering wavevector Q , given by
$$\eta = \cos ^ { - 1 } \hat { Q } \cdot \hat { b }$$
and ζ is the angle between the a -axis and the scattering plane, given by
$$\zeta = \tan ^ { - 1 } \frac { Q \cdot \hat { c } } { Q \cdot \hat { a } } \.$$
Using the diagram in Fig. 9 the transformation matrix is
obtained as
$$T ( 0 ) = \left [ \begin{array} { c c c } - \cos \zeta \cos \eta & \sin \eta & - \sin \zeta \cos \eta \\ \sin \zeta & 0 & - \cos \zeta \\ - \cos \zeta \sin \eta & - \cos \eta & - \sin \zeta \sin \eta \end{array} \right ]. \quad ( D 2 )$$
When moving away from the reference position to a finite azimuth angle Ψ the coordinates in the Laboratory frame need to be multiplied by the rotation matrix such that the resulting transformation matrix from the Crystal to the Laboratory bases in (D1) becomes
$$T ( \Psi ) = R ( \Psi ) \ T ( 0 ). \text{ \quad \ \ } \text{(D4)}$$
$$R ( \Psi ) = \left [ \begin{array} { c c c } \cos \Psi & \sin \Psi & 0 \\ - \sin \Psi & \cos \Psi & 0 \\ 0 & 0 & 1 \end{array} \right ], \quad \text{(D3)} \quad \text{Ex}$$
Explicitly, substituting (D2) and (D3) into (D4) yields
$$T ( \Psi ) = \left [ \begin{array} { c c c } \sin \Psi \sin \zeta - \cos \Psi \cos \zeta \cos \eta & \cos \Psi \sin \eta & - \cos \Psi \sin \zeta \cos \eta - \sin \Psi \cos \zeta \\ \sin \Psi \cos \zeta \cos \eta + \cos \Psi \sin \zeta & - \sin \Psi \sin \eta & \sin \Psi \sin \zeta \cos \eta - \cos \Psi \cos \zeta \\ - \cos \zeta \sin \eta & - \cos \eta & - \sin \zeta \sin \eta \end{array} \right ]. \quad \quad ( D 5 )$$
Once the structure factor components are obtained in the Laboratory basis, then the projection of the structure factor vector along the scattered beam direction is readily obtained as
$$\mathcal { F } _ { | | } = \cos \theta \ \mathcal { F } _ { 1 } - \sin \theta \ \mathcal { F } _ { 3 }. \quad \quad ( \text{D6} ) \quad \stackrel { \cdots } { \text{of t} }$$
To illustrate the azimuth dependence of the magnetic x-ray diffraction intensity we consider the magnetic peak at ( -6, 6, 8)+ q in Fig. 5a). Here only the A x and F z basis vectors contribute, therefore F = ( F x , 0, F z ) C where the structure factor components F x and F z are of the type F A and F F in (C2,3), respectively. Using (D6) combined with (D1) and (D5) we obtain
The cross-term is therefore directly sensitive to the phase difference between the two contributing basis vectors. If F z and A x are in-phase or π out-of-phase (denoted as F z ± A x ) the product F F z ∗ x is purely imaginary, so the cross-term cancels and the intensity is given by the sum of the contributions from the two basis vectors taken separately. This is clearly inconsistent with the data (purple curve in Fig. 5a)). A phase difference of ± π/ 2 between the two basis vectors (denoted as F z ± iA x ) gives a finite cross-term, which changes sign between those two cases, and the data in Fig. 5a) identifies the basis vector combination as F z + iA x (red solid line).
$$\mathcal { F } _ { | | } ( \Psi ) = g _ { x } ( \Psi ) \ \mathcal { F } _ { x } + g _ { z } ( \Psi ) \ \mathcal { F } _ { z },$$
where the geometric factors are
$$g _ { x } ( \Psi ) \ = \ & ( \sin \Psi \sin \zeta - \cos \Psi \cos \zeta \cos \eta ) \cos \theta + \\ & \cos \zeta \sin \eta \sin \theta \\ g _ { z } ( \Psi ) \ = \ & \sin \zeta \sin \eta \sin \theta - \\ & \quad ( \cos \Psi \sin \zeta \cos \eta + \sin \Psi \cos \zeta ) \cos \theta.$$
The magnetic scattering intensity is then proportional to
$$| \mathcal { F } _ { | | } ( \Psi ) | ^ { 2 } & = | g _ { x } ( \Psi ) \ \mathcal { F } _ { x } | ^ { 2 } + | g _ { z } ( \Psi ) \ \mathcal { F } _ { z } | ^ { 2 } \\ & + 2 \text{Re} ( \mathcal { F } _ { z } \mathcal { F } _ { x } ^ { * } ) \ g _ { x } ( \Psi ) g _ { z } ( \Psi ),$$
where Re() means the real part. The first two terms give the sum of the separate contributions from the two basis vectors, whereas the last (cross-term) is sensitive to the relative phases between the two basis vectors. Using eqs. (3,C2) the cross-term is proportional to
$$\mathcal { F } _ { z } \mathcal { F } _ { x } ^ { * } \in \begin{cases} \mp i \sin \frac { \pi l } { 6 } \cos \frac { \pi l } { 6 } &, & F _ { z } \pm A _ { x } \\ \pm \sin \frac { \pi l } { 6 } \cos \frac { \pi l } { 6 } &, & F _ { z } \pm i A _ { x }. \end{cases}$$
TABLE V: Equivalence between the iridium sites in γ -Li 2 IrO 3 ( Cccm space group) with those in β -Li 2 IrO 3 ( Fddd space group) (when both are projected onto the ac plane). Last column shows additional translations to be applied to primitive lattice sites in the Fddd structure (column 2) in addition to a common translation by (-1/8,0,1/8) to get the positions in the Cccm structure (column 1). For each structure the 'ideal' iridium positions have been used.
| Site Cccm | Site Fddd | Translation |
|-------------|-------------|---------------|
| 1 | 4 | -(0,0,1) |
| 2 | 3 | (0,0,0) |
| 3 | 3 | (0,1/2,1/2) |
| 4 | 4 | -(0,1/2,1/2) |
| 1 ′ | 2 | (1/2,1/2,0) |
| 2 ′ | 1 | (1/2,0,-1/2) |
| 3 ′ | 1 | (1/2,1/2,0) |
| 4 ′ | 2 | (1/2,1,1/2) |
## Appendix E: Mapping of magnetic basis vectors between β - and γ -Li IrO 2 3
In this section we describe a formal mapping of the magnetic basis vectors between the notation used for the β - (space group Fddd ) and γ -polytypes of Li 2 IrO 3 (space group Cccm ). The iridium lattice arrangement in both crystal structures is made up of zig-zag chains verticallylinked along the c -axis, with the only difference being that in the β -polytype successive chains along c alternate in orientation between the a ± b directions, whereas in the γ -polytype a pair of successive chains are parallel and this direction then alternates for the next pair of parallel chains and so on, between the a ± b directions. When both structures are viewed in projection onto the ac plane [as shown in Fig. 6a) and Fig. 4 of Ref. 20] the two iridium lattice arrangements then look identical. This suggests that a formal mapping can be constructed between the notations used to describe the sites of the primitive cells of the two materials, and also between the magnetic basis vectors, such that one can then describe a magnetic structure equivalently in the notation of one or the other space group (of course the magnetic structures would only be 'equivalent' up to the projection of the sites onto the b -axis). Table V lists the equivalence between the sites in the primitive cell of the two structures (up to the y -coordinate of sites), where for the γ -polytype we follow the notation in Ref. 20 (two iridium sublattices with sites 1-4 and 1 -4 ). ′ ′ Using this site equivalence one can construct a mapping of the magnetic basis vectors as listed in Table VI. The magnetic structure described by
TABLE VI: Mapping of magnetic basis vectors between the Fddd and Cccm notations.
| Basis vector | Basis vector |
|---------------------|-------------------------------|
| Fddd (4-site [1-4]) | Cccm (8-site,[1-4,1 ′ -4 ′ ]) |
| A [+ -- +] | ( A , - A ) [+ -- + - ++ - ] |
| C [+ + -- ] | ( - F , F ) [ ---- ++++] |
| F [+ +++] | ( F , F ) [+ +++++++] |
the basis vector combination ( iA x , iC y , F z ) in the Fddd setting then corresponds, in the notation of the Cccm setting, to i ( A , -A ) x , i ( -1) m ( F , -F ) y , ( F F , ) z with m = 1. | 10.1103/PhysRevB.90.205116 | [
"A. Biffin",
"R. D. Johnson",
"Sungkyun Choi",
"F. Freund",
"S. Manni",
"A. Bombardi",
"P. Manuel",
"P. Gegenwart",
"R. Coldea"
] | 2014-08-01T17:56:02+00:00 | 2014-08-01T17:56:02+00:00 | [
"cond-mat.str-el"
] | Unconventional magnetic order on the hyperhoneycomb Kitaev lattice in $β$-Li2IrO3: full solution via magnetic resonant x-ray diffraction | The recently-synthesized iridate $\beta$-Li$_2$IrO$_3$ has been proposed as a
candidate to display novel magnetic behavior stabilized by frustration effects
from bond-dependent, anisotropic interactions (Kitaev model) on a
three-dimensional "hyperhoneycomb" lattice. Here we report a combined study
using neutron powder diffraction and magnetic resonant x-ray diffraction to
solve the complete magnetic structure. We find a complex, incommensurate
magnetic order with non-coplanar and counter-rotating Ir moments, which
surprisingly shares many of its features with the related structural polytype
"stripyhoneycomb" $\gamma$-Li$_2$IrO$_3$, where dominant Kitaev interactions
have been invoked to explain the stability of the observed magnetic structure.
The similarities of behavior between those two structural polytypes, which have
different global lattice topologies but the same local connectivity, is
strongly suggestive that the same magnetic interactions and the same underlying
mechanism governs the stability of the magnetic order in both materials,
indicating that both $\beta$- and $\gamma$-Li$_2$IrO$_3$ are strong candidates
to realize dominant Kitaev interactions in a solid state material. |
1408.0247v2 | Revised annotations, sex-biased expression, and lineage-specific genes in the Drosophila melanogaster group
Rebekah L. Rogers , Ling Shao , Jaleal S. Sanjak 1 1 1 , Peter Andolfatto , 2 and Kevin R. Thornton 1
## Research Article
- 1) Ecology and Evolutionary Biology, University of California, Irvine
- 2) Ecology and Evolutionary Biology and the Lewis Sigler Institute for Integrative Genomics, Princeton University
Running head: Gene annotations and sex-biased expression in Drosophila
Key words: Gene annotation, sex-specific gene expression, polycistronic genes, gene fusion, D. simulans , D. yakuba , D. ananassae .
Corresponding author: Rebekah L. Rogers, Dept. of Ecology and Evolutionary Biology, 5323 McGaugh Hall, University of California, Irvine, CA 92697
Phone:
949-824-0614
Fax:
949-824-2181
Email:
[email protected]
## Abstract
Here, we provide revised gene models for D. ananassae , D. yakuba , and D. simulans , which include UTRs and empirically verified intron-exon boundaries, as well as ortholog groups identified using a fuzzy reciprocal-best-hit blast comparison. Using these revised annotations, we perform differential expression testing using the cufflinks suite to provide a broad overview of differential expression between reproductive tissues and the carcass. We identify thousands of genes that are differentially expressed across tissues in D. yakuba and D. simulans , with roughly 60% agreement in expression patterns of orthologs in D. yakuba and D. simulans . We identify several cases of putative polycistronic transcripts, pointing to a combination of transcriptional read-through in the genome as well as putative gene fusion and fission events across taxa. We furthermore identify hundreds of lineage specific genes in each species with no blast hits among transcripts of any other Drosophila species, which are candidates for neofunctionalized proteins and a potential source of genetic novelty.
## Introduction
Accurate models of gene structure including UTRs, intron-exon boundaries, as well as coding sequences are essential for proper interpretation of molecular genetics ( Fire et al. 1998, Jinek et al. 2012), demographic inference ( Halligan and Keightley 2006, Parsch et al. 2010, Clemente and Vogl 2012), tests of selection ( McDonald and Kreitman 1991), and comparative genomics ( Chen et al. 2014). The Drosophila offer an excellent model for comparative genomics, with high-quality sequenced genomes for 12 species( Drosophila Twelve Genomes Consortium 2007) as well as draft genomes for an additional 8 species ( Chen et al. 2014) spanning a total of 63 MY ( Tamura et al. 2004). Previous gene models provided for the 12 Drosophila genomes focused on gene prediction with the aid of homology to establish putative annotations of coding sequences across taxa with 15,000-16,000 genes for most species ( Drosophila Twelve Genomes Consortium 2007). These gene models produce reliable annotations for conserved genes as well as genes that are present in multiple species but will lack lineage specific genes where sources of genetic novelty add to the genome, will misidentify rapidly evolving sequences, and will neglect isoforms that may offer alternative molecular functions.
Recent work has validated gene models in D. melanogaster through cross-species comparisons ( Chen et al. 2014). While aligned CDS sequences often display patterns of
expression consistent with conservation, gene structure varies across taxa at UTRs, introns, and noncoding DNA ( Chen et al. 2014). Furthermore, any gene families and functional classes subject to rapid evolution in gene structure which is unlikely to be reflected in homology based annotations ( Wasbrough et al. 2010) and de novo genes will also be absent in spite of their role in developing functional diversity ( Zhao et al. 2014).
Beyond the evolution of gene structure, expression patterns of genes across tissues is expected to influence their evolutionary constraints and rates of sequence evolution Van Dyken and Wade (2010) as well as their functional roles within the organism. Sex-biased genes are often rapidly evolving Ranz et al. (2003), Ellegren and Parsch (2007) especially among male reproductive proteins Zhang et al. (2007), Meisel (2011), Haerty et al. (2007). Moreover, sex specific expression patterns can influence the extent to which genes contribute to sexual antagonism, influencing their role in the evolution of sexual dimorphism Mank et al. (2008). Correctly identifying sex-specific and tissue specific expression is therefore essential for studying the molecular and evolutionary impacts of genes within species. Moreover, changes in tissue specific or sex-specific expression may point to rapid evolution of gene functions and candidate loci to search for signals of evolutionary change and differential selective effects between the sexes.
Here, we describe RNA-seq based gene annotations for Drosophila outgroup species D. simulans , D. yakuba , and D. ananassae based on whole transcriptome sequencing of male and female adult tissues. These revised gene models capture hundreds of lineage specific genes on major chromosomal arms in D. yakuba and D. simulans . We also describe 5 ′ and 3 ′ UTRs, and intron-exon structure for genes throughout the D. melanogaster group. Finally, we describe sex-biased expression across species, identifying thousands of genes that are differentially expressed across tissues. These revised gene models as well as results of sex-biased and tissue-biased differential expression testing should serve as a resource for the Drosophila evolutionary and molecular genetics community interested in evolution, conservation, and gene expression.
## Methods
## Sample preparation
Fly stocks were incubated under controlled conditions at 25 ℃ and ∼ 40% humidity. Virgin flies were collected within 2 hours of eclosion, then aged 2-5 days post eclosion before dissection. We dissected samples in isotonic Ringers solution, using female ovaries and headless gonadectomized carcass from two adult flies as well as testes plus glands and male headless gonadectomized carcass for four adult flies for each sample RNA prep. We collected three biological replicates of the D. yakuba reference, three biological replicates of the D. simulans w 501 reference, and one replicate of the D. ananassae reference and one replicate of the D. melanogaster reference (stock numbers in Table 1). Samples were flash frozen in liquid nitrogen immediately after dissection, and and stored in 0.2ml Trizol at -80 ℃ . All samples were homogenized in 0.5ml Trizol Reagent (Invitrogen) with plastic pestle in 1.5ml
tube, mixed with 0.1ml chloroform, and centrifuged 12,000g 15min at 4oC, as Trizol RNA extraction protocol. The RNAs in the supernatant about 0.4ml were then collected and purified with Direct-Zol RNA MiniPrep Kit (Zymo), following the standard protocol. The total RNAs were eluted in 65 µ L RNase-Free H O. About 1 2 µ g purified RNAs were treated with 2 µ L Turbo DNase (Invitrogen) in 65 µ L reaction, incubated 15min at room temperature with gentle shaking. These RNAs were further purified with RNA Clean and Concentrator-5 (Zymo). One extra wash with fresh 80% ethanol after the final wash step was added into the original protocol. The treated RNAs were eluted with 15 µ L RNAse-Free H O, and stored 2 at -80 ℃ .
The amplified cDNAs were prepared from 100ng DNase treated RNA with Ovation RNA-Seq System V2 (Nugen) and modified protocol. The preparations followed the protocol to the step of SPIA Amplification (Single Primer Isothermal Amplification). The amplified cDNAs were first purified with Purelink PCR Purification Kit (Invitrogen, HC Binding Buffer) and eluted in 100 µ L EB (Invitrogen). These cDNAs were purified again to 25 mu L EB with DNA Clean and Concentrator -5 Kit (Zymo) for Nextera library preparation. About 43ng cDNAs were used to construct libraries with Nextera DNA Sample Preparation Kit (Illumina) and modified protocol. After Tagmentation, Purelink PCR Purification Kit with HC Binding Buffer was used for purification and eluted with 30 µ L EB or H2O. The products (libraries) of final PCR amplification were purified with DNA Clean and Concentractor-5 and eluted in 20 µ L EB. The average library lengths about 500bp were estimated from profiles on a Bioanalyzer 2100 (Agilent) with DNA HS Assay. All libraries were normalized to 2-10nM based on real-time PCR method with Kapa Library Quant Kits (Kapa Biosystems). The qualities and quantities of these RNAs, cDNAs and final libraries were measured from Bioanalyzer with RNA HS or DNA HS Assays and Qubit (Invitrogen) with RNA HS or DNA HS Reagents, respectively. Samples were barcoded and sequenced using 76 bp reads in 4-plex on an Illumina HiSeq 2500 using standard Illumina barcodes, resulting in high coverage (Figure S1-S6). Coverage in RNA-seq data is dependent on expression level, but the majority of sites have coverage less than 50 reads. One replicate of female tissues was sequenced using single end reads. All other libraries were sequenced using paired end reads and dual indexing. Samples were sequenced as they became available, across the equivalent of one eight lane flow cell for the 32 samples in total.
## Trinity and Augustus gene annotation
RNA sequencing data for ovary, female carcass, testes, and male carcass of D. yakuba , D. simulans , and D. ananassae were concatenated into a single fastq file, and digitally normalized to remove excess redundant reads for highly expressed transcripts. This step results in tractable runtimes with no loss of information in transcript annotations. We ran Trinity http://trinityrnaseq.sourceforge.net/ ( Haas et al. 2013, Grabherr et al. 2011). For a single sample:
1. FASTQ files for left and right reads were concatenated. 2. The concatanated files were subject to 'digital normalization' using the following command in the Trinity package:
normalize\_by\_kmer\_coverage.pl --seqType fa --JM 100G --max\_cov 30 --left left.fa
--right right.fa --pairs\_together --PARALLEL\_STATS --JELLY\_CPU $CORES , where $ CORES is how many CPU we had available
- 3. The resulting normalized left and right fastq files where then used in the following command: Trinity.pl --seqType fa --bflyHeapSpaceInit 1G --bflyHeapSpaceMax 8G --JM 7G --left $LEFTFILE --right $RIGHTFILE --output trinity\_output --min\_contig\_length 300 --CPU $CORES --inchworm\_cpu $CORES --bflyCPU $CORES , where the variables are the normalized fastq files and the number of cores available.
- 4. Further detail is at https://github.com/ThorntonLab/annotation\_methods
The resulting annotations were used as input in the Augustus v2.5.5 gene prediction software http://bioinf.uni-greifswald.de/augustus/ with command line options --species=fly --hintsfile=$INFILE\_BASE.hints.E.gff
--extrinsicCfgFile=augustus.2.5.5/config/extrinsic/extrinsic.ME.cfg $INFILE --gff3=on --uniqueGeneId=true > $INFILE\_BASE.gff3 .
## Alignment and annotation
We matched Augustus gene models to previous annotations from FlyBase r1.3 for D. ananassae , and D. yakuba , or to the D. simulans w 501 reference annotations. CDS sequences were required to physically overlap with the location of a current FlyBase or w 501 gene model and were required to have matches to 85% or more CDS sequence with 90% or greater amino acid similarity in an all-by-all BLASTp of translated sequences at a cutoff of E ≤ 10 -10 with low-complexity filters turned off (-F F). We mapped RNA-seq data to known gene models annotated in FlyBase, D. yakuba r1.3, D. ananassae r1.3, D. melanogaster r.5.45, and the D. simulans w 501 gene models annotated by Hu et al. ( Hu et al. 2012). GFF Files were reformatted using gffread from Cufflinks suite. Sequences were mapped to the genome using Tophat v.2.0.6 ( Trapnell et al. 2009, Kim et al. 2013) and Bowtie2 v.2.0.2 ( Langmead et al. 2009), using reference annotations as a guide, with no attempt to identify novel transcripts using reads which fell outside reference annotations (-G) and all other parameters set to default. We used Cufflinks 2.0.2 ( Trapnell et al. 2012) to calculate expression levels across genes and transcripts, normalizing expression by the upper quantile (-N) and ignoring reads which fall outside known gene models (-G) with all other options set to default. Orphaned gene models from FlyBase or from Hu et al. (2012) which had FPKM ≥ 2 but which had no assembled gene model match from Augustus were included in the final annotations used for differential expression testing. Some annotations contain polycistronic transcripts encompassing multiple independent open reading frames. A portion of these polycistronic transcripts may reflect only low-level polycistronic transcription, rather than polycistronic transcripts serving as the dominant isoforms but the rate of polycistronic transcription cannot be readily determined with available data. For genes with polycistronic transcripts but no 1:1 transcript match with FlyBase gene models, we included annotations for both the polycistronic Augustus gene models and the gene models from FlyBase supporting independent transcripts.
## Sex-bias and Tissue Bias
The union of gene models from Augustus and orphaned gene models from FlyBase were combined into a single GFF containing transcript and CDS annotations for each species. We then re-mapped RNA-seq reads to gene models in the reannotated GFF for D. yakuba , D. ananassae , and D. simulans , as well as unmodified gene models for D. melanogaster r.5.45 with Tophat and performed differential expression testing at an FDR ≤ 0.1 normalizing expression by the upper quantile (-N) and ignoring reads which fall outside known gene models (-G) with all other options set to default using the Cufflinks suite according to the same criteria described above. We compared female carcass to female ovaries, male carcass to male testes, female carcass to male carcass, and female ovaries to male testes for each species, grouping replicates for reference genomes.
## Orthologs and lineage specific genes
Orthologs were identified using fuzzy reciprocal best hit BLASTp comparisons of all translations across reference genomes for gene model predictions of D. ananassae , D. yakuba , D. simulans , as well as D. melanogaster . Orthologs are similar to those previously used to annotate gene families in Drosophila ( Drosophila Twelve Genomes Consortium 2007, Hahn et al. 2007). Putative orthologs must be putative reciprocal hits of the same rank order, where genes with an E-value within a single log-unit of one another are assigned the same rank, using the best E-Value for a gene with a cutoff of E ≤ 10 -10 . Lineage specific genes were defined as genes with no hit in an all-by-all BLASTp of translations against translations of the other outgroups (e.g. D. yakuba , D. ananassae , and D. melanogaster for D. simulans ) at a cutoff of E ≤ 10 -10 with low-complexity filters turned off (-F F).
## Gene ontology
We used DAVID gene ontology analysis software http://david.abcc.ncifcrf.gov/ to determine whether any functional categories were overrepresented among genes with sex specific or tissue specific expression. Functional data for D. ananassae , D. yakuba and D. simulans are not readily available in many cases, and thus we identified functional classes in the D. melanogaster orthologs as classified in Flybase. Gene ontology clustering threshold was set to Low. Genes with tissue specific expression were based on genes with differential expression from cufflinks for comparisons of female carcass vs. female ovaries and male carcass vs. male testes at a genomewide FDR ≤ 0 10, according to cufflinks default settings. .
## Results
## Annotations
For moderately to highly expressed genes we recover gene structure with intron-exon boundaries and UTR sequences for full length transcripts including novel exons which were
previously unannotated based on comparative genomics (Figure 1-2). Many genes are part of polycistronic transcripts including one from D. melanogaster . We identify 2529 putative polycistronic transcripts in D. yakuba , 2379 in D. ananassae , and 561 in D. simulans . For such genes we offer gene models from FlyBase as well as fused models from Augustus. The extent to which such transcription of multiple genes is functional as opposed to a stochastic byproduct of transcriptional errors is unclear. We also include FlyBase gene models expressed in the reference genomes with no 1:1 match in gene models from Augustus. The addition of FlyBase gene models results in an additional 4265 annotations in D. yakuba , 5367 in D. ananassae , and 7419 in D. simulans . We identify a total of 22,989 transcripts for 16,278 genes in D. simulans , 20,315 transcripts for 17,579 genes in D. yakuba , and 22,420 transcripts for 20,580 genes in D. ananassae (Table 2). Compared to D. yakuba , for D. ananassae an additional 1,173 FlyBase gene models failed to match sufficiently well with RNA-seq supported gene models and these were added to the annotations, explaining some of the excess in the number of genes and also highlighting the difficulties of annotation through comparative genomics across large phylogenetic distances. In D. ananassae , 72% of gene models have RNA-seq data supporting 60% or more gene features (exons, UTRs) compared to 79.4% in D. yakuba and 80.1% in D. simulans (Table 3).
As defined by a fuzzy reciprocal best-hit blast ( Drosophila Twelve Genomes Consortium 2007, Hahn et al. 2007) we identify 12,127 genes in D. melanogaster with first-order orthologs in D. simulans , 11,425 with first-order orthologs in D. yakuba , and 11,348 with first-order orthologs in D. ananassae (Table 4). The increase in the number of genes with orthologs in D. simulans is the product of improved annotations as well as the improved assembly of the w 501 D. simulans reference( Hu et al. 2012). We observe a 1:1 concordance for 48% of FlyBase gene models in D. yakuba and 46% of FlyBase gene models for D. ananassae . These annotations typically include UTR sequences in addition to empirically supported intron-exon and coding sequence boundaries, an improvement over previous gene models from release r1.3 which lack UTRs ( Drosophila Twelve Genomes Consortium 2007). We further identify thousands of lineage specific genes in each species of Drosophila with no matching gene model in other outgroup species. While lineage specific genes identified on minor chromosomes could result from assembly issues, we identify hundreds on major chromosomal arms (Table 5) suggesting that many of these are in fact cases of lineage-specific gene formation.
## Sex-biased and tissue-biased expression
We observe thousands of genes with sex-biased or tissue-biased expression in D. yakuba and D. simulans , but hundreds of genes with sex-biased or tissue-biased expression in D. melanogaster and D. ananassae , a direct product of increasing power to detect differences in RNA levels with biological replicates. Gene ontology categories overrepresented between ovary and female carcass reflect differences in genes involved in reproduction, chromosome segregation and DNA synthesis or repair, while genes differentially expressed between testes and carcass reflect sperm development, cell division, and energy production (see Supplementary Information). We used reciprocal best hit orthologs to identify genes with
similar regulation across species, focusing on D. yakuba and D. simulans where biological replicates increase power for differential expression testing (Table 6). A total of 10,369 genes in D. yakuba have reciprocal best hit orthologs in D. simulans , and were retained to compare differential expression across the two species. We have collected replicates for both D. yakuba and D. simulans , contributing to the greater power in differential expression testing. Roughly 60% of genes with tissue biased expression in D. yakuba that have a reciprocal best hit ortholog in D. simulans exhibit the same tissue specific bias in D. simulans , with marginally greater agreement in genes biased toward the carcass than the reproductive tissues in both males and females (Figure 3). We additionally observe evidence of differential expression in at least one of the four comparisons of male and female germline or somatic tissues for 118 lineage specific genes in D. ananassae , 334 in D. yakuba , and 222 in D. simulans (Table 5), suggesting that they are not solely artifacts of gene annotation software.
## Discussion
## Gene models and ortholog calls
Correct interpretations of gene expression changes and gene family evolution depend on accurate gene models. Among the Drosophila , D. melanogaster has received the most attention with empirically verified gene annotations through high throughput EST and RNA-seq data as well as detailed manual or molecular curation of single genes. Until present, gene models for outgroup species offer only CDS sequences, with no information concerning 5 ′ or 3 ′ UTRs or alternative isoforms even for well-studied genes such as Adh . Establishing more complete gene models based on RNA-seq data will allow us to correctly identify coding and non-coding sequences during functional assays, correctly identify putatively neutral vs non-neutral mutations, and correctly define new mutations including the origins of new genes, expansion of gene families, and gross modification of coding sequences through rearrangement, duplication, and deletion. Moreover, identifying putative isoforms provides a more complete portrait of gene structure and function across species. Here, we provide updated gene annotations based on high coverage RNA-seq data for the reference genomes of 3 species of Drosophila : D. ananassae , D. yakuba , and D. simulans , which should serve as an excellent springboard and initial resource to the Drosophila molecular and evolutionary genetic community.
Additionally, lineage specific genes are unlikely to be identified in previous annotation efforts that focused largely on conserved amino acid sequences. With RNA-seq annotations we can identify lineage specific genes, which are expected to be important in the evolution of genomes and emergence of genetic novelty. The increased power to identify genes independently from conservation is a major step toward studying the evolution of genome content and rapidly evolving gene sequences. The ability to identify transcribed genes independently from conservation will facilitate evolutionary and functional analyses of the most rapidly changing segments of the genome. There may be additional gene models and isoforms in other tissue types or timepoints which deserves to be explored. Particularly, these
annotations are unlikely to reflect the full diversity of sequences and isoforms expressed during embryonic development post-fertilization, pupal development, or in larvae. These alternative timepoints deserve future exploration and the gene models offered here will serve as a lower bound on the full diversity of sequences that are transcribed within each species.
## Polycistronic genes
We observe thousands of putatively polycistronic annotations in D. ananassae and D. yakuba , as well as hundreds in D. simulans w 501 where fewer gene model annotations were previously aligned and power to identify polycistronic genes is limited. In D. melanogaster , hundreds of genes are known to show signs of polycistronic transcription ( Lin et al. 2007) offering a means of co-regulation across genes with similar functions ( Slone et al. 2007, Blumenthal 1998). With very high sequencing coverage we are able to recover a greater number of polycistronic transcripts, though some of these may be false positives resulting from annotation algorithms. Some genes may differ in the frequency with which they are transcribed as polycistronic vs. independent transcripts, but these results imply that at least low levels of polycistronic transcription may be common for many genes in the genome and future validation may explore the extent of their functionality. Adh and Adhr show evidence of differing polycistronic status across Drosophila species ( Betran and Ashburner 2000) and plant genomes are thought to split and fuse genes at rates of roughly 10 -11 -10 -10 per gene per year ( Nakamura et al. 2007). Switching the rates at which genes are co-transcribed has the potential to alter regulatory patterns across species ( Slone et al. 2007, Blumenthal 1998) and thereby produce novel phenotypes. These differences in polycistronic transcription therefore represent potential sources of genetic change that may be important in evolutionary change.
## Tissue specific expression
We identify thousands of genes that are differentially expressed across tissues in D. yakuba and D. simulans , where biological replicate samples for each tissue are available. A large fraction of the genome appears to display tissue biased expression between germiline and carcass, with many fewer genes showing differential expression between male and female gonadectomized carcass, consistent with early microarry-based assays in D. melanogaster ( Parisi et al. 2003; 2004). We have sequenced samples to extremely high coverage, generating transcript annotations with 5 ′ and 3 ′ UTRs with empirically supported intron-exon structures, improving accuracy in differential expression testing. Biological replicates for D. yakuba and D. simulans result in much greater power to identify differentially expressed genes in comparison with D. ananassae and D. melanogaster . However, even in D. ananassae and D. melanogaster where only one replicate was available per tissue, we are still able to identify hundreds of genes that are differentially expressed across tissues with high coverage.
A comparison of orthologs between D. yakuba and D. simulans where biological replicates should result in sufficient power to detect differential expression reliably across both species
reveals that roughly 60% of genes with reciprocal best hit orthologs exhibit similar tissue biased expression between the two species. The remaining 30% represent either genes that are differentially regulated between tissues but with effect sizes beyond the limits of detection, or genes that have evolved independent expression patterns between the two species. We also observe tissue biased expression in hundreds of lineage specific genes, which may represent candidates for neofunctionalization and new gene origination. These genes that display changes in sex-biased or tissue-biased expression across taxa, as well as lineage specific genes that exhibit tissue-biased and sex-biased expression are important candidate loci for evolutionary change in genome content and function that will be useful in exploring the functional and selective impacts of genomic changes.
## Data access
Gene annotations, ortholog calls, gene ontology calls, and Cuffdiff differential expression testing output files for all samples are available at http://github.com/ThorntonLab/GFF . RNA-seq based annotations as well as first order orthologs in comparison to D. melanogaster can be viewed in the UCSC browser on the Thornton Lab public track hub at http:// genome.ucsc.edu . Sequencing fastq files were deposited in SRA with accession numbers PRJNA196536, PRJNA193071, PRJNA257286, and PRJNA257287.
## Author Contributions
RLR, PA, and KRT designed experiments. LS and RLR performed experiments and collected data. RLR, JS, and KRT performed analyses. RLR and KRT wrote the manuscript with methodological input from LS.
## Acknowledgements
We would like to thank Rahul Warrior for offering incubator space for fly cultures and Rachel Martin for an emergency supply of liquid nitrogen. RLR is supported by NIH Ruth Kirschstein National Research Service Award F32-GM099377. Research funds were provided by NIH grant R01-GM085183 to KRT and R01-GM083228 to PA. All sequencing was performed at the UC Irvine High Throughput Genomics facility, which is supported by the National Cancer Institute of the National Institutes of Health under Award Number P30CA062203. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
## Bibliography
Betran, E. , and M. Ashburner , 2000 Duplication, dicistronic transcription, and subsequent evolution of the Alcohol dehydrogenase and Alcohol dehydrogenase-related genes in Drosophila. Mol. Biol. Evol. 17 : 1344-1352.
Blumenthal, T. , 1998 Gene clusters and polycistronic transcription in eukaryotes. Bioessays 20 : 480-487.
Chen, Z. X. , D. Sturgill , J. Qu , H. Jiang , S. Park , et al. , 2014 Comparative validation of the D. melanogaster modENCODE transcriptome annotation. Genome Res. 24 : 1209-1223.
Clemente, F. , and C. Vogl , 2012 Unconstrained evolution in short introns? - an analysis of genome-wide polymorphism and divergence data from Drosophila. J. Evol. Biol. 25 : 1975-1990.
Drosophila Twelve Genomes Consortium , 2007 Evolution of genes and genomes on the Drosophila phylogeny. Nature 450 : 203-218.
Ellegren, H. , and J. Parsch , 2007 The evolution of sex-biased genes and sex-biased gene expression. Nat. Rev. Genet. 8 : 689-698.
Fire, A. , S. Xu , M. K. Montgomery S. A. Kostas S. E. Driver , , , et al. , 1998 Potent and specific genetic interference by double-stranded RNA in Caenorhabditis elegans. Nature 391 : 806-811.
Grabherr, M. G. , B. J. Haas , M. Yassour , J. Z. Levin , D. A. Thompson , et al. , 2011 Full-length transcriptome assembly from RNA-Seq data without a reference genome. Nat. Biotechnol. 29 : 644-652.
Haas, B. J. , A. Papanicolaou , M. Yassour , M. Grabherr , P. D. Blood , et al. , 2013 De novo transcript sequence reconstruction from RNA-seq using the Trinity platform for reference generation and analysis. Nat Protoc 8 : 1494-1512.
Haerty, W. , S. Jagadeeshan , R. J. Kulathinal , A. Wong , K. Ravi Ram , et al. , 2007 Evolution in the fast lane: rapidly evolving sex-related genes in Drosophila. Genetics 177 : 1321-1335.
| Hahn, M. W. , M. V. Han , and S. G. Han , 2007 Gene family evolution across 12 Drosophila genomes. PLoS Genetics 3 : e197. |
|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| Halligan, D. L. , and P. D. Keightley , 2006 Ubiquitous selective constraints in the Drosophila genome revealed by a genome-wide interspecies comparison. Genome Res. 16 : 875-884. |
| Hu, T. T. , M. B. Eisen , K. R. Thornton , and P. Andolfatto , 2012 A second generation assembly of the Drosophila simulans genome provides new insights into patterns of lineage-specific divergence. Genome Res. . |
| Jinek, M. , K. Chylinski , I. Fonfara , M. Hauer , J. A. Doudna , et al. , 2012 A programmable dual-RNA-guided DNA endonuclease in adaptive bacterial immunity. Science 337 : 816-821. |
| Kim, D. , G. Pertea , C. Trapnell , H. Pimentel , R. Kelley , et al. , 2013 TopHat2: accurate alignment of transcriptomes in the presence of insertions, deletions and gene fusions. Genome Biol. 14 : R36. |
| Langmead, B. , C. Trapnell , M. Pop , and S. L. Salzberg , 2009 Ultrafast and memory-efficient alignment of short DNA sequences to the human genome. Genome Biol. 10 : R25. |
| Lin, M. F. , J. W. Carlson , M. A. Crosby , B. B. Matthews , C. Yu , et al. , 2007 Revisiting the protein-coding gene catalog of Drosophila melanogaster using 12 fly genomes. Genome Res. 17 : 1823-1836. |
| Mank, J. E. , L. Hultin-Rosenberg , M. Zwahlen , and H. Ellegren , 2008 Pleiotropic constraint hampers the resolution of sexual antagonism in vertebrate gene expression. Am. Nat. 171 : 35-43. |
| McDonald, J. H. , and M. Kreitman , 1991 Adaptive protein evolution at the Adh locus in Drosophila. Nature 351 : 652-654. |
| Meisel, R. P. , 2011 Towards a more nuanced understanding of the relationship between sex-biased gene expression and rates of protein-coding sequence evolution. Mol. Biol. Evol. 28 : 1893-1900. |
| Nakamura, Y. , T. Itoh , and W. Martin , 2007 Rate and polarity of gene fusion and fission in Oryza sativa and Arabidopsis thaliana. Mol. Biol. Evol. 24 : 110-121. |
| Parisi, M. , R. Nuttall , P. Edwards , J. Minor , D. Naiman , et al. , 2004 A survey of ovary-, testis-, and soma-biased gene expression in Drosophila melanogaster adults. Genome Biol. 5 : R40. |
| Parisi, M. , R. Nuttall , D. Naiman , G. Bouffard , J. Malley , et al. , 2003 Paucity of genes on the Drosophila X chromosome showing male-biased expression. Science 299 : 697-700. |
|---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| Parsch, J. , S. Novozhilov , S. S. Saminadin-Peter , K. M. Wong , and P. Andolfatto , 2010 On the utility of short intron sequences as a reference for the detection of positive and negative selection in Drosophila. Mol. Biol. Evol. 27 : 1226-1234. |
| Ranz, J. M. , C. I. Castillo-Davis , C. D. Meiklejohn , and D. L. Hartl , 2003 Sex-dependent gene expression and evolution of the Drosophila transcriptome. Science 300 : 1742-1745. |
| Slone, J. , J. Daniels , and H. Amrein , 2007 Sugar receptors in Drosophila. Curr. Biol. 17 : 1809-1816. |
| Tamura, K. , S. Subramanian , and S. Kumar , 2004 Temporal patterns of fruit fly (Drosophila) evolution revealed by mutation clocks. Mol. Biol. Evol. 21 : 36-44. |
| Trapnell, C. , L. Pachter , and S. L. Salzberg , 2009 TopHat: discovering splice junctions with RNA-Seq. Bioinformatics 25 : 1105-1111. |
| Trapnell, C. , A. Roberts , L. Goff , G. Pertea , D. Kim , et al. , 2012 Differential gene and transcript expression analysis of RNA-seq experiments with TopHat and Cufflinks. Nat Protoc 7 : 562-578. |
| Van Dyken, J. D. , and M. J. Wade , 2010 The genetic signature of conditional expression. Genetics 184 : 557-570. |
| Wasbrough, E. R. , S. Dorus , S. Hester , J. Howard-Murkin , K. Lilley , et al. , 2010 The Drosophila melanogaster sperm proteome-II (DmSP-II). J Proteomics 73 : 2171-2185. |
| Zhang, Y. , D. Sturgill , M. Parisi , S. Kumar , and B. Oliver , 2007 Constraint and turnover in sex-biased gene expression in the genus Drosophila. Nature 450 : 233-237. |
| Zhao, L. , P. Saelao , C. D. Jones , and D. J. Begun , 2014 Origin and spread of de novo genes in Drosophila melanogaster populations. Science 343 : 769-772. |
Table 1: Fly stocks used for RNA-seq Species Strain
| D. melanogaster | 14021-0231.36 501 |
|-------------------|---------------------|
| D. simulans | w |
| D. yakuba | 14021-0261.01 |
| D. ananassae | 14024-0371.13 |
Table 2: Number of transcripts and genes identified
| Species | | Transcripts | Genes | Genes | Genes |
|--------------|------------------|---------------|---------|---------|---------|
| | Previous Release | Revised | FlyBase | Revised | FlyBase |
| D. simulans | r1.3 | 18,781 | 15,415 | 16,278 | 15,413 |
| D. yakuba | r1.3 | 20,239 | 16,082 | 17,579 | 16,077 |
| D. ananassae | r1.3 | 22,418 | 15,070 | 20,580 | 15,069 |
Table 3: Percent of revised gene models with ≥ 60% of features supported by RNA-seq data
| Species | Percent Suported |
|--------------|--------------------|
| D. ananassae | 72.3% |
| D. yakuba | 79.4% |
| D. simulans | 80.05% |
Table 4: Genes with a first order ortholog identified
| Genes in | Genes in | With an ortholog in | Revised | FlyBase |
|------------|--------------|-----------------------|-----------|-----------|
| D. | melanogaster | D. simulans | 12,199 | 10,705 |
| D. | melanogaster | D. yakuba | 11,472 | 11,556 |
| D. | melanogaster | D. ananassae | 11,451 | 10,938 |
| D. | simulans | D. melanogaster | 13,295 | - |
| D. | simulans | D. yakuba | 12,299 | - |
| D. | simulans | D. ananassae | 11,994 | - |
| D. | yakuba | D. melanogaster | 12,868 | - |
| D. | yakuba | D. simulans | 12,831 | - |
| D. | yakuba | D. ananassae | 12,337 | - |
| D. | ananassae | D. melanogaster | 12,897 | - |
| D. | ananassae | D. simulans | 12,612 | - |
| D. | ananassae | D. yakuba | 12,723 | - |
Table 5: Putative Lineage Specific Genes on Major Chromosomes
| Species | Major chromosomes | Total | Diff Exp |
|--------------|---------------------|---------|------------|
| D. ananassae | - | 2977 | 118 |
| D. yakuba | 230 | 1340 | 334 |
| D. simulans | 369 | 1314 | 222 |
Table 6: Differentially Expressed Genes By Tissue and Species at FDR ≤ 0.1
| Species | Tissue | Tissue | Significant | Tested |
|-----------------|--------------|----------------|---------------|----------|
| D. ananassae | Female Ovary | Female Carcass | 203 | 7537 |
| | Female Ovary | Male Testes | 200 | 8282 |
| | Male Carcass | Male Testes | 1013 | 8349 |
| | Male Carcass | Female Carcass | 175 | 8417 |
| D. yakuba | Female Ovary | Female Carcass | 5420 | 8689 |
| | Female Ovary | Male Testes | 5868 | 10202 |
| | Male Carcass | Male Testes | 3065 | 10412 |
| | Male Carcass | Female Carcass | 724 | 9430 |
| D. simulans | Female Ovary | Female Carcass | 5053 | 8967 |
| | Female Ovary | Male Testes | 5741 | 10222 |
| | Male Carcass | Male Testes | 4628 | 10679 |
| | Male Carcass | Female Carcass | 611 | 9566 |
| D. melanogaster | Female Ovary | Female Carcass | 112 | 11326 |
| | Female Ovary | Male Testes | 370 | 12890 |
| | Male Carcass | Male Testes | 220 | 13268 |
| | Male Carcass | Female Carcass | 286 | 12502 |
## Supplementary Information
## Expression of Adh in Female Carcass
l
l
l
l
l
l
l
l
l
l
l l
l
l l l
l l
l
l
l l
l
l
l
l
l
l
l
l
l
l l
l
l l
l
l
l
l
l
l
l
ll
l
l
l
l
l
l
l
l
l
l
l
## Expression of Adh in Female Ovaries
l
ll
l
l
l
l
l
l
l l
l
l
l
l
l
l l
l
l
l l
l
l
l
l
l
l
l
l
l
l
l
l
l
lllll
l
lll
l
Figure 1: Quantile normalized coverage for reference strains at the Adh locus in D. yakuba . Coverage shows clear distinctions between introns and exons, and coverage that spans both 5 ′ and 3 ′ UTRs in ovaries and carcass. Low coverage of intron sequence points to partial sequencing of low levels of unprocessed transcripts.
l
l
l
l
l
l
l
l
l
l
l
l
l
l
l l
l
l
l
l
l
l
l
l l l l
l
l
l
l
l
l
l
l
l ll l
l
ll
l
l
l
l l
l
l
l l
l
l l ll
l
l
ll
l
l
l
l
ll
l
l
l
l
l
l
ll l l
l
l
l l
l
lll
l
l
ll
l
l
l
Figure 2: Quantile normalized RNA-seq data for three replicates of the D. yakuba reference with an example of flybase gene model (white) and revised gene model (grey).
Figure 3: Genes with tissue biased expression in both D. yakuba and D. simulans in A) female ovary B) female carcass C) male testes D) male carcass. Numbers shown include only genes with a reciprocal best hit ortholog in the sister species.
## Female Carcass I
Female Ovary I
## Female Carcass II
## Female Ovary II
## Female Carcass III
## Female Ovary III
Figure S1: Raw coverage for sites with coverage sequencing depth between 1 and 1000 reads in RNA-seq data for replicates of female tissues in D. simulans .
## Male Testes I
## Male Carcass II
## Male Carcass III
## Male Testes II
## MaleTestes III
Figure S2: Raw coverage for sites with coverage sequencing depth between 1 and 1000 reads in RNA-seq data for replicates of male tissues in D. simulans .
Figure S3: Raw coverage for sites with coverage sequencing depth between 1 and 1000 reads in RNA-seq data for replicates of female tissues in D. yakuba .
Figure S4: Raw coverage for sites with coverage sequencing depth between 1 and 1000 reads in RNA-seq data for replicates of male tissues in D. simulans .
## Female Carcass
## Female Ovary
## Male Carcass
## Male Testes
Figure S5: Raw coverage for sites with coverage sequencing depth between 1 and 1000 reads in RNA-seq data for tissues in D. ananassae .
## Female Carcass
## Male Carcass
## Female Ovary
## Male Testes
Figure S6: Raw coverage for sites with coverage sequencing depth between 1 and 1000 reads in RNA-seq data for tissues in D. melanogaster .
 | 10.1534/g3.114.013532 | [
"Rebekah L. Rogers",
"Ling Shao",
"Jaleal S. Sanjak",
"Peter Andolfatto",
"Kevin R. Thornton"
] | 2014-08-01T17:59:53+00:00 | 2014-10-01T22:24:31+00:00 | [
"q-bio.GN"
] | Revised Annotations, Sex-Biased Expression, and Lineage-Specific Genes in the Drosophila melanogaster group | Here, we provide revised gene models for D. ananassae, D. yakuba, and D.
simulans, which include UTRs and empirically verified intron-exon boundaries,
as well as ortholog groups identified using a fuzzy reciprocal-best-hit blast
comparison. Using these revised annotations, we perform differential expression
testing using the cufflinks suite to provide a broad overview of differential
expression between reproductive tissues and the carcass. We identify thousands
of genes that are differentially expressed across tissues in D. yakuba and D.
simulans, with roughly 60% agreement in expression patterns of orthologs in D.
yakuba and D. simulans. We identify several cases of putative polycistronic
transcripts, pointing to a combination of transcriptional read-through in the
genome as well as putative gene fusion and fission events across taxa. We
furthermore identify hundreds of lineage specific genes in each species with no
blast hits among transcripts of any other Drosophila species, which are
candidates for neofunctionalized proteins and a potential source of genetic
novelty. |
1408.0248v2 | ## CLASSICAL AND MOTIVIC ADAMS-NOVIKOV CHARTS
DANIEL C. ISAKSEN
Abstract. This document contains large-format Adams-Novikov charts that compute the classical 2-complete stable homotopy groups. The charts are essentially complete through the 60-stem. We believe that these are the most accurate and extensive charts of their kind. We also include a motivic Adams-Novikov E ∞ chart.
This document contains large-format Adams-Novikov charts that compute the classical 2-complete stable homotopy groups. The charts are essentially complete through the 60-stem. We believe that these are the most accurate and extensive charts of their kind. We also include a motivic Adams-Novikov E ∞ chart. The charts are intended to be viewed electronically. The author can supply versions that are suitable for printing.
Justifications for these calculations appear in [1] and in the forthcoming publication [2]. See also [3] and [4] for classical calculations in a smaller range.
This document is the second version of these charts.
- 1. The E 2 -page of the Adams-Novikov spectral sequence
This chart shows the E 2 -page of the classical Adams-Novikov spectral sequence through the 60-stem. For legibility, the chart is split into two pieces. The well-understood α family and its α 1 multiples are depicted on a separate grid.
Except for the α family and its α 1 multiples, the names of elements are consistent with the names of elements in the Adams spectral sequence for the cofiber of τ . These names are incompatible with the Greek letter system.
- (1) Solid dots indicate copies of Z / 2.
- (2) Open circles indicate copies of Z / 4.
- (3) Solid squares indicate copies of Z / 8.
- (4) Open squares indicate copies of Z / 2 k for some k ≥ 4. The default value of k is 4, unless shown otherwise on the chart. The value k = ∞ indicates a copy of Z 2 .
- (5) Lines of slope 1 indicate α 1 multiplications.
- (6) Arrows of slope 1 indicate infinitely many α 1 multiplications.
- (7) Lines of slope 1 / 3 indicate α 2 / 2 multiplications.
- (8) Magenta lines indicate that an extension equals 2 times a generator. For example, α 2 2 / 2 d 0 equals 2 · h g 0 in the 20-stem.
- (9) Orange lines indicate that an extension equals 4 times a generator. For example, α 3 1 · τg equals 4 · h g 2 in the 23-stem.
- (10) Red lines of slope -3 indicate Adams-Novikov d 3 differentials.
- (11) Blue lines of slope -5 indicate Adams-Novikov d 5 differentials.
- (12) Green lines of slope -7 indicate Adams-Novikov d 7 differentials.
- (13) Purple lines of slope -9 indicate Adams-Novikov d 9 differentials.
- 2. The E ∞ -page of the Adams-Novikov spectral sequence
This chart indicates the E ∞ -page of the classical Adams-Novikov spectral sequence, together with hidden extensions by 2, η , and ν . Beyond the 48-stem, there are some uncertainties. See Section 1 for instructions on interpreting the chart. In addition,
- (1) Olive lines indicate hidden 2 extensions.
- (2) Purple lines indicate hidden η extensions.
- (3) Brown lines indicate hidden ν extensions.
- (4) Dashed lines indicate possible hidden extensions.
- 3. The E ∞ -page of the motivic Adams-Novikov spectral sequence
This chart indicates the E ∞ -page of the motivic Adams-Novikov spectral sequence. For clarity, hidden extensions are not shown on this chart.
See Section 1 for instructions on interpreting the chart. In addition,
(1) Red dots indicate copies of Z / 2[ τ /τ ] .
(2) Blue dots indicate copies of Z / 2[ τ /τ ] 2
.
(3) Green dots indicate copies Z / 2[ τ /τ ] 3 .
(4) Purple dots indicate copies of Z / 2[ τ /τ ] 4 .
(5) Green squares with blue centers indicate copies of Z / 8[ τ / τ ] ( 3 , 2 τ 2 ). This occurs only once in the 43-stem.
## References
[1] Daniel C. Isaksen, Stable stems (2014), submitted.
[2] Daniel C. Isaksen, Guozhen Wang, and Zhouli Xu, More stable stems , in preparation.
[3] Douglas C. Ravenel, Complex cobordism and stable homotopy groups of spheres , Pure and Applied Mathematics, vol. 121, Academic Press, Inc., Orlando, FL, 1986. MR860042 (87j:55003)
[4] Raphael Zahler, The Adams-Novikov spectral sequence for the spheres , Ann. of Math. (2) 96 (1972), 480-504. MR0319197
1
2
3
0
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30
32
34
36
38
40
42
44
46
48
50
52
54
56
58
60
12
6
10
8
6
4
2
0
0
3
## The α family in the classical Adams-Novikov E ∞ -page
4
3
30
0
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30
32
34
36
38
40
42
44
46
48
50
52
54
56
58
60
12
10
8
6
4
2
0
0 | null | [
"Daniel C. Isaksen"
] | 2014-08-01T18:00:03+00:00 | 2020-01-17T21:01:22+00:00 | [
"math.AT",
"55T15, 55Q45, 14F42"
] | Classical and motivic Adams-Novikov charts | This document contains large-format Adams-Novikov charts that compute the
classical 2-complete stable homotopy groups. The charts are essentially
complete through the 60-stem. We believe that these are the most accurate and
extensive charts of their kind. We also include a motivic Adams-Novikov
E-infinity chart. |
1408.0250v2 | ## STRONG HYPERBOLICITY
## BOGDAN NICA AND J ´ N ˇ PAKULA A S
Abstract. We propose the metric notion of strong hyperbolicity as a way of obtaining hyperbolicity with sharp additional properties. Specifically, strongly hyperbolic spaces are Gromov hyperbolic spaces that are metrically well-behaved at infinity, and, under weak geodesic assumptions, they are strongly bolic as well. We show that CAT( -1) spaces are strongly hyperbolic. On the way, we determine the best constant of hyperbolicity for the standard hyperbolic plane H 2 . We also show that the Green metric defined by a random walk on a hyperbolic group is strongly hyperbolic. A measure-theoretic consequence at the boundary is that the harmonic measure defined by a random walk is a visual Hausdorff measure.
## 1. Introduction
Hyperbolicity for metric spaces, as introduced by Gromov [9], is a coarse notion of negative curvature which yields a very satisfactory theory. For certain analytic purposes, however, hyperbolicity by itself is not enough, and one needs certain enhancements.
After the groundbreaking work of Lafforgue [14], the resolution of the Baum - Connes conjecture for hyperbolic groups hinged on showing the following: every hyperbolic group Γ admits a geometric action on a roughly geodesic, strongly bolic hyperbolic space. Here the enhancement is strong bolicity (see Section 4), for which one is willing to pay the small price of weakening the usual geodesic assumption. This is achieved by Mineyev and Yu in [17], and the desired space is Γ itself endowed with a new metric.
The enhancement needed in [18] is a boundary metric structure with sharp properties. Let us say that a hyperbolic space X is a good hyperbolic space if the following hold: the Gromov product ( · , · ) o extends continuously to the bordification X ∪ ∂X for each basepoint o ∈ X , and there is some /epsilon1 > 0 such that exp( -/epsilon1 ( · , · ) o ) is a metric on the boundary ∂X , again for each basepoint o ∈ X . For example, CAT( -1) spaces are good hyperbolic spaces, and one can take /epsilon1 = 1 (Bourdon [6]). Now the construction of [18] can be summarized as follows: from a geometric action of a hyperbolic group Γ on a roughly geodesic, good hyperbolic space X , one obtains a proper isometric action of Γ on an L p -space associated to the double boundary ∂X × ∂X . The concrete X used in [18] is Γ itself, equipped with a metric constructed by Mineyev [15, 16] and which is a slightly improved version of the metric used in [17].
In this paper we introduce a certain metric condition that we call strong hyperbolicity . The definition is given in Section 4 and we will not repeat here. The main point about strong hyperbolicity is that there are useful consequences flowing out and important examples flowing in.
As far as consequences are concerned, the following is a fairly easy outcome of the definition.
Theorem 1.1. A strongly hyperbolic space is a good hyperbolic space. A roughly geodesic, strongly hyperbolic space is strongly bolic.
Date : June 30, 2015.
2000 Mathematics Subject Classification. 20F67.
Key words and phrases. Hyperbolic group, Green metric, CAT( -1) space, harmonic measure.
As for examples, we show the following.
Theorem 1.2. CAT( -1) spaces are strongly hyperbolic.
Theorem 1.3. The Green metric arising from a random walk on a hyperbolic group is strongly hyperbolic.
Theorem 1.2 implies Bourdon's result mentioned above, but also one of the main results of Foertsch and Schroeder from [7]. From Theorem 1.2 and its proof, we also deduce that the hyperbolic plane H 2 is log 2-hyperbolic, in the sense of Gromov's original definition, and this is optimal (Corollary 5.4). Quite surprisingly, this was not known before.
Good hyperbolicity for the Green metric on a hyperbolic group answers positively a question raised in [18, Remark 19]. As a consequence, we find that, on the boundary, the harmonic measure defined by a random walk coincides with the normalized Hausdorff measure defined by a Green visual metric (Corollary 6.4). Strong bolicity of the Green metric was recently proved by Ha¨ ıssinsky and Mathieu in [12] by a different argument. The upshot is that the Green metric provides a more natural, and much simpler alternative to the hat metric constructed by Mineyev and Yu in [17]. Strong hyperbolicity for the Green metric, established herein, makes a similar point relative to Mineyev's improved hat metric from [15, 16].
## 2. Hyperbolic spaces - An elliptical reminder
## 2.1. Hyperbolic spaces and their boundaries. We begin by recalling some basic notions and facts.
Definition 2.1 (Gromov) . A metric space X is δ -hyperbolic , where δ ≥ 0, if for all x, y, z, o ∈ X ; equivalently, the four-point condition
$$( x, y ) _ { o } \geq \min \{ ( x, z ) _ { o }, ( y, z ) _ { o } \} - \delta \\........................................................................................................................................................................................................$$
$$| x, y | + | z, t | & \leq \max \left \{ | x, z | + | y, t |, | z, y | + | x, t | \right \} + 2 \delta \\ \dots \sim v$$
holds for x, y, z, t ∈ X .
We write | x, y | for the distance between two points x, y ∈ X . Recall that the Gromov product is defined by the formula
$$( x, y ) _ { o } = \frac { 1 } { 2 } \Big ( | x, o | + | y, o | - | x, y | \Big ). \\ \vdots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots$$
When X is a CAT( -1) space, that is, a metric space of actual negative curvature, the Gromov product can be extended from X to the bordification X ∪ ∂X by setting ( ξ, ξ ′ ) o := lim ( x , x i ′ i ) o for x i → ξ and x ′ i → ξ ′ , respectively ( x, ξ ) o := lim ( x, x i ) o for x i → ξ . But for a general hyperbolic space X , such a process is not well-defined in general: the limit may not exist, and even if it does, it might depend on the choice of a representing sequence. The traditional procedure of getting around this problem runs as follows. The Gromov product on ∂X × ∂X is defined by setting ( ξ, ξ ′ ) o := inf { lim inf ( x , x i ′ i ) o : x i → ξ, x ′ i → ξ ′ } . This is a finite quantity, except for ( ξ, ξ ) o = ∞ . Similarly, the Gromov product on X × ∂X is defined by ( x, ξ ) o := inf { lim inf ( x, x i ) o : x i → } ξ . It turns out that the other reasonable choices,
Let X be a hyperbolic space, and fix a basepoint o ∈ X . A sequence ( x i ) in X converges to infinity if ( x , x i j ) o →∞ as i, j →∞ . Two sequences ( x i ), ( y i ) converging to infinity are declared to be equivalent if ( x , y i i ) o →∞ as i →∞ . The (Gromov) boundary of X , denoted ∂X , is the set of equivalence classes of sequences converging to infinity. We also say that a sequence ( x i ) converging to infinity converges to its equivalence class representing a point in ∂X . Since a change of basepoint modifies the Gromov product by a uniformly bounded amount, all these notions are actually independent of the chosen o ∈ X .
namely taking sup instead of inf, or lim sup instead of lim inf, are all uniformly close to each other. Although rather ad-hoc, such an extension of the Gromov product has at least the benefit of being canonical at the topological level: it defines a topology on the bordification X ∪ ∂X , for which convergence at infinity takes on an actual meaning. If X is proper then both X ∪ ∂X and ∂X are compact.
It turns out that the topology on ∂X can be metrized. Namely, for small enough /epsilon1 > 0 there exists a metric d /epsilon1 on ∂X , such that d /epsilon1 is within constant positive multiples of exp( -/epsilon1 ( · , · ) o ). When X is a CAT( -1) space, exp( - · · ( , ) o ) is already a metric on ∂X , but this is not true in general [16, Lemma 7]. To summarize, the fuzziness of the Gromov product at the boundary is resolved by a non-canonical choice, which in turn leads to a quasified metric structure on the boundary.
- 2.2. Roughly geodesic spaces. The classical theory of hyperbolic spaces works under the assumption that the spaces are geodesic. But the theory is just as solid for roughly geodesic spaces. These are defined as follows.
Definition 2.2. A metric space X is roughly geodesic if there is a constant C ≥ 0 such that the following holds: for every x, y ∈ X , there is a (not necessarily continuous) map γ : [ a, b ] → X such that γ a ( ) = x , γ b ( ) = y , and | s -s ′ | -C ≤ | γ s , γ ( ) ( s ′ ) | ≤ | s -s ′ | + C for all s, s ′ ∈ [ a, b ].
The viewpoint that roughly geodesic spaces provide a natural relaxation of the geodesic assumption for hyperbolic spaces was first advocated by Bonk and Schramm in [5]. We remark that the notion of weakly geodesic space used by Lafforgue [14] and Mineyev - Yu [17], although weaker in general, is equivalent to the notion of roughly geodesic space within the hyperbolic world.
Hyperbolicity is a quasi-isometry invariant for roughly geodesic spaces, and the Schwarz -Milnor lemma holds in the roughly geodesic context. So we may say that a group is hyperbolic if it admits a geometric action (that is, isometric, proper, and cocompact) on a roughly geodesic, proper hyperbolic space.
Throughout this paper, hyperbolic groups are assumed to be non-elementary.
## 3. Good hyperbolic spaces
Let us recall the definition we made in the Introduction, this time in a quantified form.
Definition 3.1. We say that a hyperbolic space X is /epsilon1 -good , where /epsilon1 > 0, if the following two properties hold for each basepoint o ∈ X :
- · the Gromov product ( · , · ) o on X extends continuously to the bordification X ∪ ∂X , and
- · exp( -/epsilon1 ( · , · ) o ) is a metric on the boundary ∂X .
Let X be a good hyperbolic space. The continuity of the Gromov product implies that we have a well-defined notion of Busemann function
$$\beta ( y, z ; \xi ) = 2 ( \xi, z ) _ { y } - | y, z | = \lim _ { x \to \xi } \left ( | x, y | - | x, z | \right ) \quad ( y, z \in X, \xi \in \partial X ).$$
Incidentally, it can be checked that the Busemann boundary of X coincides with the Gromov boundary ∂X when X is proper. (We owe this remark to Peter Ha¨ ıssinsky.) This topological consequence can be very useful - see, for instance, [2] and [12] - but we will not need it here.
For the purposes of this paper, the main feature of a good hyperbolic space is its conformal structure on the boundary - as opposed to mere quasi-conformality for general hyperbolic
spaces. This is discussed in distinct but related ways in [16, Sections 7 and 8], [10, Section 5.1], [18, Section 4]. See also [6, Sections 2.6 and 2.7] for the CAT( -1) case. Let g be an isometry of X . For all x, x ′ ∈ X we have that
Letting x → ξ ∈ ∂X respectively x ′ → ξ ′ ∈ ∂X , and using the notation d /epsilon1,o for the visual metric exp( -/epsilon1 ( · , · ) o ) on ∂X , we obtain the relation
$$- 2 ( g x, g x ^ { \prime } ) _ { o } = | x, o | - | x, g ^ { - 1 } o | + | x ^ { \prime }, o | - | x ^ { \prime }, g ^ { - 1 } o | - 2 ( x, x ^ { \prime } ) _ { o }. \\ \iota \dots \dots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots.$$
$$d _ { \epsilon, o } ^ { 2 } ( g \xi, g \xi ^ { \prime } ) = \exp \left ( \epsilon \, \beta ( o, g ^ { - 1 } o ; \xi ) \right ) \exp \left ( \epsilon \, \beta ( o, g ^ { - 1 } o ; \xi ^ { \prime } ) \right ) d _ { \epsilon, o } ^ { 2 } ( \xi, \xi ^ { \prime } )$$
for all ξ, ξ ′ ∈ ∂X . This is a strong form of metric conformality for the action of Isom( X ) on ∂X . Minor additional assumptions on ∂X [18, Lemma 8] then guarantee the following measure-theoretic conformality :
$$\frac { \dot { d } g ^ { * } \mu _ { o } } { d \mu _ { o } } ( \xi ) = \exp \left ( \epsilon D _ { \epsilon } \, \beta ( o, g ^ { - 1 } o ; \xi ) \right )$$
where D /epsilon1 is the Hausdorff dimension of d /epsilon1,o (independent of the basepoint o ), and µ o is the Hausdorff measure defined by d /epsilon1,o (independent of the visual parameter /epsilon1 ).
For applications, the most interesting case is when X is a proper, roughly geodesic, good hyperbolic space endowed with a geometric action of a (hyperbolic) group Γ. By the above discussion, Γ acts conformally on ∂X .
## 4. Strongly hyperbolic spaces
$$\text{Definition 4.1.\ } \text{We say that a metric space } X \text{ is strongly hyperbolic with parameter } \epsilon > 0 \text{ if} \\ \exp ( - \epsilon \left ( x, y _ { o } \right ) \leq \exp ( - \epsilon \left ( x, z \right ) _ { o } ) + \exp ( - \epsilon \left ( z, y \right ) _ { o } ). \\ \text{for all } x. u. z. o \in X \colon \text{envilalently, the four-noint condition}$$
$$\exp \left ( \frac { 1 } { 2 } \epsilon \left ( | x, y | + | z, t | \right ) \right ) & \leq \exp \left ( \frac { 1 } { 2 } \epsilon \left ( | x, z | + | y, t | \right ) \right ) + \exp \left ( \frac { 1 } { 2 } \epsilon \left ( | z, y | + | x, t | \right ) \right ) \\ \intertext { a n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d nd e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e d n d e s t e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e d n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e nd e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e s t e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e nd e n d e e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d i n e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e n d e p h o n n$$
for all x, y, z, o ∈ X ; equivalently, the four-point condition holds for all x, y, z, t ∈ X .
One motivation for considering this notion is that it grants the enhancements of hyperbolicity that we have mentioned. Here is the quantified version of Theorem 1.1.
Theorem 4.2. Let X be a strongly hyperbolic space, with parameter /epsilon1 . Then X is an /epsilon1 -good, (log 2) //epsilon1 -hyperbolic space. Furthermore, X is strongly bolic provided X is roughly geodesic.
The proof is straightforward. We include the details.
Proof. From
$$\exp ( - \epsilon \left ( x, y \right ) _ { o } ) & \leq 2 \max \left \{ \exp ( - \epsilon \left ( x, z \right ) _ { o } ), \exp ( - \epsilon \left ( z, y \right ) _ { o } ) \right \} \\ & = 2 \exp \left ( - \epsilon \min \{ \left ( x, z \right ) _ { o }, \left ( z, y \right ) _ { o } \} \right ) \\ \text{that } X \text{ is } \log 2 \right \} / \epsilon \text{hyperbolic}$$
we see that X is (log 2) //epsilon1 -hyperbolic.
Next we check that the Gromov product ( · , · ) o extends continuously to the bordification X = X ∪ ∂X ; it will then follow that exp( -/epsilon1 ( · , · ) o ) is a metric on ∂X . Let z ∈ X and ξ ∈ ∂X . If ( x i ) is a sequence in X converging to ξ , then
$$\left | \exp ( - \epsilon \left ( z, x _ { i } \$$
so we may put ( z, ξ ) o := lim ( z, x i ) o . To see that this is well-defined, let ( y i ) be another sequence in X converging to ξ ; then
$$\left | \exp ( - \epsilon \left ( z, x _ { i } \$$
Thus, we have a continuous extension of ( · , · ) o to X × X . Similar arguments lead to a continuous extension of ( · , · ) o to X × X .
$$\begin{array} {$$
For hyperbolic spaces which are roughly geodesic, strong bolicity in the sense of Lafforgue [14] amounts to the following: for every η, r > 0 there exists R > 0 such that | x, y | + | z, t | ≤ r and | x, z | + | y, t | ≥ R imply | x, t | + | y, z | ≤ | x, z | + | y, t | + . Since η
1 1 . /square it suffices to choose R > 0 such that 1 + exp 2 /epsilon1 ( r -R ) ≤ exp 2 /epsilon1η
The next two sections are devoted to interesting examples of strongly hyperbolic spaces.
## 5. Strong hyperbolicity for CAT( -1) spaces
It is known that CAT( -1) spaces are strongly bolic and good hyperbolic spaces. The next result strengthens and unifies these facts.
Theorem 5.1. CAT( -1) spaces are strongly hyperbolic with parameter 1 .
Let us recall the four-point characterization of the CAT( -1) property for a geodesic space X . Informally, every quadrilateral in X has a comparison quadrilateral in the hyperbolic plane H 2 whose diagonals are at least as long as the diagonals of the original quadrilateral in X . Formally, for every choice of points x , x 1 2 , x 3 , x 4 ∈ X , there are points ˜ x , x 1 ˜ 2 , x ˜ 3 , x ˜ 4 ∈ H 2 such that
$$| x _ { 1 }, x _ { 2 } | = | \tilde { x } _$$
It therefore suffices to prove Theorem 5.1 for H 2 , the model for metric spaces of curvature at most -1.
while | x , x | ≤ | ˜ x , x ˜ | and | x , x | ≤ | ˜ x , x ˜ |
Proposition 5.2. The hyperbolic plane H 2 is strongly hyperbolic with parameter 1 .
Proof. We use the upper half-space model { w ∈ C : Im w > 0 } , with distance given by
$$\cosh | w _ { 1 }, w _ { 2 } | = 1 + \frac {$$
that is,
$$| w _ { 1 }, w _ { 2 } | = 2 \log \frac { | w _ { 1 } - w _ { 2 } | + | w _ { 1 } - \overline { w } _ { 2 } | } { 2 \sqrt { \text{Im} \, w _ { 1 } } \sqrt { \text{Im} \, w _ { 2 } } }.$$
Using the notation ‖ w , w 1 2 ‖ = | w 1 -w 2 | + | w 1 -w 2 | , the four-point formulation of strong hyperbolicity with parameter 1 becomes simply
$$\| w _ { 1 }, w _ { 3 } \| \| w _ { 2 }, w _ { 4 } \| \leq \| w _ { 1 }, w _ { 2 } \| \| w _ { 3 }, w _ { 4 } \| + \| w _ { 1 }, w _ { 4 } \| \| w _ { 2 }, w _ { 3 } \|.$$
This brings to mind Ptolemy's inequality in the plane:
$$| w _ { 1 } - w _ { 3 } | \, | w _ { 2 } - w _ { 4 } | \leq | w _ { 1 } - w _ { 2 } | \, | w _ { 3 } - w _ { 4 } | + | w _ { 1 } - w _ { 4 } | \, | w _ { 2 } - w _ { 3 } |.$$
And indeed, several applications of Ptolemy's inequality yield the desired inequality. Here is a more conceptual formulation. Let X be a metric space which is ptolemaic, i.e., | x , x 1 3 || x , x 2 4 | ≤ | x , x 1 2 | | x , x 3 4 | + | x , x 1 4 | | x , x 2 3 | for all x , x 1 2 , x 3 , x 4 ∈ X , and let G be a finite group of isometries of X . (In our case, X is the Euclidean plane and G is the cyclic group of order 2 generated by the reflection in the x -axis.) Then, on the quotient X/G , the formula
/negationslash
$$\left \| [ x ], [ y ] \right \| = \frac { 1 } { | G | } \sum _ { g \in G } ^ { \mu } | x, g y | \quad \text{ for } [ x ] \neq [ y ]$$
defines a ptolemaic metric. Both the triangle inequality and Ptolemy's inequality are established by averaging over the G -orbits of all but one of the variables. /square
Remark 5.3. Let X be strongly hyperbolic with parameter /epsilon1 , and fix a basepoint o ∈ X . On X , let us write d /epsilon1 ( x, y ) for exp( -/epsilon1 ( x, y ) o ). The strong hyperbolicity assumption means that d /epsilon1 satisfies the triangle inequality; however, d /epsilon1 fails to be an actual metric on X because d /epsilon1 ( x, x ) > 0 for each x ∈ X . Now observe that d /epsilon1 fulfills Ptolemy's inequality for this is simply another way of writing the four-point formulation of strong hyperbolicity with parameter 1. Consequently, at infinity, the visual metric d /epsilon1 = exp( -/epsilon1 ( · , · ) o ) on ∂X is ptolemaic.
$$d _ { \epsilon } ( x _ { 1 }, x _ { 3 } ) \, d _ { \epsilon } ( x _ { 2 }, x _ { 4 } ) \leq d _ { \epsilon } ( x _ { 1 }, x _ { 2 } ) \, d _ { \epsilon } ( x _ { 3 }, x _ { 4 } ) + d _ { \epsilon } ( x _ { 1 }, x _ { 4 } ) \, d _ { \epsilon } ( x _ { 2 }, x _ { 3 } ), \\ \int \dots \colon \dots \dots \int \dots \int \dots \int \dots \int \dots \int \dots \int \dots \int \dots \int \dots \int \dots \int \dots \int \dots \int \dots \int \dots \int \dots \int \dots \end{pmatrix}$$
Foertsch and Schroeder have shown [7, Theorems 1 and 16] that, for a CAT( -1) space X , the visual metric d /epsilon1 = exp( - · · ( , ) o ) on the boundary ∂X is ptolemaic. In light of Theorem 5.1, the Foertsch - Schroeder result is a particular case of the fact pointed out above.
As another consequence of Proposition 5.2, we find the best constant of hyperbolicity (in the sense of Definition 2.1) for H 2 . This answers a folklore question.
Corollary 5.4. H 2 is log 2 -hyperbolic, and this is optimal.
Proof. That H 2 is log 2-hyperbolic follows from Theorem 4.2. To see that this is best possible, let H 2 be δ -hyperbolic for some δ ≥ 0. Sticking to the upper half-space model, the hyperbolic inequality at infinity says that ( a, b ) i ≥ min ( { a, c ) i , ( c, b ) i }-δ whenever a, b, c ∈ R are distinct. At infinity, the Gromov product with respect to i is given by
$$( a, b ) _ { i } = \log \frac { | a - i | \, | b - i | } { | a - b | } \\ > 0 \ h = - a \ e = 0 \ \ w e \ a r e \ \text{mic} k$$
For comparison, we recall the well-known fact that H 2 is log(1 + √ 2)-hyperbolic for the Rips definition, in terms of thin triangles, and this is optimal.
for distinct a, b ∈ R . For a > 0, b = -a , c = 0, we are quickly led to | a - | ≥ i 2 exp( -δ ). Hence exp δ ≥ 2, by letting a tend to 0. /square
## 6. Strong hyperbolicity of the Green metric on a hyperbolic group
Another example of a strongly hyperbolic metric is Mineyev's 'hat' metric from [15, 16]. In fact, our notion of strong hyperbolicity is inspired by the proof of [16, Theorem 5]. Let us recall that Mineyev's hat metric is a (type of) metric defined on any hyperbolic group Γ, with the following properties: it is Γ-invariant, quasi-isometric to any word-metric on Γ, roughly geodesic, and - in our terminology - strongly hyperbolic. Of course, the latter property is the crucial one.
Mineyev's hat metric is a very useful tool, but it has the conceptual downside of being custom-made. One would like a 'natural' type of metric doing the same job - that of making hyperbolic groups strongly hyperbolic. We show that this is the case for the Green metric coming from a random walk.
Theorem 6.1. Let Γ be a hyperbolic group. Then the Green metric defined by a symmetric and finitely supported random walk on Γ is strongly hyperbolic.
We recall the essential facts on the Green metric, and we refer to [4] and [11] for more details. A probability measure µ on Γ defines a random walk on Γ with transition probabilities p x, y ( ) = µ x ( -1 y ). The probability measure µ , or the corresponding random walk, is
symmetric if µ x ( ) = µ x ( -1 ) for every x ∈ Γ, and finitely supported if the support of µ is a finite generating set for Γ.
Let F x, y ( ) be the probability that the random walk started at x ever hits y . The Green metric on Γ, first introduced by Blach´re and Brofferio [3], is given by e
$$| x, y | _ { G } & = - \log F ( x, y ). \\ \sim \quad. \quad. \quad \sim \quad \sim$$
The name has to do with the Green function G x, y ( ) = ∑ n ≥ 0 µ n ( x -1 y ), where µ n denotes the n -th convolution power of µ , counting the expected number of visits of y starting from x . It turns out that F and G are proportional: G e, e F ( ) ( x, y ) = G x, y ( ).
The Green metric is Γ-invariant, quasi-isometric to any word-metric on Γ, and roughly geodesic. The first two properties are generic, in the sense that they only use the nonamenability of Γ. The latter property follows from a result of Ancona [1] saying that geodesics for the word-metric are rough geodesics for the Green metric, and it relies on the hyperbolicity Γ. It follows that the Green metric is a hyperbolic metric, but our aim is to show more.
We remark that strong bolicity for the Green metric was first shown by Ha¨ ıssinsky and Mathieu in [12]. For their short proof it suffices to know Ancona's result that the Martin boundary of the random walk coincides with the Gromov boundary. The fact that the Green metric on a hyperbolic group is a good hyperbolic metric answers a question raised in [18], and for the purposes of [18] it means that the Green metric can be used in place of Mineyev's hat metric from [15, 16].
- 6.1. Proof of Theorem 6.1. The next lemma spells out what else is needed, on top of hyperbolicity, in order to have strong hyperbolicity.
Lemma 6.2. A hyperbolic space X is strongly hyperbolic if and only if the following holds:
(EB) there exist L, λ > 0 and R > 0 0 such that ∣ ∣ | x, y | + | z, t | - | x, t | - | z, y | ∣ ∣ ≤ L exp( -λR ) whenever x, y, z, t ∈ X satisfy | x, y | + | z, t | - | x, z | - | y, t | ≥ R ≥ R 0 .
The (EB) condition may be interpreted as an exponentially strong bolicity, since merely knowing that | x, y | + | z, t | - | x, t | - | z, y | → 0 as | x, y | + | z, t | - | x, z | - | y, t | → ∞ already implies strong bolicity (for hyperbolic spaces which are roughly geodesic).
Proof. For the sake of notational simplicity, let us write
$$A \coloneqq \frac { 1 } { 2 } ( | x, y | + | z, t | - | x, z | - | y, t | ), \quad B \coloneqq \frac { 1 } { 2 } ( | x, y | + | z, t | - | x, t | - | z, y | ).$$
( ⇐ ) The four-point condition for strong hyperbolicity with parameter /epsilon1 amounts to showing that, for some /epsilon1 > 0, we have 1 ≤ exp( -/epsilon1A )+exp( -/epsilon1B ) . We may assume that A,B ≥ 0, and A ≥ B . We rewrite the desired inequality as /epsilon1B ≤ -log(1 -exp( -/epsilon1A )). As a ≤ -log(1 -a ) for a ∈ [0 , 1), it suffices to obtain /epsilon1B ≤ exp( -/epsilon1A ). When A ≥ R / 0 2, this can be achieved for /epsilon1 ≤ min 2 { λ, 2 /L } by using (EB) . When A < R / 0 2, we use the assumption that X is hyperbolic: B = min { A,B } ≤ δ , a hyperbolicity constant of X . So it suffices to have /epsilon1δ ≤ exp( -/epsilon1R 0 / 2), and this can be achieved for, say, /epsilon1 ≤ min 1 (2 ) (2 log 2) { / δ , /R 0 } .
( ⇒ ) Strong hyperbolicity with parameter /epsilon1 yields 1 + exp( -/epsilon1A ) ≥ exp( -/epsilon1B ) ≥ 1 -exp( -/epsilon1A ). When A ≥ 0, this means that -log(1 + exp( -/epsilon1A )) ≤ /epsilon1B ≤ -log(1 -exp( -/epsilon1A )). Now log(1 + a ) ≤ 2 a and -log(1 -a ) ≤ 2 a for small enough a ∈ [0 , 1). Thus, there is A > 0 0 such that | /epsilon1B | ≤ 2 exp( -/epsilon1A ) whenever A ≥ A 0 . We deduce (EB) with L = 4 //epsilon1 , λ = /epsilon1/ 2, R 0 = 2 A 0 . /square
The following is the key estimate for the proof of Theorem 6.1. It is a particular instance of Gou¨ ezel's strong uniform Ancona inequalities [8, Definition 2.8, Theorem 2.9]. It can also be
deduced from an adaptation of [13, Lemma 3.2], and that was, in fact, our original approach before becoming aware of the very recent [8].
Lemma 6.3. There exist L, λ > 0 and R > 0 0 such that
$$\left | \frac { G ( x, y ) \, G ( z, t ) } { G ( x, t ) \, G ( z, y ) } & - 1 \right | \leq L \exp ( - \lambda R ) \\. \quad. \quad \tau \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \tcdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t \cdot \ t$$
whenever the distance between geodesics [ x, z ] and [ y, t ] is at least R ≥ R 0 .
We emphasize that, in the above lemma, geodesics are taken with respect to the wordmetric on Γ. Let us derive condition (EB) for the Green metric on Γ. On the one hand, as | log a | is asymptotic to | a - | 1 for a → 1, the conclusion of Lemma 6.3 can be stated as
$$\left | | x, y | _ { G } + | z, t | _ { G } - | x, t | _ { G } - | z, y | _ { G } \right | & = \left | \log \frac { G ( x, y ) \, G ( z, t ) } { G ( x, t ) \, G ( z, y ) } \right | \leq L \exp ( - \lambda R ), \\ \dots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots\cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cd� \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \colon \infty \, \infty \, \infty$$
up to increasing R 0 if necessary. (Compare, at this point, with [15, Theorem 32d].)
On the other hand, we claim that there is some spatial constant C > 0 such that
$$C \text{ dist} \left ( [ x, z ], [ y, t ] \right ) + C \ge$$
where, once again, the left hand side is in terms of the word metric. Let p and q be points on [ x, z ], respectively [ y, t ], with | p, q | = dist [ ( x, z , ] [ y, t ] ) . (See the figure below.) We have | p, q | G ≤ C 1 | p, q | + C 1 for some spatial constant C 1 > 0, by the quasi-isometry of the word and the Green metrics. Also, there is a spatial constant C 2 > 0 such that
$$| x, p | _ { G } + | p, z | _ { G } \leq | x, z |$$
since geodesics for the word metric are rough geodesics for the Green metric. It follows that
$$| x, y | _ { G } + | z, t | _ { G } & \leq \left ( |$$
and the claim is proved. This completes the verification of condition (EB) for the Green metric.
6.2. Harmonic measure. We will now explain a consequence of Theorem 6.1 which concerns the harmonic measure on the boundary. In what follows, we write ( · , · ) G for the Gromov product with respect to the Green metric, and having the identity element e ∈ Γ as the basepoint.
The random walk ( Z n ) started at e ∈ Γ converges almost surely to a boundary point Z ∞ . The harmonic measure ν is the probability measure on ∂ Γ defined by the condition that ν A ( ) is the probability that Z ∞ is in A ⊆ ∂ Γ. By [4, Section 3.4], see also the proof of Proposition 3.6 in [11], one knows that ν is a Γ-conformal measure:
$$( 1 ) \quad \begin{array} {$$
where β G is the Busemann function with respect to the Green metric. Recall that
$$\beta _ { G } ( e, g ; \xi ) = 2 ( g, \xi ) _ { G } -$$
In [4] and [11], the existence of the above limit follows from the identification of the Gromov boundary with the Martin boundary of the random walk. Herein, we see it as a manifestation of the good hyperbolicity of the Green metric.
On the other hand, good hyperbolicity of the Green metric means that the boundary visual structure induced by the Green metric is conformal, as explained in Section 3. We endow ∂ Γ with a visual metric d /epsilon1 = exp( -/epsilon1 ( · , · ) G ) for small enough /epsilon1 > 0, and we let µ denote the normalized Hausdorff measure (independently of /epsilon1 ). Then µ is Γ-conformal, with
$$\stackrel { \cdot } { \frac { \delta g ^ { * } \mu } { \mathrm d \mu } } ( \xi ) = \exp \beta _ { G } ( e, g ^ { - 1 } ; \xi )$$
since the Hausdorff dimension of d /epsilon1 is 1 //epsilon1 by [4, Theorem 1.1].
Now [4, Theorem 2.7] says that any two Γ-quasi-conformal measures on ∂ Γ are equivalent ergodic measures, with Radon - Nikodym derivative bounded above and below by positive constants. Equations (1) and (2) imply that the Radon - Nikodym derivative d ν/ d µ is Γinvariant, hence constant a.e. by ergodicity. We conclude:
Corollary 6.4. Let Γ be a hyperbolic group. Then the harmonic measure on ∂ Γ defined by a symmetric and finitely supported random walk on Γ equals the Hausdorff probability measure defined by any Green visual metric d /epsilon1 = exp( -/epsilon1 ( · , · ) G ) .
Acknowledgments. We thank Peter Ha¨ ıssinsky for useful comments. B.N. acknowledges support from the Alexander von Humboldt Foundation.
## References
- [1] A. Ancona: Th´ eorie du potentiel sur les graphes et les vari´t´s e e , in ´ cole d'´t´ de Probabilit´s de SaintE e e e Flour XVIII-1988 , Lecture Notes in Math. 1427, Springer 1990, 1-112
- [2] M. Bj¨ orklund: Central limit theorems for Gromov hyperbolic groups , J. Theoret. Probab. 23 (2010), no. 3, 871-887
- [3] S. Blach`re, S. Brofferio: e Internal diffusion limited aggregation on discrete groups having exponential growth , Probab. Theory Related Fields 137 (2007), no. 3-4, 323-343
- [4] S. Blach`re, P. Ha¨ ıssinsky, P. Mathieu: e Harmonic measures versus quasiconformal measures for hyperbolic groups , Ann. Sci. Ec. Norm. Sup´r. (4) 44 (2011), 683-721 ´ e
- [5] M. Bonk, O. Schramm: Embeddings of Gromov hyperbolic spaces , Geom. Funct. Anal. 10 (2000), no. 2, 266-306
- [6] M. Bourdon: Structure conforme au bord et flot g´od´sique d'un e e CAT( -1) -espace , Enseign. Math. (2) 41 (1995), no. 1-2, 63-102
- [7] T. Foertsch, V. Schroeder: Hyperbolicity, CAT( -1) -spaces and the Ptolemy inequality , Math. Ann. 350 (2011), no. 2, 339-356
- [8] S. Gou¨ ezel: Local limit theorem for symmetric random walks in Gromov-hyperbolic groups , J. Amer. Math. Soc. 27 (2014), no. 3, 893-928
- [9] M. Gromov: Hyperbolic groups , in Essays in group theory , Publ. MSRI 8, Springer 1987, 75-263
- [10] P. Ha¨ ıssinsky: G´ eom´trie e quasiconforme, analyse au bord des espaces m´ etriques hyperboliques et rigidit´s e (d'apr`s e Mostow, Pansu, Bourdon, Pajot, Bonk, Kleiner) , S´ eminaire Bourbaki 2007/2008, Exp. No. 993, Ast´risque 326 (2009), 321-362 e
- [11] P. Ha¨ ıssinsky: Marches al´atoires sur les groupes hyperboliques e , in G´ eom´trie ergodique e , F. Dal'Bo ed., Monographie de l'Enseignement Math´ ematique 43 (2013), 199-265
- [12] P. Ha¨ ıssinsky, P. Mathieu: La conjecture de Baum-Connes pour les groupes hyperboliques par les marches al´atoires e , preprint 2011
- [13] M. Izumi, S. Neshveyev, R. Okayasu: The ratio set of the harmonic measure of a random walk on a hyperbolic group , Israel J. Math. 163 (2008), 285-316
- [14] V. Lafforgue: K-th´orie bivariante pour les alg`bres de Banach et conjecture de Baum-Connes e e , Invent. Math. 149 (2002), no. 1, 1-95
- [15] I. Mineyev: Flows and joins of metric spaces , Geom. Topol. 9 (2005), 403-482
- [16] I. Mineyev: Metric conformal structures and hyperbolic dimension , Conform. Geom. Dyn. 11 (2007), 137-163
- [17] I. Mineyev, G. Yu: The Baum-Connes conjecture for hyperbolic groups , Invent. Math. 149 (2002), no. 1, 97-122
- [18] B. Nica: Proper isometric actions of hyperbolic groups on L p -spaces , Compos. Math. 149 (2013), no. 5, 773-792
- (B.N.) Mathematisches Institut, Georg-August Universit¨t G¨ttingen, Germany a o E-mail address : [email protected]
- (J.S.) Mathematical Sciences, University of Southampton, UK
- E-mail address : [email protected] | null | [
"Bogdan Nica",
"Jan Spakula"
] | 2014-08-01T18:06:37+00:00 | 2015-06-29T09:50:25+00:00 | [
"math.GR",
"math.MG",
"20F67"
] | Strong hyperbolicity | We propose the metric notion of strong hyperbolicity as a way of obtaining
hyperbolicity with sharp additional properties. Specifically, strongly
hyperbolic spaces are Gromov hyperbolic spaces that are metrically well-behaved
at infinity, and, under weak geodesic assumptions, they are strongly bolic as
well. We show that CAT(-1) spaces are strongly hyperbolic. On the way, we
determine the best constant of hyperbolicity for the standard hyperbolic plane.
We also show that the Green metric defined by a random walk on a hyperbolic
group is strongly hyperbolic. A measure-theoretic consequence at the boundary
is that the harmonic measure defined by a random walk is a visual Hausdorff
measure. |
1408.0251v2 | ## MODELING CASSAVA YIELD: A RESPONSE SURFACE
APPROACH
## A.O. Bello Federal University of Technology, Nigeria
## A BSTRACT
This paper reports on application of theory of experimental design usingechniques in R graphical t programming language and application of nonlinear bootstrap regression method to demonstrate the invariant property of parameter estimates of the Inverse polynomial Model IPM in a nonlinear surface.
## K EYWORDS
Design properties, riance prediction function, Bootstrap non-linear regression, Cassava Va
## 1. INTRODUCTION
Yield has always being a major yard stick for measuring successful agricultural method and economic indices.
There has being increased demand for export of cassava to China, US and other developed nations of the world whose production of currency papers, starch related products, recently fas increasing bio-fuel energy and others, largely depends on Africa's production of cassava. The demand for Cassava has globally increased and it overshoot supply, the of drop in occurrence yield has put a lot of pressure on production of Cassava and the present increase in cultivation not enough to curb demand, according to Food and Agriculture Organization of the United Nationsdatabase FAOSTAT (2009.
Achieving the production efficiency of cassava is the problem that this research tends to employ statistical tools to approach, focusing on the area of modelling Cassava yield plantation an highlighting the respective point of intercept and gradient levels of fertilizer and crop spac with the variety of cassava that is capable of delivering the cassava production efficiency. T soil, cultural practices, and socio-economic factors also play a key role in cassava production. many tropical countries, soil fertility is usually poor and requires fertilization to achieve a level of cassava yield Muengula-Manyi1, et al., 2012 morusi2011 M. . O on the other hand reported that the incidence of CMD infection that affects cassava yield was depressed by high dosage of NPK (100 kg ha-1 when compared to plants in soil with low NPK (10 kg ha -1.
## 1.2 The World Bio-Fuel Target: The Pressing Pressure for Highva Yield Cassa .
According to Rishi Sidhu2011, cassava is the newest addition to io-fuel line-up, joining the b other bio-chemical substances, and has doubled in price because of its new role, quoting New
International Journal on Computational Sciences & Applications (IJCSA) Vol.4, No.3, June 2014
York Times.To find a new source of energy, Chinese scientists perfected a process to make fuel from cassava. he European Union has aid that 10 percent of transportation fuel must come from T s renewable sources like bio or wind power by 2020, and the U.S. requires bio to - fuel - fuel use reach 36 billion gallons annually by 2022. These targets, according D. to Timothy Searchinger(2013) r esearch scholar at Princeton University. to have quick increase in a Failure yield would not only affect the bio arget but also amount to cassava food and its by - fuel t - products shortage.
## 2.METHODOLOGY
RSM involves several experiments, and using the resul e experiment to provide direction ts of on for what to do next. This next action could be to focus the experiment around a different set conditions, or to collect more data in the current experimental region in order to fit a higher -model or confir what we seem to have found ( sell V. Lenth 2012 . m Rus , )
$$y = f ( x ) \beta + \varepsilon \text{ \quad \ \$$
Where ), ..., ( , 2 , 1 k x x x x glyph<UNKNOWN> s is the coefficient estimates,r term) ~ (0; 1) and (erro N independent. f is the function that describes the form in which the response and the input variable are related, its mathematical form is not known in practice. It is often approximate within the experimental region. The purpoonsidering a model such as is to establish a se of c Eq. 1 relationship, albeit approximations, that can be used to predict response values for given setti of the control variables, determine significance of the factors. The totality of the predictors' settings is denoted by D, considering n matrix form; Eq. 1, i
$$y = X \beta + \varepsilon$$
Ordinary least quares estimator ofis - Y X X X ' ) ' ( ˆ 1 glyph<UNKNOWN> glyph<UNKNOWN>
Y, estimates riance va 1 2 ) ' ( ) ˆ ( glyph<UNKNOWN> glyph<UNKNOWN> X X Var and the predicted response given glyph<UNKNOWN> glyph<UNKNOWN> glyph<UNKNOWN> ) ( ˆ , ) ( ˆ x f x y with variance as
$$V a r [ \hat { y } ( x ) ] = \delta ^ { 2 }$$
The proper choice of design is very important in any response surface investigation. This is tr because the quality of prediction, as measured by the size of the prediction variance, depends o the design matrix D as can be seen from formulaIt is therefore imperative that the ). ˆ ( Var prediction variance in Eq. 3 be as small as possi ed that the postulated model in Eq. 1 ble provid does not suffer from lack of fit. [SeeKhuri et al (2010) Khuri, AI, Cornell (1996) Andre I and for details].
## 2.1Orthogonality
A design D is said to be orthogonal if the matrix XX is diagonal, where X is t in he model matrix Eq. 2. The elements ofwill be uncorrelated because the off diagonal elements of ˆ ) will ˆ ( Var be zero.
International Journal on Computational Sciences & Applications (IJCSA) Vol.4, No.3, June 2014
## 2.2Rotatability:
A design D is said to be rotatable if the prediction variance in Eq. 3 ispoints that constant at all are equidistant from the design centre. The prediction variance remains unchanged under any rotation of the coordinate axes.
## C. Uniform Precision:
A rotatable design is said to have the additional uniform precision property i the f Var [ (x)] at y ˆ origin is equal to its value at a distance of one from the origin. This helps in producing so stability in the prediction variance in the vicinity of the design centre
$$\text{any in the creation variance} \, \begin{matrix} \\ \\ \\$$
## 2.2The Choice Of Design and Model :
The problem of establishing cassava production efficiency in a world of inadequate resources, where resources such as maner, seedling, fertilizers, time etc cannot be risked to test run all - pow possible yield in a realplanting venture emphasis on tropical regions of Africa, where other life agricultural factors are asymptotically the same. . et (2000 ,noted that the cost of Parsad, R al ) constructing a response surface affects the choice of design of experiment and this can have large influence on accuracy of the approximation. These call for caution in making choice of design and model.
In the quest of modelling cassava yield effectively, a good choice of design and model (methodology) is vital. The above challenges sug need for a methodology that has the gest the ability to develop a smaller design that can lead to a model that will adequately suggest a possible yields per planting input factors. A design that will help to reduce the number o experimental runs which uld be costly or unattainable. The ability to model all possible yields co per planting input will help to estimate the quantity of harvest that will eventually be due exports and local consumption. Also this will ensure good planning for each farming a nd economic year.
## 2.3The Design
The first rder design proposed for fitting first odel [factorial Design]. second - order - o - degree m -The design proposed for fitting second model [The Central Composite design (CCD)] - d gree --Orthogonality, Rotatability and Unirecision are desired properties in any design. form P CCD consist the full factorial runs (or fractional factorial of resolution V) with nf runs, also "cube point", (ii) The axial or star runs consisting n on the axis of each factor at a = 2k points distance from the center of the design and, (iii) The center runs 0) without replication. 0 n (0, 0...,
The total number of runs 0 n n n N a f glyph<UNKNOWN> glyph<UNKNOWN> glyph<UNKNOWN> (See Klaus Hinkelmann, Design and Analysis of Experiments Volume 1 Second Editir details). on fo
$$b = \sum _ { i = 1 } ^ { N } x _ { i } ^ {$$
International Journal on Computational Sciences & Applications (IJCSA) Vol.4, No.3, June 2014
$$\begin{array} { c } \cdot & \cdot \\ \text{For appropriate distance from the center } \alpha = ( n f ) ^ { \sqrt { 4 } } \\ b ^ { 2 } = N c \Leftrightarrow [ n f + 2 \alpha ^ { 2 } ] ^ { 2 } = [ n f + n _ { a } + n _ { 0 } ] n _ { f } \\ & \text{and} \\ d = 2 c & ( 6 ) \\ 2. 4 \text{ The Inverse Polynomial Model (IPM) A Class of } \Gamma \end{array}$$
## 2.4The Inverse Polynomial Model (IPM) A Class Nonlinear Model of
The general forms of Inverse Polynomials of (Nelder, 1966; Nduka, 1994; Holger Dette, 2007) are family of nonear function. - l i
$$& \frac { x _ { 1 }, x _ { 2 }, \dots, x _ { k } } { y } = p o l y n o m i a l \, \text{in} \ x _ { 1 }, x _ { 2 }, \dots, x _ { k } \ \text{(Nelder1964)} \\ & \prod _ { i = 1 } ^ { k } \frac { x _ { i } } { q } = \prod _ { i = 1 } ^ { k } ( \beta _ { 0 } + \beta _ { 1 i } x _ { i } + \beta _ { 2 i } x _ { i } ^ { 2 } + \dots + \beta _ { p - 1 i } x _ { i } ^ { p - 1 } ) \quad \quad ( 7 )$$
Where, k = number of factors p = number of levels of factor i. eld per unit area. q = the yi
Intrinsically Non near Model; Non inearity i parameters and when linearity cannot be - li - l n achieved through transformation
A regression model is called nonr, if the derivatives of the model with respect to the model - li ea parameters depend on one or more parameters. Method of Gaussis more adequ te in - Newton a estimation of inverse polynomial model; it is preferable because of its faster convergence i inverse polynomials, also the boundedness, invariant of parameter estimates, distribution free asymmetry, and quick location of optimum properties favoureice of Inverse Polynomial d the cho model. [See Nduka, E. C. 1994 nd 1997]. a
*First Order Model;
$$\prod _ { i = 1 } ^ { k } \frac { \stackrel { \cdot } { x _ { i } } } { q } = ( \beta _ { _ { 0 1 } } + \beta _ { _ { i 1 } } x _ { _ { 1 } } ) ( \beta _ { _ { 0 2 } } + \beta _ { _ { i 2 } } x _ { _ { 2 } } ) \ \text{ also, let } q = y$$
$$\frac { x _ { 1 } x _ { 2 } } { q } = \beta _ { 0 1 } \beta _ { 0 2 } + \beta _ { 0 1 } \beta _ { 1 2 } x _ { 2 } + \beta _ { 1 1 } \beta _ { 0 2 } x _ { 1 } + \beta _ { 1 1 } \beta _ { 1 2 } x _ { 1 } x _ { 2 }$$
$$y ^ { - 1 } = \beta _ { 1 1 } + \beta _ { 0 1 } x _ { 1 } ^ { - 1 } + + \beta _ { 1 0 } x _ { 2 } ^ { - 1 } + + \beta _ { 0 0 } ( x _ { 1 } x _ { 2 } ) ^ { - 1 }$$
*Second Order Form Quadratic Surface (k=2, p=4)
$$\frac { x x } { q } = ( \beta _ { 0 1 } + \beta _ { 1 1 } x + \beta _ { 2 1 } x ^ { 2 } ) \| \beta _ { 0 2 } + \beta _ { 1 2 } x + \beta _ { 2 2 } x ^ { 2 } ) \| \beta _ { 0 3 } + \beta _ { 1 3 } x + \beta _ { 2 3 } x ^ { 2 } ) \| \beta _ { 0 4 } + \beta _ { 1 4 } x + \beta _ { 2 4 } x ^ { 2 } )$$
above is rewritten considering only up to the 2nd order as ;
$$& y ^ { - 1 } = \beta _ { 1 1 } + \beta _ { 0 1 } x _ { 1 } ^ { - 1 } + \beta _ { 1 0 } x _ { 2 } ^ { - 1 } + \beta _ { 0 2 } x _ { 2 } x _ { 1 } ^ { - 1 } + \beta _ { 2 0 } x _ { 1 } x _ { 2 } ^ { - 1 } + \beta _ { 0 0 } ( x _ { 1 } x _ { 2 } ) ^ { - 1 } \\ & ( 1 0 )$$
## 3.PARAMETER ESTIMATION
## Nonlinear Estimation
The normal equation is extremely d ult to solve, if it have multiple solution to multiple iffic stationary values of functionIt is said to have no closed form solution. Thereby the ). ˆ ( S procedure of iterative method must be approached in order to man over the mathematical intractable problem.
## 3.1InversePolynomial Nonlinear Estimation
Considering the parameter the estimate ( of , ˆ ) glyph<UNKNOWN> is obtained by differentiating equation S ( ˆ ) and equated to zero, giving rise to Jequations and solve for the Considering - normal glyph<UNKNOWN> s. determistic omponent of our model in Eq. 1. T m of square error for non linear model is c he su defined as
$$S ( \theta ) = \sum _ { i = 1 } ^ { n } [ y _ { i } - f ( x _ { i }, \theta ) ] ^ { 2 } \, \quad \quad ( 1 1 )$$
When the J - normal equation becomes mathematical intractable, an iterative procedure is required.
$$S ( \theta ) = \sum _ { i = 1 } ^ { n } [ \begin{matrix} y _ { i } - f ( x _ { i }, \theta _ { j } ) \\ \end{matrix} ] ^ { 2 }$$
$$\text{Let } \beta _ { j } = \theta _ { j }$$
$$S ( \theta ) = \sum _ { i = 1 } ^ { n } [ y _ { i } - ( \theta _ { 0 0 } ( x _ { 1 } x _ { 2 } ) ^ { - 1 } + \theta _ { 0 1 } x _ { 2 } ^ { 1 } + \theta _ { 1 1 } ) ^ { - 1 } ] ^ { 2 } \, | \frac { \partial f ( x _ { i, } \theta ) } { \partial \theta _ { j } } | _ { ( \theta _ { j } - \theta _ { j } ^ { \theta _ { j } } ) }$$
$$\text{Combining } \frac { \partial S ( \theta ) } { \partial \theta _ { 0 0 } }, \frac { \partial S ( \theta ) } { \partial \theta _ { 0 1 } }, \frac { \partial S ( \theta ) } { \partial \theta _ { 1 0 } } \text{ and } \frac { \partial S ( \theta ) } { \partial \theta _ { 1 1 } } \text{ we the normal equation;}$$
$$& \sum _ { i = 1 } ^ { n } [ y _ { i } - f ( x _ { i } \theta ) ] [ \theta _ { 0 0 } ( x _ { 1 } x _ { 2 } ) ^ { - 1 } + \theta _ { 0 1 } x _ { 1 } ^ { - 1 } + \theta _ { 1 0 } x _ { 2 } ^ { - 1 } + \theta _ { 1 1 } ) ^ { - 2 } ] ( x _ { 1 } x _ { 2 } ) ^ { - 1 } = 0 \\ & \sum _ { i = 1 } ^ { n } [ y _ { i } - f ( x _ { i }, \theta ) ] [ \theta _ { 0 0 } ( x _ { 1 } x _ { 2 } ) ^ { - 1 } + \theta _ { 0 1 } x _ { 1 } ^ { - 1 } + \theta _ { 1 0 } x _ { 2 } ^ { - 1 } + \theta _ { 1 1 } ) ^ { - 2 } ] x _ { 1 } ^ { - 1 } = 0 \\ & \sum _ { i = 1 } ^ { n } [ y _ { i } - f ( x _ { i }, \theta ) ] [ \theta _ { 0 0 } ( x _ { 1 } x _ { 2 } ) ^ { - 1 } + \theta _ { 0 1 } x _ { 1 } ^ { - 1 } + \theta _ { 1 0 } x _ { 2 } ^ { - 1 } + \theta _ { 1 1 } ) ^ { - 2 } ] x _ { 2 } ^ { - 1 } = 0 \\ & \frac { \partial S ( \theta ) } { \partial \theta _ { 1 1 } } = 0 \\ & \frac { \partial S ( \theta ) } { \partial \theta _ { 1 1 } } = 0 \cdots.$$
$$\frac { \partial S ( \theta ) } { \partial \theta _ { 1 1 } }$$
The normal equation has no close form solution.
## 3.2 Gauss Newton Numerical Method
Consider the de rministic component of equation (1), by the Taylor first ximation; te - order appro
$$( 1 2 )$$
by substitution,
$$y \, =$$
International Journal on Computational Sciences & Applications (IJCSA) Vol.4, No.3, June 2014
$$y \, =$$
For the j parameters i observation, we have; and th
$$y \text{-} f _ { 1 } ^ { 0 } = \Sigma _ { j = n } ^ { j - 1 } \frac { g _ { j } | g _ { i } | g _ { j } } { \omega. } | ( \theta - \theta ^ { 0 } ) + \varepsilon \quad \ ( 1 3 )$$
$$\text{Note that $f_{1}^{0}$}$$
$$\text{Let } \beta _ { i } ^ { o } = ( \theta _ { j } - \theta _ { j } ^ { o } ) \quad \text{and } \, z _ { i j }$$
Eq. 13 can be written as
$$y _ { i } - f _ { i } ^ { o } = \sum _ { i = i } ^ { k } \beta _ { j } Z _ { j i } + \varepsilon _ { i } \\. \quad. \quad. \.$$
The above is also the same in matrix form below
$$Y - f ^ { 0 } = Z ^ { 0 } \beta + \varepsilon$$
Column vector 0 f Y glyph<UNKNOWN> (k X1) , Z(jxk) 0 is determinant matrix , is the coefficient matrix and glyph<UNKNOWN> (jx1) glyph<UNKNOWN> is the error term(kx1). The least square estimator is of form
$$\hat { \beta } = ( Z ^ { o } \cdot Z ^ { o } ) Z ^ { o } ( Y - f ^ { 0 } ) \quad \ \ ( 1 5 )$$
$$\min \text{izing the sum of square } \ S ^ { 0 } ( \theta ) = \sum _ { i = 1 } ^ { n } [ y _ { i } - f _ { i } ^ { o } - \sum \beta _ { j } Z _ { i j } ] ^ { 2 }$$
The initial value for and is supplied either through theoreti wledge of subject, cal kno intelligent guess, self er function in R or grid search function in SAS. A good choice of - start starting value will ensure rapid convergence if not there may not be any convergence in the serie (See Schabenberger Oliver,"Nonlinea gression in SAS"). r Re Condition for Convergence
$$| \frac { ( \theta _ { i ( j + 1 ) } - \theta _ { i j } ) } { \theta _ { i j } } | < \delta, = 0. 0 0 0 0 1 \, \text{or} \, 1. 0 x 1 0 ^ { - 6 } & & i = 1, 2, \dots, n$$
## 3.3Gauss Newton Nonlinear Least Square Estimation of IPM
Starting value fors i.e ( , glyph<UNKNOWN> j 00 10 01 , and 11 ) is supplied for 1st iteration using equation[19], note and 2 11 1 2 10 1 1 01 1 2 1 00 ) ) ( glyph<UNKNOWN> glyph<UNKNOWN> glyph<UNKNOWN> glyph<UNKNOWN> glyph<UNKNOWN> glyph<UNKNOWN> glyph<UNKNOWN> glyph<UNKNOWN> x x x x
## 3.4The general Bootstrap Algorithm .
- 1. Generate a sample x size k from distributioth replacement. *' of n F wi
- 2. Compute j for this bootstrap sample.
- 3. Repeat steps 1 and 2, B time.
By this procedure we end up with bootstrap values
$$\varphi \theta _ { b j } ^ { * } = [ \theta _ { b 0 } ^ { * }, \theta _ { b 1 } ^ { * }, \dots, \overset { \nu } { \theta _ { b k } ^ { * } } ] ^ { \cdot } \ j = 1, \dots, \overset { \nu } { B }$$
Afterward, we can use these bootstrap values to ca ll the quantities of interest. Generally, lculate a 500 or 1000 samples will be sufficient [See s Karlis (2004) for details] Dimitri .
## 4.NUMERICAL ANALYSIS.
The data considered for this work is an experimental data of two factorial design, graded into fo levels, having crop spacing (meters) and inorganic fertilizer application per crop stand (decilitres). Cassava varieties TMS 30572 were exposed to the factors (crop spacing and fertilizer). Plant and soil analysis were also carried out to ensure the requi red agricultural standards. The responses generated were measured in yield (kilograms) and dimension (centimetres) as dual responses Yield and Dim respectively.
## 4. 1 Bootstrap Re - sampling Method
The term bootstrap, coined by B. Efron (1986, and 1997), refe g the sample to learn rs to usin about the sampling distribution of a statistic without reference to external assumptions while i a method for improving estimators, it is also well known as a method for estimating standard errors, bias and constructing dence intervals for parameters. It substitute considerate amount confi of computation in place of theoretical analysis ( d Tibshirani,1986) B.Efron an .
TABLE I. R - NLS ()ITERATION R ESULT
| IPM | S | iteration | SSE | Convergence Tolerance | Pr>F |
|-------|-------|-------------|--------|-------------------------|--------|
| Yield | 6 | 10 | 2.25 | 7.47e07 - | 0.001 |
| Dim | 6 | 12 | 349.2 | 5.62e07 - | 0.001 |
## 4. 2 Model Adequacy
The Shapiro Wilk normality test is conducted for cassavaYield and cassavaDim models -respectively; standardRes results W = 0:9386, W = 0:9767 respective closeness of standardRes values to '1' indicated the assumption of normally distributed errors is not violated.
TABLE II. BOOTSTRAP RESULT WIT H YIELD
| Parameter | Bootstrap Sample | Bootstrap Sample | Observed Sample | Observed Sample |
|-------------|--------------------|--------------------|-------------------|-------------------|
| | Estimates | C.I (2.5% 97.5%) | Estimates | C.I (2.5% 97.5%) |
| 11 | 0.356039089 | 0. 2 0.6 | 0.349449160 | 0.1 0.62 |
| 01 | - 0.00922816 | - 0.1 0.12 | - 0.00844514 | - 0.09 0.07 |
| 10 | - 0.22006195 | - 0.54 0.09 | - 0.21609962 | - 0.66 0.21 |
| 00 | 0.011454663 | - 0.05 0.078 | 0.011312689 | - 0.07 0.099 |
| 20 | 0.202227135 | - 0.242 0.617 | 0.197842082 | - 0.386 0.793 |
International Journal on Computational Sciences & Applications (IJCSA) Vol.4, No.3, June 2014
First and second CassavaYield Model;
```
1 ^ { s t } = [ 0. 3 5 6 - 0. 0 0 9 \xi _ { 1 } ^ { 1 } - 0. 2 2 0 x _ { 2 } ^ { 1 } + 0. 0 1 1 6 \xi _ { 1 } x _ { 2 } ) ^ { - 1 } ] ^ { 1 } \\ 2 ^ { n d } = & [ 0. 3 5 6 - 0. 0 0 9 2 x _ { 1 } ^ { - 1 } - 0. 2 2 0 x _ { 2 } ^ { - 1 } + 0. 0 1 1 5 ( x _ { 1 } x _ { 2 } ) ^ { - 1 } + 0. 2 0 2 2 x _ { 1 } x _ { 2 } ^ { - 1 } ] ^ { - 1 }
```
## 5.DISCUSSION OF RESULTS
The various residual plots and test has giving a basis to s e inverse polynomial model ubmit that th is appropriate for fitting cassava yield (kg) and cassava dimension (cm).
It was also observed that the residuals does not sum up to zero as of the cases of the line surfaces. This is expected of a non resu . The results of bootstrap and the observed data - linear lt show closer agreement, which produce a negligible bias value. The bootstrap C.I. and original sample C.I also agree quite well demonstrating the precision of the model coefficient estimates The strong sig ficance of crop spacing in the model of cassava plantation has shown that ni crowding of plants or mixed planting should be discouraged. The obtained non mates - linear esti were reintroduced as the starting values in model and the same estimates value were ob tained with their sum of squares error unchanged and zero iteration number, indicating a global minimum estimates were obtained.
## 5.1CONCLUSION
It is acceptable to say the cassavaYield model is most adequate, robust and best to fit cassa yield and good rop spacing affects cassava yield in a direct proportion. c
## ACKNOWLEDGMENT
The author will like to express his appreciation to Prof. T.A. Bamiduro, University of Ibadan, Nigeria for his helpful comments and suggestions.
## REFERENCES
- [1] Andre I Khuri anSiuli Mukhopodhgay Advance Review Response Surface methodology Volume 2 d March/April 2010;JohnWiley and sons Inc.
- [2] B.Efron and Tibshirani(1986) Bootstrap Methods for standard errors, confidence interval and other measure of Statistical Accuracy; Statisience 1986 Vol No. 1, pg 54 tical Sc -77
- [3] B.Efron and Tibshirani(1997) Improvements of cross tion; Bootstrap Method;Journal of the -Valida American Statistics Association 92(438)pg 548 -560
- [4] Christian Ritz and Jens Carl S.(2008) Non Regression with Springer Science+Business -linear R Media LLC
- [5] Dimitris Karlis (April 2004)Department of Statistics Athens University of Economics Lefkada, An introduction to Bootstrap Methods 17th Conference of Greek Statistical Society http://stat -athens.aueb.gr/~karlis/le/boot.pdf fkada
- [6] FAOSTAT (2009);http://faostat.fao.org Food and agriculture Organization (FAO) of the United Nations, Rome, Italy.
- [7] Gary W. Oehlert (University of Minnesota) A First Course in Design and Analysis of Experiments 2000 Edition.
- [8] Holger Dette and Christine Kiss (2007) Optimal Experimental Designs for Inverse Quadratic Regression Models; Ruhr niversity Bochum -U
- [9] J. A. Nelder(1966) Weight Regression Quantal Response Data, and Inverse Polynomials National vegetable Research Station, Welle ne, Warwick, England. sbour
International Journal on Computational Sciences & Applications (IJCSA) Vol.4, No.3, June 2014
- [10] J. A. Nelder(1968) Inverse Polynomials, A useful Group of Multi sponse Functions; The -Factor Re National vegetable Research Station, Wellesbourne, Warwick, England
- [11] Jennifer Abraham Cassava: A Plethora Of Opportunities -C opyright © 2013 Punch Nigeria Limited · January 10, 2013
- [12] Klaus Hinkelmann ,(State University Department of Statistics Blacksburg, VA)and Oscar Kempthorne (Iowa State University Department of Statistics Ames, IA)d Analysis of - Design an Experiment s Volume 1 Introduction to Experimental Design Second Edition
- [13] Khuri, AI, Cornell, JA. Response Surfaces. 2nd edition.New York: Dekker; 1996.
- [14] M. Muengula-Manyi1, et al., 2012 Application of response surface methodology to the optimization of amy lase production by Aspergillus oryzae MTCC 1847 American Journal Of Experimental Agriculture 2(3): 336, 2012 Science Domain International -350
- [15] Nduka, E.C. and T.A. Bamiduro, 1997. A new generalized inverse polynomial for response surface designs. J. NiStat. Assoc., 11: 41 g. -48.
- [16] Nduka, E. C. (1994) Inverse polynomials in the Exploration of Response Surfaces, A Phd Thesis,U. I
- [17] Omorusi, V.I., Ayanru. D.K.G. (2011). Effect of NPK fertilizer on diseases, pests, and mycorrhizal symbiosis in Cassava Int. J. Agric. Biol., 13, 391 . -395.
- [18] Parsad, R., Srivastava, R. and Batra, P.K. (2004). Designs for Fitting Response Surfaces in Agricultural Experiments. IASRI, Publication
- [19] Rajender Parsad And P.K. Batra(2000) Response Surface Design I.A.S.R.I.Avenue, New , Library Delhi - 110 012
- [20] Rishi Sidhu -Cassava, The Latest Biofuel -April12th,2011http://www.Foreignpolicyblogs.Com/2011/04/12/Cassava.
- [21] Russell V. Lenth, (The University of Iowa) Response Methods in R, and illustration Usi -Surface ng rsm Updated to version 2.00, 5 December 2012
- [22] Russell V. Lenth, (The University of Iowa)Surface Plots in the rsm Package Updated to version 2.00, 3 December 2012
California.http
- [23] Schabenberger Oliver,"Nonlinear Regression in SAS" University of : www.ats.ucla.edu/stat/mult\_pkg/faq/general/citingats.htm
- [24] Timothy D. Searchinger(2013) Bioenergy writings www.scholar.princeton.edu/tsearch/page
- [25] Whitcomb and Anderson Excerpted from manuscript for Chapter 10 of RSM Simplified, See $www.statease. om/rsm\_simplified.html$ c
## Appendix
fig1
International Journal on Computational Sciences & Applications (IJCSA) Vol.4, No.3, June 2014
fig ii
## Boostrap Residual crror
Boostrap Residual error
density-default(x
Fig III
## Author
The Author is trained in the discipline of statistics from the cradle of National diploma and Master degree in Statistics from University of Ibadan, Nigeria. He is still under traini , His fleer for Design of an Experiment and Nonlinearity studies ng produced this work. His fleer is also extended to Envirometric research.
 | null | [
"Adeshina Oyedele Bello"
] | 2014-07-31T10:29:00+00:00 | 2015-03-18T13:12:42+00:00 | [
"stat.CO",
"stat.AP"
] | Modeling Cassava Yield: A Response Surface Approach | This paper reports on application of theory of experimental design using
graphical techniques in R programming language and application of nonlinear
bootstrap regression method to demonstrate the invariant property of parameter
estimates of the Inverse polynomial Model (IPM) in a nonlinear surface. |
1408.0252v1 | ## Infrared extrapolations for atomic nuclei
R. J. Furnstahl , G. Hagen 1 2 3 , , T. Papenbrock 3 2 , , and
## K. A. Wendt 3 2 ,
1 Department of Physics, The Ohio State University, Columbus, OH 43210 2 Physics Division, Oak Ridge National Laboratory, Oak Ridge, TN 37831 USA 3 Department of Physics and Astronomy, University of Tennessee, Knoxville, TN 37996, USA
E-mail: [email protected] , [email protected] , [email protected] [email protected]
,
Abstract. Harmonic oscillator model-space truncations introduce systematic errors to the calculation of binding energies and other observables. We identify the relevant infrared scaling variable and give values for this nucleus-dependent quantity. We consider isotopes of oxygen computed with the coupled-cluster method from chiral nucleon-nucleon interactions at next-to-next-to-leading order and show that the infrared component of the error is sufficiently understood to permit controlled extrapolations. By employing oscillator spaces with relatively large frequencies, well above the energy minimum, the ultraviolet corrections can be suppressed while infrared extrapolations over tens of MeVs are accurate for ground-state energies. However, robust uncertainty quantification for extrapolated quantities that fully accounts for systematic errors is not yet developed.
PACS numbers: 21.10.Dr, 21.60.-n, 31.15.Dv, 21.30.-x
Submitted to: J. Phys. G: Nucl. Phys.
## 1. Introduction
Wave-function based methods for computing atomic nuclei, such as exact diagonalization [1, 2], coupled cluster [3, 4, 5], self-consistent Green's functions [6], or in-medium similarity renormalization group (SRG) [7, 8], require a computational effort that depends strongly on the size of the Hilbert space used. Truncating this model space introduces systematic errors that must be quantitatively analyzed for a given Hamiltonian and nucleus. By understanding these errors, we can devise controlled extrapolation techniques for energies and other observables.
Though we deal with a quantum mechanical problem, it is most instructive to look first at classical phase space. Let us consider the deuteron wave function, calculated from a realistic two-body interaction, and compute its Wigner transform. We recall that the Wigner transform
$$f ( r, p ) = \int d ^ { 3 } k \, \frac { e ^ { i r \cdot k } } { ( 2 \pi ) ^ { 3 } } \psi ^ { * } ( p + k / 2 ) \psi ( p - k / 2 )$$
is a mapping of a wave function to a phase-space distribution. Figure 1 shows the results for chiral effective field theory (EFT) Hamiltonians [9] with two different cutoffs. We see that the dominant part of the Wigner transform extends in (radial) position essentially up to the deuteron radius (though the exponential tail casts a long shadow), and in momentum space up to the cutoff of the interaction.
Generalizing this result, the number of single-particle states required to compute a nucleus with radius R from an interaction with cutoff Λ is essentially given by
$$n = \frac { 1 } { ( 2 \pi ) ^ { 3 } } \int _ { \ } d ^ { 3 } r \int _ { \ } ^ { \Lambda } d ^ { 3 } p \in ( R \Lambda ) ^ { 3 } \.$$
Here, we have not counted spin/isospin degrees of freedom. The key result is that the number of single-particle states grows as the third power of the cutoff, and it is proportional to R 3 ∝ A for a nucleus with mass number A . The proportionality constant in Eq. (2) depends on the actual basis one uses, and some efficiencies can possibly be gained from abandoning the oscillator basis in favor of Berggren bases [10], Sturmian bases [11] or discrete variable representations [12]. Equation (2) also makes clear that halo nuclei can be as expensive to compute as much heavier nuclei, and that much is to be gained by lowering the cutoff Λ of the interaction. This latter point explains why low-momentum interactions (e.g. V low k [13] or SRG [14]) are so popular.
When working with the harmonic-oscillator basis, its extent in position and momentum space must exceed the radius of the nucleus and the cutoff of the interaction, respectively. In other words, one must choose the oscillator frequency and the maximum excitation energy such that the ellipsoidal phase-space area of the oscillator covers the relevant parts of the Wigner function for the nucleus. Let b = glyph[planckover2pi1] / √ m glyph[planckover2pi1] Ω denote the oscillator length, where m is the nucleon mass. Then a simple semiclassical argument implies that for an oscillator basis with up to N excited quanta, we must demand
$$\sqrt { 2 N } \geq R / b \quad \text{and} \quad \sqrt { 2 N } \geq \Lambda b \,.$$
The ellipses corresponding to equality signs in these expressions are also shown in Fig. 1. We see that only the tails of the Wigner function - both in the directions of position and momentum - extend beyond the ellipses. Let us also note that Fig. 1 serves mainly as an illustration but should not be used for quantitative conclusions. Only in the semiclassical limit is it possible to quantitatively relate Wigner functions to classical phase-space structures [15]. The questions to address are what error results from the omission of these tails for the binding energy and other observables in a finite oscillator basis, and how quickly do these quantities converge as the number of basis states is increased?
Figure 1. (Color online) Top: ' s -wave' projection of Wigner function from deuteron wave function for a chiral N 2 LO potential with a regulator cutoff of Λ UV = 500MeV. Abscissa is r in units of fm, ordinate is momentum p in units of fm -1 . The dashed lines are semi-classical harmonic oscillator cutoffs for N = 4 and two different oscillator frequencies ( glyph[planckover2pi1] Ω = 20 and 50 MeV). The dotted lines are the same with N = 8. Bottom: same as top but for a lower regulator cutoff Λ UV = 450MeV.
## 2. Infrared cutoffs of the oscillator basis
The semi-classical estimates (3) are a useful guide, but for practical purposes we have to understand how the infrared (IR) cutoff Λ IR ≡ π/L and ultraviolet (UV) cutoff Λ UV of a finite oscillator basis can become tools for the extrapolation of observables. The leading-order estimates for these cutoffs are based on L ≈ √ 2 Nb and Λ UV ≈ √ 2 N/b , respectively [16, 17, 18]. Clearly, L and Λ UV are semiclassical estimates for the extent of the oscillator basis in position and momentum space, respectively.
The nucleon-nucleon forces (and many-body forces) used for wave-function based methods, such as chiral effective field theory (EFT) or low-momentum interactions, all have momentum-space regulators that rapidly drive them to zero above a cutoff Λ. If Λ UV is less than or comparable to Λ, the UV corrections to the energy and other observables will depend on the details of the regulators (but less on the nucleus under consideration). But when Λ UV sufficiently exceeds Λ (see Sec. 4), rapid UV convergence is observed in practical calculations, so that UV extrapolations are not needed.
In this paper, we focus solely on IR extrapolations. We note that the long wavelength structure of a bound state stems from its exponential fall-off in position space and is independent of the details of the interaction. A finite extent of the oscillator basis in position space cuts off this exponential tail and thereby yields an energy correction. References [19, 20, 21] provide a theoretical basis for IR extrapolations. It is established in these papers that a finite oscillator basis with N excited quanta and oscillator length b is, at low momenta, indistinguishable from a box of size
$$L _ { 2 } ( N, h \Omega ) \approx \sqrt { 2 ( N + 3 / 2 + 2 ) } b \,.$$
Here, the approximate sign indicates that this is the next-to-leading order approximation of the box size in the limit of N glyph[greatermuch] 1. This result can be derived from the fact that Λ 2 IR ≡ ( π/L 2 ) 2 is the lowest eigenvalue of the momentum operator squared in the finite oscillator basis [20]. A single-particle wave function of an s -wave bound state, approximated in a finite oscillator basis, only differs by high-momentum components from the same wave function approximated in a spherical well of radius L 2 . For a partial wave with angular momentum l , one has N = 2 n + l in Eq. (4). Thus, for fixed oscillator spacing glyph[planckover2pi1] Ω the length L 2 ( N, glyph[planckover2pi1] Ω) is a staircase function that increases at even (odd) values of N for even (odd) values of l .
This knowledge can be used to understand the IR extrapolations of bound states. The equivalent finite size L 2 of the oscillator basis has the effect on a position-space bound state of enforcing that the wave function has a node at L 2 . This correction to the true bound-state wave function is a long-wavelength phenomenon and can thus be understood in model-independent ways [20, 21]. The leading-order IR extrapolation formula for an s -wave bound-state energy of a single-particle system is
$$E ( L _ { 2 } ) = E _ { \infty } + \frac { \hbar { ^ { 2 } } k _ { \infty } \gamma _ { \infty } ^ { 2 } } { \mu } \exp \left ( - 2 k _ { \infty } L _ { 2 } \right ).$$
Here, k ∞ is the bound-state momentum, i.e. E ∞ = glyph[planckover2pi1] 2 k 2 ∞ / (2 µ ) is the bound-state energy, µ is the reduced mass, and γ ∞ is the asymptotic normalization constant. Higher-order
corrections, which are suppressed by powers of exp ( -2 k ∞ L 2 ), and extensions to general orbital angular momentum are also known [21].
Although the IR extrapolation formula (5) was derived for a single-particle degree of freedom (or systems that can be reduced to such), it has also been applied to bound states of many-body problems [19, 22, 23, 24, 25, 26] via
$$E ( L ) = E _ { \infty } + A _ { \infty } \exp \left ( - 2 k _ { \infty } L \right ),$$
with L N, ( glyph[planckover2pi1] Ω) still to be specified. The idea is that k ∞ can be generally interpreted as the (least) separation energy for the nucleus under consideration, so that (6) follows from the two-body derivation in Ref. [21] but now using the S matrix for the corresponding break-up reaction. Because of the many approximations involved, A ∞ , E ∞ , and k ∞ are all treated as fit parameters. To apply the extrapolation formula (6), one needs to work with bound states (i.e. the fully converged energy needs to be negative), L must exceed the radial extent of the nucleus under consideration, and the UV cutoff Λ UV must be sufficiently greater than the cutoff of the interaction. In practice, one can calculate using large values of glyph[planckover2pi1] Ω to ensure that UV corrections are much smaller than the IR corrections.
Let us consider applying Eq. (6) to compute the ground-state energies of oxygen isotopes. As a first step, we plot the energies obtained using different model spaces as a function of the effective box size L . The top panels in Fig. 2 show the results for 16 O (left) and 24 O (right) when plotted versus L = L 2 . The data points stem from model spaces with N = 8 10 12 oscillator quanta and Λ , , UV > 750 MeV, so that UV corrections are presumably small. Clearly, the data points do not fall on a single line for either nucleus, and similar results are found for 22 O. Earlier studies [20] showed that the smoothness of E L ( ), when plotted for a large set of N and glyph[planckover2pi1] Ω values for which UV corrections are small, is a sensitive diagnostic of the quality of L N, ( glyph[planckover2pi1] Ω). Thus, L = L 2 is not the accurate box size of the harmonic oscillator basis for fermionic many-body systems. In the next section we derive a more accurate value L eff by revisiting the underlying basis for the effective box size. As we will see, this value depends on the nucleus under consideration.
## 3. Box size L eff ( N, glyph[planckover2pi1] Ω) for nuclei
The paradigm for IR extrapolations in the harmonic oscillator basis can be stated as follows: For sufficiently low energies (and long wavelengths), a finite oscillator basis is indistinguishable (in the sense of an effective theory) from a spherical box with an appropriately chosen radius. This radius is determined by matching the lowest (most infrared) eigenvalue of the squared momentum operator in the oscillator basis to the lowest value in the box. To apply this matching procedure to the many-body case, we have to consider the lowest total squared momentum ∑ A i =1 p 2 i for a given nucleus. To do so, we identify the occupation numbers ν nl that give the lowest kinetic energy for that nucleus, and then equate the eigenvalue of the total squared momentum operator
Figure 2. Ground-state energies versus the scaling variable L from coupled-cluster theory in the CCSD approximation for 16 O (left) and 24 O (right) using a chiral N 2 LO potential with a regulator cutoff of Λ UV = 500MeV. The squares, diamonds and triangles stem from model spaces with N = 8 10 12, respectively. , , The top panels use the naive single-particle choice L = L 2 ( N, glyph[planckover2pi1] Ω), while the middle and bottom panels use approximate and exact L = L eff ( N, glyph[planckover2pi1] Ω), respectively, as described in Sec. 3.
for the oscillator basis to the total squared momentum for the same number of nucleons (occupying identical partial waves) in a spherical box of size L eff , i.e.
$$\sum _ { n l } \nu _ { n l } \kappa _ { l n } ^ { 2 } = \sum _ { n l } \nu _ { n l } \left ( \frac { a _ { l, n } } { L _ { \text{eff} } } \right ) ^ { 2 } \,.$$
Here, κ 2 ln is the eigenvalue of the single-particle squared momentum operator and a l,n is the ( n +1) th zero of the spherical Bessel function j l [27]. This yields
$$L _ { \text{eff} } = \left ( \frac { \sum _ { n l } \nu _ { n l } a _ { l, n } ^ { 2 } } { \sum _ { n l } \nu _ { n l } \kappa _ { l n } ^ { 2 } } \right ) ^ { 1 / 2 } \,. \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \ \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \cdot \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \ } \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \$$
We can parameterize this result using L 2 by introducing N eff , which is defined as a function of N by
$$L _ { \text{eff} } ( N, \hbar { \Omega } ) \equiv L _ { 2 } ( N _ { \text{eff} }, \hbar { \Omega } ) \.$$
We note that the approach to the many-body scaling variable L eff can be extended to any localized basis set by numerically computing the eigenvalues of the total squared momentum operator and equating them to the corresponding eigenvalues of a spherical box with radius L eff .
For an understanding of L eff (or N eff ), it is useful to consider analytical approximations to L eff . These are based on approximate expressions for the eigenvalues κ lm which are valid for model spaces with N glyph[greatermuch] 1. In the single-particle problem (or the deuteron in its center-of-mass frame), the ( m + 1) th eigenvalue κ 2 lm in a harmonic-oscillator model space with up to N quanta of excitation is given to good approximation [21] by
$$\kappa _ { l m } ^ { 2 } \approx \frac { a _ { l, m } ^ { 2 } } { 2 ( N _ { l } + 3 / 2 + 2 ) b ^ { 2 } } = \frac { a _ { l, m } ^ { 2 } } { ( L _ { 2 } ( N, \hbar { \Omega } ) ) ^ { 2 } }$$
and
$$N _ { l } = \begin{cases} \ N \, & \text{for $N$ and $l$ even or for $N$ and $l$ odd;} \\ \ N - 1 \, & \text{for $N$ even and $l$ odd or for $N$ odd and $l$ even.} \end{cases}$$
As a check, consider a single particle in a model space with N = 2 n + . We equate the l lowest eigenvalue κ 2 l 0 to the lowest eigenvalue ( a l, 0 /L ) 2 in a spherical box of radius L for angular momentum l . This yields L = L 2 ( N, glyph[planckover2pi1] Ω) as defined in Eq. (4).
Thus, approximate values for L eff are found by inserting the analytical approximations (10) into Eq. (8). The middle panels of Fig. 2 show the groundstate energies for 16 24 , O when plotted as a function of this approximate value for L eff . Compared to the top panel ( L = L 2 ), the improvement is considerable and clearly visible. The bottom panels of Fig. 2 show the ground-state energies for 16 24 , O when plotted as a function of the exact numerical value for L eff . Again, the data points fall close to a single line. For reasons we do not yet understand, the approximate L eff (middle panel) leads to the smoothest line in 16 O, while the exact L eff yields the smoothest line in 24 O.
Table 1 gives N eff for isotopes 12 14 , C and 16 22 24 , , O. These results are determined by diagonalizing the operator ∑ A i =1 p 2 i in the specified model space. We note that the 'square' model space (defined by each particle allowed up to N oscillator quanta) is used in the coupled-cluster calculations of this paper and by other methods [6, 7, 8]. This model space differs from the 'triangular' full N glyph[planckover2pi1] Ω model space used in most NCSM calculations.
Let us make some comments on the values of N eff in Table 1. We recall that N eff is computed from the exact eigenvalues κ 2 lm of the total squared momentum operator. The resulting values for N eff are a few percent smaller than the results one obtains from using the N glyph[greatermuch] 1 approximation (10) in the computation of N max . However, these latter approximations N approx eff can be understood semi-quantitatively as follows. For the nucleus 4 He (not shown) we have N approx eff = N ( N approx eff = N -1) for even (odd) N because only s -waves are occupied, so L 2 ( N, glyph[planckover2pi1] Ω) = L eff ( N, glyph[planckover2pi1] Ω). For 12 14 , C and 16 O, p -waves are at the Fermi surface, and for pure p -waves and even (odd) N one has N approx eff = N -1 ( N approx eff = N ), see Eq. (10). Thus, our values for N approx eff are closest to these extreme values for 16 O and somewhat farther away for 14 12 , C. For 22 24 , O, d -shell orbitals are at the Fermi surface for the kinetic energy, countering the effects from the lower-lying p waves and effectively increasing (decreasing) N approx eff for even (odd) N . As
Table 1. Effective excitation number N eff for isotopes 12 14 , C and 16 22 24 , , O, computed from the exact eigenvalues of the total squared momentum operator. Where no result is given, the model space is too small to accommodate A nucleons.
| N | 12 C | 14 C | N eff 16 O | 22 O | 24 O |
|-----|--------|--------|--------------|--------|--------|
| 1 | 0.365 | 0.393 | 0.413 | - | - |
| 2 | 0.745 | 0.709 | 0.684 | 0.899 | 0.939 |
| 3 | 2.515 | 2.545 | 2.566 | 1.939 | 1.844 |
| 4 | 2.902 | 2.867 | 2.842 | 3.092 | 3.137 |
| 5 | 4.586 | 4.617 | 4.639 | 4.116 | 4.033 |
| 6 | 4.976 | 4.941 | 4.917 | 5.187 | 5.236 |
| 7 | 6.629 | 6.66 | 6.682 | 6.205 | 6.127 |
| 8 | 7.02 | 6.985 | 6.961 | 7.245 | 7.296 |
| 9 | 8.658 | 8.689 | 8.711 | 8.259 | 8.185 |
| 10 | 9.049 | 9.015 | 8.99 | 9.285 | 9.337 |
| 11 | 10.678 | 10.709 | 10.732 | 10.297 | 10.225 |
| 12 | 11.07 | 11.036 | 11.011 | 11.313 | 11.366 |
| 13 | 12.693 | 12.725 | 12.748 | 12.324 | 12.254 |
| 14 | 13.085 | 13.051 | 13.027 | 13.335 | 13.389 |
| 15 | 14.705 | 14.737 | 14.76 | 14.345 | 14.276 |
| 16 | 15.097 | 15.064 | 15.04 | 15.352 | 15.406 |
| 17 | 16.715 | 16.747 | 16.77 | 16.361 | 16.293 |
| 18 | 17.107 | 17.074 | 17.05 | 17.366 | 17.421 |
| 19 | 18.723 | 18.755 | 18.778 | 18.375 | 18.307 |
| 20 | 19.115 | 19.082 | 19.058 | 19.377 | 19.433 |
naively expected, the effect is larger in 24 O than in 22 O. From these arguments, one expects N -1 ≤ N approx eff ≤ N in general.
We note that the defect N -N eff varies very little (focusing on either even or odd N ) as N varies. In practice, one might thus take a fixed defect (say from N = 10 or so) and consider the small variation of the defect a higher-order correction. We also note an odd-even staggering of N eff . This staggering is caused by Eq. (11) and has its root in the fact that L 2 increases for even (odd) values of angular momentum only at even (odd) values of N . In most NCSM and coupled-cluster calculations, practitioners limit themselves to sequences of model spaces with either even or odd values of N .
## 4. Extrapolations and the coupled-cluster method
In this section, we apply IR extrapolations to the results of the coupled-cluster method with singles and doubles (CCSD) calculations [3, 4, 28, 5]. Our interest is twofold. First, we want to study in more detail whether L eff is also the relevant length scale when
approximate many-body solutions (such as CCSD) are employed. We note that most ab initio methods that presently compute nuclei beyond the p shell employ approximate solutions of the many-body problem [29, 30, 7, 25, 24]. Second, we want to probe the extrapolation formula (6) over a large energy range and see how large the model space needs to be for a reliable and accurate extrapolation.
We compute the ground-state energies of the nuclei 16 22 24 , , O using the nextto-next-to-leading order chiral nucleon-nucleon interaction of Ref. [9] with a cutoff Λ χ = 500 MeV. This interaction has been optimized to scattering data of the nucleonnucleon system and deuteron bound-state properties. It is similar in quality to the chiral nucleon-nucleon interaction NNLO opt [31], which was optimized with respect to phase shifts.
Figure 3 shows the CCSD ground state energies for 16 O as a function of L eff (left figure) and glyph[planckover2pi1] Ω (right figure) for model spaces with N = 8 10 12. , , Our model spaces have oscillator frequencies in the interval glyph[planckover2pi1] Ω (MeV) / ∈ [16 , 70]. At fixed N , the energy computed at the highest oscillator frequency corresponds to the smallest value of L eff and exhibits the smallest UV error. In the left panel of Fig. 3 the solid data points forming the exponential envelope have negligible UV corrections and can be used in the IR extrapolation. As seen in the right panel, these frequencies are much larger than the naive estimate glyph[planckover2pi1] Ω min ≈ glyph[planckover2pi1] 2 Λ χ / mR ( ) that minimizes the energy for a nucleus of radius R and interaction with cutoff Λ χ [17]. The exponential extrapolation (6) to all solid data points is shown as a dashed line, and the asymptote E ∞ as a full line.
The extrapolation results are from a least-squares penalty-function fit with equal weighting [32] of the parameters E ∞ , k ∞ , and A ∞ . We note that the exponential extrapolation practically works over the entire range of about 60 MeV. While the energy correction can be a substantial fraction of the total binding energy, we have exp ( -2 k ∞ L eff ) glyph[lessmuch] 1 over the considered range of L eff . For fixed N , the (very small) UV corrections are expected to increase with increasing L eff , see Fig. 20 of Ref. [20] as an example. We note that all solid data points employed in the IR extrapolation in Fig. 3 are from model spaces with Λ UV > 750 MeV. The UV cutoff parameter of the nucleon-nucleon interaction used here is Λ χ ≈ 500 MeV, but this is not a sharp cutoff, which is why Λ UV must be sufficiently higher than Λ χ so that effects from the omitted UV tail are small.
Let us discuss errors and error estimates. We applied a theory, derived for the deuteron and k ∞ L eff glyph[greatermuch] 1 to many-body systems and for k ∞ L eff > 4 or so. We view the extrapolation formula (6) as a leading order and systematically improvable result applied to a complex nucleus over a wide range of k ∞ L eff , neglecting higher-order corrections. We also deal with systematic errors from the CCSD approximation (recall that L eff was worked out for an exact solution of the operator ∑ A i =1 p 2 i ). We believe that UV errors are negligible. Thus, we have systematic errors from neglected corrections in the IR and from the approximate many-body method. Both systematic errors are hard to quantify. Some aspects of these systematic errors behave as statistical errors. For instance, we deal with a relatively small scatter of our data points around the exponential extrapolation. These
eff
Figure 3. Ground-state energies in CCSD approximation for 16 O as a function of L eff (left panel) and glyph[planckover2pi1] Ω (right panel) for harmonic-oscillator spaces with N as indicated. Dashed line: exponential fit of Eq. (5) to data with N max = 12. Full line with band: asymptote E ∞ from fit with 95% confidence interval. Hollow markers: points excluded from fit. The arrow marks the minimum energy E glyph[planckover2pi1] Ωmin that is obtained when varying the oscillator spacing glyph[planckover2pi1] Ω for N max = 12.
(relatively small) errors are easy to quantify and usually returned by fitting routines in the form of standard asymptotic errors computed from the covariance matrix. At this moment, these are the only errors we quantify and present in tables below. However, we emphasize that these errors are presumably much smaller than the systematic errors.
We fit Eq. (6) to 16 O, including increasingly larger sets of data points from model spaces with N ≤ N max . Table 2 shows the results. The error estimates are from the covariance matrix. We repeat that any systematic errors from sub-leading IR corrections or due to the CCSD approximation are not included. We note that the E ∞ result for N max = 8 is within the error estimates for N max = 10 12. , Table 2 also shows the minimum energy E glyph[planckover2pi1] Ωmin that is obtained at fixed N max when varying the oscillator spacing glyph[planckover2pi1] Ω. For the smaller model space with N max = 8, the extrapolated energy E ∞ is much closer to the 'true' result (from extrapolations in larger model spaces) than the minimum energy E glyph[planckover2pi1] Ωmin , and this is the practical value of IR extrapolations. Of course, the challenge remains to give a meaningful error estimate for all model spaces.
The left panel of Fig. 4 shows a log plot of the difference ∆ E = E -E ∞ for 16 O. The dashed line is the N max = 12 exponential fit from Table 2. Deviations at the largest values of L eff reflect UV corrections (open symbols are not used in the fit) and other systematic errors discussed above. However, the consistency of the fit to Eq. (6) over a large range of ∆ E is striking. This suggests that there is enough information for reliable extrapolations even when using only calculations far from the energy minimum. This is validated for 16 O in the right panel of Fig. 4, where extrapolations are shown for
Table 2. Extrapolation parameters (and statistical error estimates) and energy minima for 16 O as a function of basis truncation N max . The neglected systematic errors are expected to dominate the error budget.
| N max | N max | 8 | 10 | 12 |
|------------------------------|------------|---------------------|---------------------|---------------------|
| E glyph[planckover2pi1] Ωmin | [MeV] | - 121 . 83 | - 122 . 46 | - 122 . 56 |
| E ∞ | [MeV] | - 122 . 62 ± 0 . 06 | - 122 . 68 ± 0 . 35 | - 122 . 73 ± 0 . 35 |
| k ∞ | [fm - 1 ] | 1 . 00 ± 0 . 00 | 0 . 99 ± 0 . 01 | 0 . 98 ± 0 . 01 |
| A ∞ | [10 4 MeV] | 6 . 95 ± 0 . 20 | 6 . 48 ± 0 . 49 | 5 . 96 ± 0 . 63 |
fixed N = 8, 10, and 12 using only points in each case with glyph[planckover2pi1] Ω ≥ 50 MeV. This plot can also be interpreted as showing that controlled and consistent (if not highly precise) extrapolations to glyph[planckover2pi1] Ω = 0 (i.e. removing the IR cutoff for a fixed N ) can be achieved if UV corrections are suppressed.
Figure 4. Left: energy difference ∆ E for the ground-state energy (in CCSD approximation) of 16 O as a function of L eff in a log plot. The symbols are as in Fig. 3, and the dashed line is the exponential fit for N max = 12 from Table 2. Right: ground-state energies as a function of glyph[planckover2pi1] Ω as in the right panel of Fig. 3. The solid lines are fits of Eq. (6) for fixed N to only the solid points. The arrow marks the minimum energy E glyph[planckover2pi1] Ωmin that is obtained when varying the oscillator spacing glyph[planckover2pi1] Ω for N max = 12.
Now we turn to the neutron-rich isotopes 22 24 , O. Figure 5 shows the CCSD groundstate energies for 22 O (left) and 24 O (right) as a function of L eff . The exponential extrapolation (6), and the extrapolated ground-state energies E ∞ = -148 41 . ± 0 60 MeV . (for 22 O) and E ∞ = -155 38 . ± 0 83 MeV (for . 24 O) are also shown. These results employ model spaces with N = 8 10 12. , , Table 3 shows the extrapolated energies (and error estimates from the χ 2 fit) when only data points with N ≤ N max are employed in the
eff
eff
Figure 5. (Color online) Ground-state energies (CCSD approximation) for 22 O (left) and 24 O (right) as a function of L 2 for harmonic oscillator spaces with N as indicated. Dashed line: Exponential fit to Eq. (6). Full line with band: Asymptote E ∞ from errors from covariance matrix. Hollow markers: Points excluded from fit. The arrow marks the minimum energy E glyph[planckover2pi1] Ωmin that is obtained when varying the oscillator spacing glyph[planckover2pi1] Ω for N max = 12.
fit. We note that the fits work well over an energy range of tens of MeVs (see the left panels of Figs. 6 and 7). Again, the value of the IR extrapolation lies in the finding that N max = 8 extrapolations yield results that are close to the 'true' ground-state energies (see the right panels of Figs. 6 and 7).
| | N max | 8 | 10 | 12 |
|------|----------------|---------------------|---------------------|---------------------|
| 22 O | E ∞ [MeV] | - 147 . 93 ± 0 . 01 | - 148 . 27 ± 0 . 47 | - 148 . 41 ± 0 . 60 |
| 22 O | k ∞ [fm - 1 ] | 0 . 91 ± 0 . 00 | 0 . 90 ± 0 . 01 | 0 . 89 ± 0 . 00 |
| 22 O | A ∞ [10 4 MeV] | 8 . 56 ± 0 . 02 | 7 . 76 ± 0 . 52 | 7 . 19 ± 0 . 33 |
| 24 O | E ∞ [MeV] | - 154 . 42 ± 0 . 10 | - 155 . 08 ± 0 . 63 | - 155 . 38 ± 0 . 83 |
| 24 O | k ∞ [fm - 1 ] | 0 . 83 ± 0 . 00 | 0 . 83 ± 0 . 01 | 0 . 82 ± 0 . 00 |
| 24 O | A ∞ [10 4 MeV] | 7 . 89 ± 0 . 11 | 7 . 06 ± 0 . 52 | 6 . 53 ± 0 . 31 |
Table 3. Extrapolated energies E ∞ for 22 24 , O as a function basis truncation N max. The energy minima for 22 24 , O in a N max = 12 model-space are found at the oscillator frequency glyph[planckover2pi1] Ω = 20MeV and are E glyph[planckover2pi1] Ω min = -148 85 MeV . and E glyph[planckover2pi1] Ω min = -156 35 MeV, respectively. .
Our study shows that IR extrapolation can be a practical tool for approximate solutions of the nuclear many-body problem. The extrapolation is accurate and reliable over a large energy range of tens of MeV. For the employed chiral interaction at NNLO, the examples of 16 22 24 , , O suggest that ground-state energies of p -shell and sd -shell nuclei
Figure 6. Left: energy difference ∆ E for the ground-state energy (in CCSD approximation) of 16 O as a function of L eff in a log plot. The symbols are as in Fig. 3, and the dashed line is the exponential fit for N max = 12 from Table 3. Right: ground-state energies as a function of glyph[planckover2pi1] Ω as in the right panel of Fig. 3. The solid lines are fits of Eq. (6) for fixed N to only the solid points. The arrow marks the minimum energy E glyph[planckover2pi1] Ωmin that is obtained when varying the oscillator spacing glyph[planckover2pi1] Ω for N max = 12.
eff
Figure 7. Left: energy difference ∆ E for the ground-state energy (in CCSD approximation) of 24 O as a function of L eff in a log plot. The symbols are as in Fig. 3, and the dashed line is the exponential fit for N max = 12 from Table 3. Right: ground-state energies as a function of glyph[planckover2pi1] Ω as in the right panel of Fig. 3. The solid lines are fits of Eq. (6) for fixed N to only the solid points. The arrow marks the minimum energy E glyph[planckover2pi1] Ωmin that is obtained when varying the oscillator spacing glyph[planckover2pi1] Ω for N max = 12.
can be extrapolated from model spaces with N = 8.
We also tried to include N = 6 data points in the extrapolation, but the data points did not fall onto the same line as the N = 8 10 12 points. , , We speculate that this is due to peculiarities of the CCSD approximation, or due to a less complete decoupling of the center of mass in small model spaces [17]. Again, this points to the need to better understand the systematic errors involved in IR extrapolations of results obtained with approximate many-body results.
## 5. Summary
We studied IR extrapolations for coupled-cluster computations of oxygen isotopes with chiral nucleon-nucleon interactions at NNLO. One of the main results is the identification of the nucleus-dependent infrared box size L eff . Our results show that IR extrapolations are feasible in practice, but we need a better understanding of systematic errors for extrapolated quantities; a Bayesian framework may be useful in this regard. Nevertheless, we demonstrated that reliable IR extrapolations can be performed in p s d --nuclei spanning over tens of MeVs. For the IR extrapolations to work in practice one needs to minimize UV corrections and work at frequencies away from the usual range about the oscillator frequency that minimizes the energy at fixed number N of oscillator quanta.
## Acknowledgments
This material is based upon work supported in part by the National Science Foundation under Grant No. PHY-1306250 (Ohio State University), by the U.S. Department of Energy, Office of Science, Office of Nuclear Physics, under Award Numbers DEFG02-96ER40963 (University of Tennessee), DE-SC0008499/DE-SC0008533 (SciDAC-3 NUCLEI Collaboration), the Field Work Proposal ERKBP57 at Oak Ridge National Laboratory, and under contract number DEAC05-00OR22725 (Oak Ridge National Laboratory).
## References
- [1] Navr´ atil P, Quaglioni S, Stetcu I and Barrett B R 2009 Journal of Physics G: Nuclear and Particle Physics 36 083101 URL http://stacks.iop.org/0954-3899/36/i=8/a=083101
- [2] Barrett B R, Navr´til P and Vary J P 2013 a Progress in Particle and Nuclear Physics 69 131 181 URL http://www.sciencedirect.com/science/article/pii/S0146641012001184
- [3] K¨ ummel H, L¨hrmann K H and Zabolitzky J G 1978 u Physics Reports 36 1 - 63 URL http: //www.sciencedirect.com/science/article/pii/0370157378900819
- [4] Bishop R F 1991 Theoretical Chemistry Accounts: Theory, Computation, and Modeling (Theoretica Chimica Acta) 80 (2) 95-148 10.1007/BF01119617 URL http://dx.doi.org/10. 1007/BF01119617
- [5] Hagen G, Papenbrock T, Hjorth-Jensen M and Dean D J 2013 ArXiv:1312.7872 ( Preprint 1312.7872 ) URL http://adsabs.harvard.edu/abs/2013arXiv1312.7872H
- [6] Dickhoff W and Barbieri C 2004 Progress in Particle and Nuclear Physics 52 377 - 496 URL http://www.sciencedirect.com/science/article/pii/S0146641004000535
- [7] Tsukiyama K, Bogner S K and Schwenk A 2011 Phys. Rev. Lett. 106 (22) 222502 URL http: //link.aps.org/doi/10.1103/PhysRevLett.106.222502
- [8] Hergert H, Binder S, Calci A, Langhammer J and Roth R 2013 Phys. Rev. Lett. 110 (24) 242501 URL http://link.aps.org/doi/10.1103/PhysRevLett.110.242501
- [9] Ekstr¨m A, Jansen G R, Wendt K A, Hagen G, Papenbrock T, Bacca S, Carlsson B and Gazit D o 2014 ArXiv:1406.4696 URL http://adsabs.harvard.edu/abs/2014arXiv1406.4696E
- [10] Michel N, Nazarewicz W, Pglyph[suppress] loszajczak M and Vertse T 2009 Journal of Physics G: Nuclear and Particle Physics 36 013101 URL http://stacks.iop.org/0954-3899/36/i=1/a=013101
- [11] Caprio M A, Maris P and Vary J P 2012 Phys. Rev. C 86 (3) 034312 URL http://link.aps. org/doi/10.1103/PhysRevC.86.034312
- [12] Bulgac A and Forbes M M 2013 Phys. Rev. C 87 (5) 051301 URL http://link.aps.org/doi/ 10.1103/PhysRevC.87.051301
- [13] Bogner S K, Kuo T T S and Schwenk A 2003 Physics Reports 386 1 -27 URL http: //www.sciencedirect.com/science/article/pii/S0370157303002953
- [14] Bogner S K, Furnstahl R J and Perry R J 2007 Phys. Rev. C 75 (6) 061001 URL http: //link.aps.org/doi/10.1103/PhysRevC.75.061001
- [15] Bohigas O, Tomsovic S and Ullmo D 1993 Physics Reports 223 43 - 133 ISSN 0370-1573 URL http://dx.doi.org/10.1016/0370-1573(93)90109-Q
- [16] Stetcu I, Barrett B and van Kolck U 2007 Physics Letters B 653 358 -362 URL http: //www.sciencedirect.com/science/article/pii/S0370269307009185
- [17] Hagen G, Papenbrock T, Dean D J and Hjorth-Jensen M 2010 Phys. Rev. C 82 (3) 034330 URL http://link.aps.org/doi/10.1103/PhysRevC.82.034330
- [18] Coon S A, Avetian M I, Kruse M K G, van Kolck U, Maris P and Vary J P 2012 Phys. Rev. C 86 (5) 054002 URL http://link.aps.org/doi/10.1103/PhysRevC.86.054002
- [19] Furnstahl R J, Hagen G and Papenbrock T 2012 Phys. Rev. C 86 (3) 031301 URL http: //link.aps.org/doi/10.1103/PhysRevC.86.031301
- [20] More S N, Ekstr¨m A, Furnstahl R J, Hagen G and Papenbrock T 2013 o Phys. Rev. C 87 (4) 044326 URL http://link.aps.org/doi/10.1103/PhysRevC.87.044326
- [21] Furnstahl R J, More S N and Papenbrock T 2014 Phys. Rev. C 89 (4) 044301 URL http: //link.aps.org/doi/10.1103/PhysRevC.89.044301
- [22] Jurgenson E D, Maris P, Furnstahl R J, Navr´til P, Ormand W E and Vary J P 2013 a Phys. Rev. C 87 (5) 054312 URL http://link.aps.org/doi/10.1103/PhysRevC.87.054312
- [23] Roth R, Calci A, Langhammer J and Binder S 2013 ArXiv:1311.3563 URL http://adsabs. harvard.edu/abs/2013arXiv1311.3563R
- [24] Som` a V, Barbieri C and Duguet T 2013 Phys. Rev. C 87 (1) 011303 URL http://link.aps.org/ doi/10.1103/PhysRevC.87.011303
- [25] Hergert H, Bogner S K, Binder S, Calci A, Langhammer J, Roth R and Schwenk A 2013 Phys. Rev. C 87 (3) 034307 URL http://link.aps.org/doi/10.1103/PhysRevC.87.034307
- [26] S¨ a¨f a D and Forss´ en C 2014 Phys. Rev. C 89 (1) 011303 URL http://link.aps.org/doi/10. 1103/PhysRevC.89.011303
- [27] Olver F W J, Lozier D W, Boisvert R F and Clark C W (eds) 2010 NIST Handbook of Mathematical Functions (New York, NY: Cambridge University Press)
- [28] Bartlett R J and Musiaglyph[suppress] l M 2007 Rev. Mod. Phys. 79 (1) 291-352 URL http://link.aps.org/ doi/10.1103/RevModPhys.79.291
- [29] Hagen G, Papenbrock T, Dean D J and Hjorth-Jensen M 2008 Phys. Rev. Lett. 101 (9) 092502 URL http://link.aps.org/doi/10.1103/PhysRevLett.101.092502
- [30] Barbieri C and Hjorth-Jensen M 2009 Phys. Rev. C 79 (6) 064313 URL http://link.aps.org/ doi/10.1103/PhysRevC.79.064313
- [31] Ekstr¨m A, Baardsen G, Forss´ en o C, Hagen G, Hjorth-Jensen M, Jansen G R, Machleidt R,
Nazarewicz W, Papenbrock T, Sarich J and Wild S M 2013 Phys. Rev. Lett. 110 (19) 192502 URL http://link.aps.org/doi/10.1103/PhysRevLett.110.192502
- [32] Dobaczewski J, Nazarewicz W and Reinhard P G 2014 Journal of Physics G: Nuclear and Particle Physics 41 074001 URL http://stacks.iop.org/0954-3899/41/i=7/a=074001 | 10.1088/0954-3899/42/3/034032 | [
"R. J. Furnstahl",
"G. Hagen",
"T. Papenbrock",
"K. A. Wendt"
] | 2014-08-01T18:06:43+00:00 | 2014-08-01T18:06:43+00:00 | [
"nucl-th"
] | Infrared extrapolations for atomic nuclei | Harmonic oscillator model-space truncations introduce systematic errors to
the calculation of binding energies and other observables. We identify the
relevant infrared scaling variable and give values for this nucleus-dependent
quantity. We consider isotopes of oxygen computed with the coupled-cluster
method from chiral nucleon-nucleon interactions at next-to-next-to-leading
order and show that the infrared component of the error is sufficiently
understood to permit controlled extrapolations. By employing oscillator spaces
with relatively large frequencies, well above the energy minimum, the
ultraviolet corrections can be suppressed while infrared extrapolations over
tens of MeVs are accurate for ground-state energies. However, robust
uncertainty quantification for extrapolated quantities that fully accounts for
systematic errors is not yet developed. |
1408.0253v2 | ## Prequantization of the Moduli Space of Flat PU( p )-Bundles with Prescribed Boundary Holonomies glyph[star]
Derek KREPSKI
Department of Mathematics, University of Manitoba, Canada
E-mail:
[email protected]
URL:
http://server.math.umanitoba.ca/ ~ dkrepski/
Received August 05, 2014, in final form November 28, 2014; Published online December 05, 2014 http://dx.doi.org/10.3842/SIGMA.2014.109
Abstract. Using the framework of quasi-Hamiltonian actions, we compute the obstruction to prequantization for the moduli space of flat PU( p )-bundles over a compact orientable surface with prescribed holonomies around boundary components, where p > 2 is prime.
Key words: quantization; moduli space of flat connections; parabolic bundles
2010 Mathematics Subject Classification: 53D50; 53D30
## 1 Introduction
Let G be a compact connected simple Lie group and Σ a compact oriented surface with s boundary components. Given conjugacy classes C 1 , . . . , C s , let M = M G (Σ; C 1 , . . . , C s ) denote the moduli space of flat G -bundles on Σ with prescribed boundary holonomies in the conjugacy classes C j . Alternatively, M may be described as the character variety of the fundamental group of Σ,
$$\mathfrak { M } = \text{Hom} _ { \mathcal { C } _ { 1 }, \dots, \mathcal { C } _ { s } } ( \pi _ { 1 } ( \Sigma ), G ) / G.$$
Here, Hom C 1 ,..., C s ( π 1 (Σ) , G ) consists of homomorphisms ρ : π 1 (Σ) → G whose restriction to (the homotopy class of) the j -th boundary circle of Σ lies in C j , and G acts by conjugation. Recall that M is a (possibly singular) symplectic space, where the symplectic form is defined by a choice of invariant inner product on the Lie algebra g of G [5, 12]. This paper considers the obstruction to the existence of a prequantization of M - that is, a prequantum (orbifold) line bundle L → M (see Section 5.2 for details) - by expressing the corresponding integrality condition on the symplectic form in terms of the choice of inner product on the simple Lie algebra g , which is hence a certain multiple k of the basic inner product.
If the underlying structure group G is simply connected, the moduli space M is connected and the obstruction to prequantization is well known - a prequantization exists if and only if k ∈ N and each conjugacy class C j corresponds to a level k weight (e.g., see [4, 6, 18]). If G is not simply connected, M may have multiple components. Moreover integrality of k is not sufficient to guarantee a prequantization even in the absence of markings/prescribed boundary holonomies: if Σ is closed and has genus at least 1, then k must be a multiple of an integer l 0 ( G ) (computed in [15] for each G ). If Σ has boundary with prescribed holonomies, only the case G = SO(3) = PU(2) has been fully resolved [17]. ∼
glyph[star] This paper is a contribution to the Special Issue on Poisson Geometry in Mathematics and Physics. The full collection is available at http://www.emis.de/journals/SIGMA/Poisson2014.html
In this paper, we describe the connected components of M for non-simply connected structure groups G/Z in Corollary 4.2 and Proposition 4.3 (where G is simply connected and Z is a subgroup of the centre of G ). The decomposition into components makes use of an action of the centre Z G ( ) on a fundamental Weyl alcove ∆ in t , the Lie algebra of a maximal torus. The action is described concretely in [23] for classical groups and Appendix A records the action for the two remaining exceptional cases.
Finally, we compute the obstruction to prequantization in Theorem 5.8 in the case G = PU( ) p ( p > 2, prime) for any number of boundary components s . We work within the theory of quasiHamiltonian group actions with group-valued moment map [2], where the moduli space M is a central example. In quasi-Hamiltonian geometry, quantization is defined as a certain element of the twisted K -theory of G [19], analogous to Spin c quantization for Hamiltonian group actions on symplectic manifolds. In this context, the obstruction to the existence of a prequantization is a cohomological obstruction (see Definition 5.1). The obstruction for other cases of non-simply connected structure group does not follow from the approach here (see Remark 5.7) and will be considered elsewhere.
## 2 Preliminaries
Notation. Unless otherwise indicated, G denotes a compact, simply connected, simple Lie group with Lie algebra g . We fix a maximal torus T ⊂ G and use the following notation:
- t - Lie algebra of T ; t ∗ - dual of the Lie algebra of T ; W = N T /T ( ) - Weyl group; I = ker exp T - integer lattice; P = I ∗ ⊂ t ∗ - (real) weight lattice; Q ⊂ t ∗ - root lattice; Q ∨ ⊂ t - coroot lattice;
- P ∨ ⊂ t - coweight lattice.
Recall that since G is simply connected, I = Q ∨ . Moreover, the coroot lattice and weight lattice are dual to each other, as are the root lattice and coweight lattice. Achoice of simple roots α , . . . , α 1 l (with l = rank( G )) spanning Q , determines the fundamental coweights λ , . . . , λ ∨ 1 ∨ l spanning P ∨ , defined by 〈 α , λ i ∨ j 〉 = δ i,j .
We let 〈- -〉 , denote the basic inner product, the invariant inner product on g normalized to make short coroots have length √ 2. With this inner product, we will often identify t ∼ = t ∗ .
Given a subgroup Z of the centre Z G ( ) of G , we shall abuse notation and denote by q : G → G/Z the resulting covering(s).
Finally, let { e , . . . , e 1 n } denote the standard basis for R n , equipped with the standard inner product that will also be denoted with angled brackets 〈- -〉 , .
Quasi-Hamiltonian group actions. We recall some basic definitions and facts from [2]. (For the remainder of this section, we may take G to be any compact Lie group with invariant inner product 〈- -〉 , on g .) Let θ L , θ R denote the left-invariant, right-invariant Maurer-Cartan forms on G , and let η = 1 12 〈 θ L , [ θ L , θ L ] 〉 denote the Cartan 3-form on G . For a G -manifold M , and ξ ∈ g , let ξ glyph[sharp] denote the generating vector field of the action. The Lie group G is itself viewed as a G -manifold for the conjugation action.
Definition 2.1 ([2]) . A quasi-Hamiltonian G -space is a triple ( M,ω, Φ) consisting of a G -manifold M , a G -invariant 2-form ω on M , and an equivariant map Φ: M → G , called the moment map, satisfying:
- i) dω +Φ ∗ η = 0,
- ii) ι ξ glyph[sharp] ω + 1 2 Φ (( ∗ θ L + θ R ) · ξ ) = 0 for all ξ ∈ g ,
- iii) at every point x ∈ M , ker ω x ∩ ker dΦ x = { } 0 .
Wewill often denote a quasi-Hamiltonian G -space ( M,ω, Φ) simply by the underlying space M when ω and Φ are understood from the context.
The fusion product of two quasi-Hamiltonian G -spaces M j with moment maps Φ j : M j → G ( j = 1 2) is the product , M 1 × M 2 , with the diagonal G -action and moment map Φ : M 1 × M 2 → G given by composing Φ 1 × Φ 2 with multiplication in G .
The symplectic quotient of a quasi-Hamiltonian G -space is the symplectic space M//G = Φ -1 (1) /G , which is a symplectic orbifold whenever the group unit 1 ∈ G is a regular value. If 1 is a singular value, then the symplectic quotient is a singular symplectic space as defined in [20].
The conjugacy classes C ⊂ G , with moment map the inclusion into G , are basic examples of quasi-Hamiltonian G -spaces. Another important example is the double D G ( ) = G × G , equipped with diagonal G -action and moment map Φ( g, h ) = ghg -1 h -1 , the group commutator. These two families of examples form the building blocks of the moduli space of flat G -bundles over a surface Σ with prescribed boundary holonomies. (See Section 4 for a sketch of this construction.)
## 3 Conjugacy classes invariant under translation by central elements
This section describes the set of conjugacy classes D ⊂ G that are invariant under translation by a subgroup Z D of the centre Z G ( ) of G . We begin with the following Lemma, which identifies such a subgroup Z D with the fundamental group of a conjugacy class in G/Z , where Z ⊂ Z G ( ).
Lemma 3.1. Let Z be a subgroup of the centre Z G ( ) of G and let C ⊂ G/Z be a conjugacy class. For any conjugacy class D ⊂ G covering C , the restriction q | D : D → C is the universal covering projection and hence the fundamental group π 1 ( C ) = ∼ Z D = { z ∈ Z : z D = D} .
Proof. The inverse image q -1 ( C ) is a disjoint union of conjugacy classes in G that cover C . Since conjugacy classes in a compact simply connected Lie group are simply connected and Z D acts freely on D , the lemma follows. glyph[squaresolid]
Recall that every element in G is conjugate to a unique element exp ξ in T , where ξ lies in a fixed (closed) alcove ∆ ⊂ t of a Weyl chamber. Therefore, the set of conjugacy classes in G is parametrized by ∆. Since the Z G ( )-action commutes with the conjugation action, we obtain an action Z G ( ) × ∆ → ∆. Next we identify this description of the action of Z G ( ) on an alcove ∆ with a more concrete description of a Z G ( )-action on ∆ given in [23, Section 4.1]. (See also [7, Section 3.1] for a similar treatment.)
Let { α , . . . , α 1 l } be a basis of simple roots for t ∗ , with highest root ˜ =: α -α 0 . Let ∆ ⊂ t be the alcove
$$\Delta = \{ \xi \in \mathfrak { t } \colon \langle \xi, \alpha _ { j } \rangle \geq 0, \, \langle \xi, \tilde { \alpha } \rangle \leq 1 \}.$$
Recall that the exponential map induces an isomorphism Z G ( ) = ∼ P ∨ /Q ∨ , and that the nonzero elements of the centre have representatives λ ∨ i ∈ P ∨ given by minimal dominant coweights. By [23, Lemma 2.3] the non-zero minimal dominant coweights λ ∨ i are dual to the special roots α i , which are those roots with coefficient 1 in the expression ˜ = α ∑ m α i i . In Proposition 4.1.4 of [23], Toledano-Laredo provides a Z G ( )-action on ∆ defined by
$$z \cdot \xi = w _ { i } \xi + \lambda _ { i } ^ { \vee },$$
where z = exp λ ∨ i , and w i ∈ W is a certain element of the Weyl group. The element w i ∈ W is the unique element that leaves ∆ ∪ { α 0 } invariant (i.e., induces an automorphism of the extended Dynkin diagram) and satisfies w α i ( 0 ) = α i (see [23, Proposition 4.1.2]). The following proposition shows these actions coincide.
Proposition 3.2. The translation action of Z G ( ) on G induces an action Z G ( ) × ∆ → ∆ and is given by the formula z · ξ = w ξ i + λ ∨ i , where z = exp λ ∨ i and w i is the unique element in W that leaves ∆ ∪ { α 0 } invariant and satisfies w α i ( 0 ) = α i .
Proof. Observe that for any element w in W wλ , ∨ i -λ ∨ i ∈ I = Q ∨ since w exp λ ∨ i = exp λ ∨ i . Therefore, w ξ i + λ ∨ i = w ξ i ( + λ ∨ i +( w -1 i λ ∨ i -λ ∨ i )). In other words, w ξ i + λ ∨ i = ˆ( w ξ + λ ∨ i ) for some ˆ in the affine Weyl group. w Letting z = exp λ ∨ i , this shows that z exp ξ = exp( ξ + λ ∨ i ) is conjugate to exp( w ξ i + λ ∨ i ), which proves the proposition. glyph[squaresolid]
In fact, as the next proposition shows, the automorphism of the Dynkin diagram induced by w i encodes the resulting permutation of the vertices of the alcove ∆.
Proposition 3.3. Let v , . . . , v 0 l denote the vertices of ∆ with v j opposite the facet parallel to ker α j . Then exp λ ∨ i · v j = v k whenever w α i j = α k , where w i is as in Proposition 3.2 .
glyph[negationslash]
Proof. Let v , . . . , v 0 l denote the vertices of ∆, where the vertex v j is opposite the facet (codimension 1 face) parallel to ker α j . That is, v 0 = 0 and for j = 0, v j satisfies:
glyph[negationslash]
$$\langle \alpha _ { 0 }, v _ { j } \rangle = - 1 \quad \text{and} \quad \langle \alpha _ { r }, v _ { j } \rangle = 0 \ \text{ if and only if } \ 0 \neq r \neq j.$$
glyph[negationslash]
glyph[negationslash]
(Hence, for j = 0 we have 〈 α , v j j 〉 = 1 m j , where m j is the coefficient of α j in the expression ˜ = α ∑ m α i i .)
Suppose that w α i 0 = α i and let w α i j = α k (where k depends on j ).
glyph[negationslash]
Consider exp λ ∨ i · v 0 . Since 〈 α , w v 0 i 0 + λ ∨ i 〉 = 〈 α , λ 0 ∨ i 〉 = -1, and (for r = 0) 〈 α , w v r i 0 + λ ∨ i 〉 = 〈 α , λ r ∨ i 〉 = δ r,k , we have exp λ ∨ i · v 0 = v i .
glyph[negationslash]
glyph[negationslash]
Next, consider exp λ ∨ i · v j = w v i j + λ ∨ i , where j = 0. If k = 0 so that w α i j = α 0 then α j = w -1 i α 0 is a special root (i.e. m j = 1) since w -1 i = w j . Therefore, 〈 α , w v 0 i j + λ ∨ i 〉 = 〈 w -1 i α , v 0 j 〉 -1 = 〈 α , v j j 〉 -1 = 0. And if r = 0,
$$\langle \alpha _ { r }, w _ { i } v _ { j } + \lambda _ { i } ^ { \vee } \rangle = \langle w _ { i } ^ { - 1 } \alpha _ { r }, v _ { j } \rangle + \langle \alpha _ { r }, \lambda _ { i } ^ { \vee } \rangle.$$
$$\iota + \lambda _ { i } \, \rangle = \langle w _ {$$
glyph[negationslash]
If r = , i w -1 i α r is a simple root other than α j ; therefore, each term above is 0. Moreover, if r = , then the above expression becomes i 〈 α , v 0 j 〉 + 〈 α , λ i ∨ i 〉 = -1 + 1 = 0. Hence we have exp λ ∨ i · v j = v 0 whenever w α i j = α 0 .
glyph[negationslash]
glyph[negationslash]
glyph[negationslash]
glyph[negationslash]
On the other hand, if k = 0 so that w α i j = α k is a simple root, then 〈 α , w v 0 i j + λ ∨ i 〉 = 〈 w -1 i α , v 0 j 〉 + 〈 α , λ 0 ∨ i 〉 = 0 -1 = -1 since the simple root w -1 i α 0 = α j . And if r = 0, we consider again the expression (3.1) and find (for the same reason as above) that (3.1) is trivial whenever r = k . If r = k , (3.1) becomes 〈 α , w v k i j + λ ∨ i 〉 = 〈 w -1 i α , v k j 〉 + 〈 α , λ k ∨ i 〉 = 〈 α , v j j 〉 glyph[negationslash] = 0. Hence we have that exp λ ∨ i · v j = v k , as required. glyph[squaresolid]
The Z G ( )-action on ∆ is explicitly described in [23] for all classical groups. (In Appendix A, we record the action of the centre on the alcove for the exceptional groups E 6 and E 7 , the remaining compact simple Lie groups with non-trivial centre.)
glyph[negationslash]
Conjugacy classes in SU( n ). We now specialize to the case G = SU( n ) and consider the action of the centre on the alcove. Identify t ∼ = t ∗ ⊂ R n as the subspace { x = ∑ x e j j : ∑ x j = 0 } and recall that the basic inner product coincides with (the restriction of) the standard inner product on R n . The roots are the vectors e i -e j with i = j . Taking the simple roots to be α i = e i -e i +1 ( i = 1 , . . . , n -1) and the resulting highest root ˜ = α e 1 -e n gives the alcove
$$\Delta = \{ x \in \mathfrak { t } \colon x _ { 1$$
Its vertices are
$$v _ { 0 } = 0 \quad \text{and} \quad v _ { j } =$$
The centre Z (SU( n )) ∼ = Z /n Z is generated by (exp of) the minimal dominant coweight λ ∨ 1 = e 1 -1 n n ∑ i =1 e i corresponding to the special root α 1 = e 1 -e 2 . Since the element w 1
inducing an automorphism of the extended Dynkin diagram for SU( n ) satisfies w α 1 0 = α 1 , by Proposition 3.3 the permutation of the vertices of ∆ induced by the action of exp λ ∨ 1 is the n -cycle ( v v 0 1 · · · v n -1 ) (since v j is the vertex opposite the facet parallel to ker α j ).
It follows that the only point in ∆ fixed by the action of Z G ( ) is the barycenter
$$\zeta _ { * } = \frac { 1 } { n } \sum _ { j =$$
Hence there is a unique conjugacy class in SU( n ) that is invariant under translation by the centre - namely, matrices in SU( n ) with eigenvalues z , . . . , z 1 n , the distinct n -th roots of ( -1) n +1 . As the next proposition shows, however, restricting the action to a proper subgroup Z ∼ = Z /ν Z ( ν n | ) of the centre results in larger Z -fixed point sets in ∆.
Proposition 3.4. Let n = νm and consider the subgroup Z /ν Z ⊂ Z /n Z ∼ = Z (SU( n )) . The Z /ν Z -fixed points in the alcove ∆ for SU( n ) consist of the convex hull of the barycenters of the faces spanned by the orbits of the vertices v , . . . , v 0 m -1 of ∆ .
Proof. Write x = ∑ t v i i in ∆ in barycentric coordinates (with t i ≥ 0 and ∑ t i = 1). Then a generator of Z /ν Z sends x to ∑ t v ′ i i , with t ′ i = t i -m mod n . Therefore x is fixed if and only if t i = t i -m mod n , and in this case we may write,
$$\begin{array} {$$
$$\dots$$
which exhibits a fixed point in the desired form.
glyph[squaresolid]
To illustrate, consider the subgroup Z ∼ = Z / 2 Z of the centre Z (SU(4)) = ∼ Z / 4 Z , which acts by transposing the vertices v 0 ↔ v 2 and v 1 ↔ v 3 . The barycenters ζ 0 , ζ 1 of the edges v v 0 2 and v v 1 3 , respectively, are fixed and thus the Z -fixed points are those on the line segment joining ζ 0 and ζ 1 (see Fig. 1).
Figure 1. Alcove for SU(4). The indicated line segment through the barycenter parametrizes the set of conjugacy classes invariant under translation by Z / 2 Z ⊂ Z (SU(4)).
## 4 Components of the moduli space with markings
In this section we recall the quasi-Hamiltonian description of the moduli space of flat bundles over a compact orientable surface with prescribed boundary holonomies. We refer to the original article [2] for the details regarding the construction sketched below.
Let Σ be a compact, oriented surface of genus h with s boundary components. For conjugacy classes C 1 , . . . , C s in G/Z , let M G/Z (Σ; C 1 , . . . , C s ) be the moduli space of flat G/Z -bundles over Σ with prescribed boundary holonomies lying in the conjugacy classes C j ( j = 1 , . . . , s ). Points in M G/Z (Σ; C 1 , . . . , C s ) are (gauge equivalence classes of) principal G/Z -bundles over Σ equipped with a flat connection whose holonomy around the j -th boundary component lies in the conjugacy class C j . This moduli space is an important example in the theory of quasiHamiltonian group actions, where it is cast a symplectic quotient of a fusion product,
$$M _ { G / Z } ( \Sigma ; \mathcal { C } _ { 1 }, \dots, \mathcal { C } _ { s } ) = ( D ( G / Z ) ^ { h } \times \mathcal { C } _ { 1 } \times \cdots \times \mathcal { C } _ { s } ) / / ( G / Z ),$$
which may have several connected components if Z is non-trivial. Extending the discussion in [17, Section 2.3], we describe the connected components of (4.1) as symplectic quotients of an auxiliary quasi-Hamiltonian G -space.
As in [17, Section 2.2], given a quasi-Hamiltonian G/Z -space N with group-valued moment map Φ : N → G/Z , let ˇ N be the fibre product defined by the Cartesian square,
/
/
GLYPH<15>
GLYPH<15>
GLYPH<15>
GLYPH<15>
/
$$\stackrel { \check { \Phi } } { \longrightarrow } G \\ \stackrel { \cdot } { \longrightarrow } \stackrel { \cdot } { \longrightarrow } G / Z$$
/
Then ˇ N is naturally a quasi-Hamiltonian G -space with moment map ˇ . The following proposiΦ tion from [17] and its Corollary summarize some properties of this construction.
Proposition 4.1 ([17, Proposition 2.2]) . Let ˇ N be the fibre product defined by (4.2) , where Φ : N → G/Z is a group-valued moment map.
- i ) We have a canonical identification of symplectic quotients ˇ N//G ∼ = N// G/Z ( ) .
- ii ) For a fusion product N = N 1 ×··· × N r of quasi-Hamiltonian G/Z -spaces, the space ˇ N is a quotient of ˇ N 1 ×··· × ˇ N r by the group { ( c 1 , . . . , c r ) ∈ Z r | r ∏ j =1 c j = e } .
- iii ) If Φ: N → G/Z lifts to a moment map Φ : ′ N → G , thus turning N into a quasiHamiltonian G -space then ˇ = N N × Z .
Corollary 4.2. Let ˇ N be the fibre product defined by (4.2) , where Φ : N → G/Z is a groupvalued moment map, and write ˇ N = ⊔ X j as a union of its connected components. Then the components of N// G/Z ( ) can be identified with the symplectic quotients X //G j .
Proof. The restrictions Φ ˇ j = ˇ Φ | X j are G -valued moment maps whose fibres are connected by [2, Theorem 7.2]. Since Φ ˇ -1 ( e ) = ⊔ ˇ Φ -1 j ( e ), it follows that ˇ N//G = ˇ Φ -1 ( e /G ) = ⊔ ˇ Φ -1 j ( e /G ) = ⊔ X //G j . The result follows from Proposition 4.1(i). glyph[squaresolid]
Hence to identify the components of (4.1), it suffices to identify the components of ˇ N//G , where N = D G/Z ( ) h ×C ×··· × C 1 s - namely, X //G j , where X j ranges over the components of ˇ . N In particular, we may view the moduli space (4.1) as a union of symplectic quotients of quasi-Hamiltonian G -spaces (as opposed to G/Z -spaces), which will be very important for the approach taken in Section 5.
With this in mind, choose conjugacy classes D ⊂ j G covering C j ( j = 1 , . . . , s ) and let
$$\widetilde { N } = D ( G ) ^ { h } \times \mathcal { D } _ { 1 } \times \cdots \times \mathcal { D } _ { s }.$$
Let
$$\Gamma = \left \{ ( \gamma _ { 1 }, \dots, \gamma _ { s } ) \in Z _ { \mathcal { D } _ { 1 } } \times \dots \times Z _ { \mathcal { D } _ { s } } \colon \prod \gamma _ { j } = 1 \right \} \subset Z ^ { s }$$
(cf. Lemma 3.1). We show next that the components of ˇ are all homeomorphic to N ˜ N/ Z ( 2 h × Γ) (generalizing the decomposition appearing in [17, Lemma 2.3] for G = SO(3)).
Proposition 4.3. Let N = D G/Z ( ) h ×C ×···×C 1 s for conjugacy classes C j ⊂ G/Z ( j = 1 , . . . , s ) and let ˇ N be the fibre product defined by (4.2) . Then ˇ N may be written as a union of its connected components,
$$\check { N } \cong \bigsqcup _ { Z / ( Z _ { \mathcal { D } _ { 1 } } \dots Z _ { \mathcal { D } _ { s } } ) } D ( G / Z ) ^ { h } \times ( \mathcal { D } _ { 1 } \times \dots \times \mathcal { D } _ { s } ) / \Gamma,$$
where D ⊂ j G are conjugacy classes covering C j ( j = 1 , . . . , s ) and Γ is as in (4.3) .
Proof. This is a straightforward application of the properties (ii) and (iii) listed in Proposition 4.1. By property (iii), ˇ ( D G/Z ) h = D G/Z ( ) h × Z , and by Lemma 3.1, ˇ C j = D × j Z/Z D j . Therefore, by property (ii),
$$\check { N } \cong D ( G / Z ) ^ { h } \times ( Z \times \mathcal { D } _ { 1 } \times Z / Z _ { \mathcal { D } _ { 1 } } \times \cdots \times \mathcal { D } _ { s } \times Z / Z _ { \mathcal { D } _ { s } } ) / \Lambda,$$
$$\text{where } \Lambda = \{ ( c _ { 0 }, \dots, c _ { s } ) \in Z ^ { s + 1 } \colon c _ { 0 } \cdots c _ { s } = 1 \}. \text{ Since }$$
$$( Z \times \mathcal { D } _ { 1 } \times Z / Z _ { \mathcal { D } _ { 1 } } \times \cdots \times \mathcal { D } _ { s } \times Z / Z _ { \mathcal { D } _ { s } } ) / \Lambda \cong ( Z \times \mathcal { D } _ { 1 } \times \cdots \times \mathcal { D } _ { s } ) / \Gamma ^ { \prime },$$
where Γ ′ = { ( γ , . . . , γ 0 s ) ∈ Z × Z D 1 ×··· × Z D s : ∏ γ j = 1 , we see that the components of } ˇ N are in bijection with Z/ Z ( D 1 · · · Z D s ).
↦
Consider the component corresponding to ¯ z ∈ Z/ Z ( D 1 · · · Z D s ) in which each point is of the form ( glyph[vector] g, [( z, x 1 , . . . , x s )] Γ ′ ), where [ ] Γ ′ denotes a Γ -orbit. ′ (Note that there is always a representative of this form with z in the first coordinate.) This component is homeomorphic to D G/Z ( ) h × D ×··· × D ( 1 s ) / Γ by the map ( glyph[vector] g, [( x , . . . , x 1 s )] Γ ) → ( glyph[vector] g, [( z, x 1 , . . . , x s )] Γ ′ ). glyph[squaresolid]
Remark 4.4. The decomposition in Proposition 4.3 is consistent with Theorem 14 in Ho-Liu's work [13] on the connected components of the moduli space for any compact connected Lie group G .
For the case G/Z = SU( ) ( p / Z /p Z ) = PU( ), where p p is prime, the decomposition above simplifies. In particular, there is only one conjugacy class D ∗ = SU( ) p · exp ζ ∗ , corresponding to the barycenter ζ ∗ ∈ ∆, invariant under the action of the centre. Let C ∗ = q ( D ∗ ) be the corresponding conjugacy class in PU( p ). Therefore, we obtain the following Corollary (cf. [17, Lemma 2.3]).
Corollary 4.5. Let p be prime and let N = D (PU( )) p h ×C × ··· × C 1 s for conjugacy classes C j ⊂ PU( ) ( p j = 1 , . . . , s ) and let ˇ N be the fibre product defined by (4.2) . Then,
$$\check { N } \cong \begin{cases} D ( \mathrm P U ( p ) ) ^ { h } \times ( \mathcal { D } _ { 1 } \times \cdots \times \mathcal { D } _ { s } ) / \Gamma & \text{if} \ \exists \, j \colon \mathcal { C } _ { j } = \mathcal { C } _ { * }, \\ D ( \mathrm P U ( p ) ) ^ { h } \times \mathcal { D } _ { 1 } \times \cdots \times \mathcal { D } _ { s } \times Z & \text{otherwise}, \end{cases}$$
where D ⊂ j SU( ) p are conjugacy classes covering C j ( j = 1 , . . . , s ) and Γ is as in (4.3) .
In particular, if (after re-labelling) C j = C ∗ for all j ≤ r ( r > 0), then we obtain
$$\check { N } & \cong D ( \mathbb { P U } ( p ) ) ^ { h } \times ( \mathcal { D } _ { * } ) ^ { r } / \Gamma \times \mathcal { D } _ { r + 1 } \times \cdots \times \mathcal { D } _ { s },$$
where, in this case, Γ = { ( γ , . . . , γ 1 r ) ∈ Z r : ∏ γ j = 1 . }
## 5 Obstruction to prequantization
## 5.1 Prequantization for quasi-Hamiltonian group actions
We recall some definitions and properties regarding prequantization of quasi-Hamiltonian group actions. Recall that the Cartan 3-form η ∈ Ω ( 3 G ) is integral - in fact, [ η ] ∈ H 3 ( G ; R ) is the image of a generator x ∈ H 3 ( G ; Z ) = ∼ Z under the coefficient homomorphism induced by Z → R . Condition (i) in Definition 2.1 says that the pair ( ω, η ) defines a relative cocycle in Ω 3 (Φ), the algebraic mapping cone of the pull-back map Φ ∗ : Ω ( ∗ G ) → Ω ( ∗ M ), and hence a cohomology class [( ω, η )] ∈ H 3 (Φ; R ). (See [8, Chapter I, Section 6] for the definition of relative cohomology.)
Definition 5.1 ([15, 19]) . Let k ∈ N . A level k prequantization of a quasi-Hamiltonian G -space ( M,ω, Φ) is an integral lift α ∈ H 3 (Φ; Z ) of the class k [( ω, η )] ∈ H 3 (Φ; R ).
Remark 5.2. The definition of prequantization in Definition 5.1 uses the assumption in this paper that G is simply connected. The general definition of prequantization [19, Definition 3.2] (with G semi-simple and compact) requires an integral lift in H 3 G (Φ; Z ) of an equivariant extension of the class k [( ω, η )]. When G is simply connected, [15, Proposition 3.5] shows that the definition above is equivalent. Our main goal is to apply the quasi-Hamiltonian viewpoint on prequantization to the moduli space of flat bundles with prescribed holonomies; therefore, by Corollary 4.2 it suffices to work on each component of the quasi-Hamiltonian G -space ˇ N in Proposition 4.3 using Definition 5.1.
We list some basic properties of level k prequantizations that we shall encounter.
- (a) If M 1 and M 2 are pre-quantized quasi-Hamiltonian G -spaces at level k , then their fusion product M 1 × M 2 inherits a prequantization at level k . Conversely, a prequantization of the product induces prequantizations of the factors. See [15, Proposition 3.8].
- (b) Alevel k prequantization of M induces a prequantization of the symplectic quotient M//G , equipped with the k -th multiple of the symplectic form.
- (c) The long exact sequence in relative cohomology gives a necessary condition k Φ ( ∗ x ) = 0 for the existence of a level k -prequantization. If H 2 ( M ; R ) = 0, k Φ ( ∗ x ) = 0 is also sufficient [15, Proposition 4.2] to conclude a level k -prequantization exists.
The following examples relate to the moduli space of flat bundles with prescribed boundary holonomies.
Example 5.3. The double D G ( ) = G × G with moment map Φ : D G ( ) → G equal to the group commutator admits a prequantization at all levels k ∈ N . For non-simply connected groups, the double D G/Z ( ) with moment map Φ : D G/Z ( ) → G the canonical lift of the group commutator admits a level k -prequantization if and only if k is a multiple of l 0 ∈ N , where l 0 is a positive integer depending on G/Z computed for all compact simple Lie groups in [15]. For G/Z = PU( n ), l 0 = n .
Example 5.4. Conjugacy classes D ⊂ G admitting a level k -prequantization are those D = G · exp ξ ( ξ ∈ ∆) with ( kξ ) glyph[flat] ∈ P [18], where ( kξ ) glyph[flat] = 〈 kξ, -〉 (i.e., a level k weight ). For simply laced groups (such as G = SU( n )), under the identification t ∼ = t ∗ , P ∨ ∼ = P . Therefore, in this case, D admits a level k -prequantization if and only if kξ ∈ P ∨ . Since exp -1 Z G ( ) = P ∨ , we see that D admits a level k -prequantization if and only if g k ∈ Z G ( ) for all g ∈ D . (So in particular if k is a multiple of the order of D [6, Definition 5.76], then D admits a level k prequantization.)
## 5.2 Quasi-Hamiltonian prequantization and symplectic quotients
To provide some context, we further elaborate on property (b) following Definition 5.1 since we view the moduli space of flat bundles as a symplectic quotient (4.1) in quasi-Hamiltonian geometry. By [15, Proposition 3.6], a level k prequantization of a quasi-Hamiltonian G -space ( M,ω, Φ) gives an integral lift of the equivariant cohomology class [ j ∗ ω ] ∈ H 2 G (Φ -1 (1) ; R ), where j : Φ -1 (1) → M denotes inclusion. Hence, there is a G -equivariant line bundle L → Φ -1 (1) with connection of curvature j ∗ ω . If 1 is a regular value, the symplectic quotient M//G = Φ -1 (1) /G is a symplectic orbifold [2] and the G -equivariant line bundle over the level set descends to a prequantum orbifold line bundle L/G → M//G . (See [1, Example 2.29] for a discussion of orbifold vector bundles in this context.)
Remark 5.5. The orbifold line bundle L/G → M//G need not be an ordinary line bundle over the underlying topological space M//G (i.e., the coarse moduli space of the orbifold). (For orbifolds that arise as quotients X/G of a smooth, proper, locally free action of a Lie group G on a smooth manifold X , this distinction is apparent from the observation that H 2 G ( X ; Z ) is not necessarily isomorphic to H 2 ( X/G ; Z ).) Some works in the literature require a prequantization to be an ordinary line bundle, and hence obtain a further obstruction to the existence of a prequantization (e.g. [14, Theorem 4.2], [10, Theorems 4.12 and 6.1], [21, Lemme 3.2]). We gratefully acknowledge the referee's comments that led to this important clarification.
## 5.3 The obstruction to prequantization for the moduli space of PU( p ) bundles, p prime
Let p be an odd prime. In this section we obtain the obstruction to prequantization for the quasi-Hamiltonian SU( p )-space ˇ , N where N = D (PU( )) p h ×C ×··· × C 1 s for conjugacy classes C j ⊂ PU( ) ( p j = 1 , . . . , s ). Let M ⊆ ˇ N be a connected component (by Corollary 4.5),
$$M = D ( \mathbf P U ( p ) ) ^ { h } \times \left ( \mathcal { D } _ { 1 } \times \dots \times \mathcal { D } _ { s } \right ) / \Gamma,$$
where Γ is as in (4.3). As we shall see in the proof of Theorem 5.8, we will find property (a) in Section 5.1 very useful in order to proceed 'factor by factor', using the decomposition (4.4).
To begin, we establish the following proposition which allows us to use property (c) in Section 5.1 to compute the obstruction to prequantization for the factor ( D ∗ ) r / Γ in (4.4).
Proposition 5.6. Let D ⊂ ∗ SU( ) p denote the conjugacy class of the barycenter ζ ∗ of the alcove ∆ and let Γ = { ( γ , . . . , γ 1 r ) ∈ Z r : ∏ γ j = 1 } with r > 1 . Then H 2 (( D ∗ ) r / Γ; R ) = 0 .
Proof. Since ( D ∗ ) r → D ( ∗ ) / Γ is a covering projection, H 2 (( D ∗ ) r / Γ; R ) = ∼ H 2 (( D ∗ ) r ; R ) Γ . By the K¨nneth Theorem, u H 2 (( D ∗ ) r ; R ) = ∼ ⊕ H 2 ( D ∗ , R ). Since the Γ-action factors through Z m , H 2 (( D ∗ ) r ; R ) Γ = ⊕ H 2 ( D ∗ ; R ) Z .
Recall that since ζ ∗ lies in the interior of the alcove, the centralizer SU( p ) exp ζ ∗ = T and hence D ∗ ∼ = SU( ) p /T . Moreover, we have H ∗ ( D ∗ ; R ) ∼ = R [ t 1 , . . . , t p ] / σ , . . . , σ ( 1 p ), where σ i 's are the elementary symmetric polynomials. In particular, we may write
$$H ^ { 2 } ( \mathcal { D } _ { * } ; \mathbb { R } ) \cong ( \mathbb { R } t _ { 1 } \oplus \cdots \oplus \mathbb { R } t _ { p } ) / ( t _ { 1 } + \cdots + t _ { p } = 0 ).$$
The Z -action on D ∗ corresponds to an action on SU( p /T ) by a cyclic subgroup of the Weyl group (e.g., see the proof of Proposition 3.2). Since the Weyl group (i.e., symmetric group Σ ) p acts by permuting the t i , Z acts by a p -cycle on the t i . Therefore, H 2 ( D ∗ ; R ) Z = 0, which establishes the result. glyph[squaresolid]
Remark 5.7. The analogue of Proposition 5.6 for the factors ( D ×···×D 1 s ) / Γ that appear in the decomposition in Proposition 4.3 need not hold when considering other non-simply connected structure groups G/Z .
Theorem 5.8. The quasi-Hamiltonian SU( ) p -space M = D (PU( )) p h × D ×···×D ( 1 s ) / Γ admits a level k -prequantization if and only if the following conditions are satisfied:
- i ) if h ≥ 1 , then k ∈ p N ;
- ii ) g k ∈ Z (SU( )) p for every g ∈ D 1 ∪ · · · ∪ D s .
Moreover, if in addition the identity matrix 1 ∈ SU( ) p is a regular value of the restriction of the group-valued moment map ˇ Φ : M → SU( ) p , the prequantization descends to a prequantization of the corresponding component of the moduli space M = M PU( ) p (Σ; C 1 , . . . , C s ) , where C j = q ( D j ) ⊂ PU( ) p .
Proof. By property (a) in Section 5.1, M admits a level k -prequantization if and only if each factor does. Since D (PU( )) admits a level p k -prequantization if and only if condition (i) is satisfied (see Example 5.3), we may assume from now on h = 0.
We first verify the necessity of condition (ii). A prequantization of M = ( D ×··· × D 1 s ) / Γ induces a prequantization of its universal cover ˜ M = D × ··· × D 1 s , and hence each D j must admit a prequantization, which is equivalent to condition (ii).
Next we verify that condition (ii) is sufficient for a level k -prequantization of M (with h = 0). As in the decomposition (4.4), write (possibly after re-labelling)
$$M = \underbrace { ( \mathcal { D } _ { * } \times \cdots \times \mathcal { D } _ { * } ) } _ { r \text{ factors} } ) / \Gamma \times \mathcal { D } _ { r + 1 } \times \cdots \times \mathcal { D } _ { s }$$
Using property (a) in Section 5.1 again, it suffices to consider the case 1 < r = s . (Note that if s = r = 1, Γ is trivial.) In this case, condition (ii) is simply that D ∗ admit a level k prequantization. Since D ∗ consists of matrices in SU( p ) conjugate to
$$\exp \zeta _ { * } = \text{diag} ( \exp \left ( \frac { p - 1 } { p } \pi \sqrt { - 1 } \right ), \exp \left ( \frac { p - 3 } { p } \pi \sqrt { - 1 } \right ), \dots, \exp \left ( \frac { 1 - p } { p } \pi \sqrt { - 1 } \right ) \right ),$$
D ∗ admits a level k -prequantization if and only if (exp ζ ∗ ) k is a scalar matrix; if and only if k is a multiple of p . By property (c) in Section 5.1 and Proposition 5.6, it suffices to show that p · ˇ Φ ∗ x = 0, where Φ : ˇ M → SU( ) is the group-valued moment map. p
By Corollary 7.6 in [3], h ∨ ˇ Φ ∗ x = W M 3 ( ), the third integral Stiefel-Whitney class, where h ∨ denotes the dual Coxeter number. Recall that W M 3 ( ) = βw 2 ( M ), where β : H 2 ( M ; Z / 2 Z ) → H 3 ( M ; Z ) is the (integral) Bockstein homomorphism and w 2 ( M ) is the second Stiefel-Whitney class. Since Γ has odd order, H 2 ( M ; Z / 2 Z ) = ∼ H 2 (( D ∗ ) r ; Z / 2 Z ) Γ , which is trivial (by an argument similar to the proof of Proposition 5.6). Since h ∨ = p , we are done.
As discussed in Section 5.2, a quasi-Hamiltonian prequantization descends to a prequantization of the symplectic quotient M . glyph[squaresolid]
## A The action of the centre on the alcove of exceptional Lie groups
Below we record the action of the centre Z G ( ) on an alcove for the exceptional Lie groups G = E 6 and G = E 7 . (The action for classical groups appears in [23].)
The vertices of the alcove were obtained using polymake [11], which outputs the vertices of a polytope presented as an intersection of half-spaces. The relevant Weyl group element from Proposition 3.2 - one which gives an automorphism of the extended Dynkin diagram - was found with the help of John Stembridge's coxeter-weyl package for Maple [22]; a direct calculation then shows that this element has the desired properties in Proposition 3.2.
Let { e , . . . , e 1 8 } denote the standard basis in R 8 , equipped with the usual inner product. Given a vector α in R 8 , s α : R 8 → R 8 denotes reflection in the subspace orthogonal to α ,
$$s _ { \alpha } ( v ) = v - \frac { 2 \langle \alpha, v \rangle } { \langle \alpha, \alpha \rangle } \alpha.$$
The notation used below is consistent with that found in [9, Planches V-VI].
G = E 6 . Let t ∼ = t ∗ ∼ { = ( x , . . . , x 1 8 ) ∈ R 8 : x 6 = x 7 = -x 8 } . The simple roots α , . . . , α 1 6 and highest root ˜ determine the half-spaces whose intersection is the alcove ∆ α ⊂ t . The vertices of ∆ (opposite the facets parallel to the corresponding root hyperplanes) are given in Table 1.
Table 1. Alcove data for E 6 .
| Simple or dominant root | Opposite vertex |
|-----------------------------------------------------------------------------------------------------------|-------------------------------------------------------------------------------------------------------------------------------|
| α 1 = ( 1 2 , - 1 2 , - 1 2 , - 1 2 , - 1 2 , - 1 2 , - 1 2 , 1 2 ) α 2 = (1 , 1 , 0 , 0 , 0 , 0 , 0 , 0) | v 1 = ( 0 , 0 , 0 , 0 , 0 , - 2 3 , - 2 3 , 2 3 ) v 2 = ( 1 4 , 1 4 , 1 4 , 1 4 , 1 4 , - 1 4 , - 1 4 , 1 4 ) 1 1 1 1 1 5 5 5 |
The non-zero elements of the centre Z E ( 6 ) ∼ = Z / 3 Z are given by (exp of) the minimal dominant coweights λ ∨ 1 = 2 3 ( e 8 -e 7 -e 6 ) and λ ∨ 6 = e 5 + 1 3 ( e 8 -e 7 -e 6 ). The corresponding elements w 1 and w 6 of the Weyl group (as in Proposition 3.2), inducing automorphisms of the extended Dynkin diagram are:
w 1 = s α 1 s α 3 s α 4 s α 2 s α 5 s α 4 s α 3 s α 1 s α 6 s α 5 s α 4 s α 2 s α 3 s α 4 s α 5 s α 6 , w 6 = s α 6 s α 5 s α 4 s α 2 s α 3 s α 1 s α 4 s α 3 s α 5 s α 4 s α 2 s α 6 s α 5 s α 4 s α 3 s α 1 .
The permutation of the vertices induced by the action of exp( λ ∨ 1 ) (encoded by the automorphism w 1 of the underlying extended Dynkin diagram) is shown schematically in Fig. 2.
Figure 2. Permutation induced by action of exp λ ∨ 1 on the vertices of the alcove for E 6 .
G = E 7 . Let t ∼ = t ∗ ∼ { = ( x , . . . , x 1 8 ) ∈ R 8 : x 7 = -x 8 } . The simple roots α , . . . , α 1 7 and highest root ˜ determine the half-spaces whose intersection is the alcove ∆ α ⊂ t . The vertices of ∆ (opposite the facets parallel to the corresponding root hyperplanes) are given in Table 2.
The non-zero element of the centre Z E ( 7 ) = ∼ Z / 2 Z is given by (exp of) the minimal dominant coweight λ ∨ 7 = e 6 + 1 2 ( e 8 -e 7 ). The corresponding element w 7 of the Weyl group (as in
Table 2. Alcove data for E 7 .
| Simple or dominant root | Opposite vertex |
|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| α 1 = ( 1 2 , - 1 2 , - 1 2 , - 1 2 , - 1 2 , - 1 2 , - 1 2 , 1 2 α 2 = (1 , 1 , 0 , 0 , 0 , 0 , 0 , 0) α 3 = ( - 1 , 1 , 0 , 0 , 0 , 0 , 0 , 0) α 4 = (0 , - 1 , 1 , 0 , 0 , 0 , 0 , 0) α 5 = (0 , 0 , - 1 , 1 , 0 , 0 , 0 , 0) α 6 = (0 , 0 , 0 , - 1 , 1 , 0 , 0 , 0) α 7 = (0 , 0 , 0 , 0 , - 1 , 1 , 0 , 0) ˜ α = (0 , 0 , 0 , 0 , 0 , 0 , - 1 , 1) | v 1 = ( 0 , 0 , 0 , 0 , 0 , 0 , - 1 2 , 1 2 ) v 2 = ( 1 4 , 1 4 , 1 4 , 1 4 , 1 4 , 1 4 , - 1 2 , 1 2 ) v 3 = ( - 1 6 , 1 6 , 1 6 , 1 6 , 1 6 , 1 6 , - 1 2 , 1 2 ) v 4 = ( 0 , 0 , 1 4 , 1 4 , 1 4 , 1 4 , - 1 2 , 1 2 ) v 5 = ( 0 , 0 , 0 , 1 3 , 1 3 , 1 3 , - 1 2 , 1 2 ) v 6 = ( 0 , 0 , 0 , 0 , 1 2 , 1 2 , - 1 2 , 1 2 ) v 7 = ( 0 , 0 , 0 , 0 , 0 , 1 , - 1 2 , 1 2 ) v 0 = 0 |
Proposition 3.2), inducing an automorphism of the extended Dynkin diagram is
```
w _ { 7 } = s _ { \alpha _ { 7 } } s _ { \alpha _ { 6 } } s _ { \alpha _ { 5 } } s _ { \alpha _ { 4 } } s _ { \alpha _ { 2 } } s _ { \alpha _ { 3 } } s _ { \alpha _ { 1 } } s _ { \alpha _ { 4 } } s _ { \alpha _ { 3 } } s _ { \alpha _ { 5 } } s _ { \alpha _ { 4 } } s _ { \alpha _ { 2 } } s _ { \alpha _ { 6 } } s _ { \alpha _ { 5 } } s _ { \alpha _ { 4 } } s _ { \alpha _ { 3 } } s _ { \alpha _ { 1 } } \\ & \times s _ { \alpha _ { 7 } } s _ { \alpha _ { 6 } } s _ { \alpha _ { 5 } } s _ { \alpha _ { 4 } } s _ { \alpha _ { 2 } } s _ { \alpha _ { 3 } } s _ { \alpha _ { 4 } } s _ { \alpha _ { 5 } } s _ { \alpha _ { 6 } } s _ { \alpha _ { 7 } }.
```
The permutation of the vertices induced by the action of exp( λ ∨ 7 ) (encoded by the automorphism w 7 of the underlying extended Dynkin diagram) is shown schematically in Fig. 3.
Figure 3. Permutation induced by action of exp λ ∨ 7 on the vertices of the alcove for E 7 .
## Acknowledgements
A portion of this work is a revised version of (previously unpublished) results from the author's Ph.D. Thesis [16], supervised by E. Meinrenken and P. Selick. I remain grateful for their guidance and support. My sincere thanks as well to the referees for their helpful suggestions for clarification.
## References
- [1] Adem A., Leida J., Ruan Y., Orbifolds and stringy topology, Cambridge Tracts in Mathematics , Vol. 171, Cambridge University Press, Cambridge, 2007.
- [2] Alekseev A., Malkin A., Meinrenken E., Lie group valued moment maps, J. Differential Geom. 48 (1998), 445-495, dg-ga/9707021.
- [3] Alekseev A., Meinrenken E., Dirac structures and Dixmier-Douady bundles, Int. Math. Res. Not. 2012 (2012), 904-956, arXiv:0907.1257.
- [4] Alekseev A., Meinrenken E., Woodward C., The Verlinde formulas as fixed point formulas, J. Symplectic Geom. 1 (2001), 1-46.
- [5] Atiyah M.F., Bott R., The Yang-Mills equations over Riemann surfaces, Philos. Trans. Roy. Soc. London Ser. A 308 (1983), 523-615.
- [6] Bismut J.-M., Labourie F., Symplectic geometry and the Verlinde formulas, in Surveys in Differential Geometry: Differential Geometry Inspired by String Theory, Surv. Differ. Geom. , Vol. 5, Int. Press, Boston, MA, 1999, 97-311.
- [7] Borel A., Friedman R., Morgan J.W., Almost commuting elements in compact Lie groups, Mem. Amer. Math. Soc. 157 (2002), x+136 pages, math.GR/9907007.
- [8] Bott R., Tu L.W., Differential forms in algebraic topology, Graduate Texts in Mathematics , Vol. 82, SpringerVerlag, New York - Berlin, 1982.
- [9] Bourbaki N., Groupes et alg`bres de Lie, Chapitres 4, 5 et 6, Masson, Paris, 1981. e
- [10] Daskalopoulos G.D., Wentworth R.A., Geometric quantization for the moduli space of vector bundles with parabolic structure, in Geometry, Topology and Physics (Campinas, 1996), de Gruyter, Berlin, 1997, 119155.
- [11] Gawrilow E., Joswig M., polymake: a framework for analyzing convex polytopes, in Polytopes - Combinatorics and Computation (Oberwolfach, 1997), DMV Sem. , Vol. 29, Birkh¨user, Basel, 2000, 43-73. a
- [12] Goldman W.M., The symplectic nature of fundamental groups of surfaces, Adv. Math. 54 (1984), 200-225.
- [13] Ho N.-K., Liu C.-C.M., Connected components of spaces of surface group representations. II, Int. Math. Res. Not. 2005 (2005), 959-979, math.SG/0406069.
- [14] Konno H., On the natural line bundle on the moduli space of stable parabolic bundles, Comm. Math. Phys. 155 (1993), 311-324.
- [15] Krepski D., Pre-quantization of the moduli space of flat G -bundles over a surface, J. Geom. Phys. 58 (2008), 1624-1637, arXiv:0708.1269.
- [16] Krepski D., Pre-quantization of the moduli space of flat G -bundles, Ph.D. Thesis, University of Toronto, 2009, arXiv:1004.2286.
- [17] Krepski D., Meinrenken E., On the Verlinde formulas for SO(3)-bundles, Q. J. Math. 64 (2013), 235-252, arXiv:1106.4854.
- [18] Meinrenken E., The basic gerbe over a compact simple Lie group, Enseign. Math. 49 (2003), 307-333, math.DG/0209194.
- [19] Meinrenken E., Twisted K -homology and group-valued moment maps, Int. Math. Res. Not. 2012 (2012), 4563-4618, arXiv:1008.1261.
- [20] Meinrenken E., Sjamaar R., Singular reduction and quantization, Topology 38 (1999), 699-762, dg-ga/9707023.
- [21] Pauly C., Espaces de modules de fibr´s paraboliques et blocs conformes, e Duke Math. J. 84 (1996), 217-235.
- [22] Stembridge J., A Maple package for root systems and finite Coxeter groups, 2004, available at http: //www.math.lsa.umich.edu/ ~ jrs/maple.html#coxeter .
- [23] Toledano Laredo V., Positive energy representations of the loop groups of non-simply connected Lie groups, Comm. Math. Phys. 207 (1999), 307-339, math.QA/0106196. | 10.3842/SIGMA.2014.109 | [
"Derek Krepski"
] | 2014-08-01T18:11:42+00:00 | 2014-12-05T05:40:02+00:00 | [
"math.SG"
] | Prequantization of the Moduli Space of Flat ${\rm PU}(p)$-Bundles with Prescribed Boundary Holonomies | Using the framework of quasi-Hamiltonian actions, we compute the obstruction
to prequantization for the moduli space of flat ${\rm PU}(p)$-bundles over a
compact orientable surface with prescribed holonomies around boundary
components, where $p>2$ is prime. |
1408.0255v1 | ## Efficient inference of protein structural ensembles
Thomas J. Lane and Christian R. Schwantes
Department of Chemistry,
Stanford University
Kyle A. Beauchamp
Computational Biology Group, Memorial Sloan-Kettering Cancer Center
Vijay S. Pande ∗
Departments of Chemistry and Computer Science and Biophysics Program, Stanford University
(Dated: August 4, 2014)
It is becoming clear that traditional, single-structure models of proteins are insufficient for understanding their biological function. Here, we outline one method for inferring, from experiments, not only the most common structure a protein adopts (native state), but the entire ensemble of conformations the system can adopt. Such ensemble models are necessary to understand intrinsically disordered proteins, enzyme catalysis, and signaling. We suggest that the most difficult aspect of generating such a model will be finding a small set of configurations to accurately model structural heterogeneity and present one way to overcome this challenge.
## CURRENT TECHNIQUES FOR ELUCIDATING ATOMISTIC STRUCTURES ARE INHERENTLY LIMITED
Structures of biomolecules have driven molecular biology and biochemistry ever since Watson and Crick solved the structure of DNA (Crick and Watson, 1953). Protein and nucleic acid structures provide a mechanistic explanation of biological phenomena, including enzyme catalysis, transfer of genetic information, and molecular signaling. Further, structure plays the medically crucial role of providing a starting point for rational drug design (Banaszak, 2000).
ordered proteins in biology (Dyson and Wright, 2005). The development of new tools to directly study conformational fluctuations could play a key role in answering questions in these fields.
Since the mid-20 th century, however, the methods of obtaining biomolecular structures remain largely unchanged. With the important exception of NMR spectroscopy (Markwick et al. , 2008), x-ray crystallography remains the workhorse of structural biology (Banaszak, 2000). Because crystals, especially under cryogenic conditions, mostly lack the conformational heterogeneity found at room temperature in the solution phase (Fraser et al. , 2011; Keedy et al. , 2014), the vast majority of high-resolution structures contain little or no information about the structural diversity of proteins.
Such dynamics are implicated in many biological questions currently under investigation, including the role fluctuations (or lack thereof) play in enzyme catalysis (Glowacki et al. , 2012; Hay and Scrutton, 2012), the conformational changes undergone during protein signaling (Bu and Callaway, 2011), and the role of intrinsically dis-
∗ Correspondance: [email protected]
Besides cyro-crystallography, room temperature crystallography and NMR have shown promise as highresolution methods for the study of structural ensembles (Baldwin and Kay, 2009; Fraser et al. , 2011; Keedy et al. , 2014; Korzhnev et al. , 2010; Sekhar et al. , 2012). Both, however, require new methods to rigorously infer ensembles from the experimental readouts obtained. Other techniques such as FRET, PRE, SAXS, IR spectroscopy (Cho, 2008), etc. do provide information about the conformational fluctuations of proteins, but are limited in either spatial resolution or total structural information imparted. One could imagine, however, that in the presence of either very strong prior information about the structure of a system, or by combining many of the above techniques into a unified model, enough information might be obtained to infer a high resolution structure. Further, combining structural information from many methods would enable one to profit from the strengths of each, e.g. balancing low resolution global information (as in SAXS) with high-resolution but local probes (FRET).
Many challenges must be overcome to build such integrated structural models. Synthesis of many experimental techniques is not trivial. Further, it can be difficult to rationalize structure in terms of experimental observables, e.g. while NMR chemical shifts contain significant information about a protein sample's structure (Robustelli et al. , 2012), one cannot simply look at an NMR spectrum and 'see' even a coarse-grained picture
of the structure. Finally, in general, the problem of determining a structure - much less a structural ensemble - from available data will be under-determined; there is simply not enough information in even the best experiments to fully specify atomic resolution of an ensemble of structures in a protein system. This makes the deduction of ensemble structural information from the raw data near-hopeless. On the other hand, the inverse problem - predicting experimental output given a model of the structure - is generally possible (though sometimes challenging). Therefore, a natural approach to the interpretation of structurally rich experiments is to generate putative models, then test those models for consistency with the data. Many such approaches have been employed in the past (Choy and Forman-Kay, 2001; De Simone et al. , 2009; Marsh and Forman-Kay, 2009; Shen et al. , 2008; Trabuco et al. , 2008; Vendruscolo, 2007; Zhang et al. , 2007)
Here, we outline a way to perform two tasks simultaneously: first, interpret complex experimental data directly as an equilibrium ensemble of structures, and second, rigorously integrate information from many different experiments into a single model. We call this combined project 'integrative modeling'. Previous work along these lines has been pursued and will be discussed momentarily.
## PROBLEM STATEMENT
Our goal is to extend structural biology by focusing not only on the single most probable equilibrium conformations of macromolecules, but the set of all 'important' conformations at equilibrium. For such a project to be successful, we need to define specifically what the final model will look like and what difficulties must, in general, be overcome by a method aiming at that endpoint.
The end result of any ensemble modeling project will be a set of conformations and their associated Boltzmann weights. The conformations should be important in the sense that they are as representative as possible of the entire Boltzmann distribution, such that for a finite set of structures { x i } , things like expectation values over the Boltzmann distribution are well approximated by a finite sum over those structures. That is, for an arbitrary function of the system coordinates g ( x ),
$$\int p ( x ) \, g ( x ) \, d x \approx \frac { 1 } { N ( k ) } \sum _ { i = 1 } ^ { k } p ( x _ { i } ) \, g ( x _ { i } ) \quad \ ( 1 ) \quad \ \ W _ { \text{bles,} } \\ \intertext { u } \, \ n \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdots }$$
Where N is an appropriate normalization constant. We could also envision extending this requirement to higher moments, replacing each g with g n ; such a model would capture not only mean properties of the system, but also low-order fluctuations around that mean. We are not suggesting (1) as a specific metric by which to judge models - indeed, we never specified g - but rather as a statement of the kinds of things we would hope ensemble models would get right.
This implies that an ensemble modeling project consists of two parts. First, one needs to be able to assign Boltzmann probabilities to given conformations - compute p ( x ) for any x . Second, one needs to generate a discrete set of structures { x i } such that things like (1) are well approximated.
Efficient generation of these structures is the truly hard part of ensemble modeling. One wants to choose a small number of structures in order to keep the model simple and interpretable, but enough structures to satisfy (1). Clearly, if a relatively small number of structures are to be chosen, the intuition built by (1) tells us that in the absence of particular constraints, we should choose those that have a high Boltzmann weight p ( x ). If, however, we know a priori that we wish to probe a particular property of the system - a specific g ( x ) - then it might also behoove us to find structures that span the range of values of that function.
Previous methods for ensemble inference have focused on the idea that a set of structures can be generated randomly or from some prior information (such as an empirical forcefield) and then weighted according to some model for p ( x ), based on experimental data, later on (Fisher and Stultz, 2011). We believe this paradigm inherently limits the efficiency of ensemble inference. If, by chance or good choice, the prior ensemble models the experimental output well, and requires only minimal re-weighting for perfect agreement, then this method is viable. This means, however, our experiments have not substantially furthered our knowledge of the structural heterogeneity of the system. Either the experimental data impart little information, or the original model of the ensemble was already quite good.
Alternatively, the experimental data may be inconsistent with the prior ensemble - in this case, if the number of structures is small, the ensemble may poorly represent the most stable regions of configuration space. If the number of structures is large, a good model may be obtained, but it is likely that the vast majority of the Boltzmann weight will be concentrated on a small subset of structures. This means that the vast majority of structures are inconsequential, but time was spent computing and evaluating them, an inefficient and unnecessary scheme.
We suggest that, as a default when modeling ensembles, structures should be chosen for inclusion in the model based on their Boltzmann weight. This can readily be done by running a Monte Carlo or molecular dynamics simulation in the potential dictated by the model for the Boltzmann weight, V ( x ) ∝ -log p ( x ). There may be cases, such as when one specific system property or observation is particularly important, when this may not be the most efficient scheme. Without additional information, however, it is an excellent place to begin.
Sampling directly from an inferred Boltzmann ensemble opens up the exciting possibility of directly interpreting experiments without the need for a specific form of the prior. If structural samples are generated from a prior, then the prior must contain significant information such that it overlaps with ensembles consistent with the experimental information included in the ensemble model. When sampling directly from a posterior Boltzmann distribution, however, this requirement is alleviated. One could imagine running simulations where the prior is a minimal potential, perhaps enforcing only protein primary sequence through bond length and angle constraints. Such a model highlights an important philosophical aspect of the protocol suggested here - we envision ensemble models as a way to infer structures from experimental data, not just a method for correcting simulation errors.
## Previous Work
While we don't wish to provide a comprehensive review of previous work in both integrating multiple experiments and ensemble modeling, there have been important advances that we build on, and understanding this previous work should help the reader place our ideas in a broader context. 1
The foundation for integrative modeling has been welllaid. As early as 2005, Rieping, Habeck and Nilges pointed out that thinking of structure inference in a Bayesian framework could be a fruitful avenue for integrating different experimental techniques into a combined model. This has lead to a body of theoretical and modeling work (Habeck et al. , 2005; MacCallum et al. , 2014; Rieping et al. , 2005), but the approach has yet to be widely employed to determine structures. Our ideas build directly off theirs, but expands these ideas to modeling entire ensembles rather than single structures, an idea that has lead to a variety of related models (Beauchamp et al. , 2014; Fisher and Stultz, 2011; Huang et al. , 2010; Olsson et al. , 2013, 2014; Pitera and Chodera, 2012).
Regarding ensemble modeling, two primary techniques have been thus far pursued. The first is to implement distance restraints into MD simulations to enforce experimental restraints. The second has been to weight or select structures from a pre-selected library of structures to generate an ensemble consistent with experiments. These techniques have successfully been able to reproduce 'test' ensembles, generated from simulated data
1 The authors admit that their coverage of the literature may not be comprehensive, and would very much welcome recommendations for citations to add to this section.
( i.e. experimental outcomes are predicted from a simulated ensemble, and then the method attempts to reproduce the ensemble only from these experimental observables). They are nicely reviewed by Fisher and Stultz (Fisher and Stultz, 2011), though work has continued since 2011, e.g. (Olsson et al. , 2013, 2014).
Our work demonstrates that these two ideas, both incorporating many experiments in a Bayesian framework and modeling ensembles, can be naturally combined. Further, we suggest that the Bayesian inference necessary to infer an ensemble model can be efficiently conducted by leveraging the well-developed techniques of molecular simulation. Recently, this idea has been implemented to study the structure of simple electrolytes (White and Voth, 2014).
## THEORY
We gently extend previous work (Beauchamp et al. , 2014; Habeck et al. , 2005; Pitera and Chodera, 2012; Rieping et al. , 2005), to prescribe an efficient protocol for modeling structural ensembles. This work contains two main advances. First, our model accounts for the error in both experiments and prediction algorithms. Second, we suggest that ensemble models be constructed by directly sampling the posterior ensemble, rather than sampling the prior ensemble and then re-weighting.
The result of this procedure is a set of structures and their relative populations that is guaranteed to reproduce the experimental measurements. If the available experimental data are insufficient to uniquely determine an ensemble, the method chooses the ensemble with quantitatively the least amount of information. This guarantees that the calculated ensemble contains only the amount of information contained in the experiments, and no more. Further, this method incorporates experimental uncertainty and error in the prediction of experimental observables through a rigorous Bayesian framework.
The basic idea is as follows. We assume we are given a set of experimental data representing ensemble averaged measurements, along with a set of prediction algorithms that can estimate those experiments given a single structural model. Using these, we can find the set of Boltzmann ensembles that reproduce our experiments when we apply those prediction algorithms - there will not, in general, be a unique ensemble capable of reproducing the experimental data. We choose one amongst these 'degenerate' ensembles by favoring the one that is closest to some informative prior, which could range from a fullfledged empirical potential (MD forcefield) to a simple model of the bonds of the primary sequence of a protein. This procedure results in a potential energy function that describes the Boltzmann ensemble. Recovering our structural ensemble then consists of nothing more than sampling that potential via some algorithm ( e.g. Metropolis
Monte Carlo or molecular dynamics) that is guaranteed to reproduce a canonical ensemble.
## Choosing Amongst Degenerate Ensembles
Our goal is to produce a structural ensemble that reproduces some experimental observations, taken from a bulk experiment. We ignore momenta and work exclusively in configuration space; a protein configuration is denoted x . Thus, given a set of experimental observations a i , our goal is to produce a Boltzmann distribution p ( x ) such that those experimental observations are reproduced, i.e. choose p ( x ) such that a i = 〈 f i 〉 , where
$$\langle f _ { i } \rangle = \int p ( { \mathbf x } ) f _ { i } ( { \mathbf x } ) \, d { \mathbf x } \quad \quad ( 2 ) \quad \mathbf v \quad \ \underset { \min } { \mu }$$
and f i ( x ) is an algorithm for calculating the experimental observable from a configuration, commonly referred to as a forward function .
In general, there may be many such distributions p ( x ) that reproduce these observables. To choose amongst them we follow Pitera and Chodera (Pitera and Chodera, 2012) and choose the ensemble that contains quantitatively the least information possible relative to some prior distribution m ( x ). 2 This additional information can be quantified by the relative entropy, D ,
$$D \left [ p ( \mathbf x ) | | m ( \mathbf x ) \right ] = \int p ( \mathbf x ) \log \frac { p ( \mathbf x ) } { m ( \mathbf x ) } \, d \mathbf x \quad \ ( 3 ) \\ \text{iso called the Kullback-Leibler divergence.\ } \text{ Thus, we} \quad.$$
also called the Kullback-Leibler divergence. Thus, we want the distribution p ( x ) that minimizes D , subject to the constraint that the experimental data are reproduced. In making this choice, we are limiting our ensemble to include only information present in the experiments we attempt to match, and no more. 3 The relative entropy can be employed post facto to quantitatively measure the structural information content of experiments included, and also should provide a mechanism for assessing what additional experiments would be best to perform to continue refining a structural model.
The information gain must be benchmarked against some 'baseline,' here denoted as the prior distribution m ( x ) that we choose. A natural choice for m ( x ) for a situation where large quantities of high-quality experimental information are available might be a potential
2 While the result of our derivation is effectively the same as (Pitera and Chodera, 2012) and (Beauchamp et al. , 2014), the philosophical choice of minimizing the relative entropy was not highlighted in those works. The derivations there involve an integral involving the differential entropy where the precise connection to the prior potential is obscured.
3 Thanks to Robert McGibbon for pointing out this is a specific case of posterior regularization (Ganchev et al. , 2010).
V 0 ( x ) that includes only bonds, angles, and sterics, enforcing the primary structure of the protein as well as basic geometric constraints (no overlapping atoms), but little more. In the case where the experiments at our disposal are fairly limited, a stronger prior, such as an empirical MD forcefield, could be employed. In principle this potential can be anything from quantum calculations to a non-interacting ideal gas.
We will next demonstrate how to find the distribution p ( x ) that minimizes (3) subject to the constraint that the Boltzmann distribution obtained gives the experimentally observed values, as in (2). In the next section, we present a more sophisticated model that includes error models for the experimental results and experimental predictors.
We solve the constrained optimization problem minimizing (3) by introducing the Lagrange functional
$$\text{mixing} \left ( 3 \right ) \text{ by introducing the Lagrange functional} \\ \Lambda \left [ p ( x ) \right ] = & \int p ( x ) \log p ( x ) \, d x - & ( 4 ) \\ & \int p ( x ) \log m ( x ) \, d x - \\ & \sum _ { i } \lambda _ { i } \left ( \int f _ { i } ( x ) p ( x ) \, d x - a _ { i } \right ) \\ \text{where the } \lambda _ { i } \text{ are Lagrange multipliers. Finding the root} \\ \text{$\mathcal{e}$} + \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \mathcal { e } \text{1}$$
where the λ i are Lagrange multipliers. Finding the root of the functional derivative δ Λ /δp gives the functional extremum of p ( x ) (Appendix A.1),
$$p _ { \lambda } ( \boldsymbol x ) = \frac { 1 } { Z ( \boldsymbol \lambda ) } \exp \left \{ \sum _ { i } \lambda _ { i } f _ { i } ( \boldsymbol x ) - V _ { 0 } ( \boldsymbol x ) \right \} \quad ( 5 ) \\ \text{where } Z \text{ is a normalization constant (nartition function)}.$$
where Z is a normalization constant (partition function), we have written the prior in terms of its corresponding potential m ( x ) ∝ -log V 0 ( x ), and have added the subscript λ to emphasize that the specific Boltzmann distribution is parameterized by the Lagrange multipliers. This shows that if we can find the Lagrange multipliers λ , we can sample p λ ( x ) by running a simulation in the potential i λ f i i ( x ) -V 0 ( x ).
∑ The Lagrange multipliers we desire are those that ensure the constraints
$$a _ { i } = \int f _ { i } ( \mathbf x ) \, p _ { \lambda } ( \mathbf x ) \, d \mathbf x \quad \text{(6)} \\ \vdots \text{This is easily done through an analog} \, \text{to}$$
are satisfied. This is easily done through an analogy to statistical mechanics (Rieping et al. , 2005). Consider the 'free energy' function
$$\Gamma ( \lambda ) = - \log Z ( \lambda ) + \sum _ { i } \lambda _ { i } a _ { i } \quad \quad ( 7 ) \\ \lim e t h o r o o t e \, \text{of} \, \text{the} \, \text{administrases} \, \text{AP} / A \,, \, / \, \Lambda \, \text{non-div}$$
by finding the roots of the derivatives ∂ Γ /∂λ i (Appendix A.2), we show that 〈 f i 〉 λ = a i for all observables i is one extremum of the function Γ( λ ). Taking second derivatives shows
$$H _ { i j } = \frac { \partial ^ { 2 } \, \Gamma ( \boldsymbol \lambda ) } { \partial \lambda _ { i } \, \partial \lambda _ { j } } = \langle f _ { i } \, f _ { j } \rangle _ { \boldsymbol \lambda } - \langle f _ { i } \rangle _ { \boldsymbol \lambda } \langle f _ { j } \rangle _ { \boldsymbol \lambda } \quad ( 8 )$$
that the Hessian of Γ( λ ) is the covariance matrix of the predicted observables f i , which ensures that the Hessian is positive semi-definite. Whenever the the f i are not linearly dependent, the Hessian will be positive definite. We assume this is the case going forward, and note it can be ensured by intelligently choosing the set of f i to include in the model.
Since the Hessian of Γ is positive definite, this function has a unique extremum, a minimum. At that minimum, 〈 f i 〉 λ = a i , i.e. the values of λ satisfy the constraints we wished to find, (6). Thus by numerically minimizing (7), a convex problem, we can find the desired Lagrange multipliers λ .
In practice, computing the gradient and Hessian of Γ( λ ) requires an expectation be evaluated over the equilibrium distribution p λ ( x ), which we have written 〈·〉 λ . This must requires samples of x to estimate the integral (6). These can be generated by protein Monte Carlo or molecular dynamics, which are guaranteed to produce correct the stationary distribution.
Notice throughout the previous discussion we have not included any treatment of possible error in either the experimental values or in the prediction algorithms f i . Next, we treat such error.
## Including Error in the Experiment and Prediction Algorithms
We consider the case where we wish to include a model for the error inherent in both experiment and the forward function predictions in our ensemble model. Suppose we have some model for the probability P ( a | D ) that a specific experimental outcome, a , is the correctly predicted ground truth given some observed experimental data D . The shorthand P ( a | D ) includes all error due to both the experiment and prediction algorithm . Appendix B provides a general example of how this function might be constructed.
Given a model for P ( a | D ), we can treat the possible experimental outcomes as a nuisance parameter, and obtain a single estimate for the Boltzmann distribution
$$p ( \mathbf x ) = \int _ { A } p _ { \lambda } ( \mathbf x ) P ( \mathbf a | D ) \, d \mathbf a \, \quad \, \quad ( 9 ) \, \\ \dots \, \dots \, \quad \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \end{cases}$$
Where A is the set of experiments our forward functions can possibly predict, A = { a : a i ∈ range[ f i ] } , excluding e.g. unphysical possibilities such as negative chemical shifts. This model includes all possible experimental 'ground truths', and weights them according to how likely each is given the observed experimental data and the quality of the forward functions.
The integral (9) cannot be evaluated directly, however, since the Boltzmann distribution for a given experimental 'ground truth', p λ , is parameterized by λ and not a . Fortunately, we just derived a method for determining a set of λ corresponding to a single a by minimizing the function Γ. Let us denote this process by M , that is M ( a ) = λ .
Then, we can write a change of variables formula
$$P ( \lambda | D ) = P ( \mathbf a | D ) \left | J _ { M ^ { - 1 } } ( \lambda ) \right | \text{ \quad \ } ( 1 0 )$$
where J M -1 is the Jacobian of the inverse of M (Appendix A.4 proves M is a bijection, and therefore invertible). We can now integrate over λ instead of a ,
$$p ( \mathbf x ) = \int _ { L } p _ { \lambda } ( \mathbf x ) \, P ( \lambda | D ) \, | J _ { M ^ { - 1 } } ( \lambda ) | \, d \lambda \quad ( 1 1 ) \\ \intertext { w h o r o \, T, i e t h o \, i m a s o \, o f \, \Delta \, M \cdot \, \Delta \, \rightharpoonup \, T. \, \ln \, \Delta \, n n o n d i v \, \Delta \, \mathbf x }$$
where L is the image of A M , : A → L . In Appendix A.3 we give an explicit formula for J M -1 .
## IMPLEMENTATION
We have begun work on an open-source software package named Odin that aims to implement this computational framework in an efficient and accessible manner ( www.github.com/tjlane/odin ).
Odin will orchestrate the following procedure in order to estimate the Boltzmann ensemble, based on the theory above. We wish to efficiently evaluate the integral (11). This can be performed via a Monte Carlo scheme, which will be efficient if we sample regions of λ -x space where the integrand is large.
To perform this, we suggest the following workflow, which is simple but has no guarantees of optimality:
- · Construct P ( λ | D ) and find it's maximum, ˆ , either λ numerically or analytically.
- · Run extensive protein simulations (MD or MC) in the ensemble p ˆ λ , and record samples { x i } . This can be done in parallel by running many independent simulations.
- · Employ Markov chain Monte Carlo to evaluate the high dimensional integral over λ in (11).
- · Run additional sampling or employ left aside samples of { x i } to check for convergence of the algorithm. It will be impossible in general to evaluate convergence, but we suggest that sampling can be deemed converged when the ensemble averages of the forward functions employed, 〈 f i 〉 , stabilize.
One possibility that may be fruitful to explore would be running many different simulations in different λ ensembles, perhaps clustered around p ˆ λ , or by changing the values of λ in serial as simulations progress. Such schemes are directly analogous to the various replicabased and annealing approaches employed in molecular simulation, but here the thermodynamic variables are the Lagrange multipliers λ (instead of e.g. temperature). This could speed sampling, but which scheme will be most efficient is very difficult to analyze and will require heuristics derived from experience.
## A Comment on Efficiency
A major shortcoming of our scheme is that the factors Z ( λ ) and J M -1 ( λ ) require integrals over the space of all x in order to evaluate. This greatly reduces the potential efficiency of the method, as samples of x are necessary to evaluate and sample p ( x ), leading to the need for an iterative scheme. The workflow suggested in the previous section calls for a single iteration, the minimum required, but it is currently unknown if this is the most efficient protocol possible.
Sadly, this problem is general to all inference schemes that attempt to model ensemble properties (such as expectations). Consider properties such as (2). If we wish to ensure that the mean value of an observable, 〈 f i 〉 , has a specific value without any a priori knowledge of the structure of f i or the value of the desired expectation 〈 f i 〉 , then it will be required that we sample values of f i ( x ) in order to evaluate the viability of the Boltzmann distribution p ( x ). This problem will be faced by any scheme that aims to reproduce ensemble properties.
We suspect that the most efficient ensemble modeling algorithms will be those equipped with efficient ways to sample both x and any auxiliary parameters (here, λ ) simultaneously in a way guaranteed to concentrate on regions of configuration space where the Boltzmann weight p ( x ) is large. Precisely how to do this for the present method is a subject of ongoing research.
## FUTURE DIRECTIONS
We have presented a method that enables the inference of ensemble structural models from diverse experiments. Our method has a number of attractive properties, including:
- · the ability to model arbitrary experimental observables,
- · a rigorous treatment of experimental and forward function error, and
- · the capability to sample directly from a posterior Boltzmann probability
The method presented here is not unique in this regard, but we feel these three properties form a minimum set of requirements for ensemble modeling techniques, and thus believe the method presented here a viable option for ensemble modeling.
A number of further avenues remain for further research in the nascent field of ensemble modeling. Any such model will be limited by the quality of the forward functions employed. Research into efficient, accurate predictors is mandatory. Software implementations of these methods and standardized ways of processing experimental data need to be developed and curated by a nascent community.
We have suggested that methods employing long MC or MD runs will be necessary to compute expectations of the form (6), which might make such calculations intractable for a large portion of the structural biology community. The good news is that by leveraging developments from protein simulation community, advances in hardware, software and equilibrium sampling techniques can be leveraged to speed this process; it should be tractable at present for small systems (100 residues or less).
Finally, as discussed previously, research remains to be done into efficient methods for sampling from p ( x ) directly. Such sampling schemes will alleviate the need for highly informative priors and should generate large gains in modeling efficiency. We hope that methods for performing MCMC sampling, developed in statistics, might be reappropriated for this purpose.
These challenges granted, we believe this scheme or some other closely related one holds great hope for the integrative modeling of structural ensembles.
## ACKNOWLEDGMENTS
We are grateful to John Chodera for sharing his notes on this subject matter, which led quickly to the extension of his work with Jed Pitera (Pitera and Chodera, 2012) to include error. Discussions with Seb Doniach, Rhiju Das, and Robert McGibbon have also been invaluable.
## Appendix A: Details of the Derivation of p λ
## 1. Derivation of Equation (5)
We begin with the Lagrange functional (4). We wish to find the functional extremum. Analogous to standard (non-functional) Lagrange multipliers, taking derivatives with respect to the conjugate variables λ i , 0 = ∂ Λ /∂λ i , reproduces the constraints (2). The functional derivative ∂ Λ /∂p is defined as
$$\int \frac { \partial \Lambda [ p ] } { \partial p ( { \mathbf x } ) } \, \phi ( { \mathbf x } ) \, d { \mathbf x } = \frac { d } { d \epsilon } \Lambda [ p + \epsilon \phi ] \Big | _ { \epsilon = 0 } \\ \text{using the left hand side,}$$
evaluating the left hand side,
$$\max \, \text{$me$} \, \text{$me$} \, \text{$me$}, \\ \frac { d } { d \epsilon } \Lambda [ p + \epsilon \phi ] \Big | _ { \epsilon = 0 } = & \int \phi + \phi \log p \, d x \\ & - \int \phi \log m \, d x \\ & - \sum _ { i } \lambda _ { i } \int f _ { i } \, \phi \, d x$$
which, by the functional derivative definition, shows
$$\frac { \partial \Lambda [ p ] } { \partial p } = 1 + \log p - \log m - \sum _ { i } \lambda _ { i } f _ { i }$$
at the extremum ∂ Λ /∂p = 0,
$$\log p = \log m + \sum _ { i } \lambda _ { i } f _ { i } - 1$$
Because the resulting p is parameterized in terms of λ , and is not normalized, we will denote its normalized form by p λ ≡ p/ ∫ p d x . This allows us to write
$$p _ { \lambda } ( { \boldsymbol x } ) \in m ( { \boldsymbol x } ) \exp \left \{ \sum _ { i } \lambda _ { i } f _ { i } ( { \boldsymbol x } ) - 1 \right \} \quad ( A 1 ) \quad \text{W} \quad \text{of } e. \\ \text{mult}$$
Defining a normalization constant (partition function) Z ( λ ) = e -1 ∫ m ( x ) exp { ∑ i λ f i i ( x ) } d x , we obtain
$$p _ { \lambda } ( \mathbf x ) = \frac { m ( \mathbf x ) } { Z ( \lambda ) } \exp \left \{ \sum _ { i } \lambda _ { i } f _ { i } ( \mathbf x ) \right \} \quad ( \mathbf A 2 ) \quad \text{show} \\ \quad \text{$i$ $p$} \\.$$
Finally, rewriting m ( x ) = -log V 0 ( x ) gives (5).
## 2. Derivation of ∇ Γ
Beginning with (7), we show 〈 f i 〉 λ = a i is a local extremum of Γ. A similar calculation can be used to derive the Hessian H ( λ ) and later the Jacobian J M -1 by taking second derivatives.
$$\text{second derivatives.} \\ 0 & = \frac { \partial \Gamma } { \partial \lambda _ { i } } \\ & = - \frac { 1 } { Z ( \lambda ) } \frac { \partial Z ( \lambda ) } { \partial \lambda _ { i } } + a _ { i } \\ & = - \frac { 1 } { Z ( \lambda ) } \frac { \partial } { \partial \lambda _ { i } } \left ( \int \exp \left \{ \sum _ { i } \lambda _ { i } f _ { i } ( x ) - V _ { 0 } ( x ) \right \} d x \right ) + a _ { i } \quad \text{report} \\ & = - \frac { 1 } { Z ( \lambda ) } \int f _ { i } ( x ) \exp \left \{ \sum _ { i } \lambda _ { i } f _ { i } ( x ) - V _ { 0 } ( x ) \right \} d x + a _ { i } \quad \text{single} \\ & = - \langle f _ { i } \rangle \lambda + a _ { i } \\ \text{The calculation may be familiar from identical mean} \quad \text{where}$$
This calculation may be familiar from statistical mechanics, as it is the method employed to find equilibrium values of thermodynamic parameters by minimizing a free energy function.
## 3. Explicit form of the Jacobian
We compute the Jacobian J M -1 for reference. One element of the Jacobian matrix can be written
$$\det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \\ [ J _ { M ^ { - 1 } } ] _ { i j } & = \frac { \partial M _ { i } ^ { - 1 } ( \lambda ) } { \partial \lambda _ { j } } \\ & = \frac { \partial } { \partial \lambda _ { j } } \left ( \int \frac { f _ { i } ( \mathbf x ) } { Z ( \lambda ) } \exp \left \{ \sum _ { i } \lambda _ { i } f _ { i } ( \mathbf x ) - V _ { 0 } ( \mathbf x ) \right \} d \mathbf x \right ) \\ & = \langle f _ { i } \, f _ { j } \rangle \lambda - \langle f _ { i } \rangle \lambda \langle f _ { j } \rangle \lambda } \\ \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \dotdet \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det\det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \dot \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \ delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta \delta
\det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \det \dots$$
Note this is the exact same computation as in (8).
## 4. Proof M is Invertible
We demonstrate that the process of translating a set of experimental observations a into a set of Lagrange multipliers λ is bijective for possible values of a .
We denote this procedure by M ( a ) = λ ; previously, we showed that this could be done numerically by minimizing Γ in eq. (7). Recall A = { a : a i ∈ range[ f i ] } , and L is the image of A under operation M . Then we will show M : A → L is bijective so long as the Hessian (8) is positive definite everywhere in λ .
For a given λ ∈ L , we know the corresponding a is given by 〈 f 〉 λ . Suppose there existed a ′ such that M ( a ′ ) = M ( a ). Since the Hessian of Γ in (8) is positive definite, then M ( a ′ ) = M ( a ) = λ . Further, it would also be the case that 〈 f 〉 λ = a ′ = a . This shows that M is one-to-one. Since L is defined as the range of M , it is also onto. Therefore, M is a bijection, and has an inverse.
## Appendix B: Example of Constructing P ( a | D )
We stated previously that the error model written P ( a | D ) should include errors for both the experiment and prediction models employed. It is generally straightforward to write down the experimental error model as the probability of a specific ground truth a ′ given some experimental data, D . Prediction algorithms also usually report the probability that a given value a is the ground truth given the prediction was some other value, say a ′ .
These two error models can then be composed into a single model
$$p ( \mathbf a | D ) = \int P _ { f } ( \mathbf a | \mathbf a ^ { \prime } ) \, P _ { e } ( \mathbf a ^ { \prime } | D ) \, d \mathbf a ^ { \prime } \\ \cdot \cdot \cdot \cdot \cdot$$
where P f ( a a | ′ ) and P e ( a ′ | D ) represent the error models for the predictors and experiments, respectively.
Let us provide a concrete example of this construction. First we consider P e ( a ′ | D ). Suppose the experiments are
well-modeled by a multivariate normal. Then we could write a likelihood function
$$\mathcal { L } ( D | \mu, \Sigma ) \in \prod _ { \mathrm d \in D } \exp \left \{ - \frac { 1 } { 2 } ( \mathrm d - \mu ) ^ { T } \Sigma ^ { - 1 } ( \mathrm d - \mu ) \right \} \quad \mathrm d \, \\ \mathcal { \Phi } \quad. \quad. \quad. \quad. \quad. \quad. \quad.$$
If we assume the parameters µ and Σ have uniform priors, then we can invoke Bayes' theorem to write
$$P _ { e } ( \mathbf a ^ { \prime } | D ) \, \in \, & \iint \mathcal { L } ( \mathbf a ^ { \prime } | \mu, \Sigma ) \mathcal { L } ( D | \mu, \Sigma ) \, d \mu \, d \Sigma \\ \nu \, \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \, & \cdot \end{array}$$
Further suppose that we have a model of the error of the predictors { f i } , P f ( a a | ′ ). To continue our concrete example, consider the basic case where the experimental predictors have a Gaussian spread, defined by covariance matrix S , whatever the predicted value may be. Then we could write
$$P _ { f } ( a | a ^ { \prime } ) \in \exp \left \{ - \frac { 1 } { 2 } ( a - a ^ { \prime } ) ^ { T } S ^ { - 1 } ( a - a ^ { \prime } ) \right \}$$
By combining the previous steps, we could construct our final model for P ( a | D ), up to a normalization constant Z ,
$$P ( \mathfrak { a } | D ) & = \frac { 1 } { Z } \int P _ { f } ( \mathfrak { a } | \mathfrak { a } ^ { \prime } ) \mathcal { L } ( \mathfrak { a } ^ { \prime } | \mu, \Sigma ) \mathcal { L } ( D | \mu, \Sigma ) \, d \mu \, d \Sigma \, d \mathfrak { a } ^ { \prime } \quad \begin{array} { c } \text{Olsso} \\ \text{Olsso} \\ \text{J. } \end{array} \,,$$
such a model could be evaluated numerically e.g. by MCMC.
This provides an illustration the construction of P ( a | D ); the specifics of this model will necessarily vary based on the application at hand.
## REFERENCES
Baldwin, A. J., and L. E. Kay (2009), Nature Chemical Biology 5 (11), 808.
- Banaszak, L. (2000), Foundations of structural biology , first edition ed. (Elsevier Science & Technology Books).
- Beauchamp, K. A., V. S. Pande, and R. Das (2014), Biophysj 106 (6), 1381.
- Bu, Z., and D. J. E. Callaway (2011), Adv Protein Chem Struct Biol 83 , 163.
- Cho, M. (2008), Chem. Rev. 108 (4), 1331.
- Choy, W.-Y., and J. D. Forman-Kay (2001), Journal of Molecular Biology 308 (5), 1011.
- Crick, F., and J. Watson (1953), Nature 171 (4356), 737.
- De Simone, A., B. Richter, X. Salvatella, and M. Vendruscolo (2009), J. Am. Chem. Soc. 131 (11), 3810.
- Dyson, H. J., and P. E. Wright (2005), Nat Rev Mol Cell Biol 6 (3), 197.
Fisher, C. K., and C. M. Stultz (2011), Current Opinion in Structural Biology 21 (3), 426.
- Fraser, J. S., H. van den Bedem, A. J. Samelson, P. T. Lang, J. M. Holton, N. Echols, and T. Alber (2011), Proceedings of the National Academy of Sciences 108 (39), 16247.
- Ganchev, K., J. Gra¸ ca, J. Gillenwater, and B. Taskar (2010), The Journal of Machine Learning Research 99 , 2001.
- Glowacki, D. R., J. N. Harvey, and A. J. Mulholland (2012), Nature Chemistry 4 (3), 169.
Habeck, M., M. Nilges, and W. Rieping (2005), Phys. Rev.
E
72
(3), 031912.
- Hay, S., and N. S. Scrutton (2012), Nature Chemistry 4 (3), 161.
- Huang, X., Y. Yao, G. R. Bowman, J. Sun, L. J. Guibas,
- G. Carlsson, and V. S. Pande (2010), Pac Symp Biocomput , 228.
- Keedy, D. A., H. van den Bedem, D. A. Sivak, G. A. Petsko, D. Ringe, M. A. Wilson, and J. S. Fraser (2014), Structure/Folding and Design , 1.
- Korzhnev, D. M., T. L. Religa, W. Banachewicz, A. R. Fersht, and L. E. Kay (2010), Science 329 (5997), 1312.
- MacCallum, J. L., A. Perez, and K. A. Dill (2014), Biophysj 106 (S1), 655a.
- Markwick, P. R. L., T. Malliavin, and M. Nilges (2008), PLoS Comput Biol 4 (9), e1000168.
- Marsh, J. A., and J. D. Forman-Kay (2009), Journal of Molecular Biology 391 (2), 359.
- Olsson, S., J. Frellsen, W. Boomsma, K. V. Mardia, and T. Hamelryck (2013), PLoS ONE 8 (11), e79439.
- Olsson, S., B. R. V¨ ogeli, A. Cavalli, W. Boomsma, J. Ferkinghoff-Borg, K. Lindorff-Larsen, and T. Hamelryck (2014), J. Chem. Theory Comput. , 140630122258002.
- Pitera, J. W., and J. D. Chodera (2012), J. Chem. Theory Comput. 8 (10), 3445.
- Rieping, W., M. Habeck, and M. Nilges (2005), Science 309 (5732), 303.
- Robustelli, P., K. A. Stafford, and A. G. Palmer III (2012), J. Am. Chem. Soc. , 120328100428005.
- Sekhar, A., P. Vallurupalli, and L. E. Kay (2012), Proceedings of the National Academy of Sciences 109 (47), 19268.
Shen,
Y.,
O. Lange,
F.
Delaglio,
P.
Rossi,
J.
M. Aramini,
- G. Liu, A. Eletsky, Y. Wu, K. K. Singarapu, A. Lemak, A. Ignatchenko, C. H. Arrowsmith, T. Szyperski, G. T. Montelione, D. Baker, and A. Bax (2008), P Natl Acad Sci Usa 105 (12), 4685.
- Trabuco, L. G., E. Villa, K. Mitra, J. Frank, and K. Schulten (2008), Structure 16 (5), 673.
- Vendruscolo, M. (2007), Current Opinion in Structural Biology 17 (1), 15.
- White, A. D., and G. A. Voth (2014), J. Chem. Theory Comput. , 140702123701004.
- Zhang, Q., A. C. Stelzer, C. K. Fisher, and H. M. Al-Hashimi (2007), Nature 450 (7173), 1263. | null | [
"Thomas J. Lane",
"Christian R. Schwantes",
"Kyle A. Beauchamp",
"Vijay S. Pande"
] | 2014-08-01T18:13:48+00:00 | 2014-08-01T18:13:48+00:00 | [
"physics.bio-ph",
"q-bio.BM"
] | Efficient inference of protein structural ensembles | It is becoming clear that traditional, single-structure models of proteins
are insufficient for understanding their biological function. Here, we outline
one method for inferring, from experiments, not only the most common structure
a protein adopts (native state), but the entire ensemble of conformations the
system can adopt. Such ensemble mod- els are necessary to understand
intrinsically disordered proteins, enzyme catalysis, and signaling. We suggest
that the most difficult aspect of generating such a model will be finding a
small set of configurations to accurately model structural heterogeneity and
present one way to overcome this challenge. |
1408.0256v1 | ## Ergodicity of a Time-Reversibly Thermostated Harmonic Oscillator and the 2014 Ian Snook Prize
William Graham Hoover and Carol Griswold Hoover Ruby Valley Research Institute Highway Contract 60, Box 601 Ruby Valley, Nevada 89833 (Dated: August 4, 2014)
## Abstract
Shuichi Nos´ opened up a new world of atomistic simulation in 1984. e He formulated a Hamiltonian tailored to generate Gibbs' canonical distribution dynamically. This clever idea bridged the gap between microcanonical molecular dynamics and canonical statistical mechanics. Until then the canonical distribution was explored with Monte Carlo sampling. Nos´'s dynamical Hamiltonian e bridge requires the 'ergodic' support of a space-filling structure in order to reproduce the entire distribution. For sufficiently small systems, such as the harmonic oscillator, Nos´'s dynamical ape proach failed to agree with Gibbs' sampling and instead showed a complex structure, partitioned into a chaotic sea, islands, and chains of islands, that is familiar textbook fare from investigations of Hamiltonian chaos. In trying to enhance small-system ergodicity several more complicated 'thermostated' equations of motion were developed. All were consistent with the canonical Gaussian distribution for the oscillator coordinate and momentum. The ergodicity of the various approaches has undergone several investigations, with somewhat inconclusive ( contradictory ) results. Here we illustrate several ways to test ergodicity and challenge the reader to find even more convincing algorithms or an entirely new approach to this problem.
## PACS numbers:
Keywords: Ergodicity, Chaos, Algorithms, Dynamical Systems
## I. INTRODUCTION
In 1989 Shuichi Nos´ won IBM-Japan's science prize for his 1984 discovery e 1,2 of a dynamical Hamiltonian approach to Gibbs' canonical ensemble. He added a new 'time-scaling' variable s ( and its conjugate momentum p s ) to the usual list of # Cartesian degrees of freedom, coupling the kinetic energy to a logarithmic temperature-dependent potential for s :
$$\sigma \. \\ \mathcal { H } _ { N } ( q, p, s, p _ { s } ) = \sum ^ { \# } ( p ^ { 2 } / 2 m s ^ { 2 } ) + \Phi ( q ) + ( p _ { s } ^ { 2 } / 2 M ) + ( \# + 1 ) k T \ln ( s ) \.$$
Multiplying the resulting equations of motion by s ( 'scaling the time' ) gives a set of (2# + 2) dynamical motion equations fully consistent with Gibbs' canonical ensemble ,
$$f ( q, p ) \in e ^ { - \mathcal { H } / k T }$$
provided that we replace ( p/s ) by p in the scaled equations of motion.
Hoover 3 soon pointed out the lack of ergodicity for this approach, applied to a harmonic oscillator, and at the same time simplified the derivation of a set of (2# + 1) equations of motion only slightly different to Nos´'s : e
$$\{ \, \dot { q } = ( p / m ) \ ; \, \dot { p } = F - \zeta p \, \} \ ; \, \dot { \zeta } = \sum ^ { \# } [ \ ( p ^ { 2 } / m k T ) - 1 \ ] / \tau ^ { 2 } \.$$
The friction coefficient ζ in these 'Nos´ e-Hoover' equations is proportional to Nos´'s e p s . These equations are much better behaved numerically and also make the extra s variable redundant. Carl Dettmann showed 4 that these same equations follow from a Hamiltonian like Nos´'s provided that the Dettmann Hamiltonian is arbitrarily set equal to zero and the e number of degrees of freedom is # rather than # + 1 ,
$$\mathcal { H } _ { D } \equiv s \mathcal { H } _ { N } \equiv 0 \.$$
Hoover and Harald Posch and Franz Vesely 5,6 investigated the details of these dynamics for the harmonic oscillator and discovered an infinite variety of periodic, toroidal, and chaotic solutions, with the sum of all of these disparate parts equal to the canonical distribution, the product of three Gaussian functions, in q , in p , and in ζ for the oscillator.
Bauer, Bulgac, and Kusnezov 7,8 generalized the thermostating approach to include two or more control variables, even managing to reproduce Brownian motion by using three of them. Soon after, Martyna, Klein, and Tuckerman 9 suggested the use of a chain of
thermostat variables to enhance ergodicity. Their equations of motion ( for the shortest chain and with all the free parameters equal to unity ) are :
$$\{ \ \dot { q } = p \ ; \ \dot { p } = - q - \zeta p \ ; \ \dot { \zeta } = p ^ { 2 } - 1 - \xi \zeta \ ; \ \dot { \xi } = \zeta ^ { 2 } - 1 \ \} \. \ [ \ M K T \ ]$$
Presently, Hoover and Holian suggested a different dual-control approach 10 , fixing both the second and fourth velocity moments :
$$\{ \ \dot { q } = p \ ; \ \dot { p } = - q - \zeta p - \xi p ^ { 3 } \ ; \ \dot { \zeta } = p ^ { 2 } - 1 \ ; \ \dot { \xi } = p ^ { 4 } - 3 p ^ { 2 } \ \} \. \ [ \ H H \ ]$$
Equations controlling the sixth or higher moments tend to be too stiff for practical use.
Very recently Patra and Bhattacharya 11 took a different direction, stimulated by work on the 'configurational temperature', as opposed to the usual kinetic one. For the oscillator the configurational temperature 12 is proportional to the potential energy, 〈 q 2 〉 . Patra and Bhattacharya had the clever idea of imposing both temperatures simultaneously on the oscillator. Their feedback equations use one friction coefficient, ζ , to control kinetic temperature, and the other, ξ , to control configurational temperature , 〈 q 2 〉 :
$$\{ \ \dot { q } = p - \xi q \ ; \ \dot { p } = - q - \zeta p \ ; \ \dot { \zeta } = p ^ { 2 } - 1 \ ; \ \dot { \xi } = q ^ { 2 } - 1 \ \} \. \ [ \ P B \ ]$$
We wish to emphasize that all three approaches followed here, Martyna-Klein-Tuckerman, Hoover-Holian, and Patra-Bhattacharya, if ergodic, provide exactly the same fourdimensional Gaussian distribution ,
$$( 2 \pi ) ^ { 2 } f ( q, p, \zeta, \xi ) \equiv e ^ { - ( q ^ { 2 } / 2 ) } e ^ { - ( p ^ { 2 } / 2 ) } e ^ { - ( \zeta ^ { 2 } / 2 ) } e ^ { - ( \xi ^ { 2 } / 2 ) } \.$$
To prove that this is so it is only necessary to evaluate Liouville's continuity equation for the flow in phase space, showing that the four-dimensional Gaussian distribution is stationary :
$$( \partial f / \partial t ) \equiv - f [ \, ( \partial \dot { q } / \partial q ) + ( \partial \dot { p } / \partial p ) + ( \partial \dot { \zeta } / \partial \zeta ) + ( \partial \dot { \xi } / \partial \xi ) \ ]$$
$$& ( \partial f / \partial t ) \equiv - f [ \ ( \partial \dot { q } / \partial q ) + ( \partial \dot { p } / \partial p ) + ( \partial \dot { \zeta } / \partial \zeta ) + ( \partial \dot { \xi } / \partial \xi ) \ ] \\ & \quad - \dot { q } ( \partial f / \partial q ) - \dot { p } ( \partial f / \partial p ) - \dot { \zeta } ( \partial f / \partial \zeta ) - \dot { \xi } ( \partial f / \partial \xi ) \equiv 0 \.$$
Let us turn to the question of establishing ergodicity for the three dynamical models, MKT, HH, and PB . Our own approach, because we know it best, is computational. It is possible that a more convincing 'rigor mortis' approach could be developed from a mathematical standpoint. At the moment any 'proof' of ergodicity is an 'open problem'. In seven brief sections we describe some of the methods that have been applied to this problem 13 , ending with a challenge for the reader.
(q,p) Projections for the MKT, HH, and PB Oscillator Models
FIG. 1: Chaotic sea for the Martyna-Klein-Tuckerman, Hoover-Holian, and Patra-Bhattacharya oscillators. The initial condition for these trajectories was ( q, p, ζ, ξ ) = (0 5 0 0) in all three cases. , , , 100,000 equally spaced ( q, p ) points are plotted for each model. The Patra-Bhattacharya oscillator has clear deviations from hyperspherical symmetry. The fourth-order Runge-Kutta timestep is 0.001 for all of the computations in our work. Here ten million timesteps were used. Note the holes in the MKT and HH projections and the lack of circular symmetry in the PB projection.
## II. TIME AVERAGES OF THE MOMENTS
Because kinetic temperature is the second velocity moment it is usual to confirm that the first few velocity moments agree with the canonical distribution. A short fourth-order Runge-Kutta calculation, adding ( p , p 2 4 , p 6 ) to the list of righthandsides being integrated, shows that the second, fourth, and sixth moments of the MKT and HH equations match the values expected from a Gaussian, (1, 3, 15) to three or four significant figures while the PB model gives instead (1.000, 3.8, 23.5) . These results are conclusive evidence that only the MKT and HH equations are candidates for ergodicity. Nevertheless we will apply each of our numerical tests to all three sets of motion equations, in order to see the difference between the difficult and the easy.
## III. PROJECTIONS
The Projections of the four-dimensional dynamics onto any plane, such as the ( q, p ) , ( q, ζ ) , ( q, ξ ) , ( p, ζ ) , ( p, ξ ) , ( ζ, ξ ) planes, should look Gaussian without any holes . For simplicity we use the initial condition p = 5 with the other variables initially zero, promoting chaos for all three sets of equations. The MKT and HH equations both appear to have a small hole near the origin of the ( qp ) plane. The PB equations produce a very different pattern, showing a strong positive correlation between the two variables, another indication that these equations though sometimes chaotic, are not ergodic. See Figure 1 for a comparison of the three models in the ( q, p ) planar projection. These results suggest two further checks, a look at the density near the ( q, p ) origin and a search for periodic orbits, which would be necessary in order that any holes form in the Gaussian distributions.
## IV. DENSITY
The probability density for the four-dimensional Gaussian at a radius of 0 1 has decreased . from its maximum value at the (0 0 0 0) origin by a factor of , , ,
$$[ \ \rho ( r = 0. 1 ) / \rho ( r = 0. 0 ) \ ] = e ^ { - r ^ { 2 } / 2 } = e ^ { - 0. 0 0 5 } \simeq 1. 0 0 0 0 0 - 0. 0 0 5 0 0 + 0. 0 0 0 0 1 = 0. 9 9 5 0 1$$
so that the densities determined from trajectories should hardly vary within that fourdimensional sphere provided that the flow is ergodic . Measuring density requires considerably longer runs because the time interval between visits near the origin increases at least as rapidly as ( r -4 ) . We measured the probability densities for the MKT and HH algorithms inside ( q, p ) circles of area 0 01 . π × (1 , 1 / 2 1 , / 4 1 , / 8 1 , / 16) using a billion timesteps, and found no significant difference in density over that range. Evidently the MKT and HH algorithms behave as though they were ergodic. The difference between those densities and the PB chaotic density was approximately a factor of two, indicating that approximately half the PB measure is chaotic and the other half quasiperiodic, and so omitted from this chaotic simulation.
It is worth mentioning that the MKT equations have two fixed points, though in the end neither one is actually attractive. In the special case that q and p both vanish so that ˙ q and ˙ p are likewise zero, the remaining equations of motion are :
$$\dot { \zeta } = - 1 - \xi \zeta \ ; \ \dot { \xi } = \zeta ^ { 2 } - 1 \.$$
These equations describe a flow from the unstable two-dimensional fixed point ( -1 +1) , to the stable ( only in two dimensions ) fixed point (+1 , -1) . These flow equations are isomorphic to those of a one-dimensional particle in a constant field, with ζ playing the rˆle o of momentum and ξ acting as the friction coefficient. Patra and Bhattacharya's assertion that the MKT equations are not ergodic relies on analyses in the vicinity of these two fixed points. That type of work is complicated by the fact that the four-dimensional flow becomes more and more intermittent as one nears either of these two points. Nevertheless we believe that our analysis in this Section and particularly in Section VII below casts doubt on their claim ( hence our offering of the 2014 Ian Snook Prize on that subject, as is described in Section IX ) .
## V. PERIODIC ORBITS
Choosing an initial condition somewhat closer to the origin
$$\{ \ q, p, \zeta, \xi \ \} = \{ \ 0. 6 8, 0. 6 8, 0, 0 \ \}$$
gives ( q, p, ζ ) projection plots which 'look' spherical for the MKT and HH equations. The PB equations, on the other hand, provide a clearly-defined torus, shown in Figure 2 . The PB dynamics has been projected into three-dimensional ( q, p, ζ ) space. The PB projection shows that the phase-space distribution has at least one large cavity in the chaotic sea and that the cavity contains a family of nested tori.
## VI. ENSEMBLE TESTS
Because a lack of ergodicity implies segregated regions in the ( q, p, ζ, ξ ) space it is a tempting idea to study ensembles of initial conditions, expecting to find two or more distinct longtime-average values rather than a Gaussian Central-Limit-Theorem distribution around a single ergodic average value. In addition to moments, and their correlations, the largest Lyapunov exponent should be a particularly 'good' property to follow. Although some complicated attractors ( such as Rayleigh-B´nard flows ) have more than a single chaotic e strange-attractor region such a situation is implausible for the simple oscillator. Therefore
FIG. 2: Patra-Bhattacharya torus with initial conditions ( q, p, ζ, ξ ) = (0 68 0 68 0 0). . , . , , Choosing (0.5,0.5,0.0,0.0) generates a very thin torus close to the periodic orbit at the center of the larger torus shown in the figure. 50,000 points are plotted.
it seems likely that a random or a grid-based selection of initial points would produce at least a bimodal distribution of values for a nonergodic set of equations. Some ensemble tests of this kind were carried out in References 6 and 13.
## VII. MANY LYAPUNOV EXPONENTS
Taking up the ensemble idea we choose 1000 different initial conditions in the fourdimensional hypercube of sidelength four centered on the origin, following each of them for 10 6 timesteps. Though there is no reason for the HH and MKT exponents to agree there turns out to be good agreement between them, as Figure 3 shows. On the other hand the PB equations' Lyapunov exponents separate into two distinct values, 0.0000 and 0.14, characteristic of their separate portions of the phase space. We view the density and Lyapunov tests as the most convincing evidence that both the MKT and the HH algorithms are ergodic.
## VIII. HISTORY OF THE EXTREMAL LYAPUNOV EXPONENTS
To nail this conclusion down we took the two extreme cases ( the maximum and minimum Lyapunov exponent out of 1000 simulations ) and ran them 1000 times longer. Figure 3 shows the behavior of the ensemble of initial conditions for the Martyna-Klein-Tuckerman and Hoover-Holian models. Figure 4 shows a comparison of the maximum and minimum
## Exponents for Oscillator Models
FIG. 3: Time-averaged Lyapunov exponents for the Martyna-Klein-Tuckerman, Hoover-Holian, and Patra-Bhattacharya oscillators. 1000 trajectories are traced from points located randomly within a 4 × 4 × 4 × 4 hypercube centered at the coordinate origin. Although there is no reason that the MKT and HH results should agree, or nearly so, the two models have similar Lyapunov exponents. The Patra-Bhattacharya motion equations lead to two clumps of exponents at long times, one at 0.0000 and the other at 0.14.
Lyapunov-exponent calculations, carried out for a time 1000 times longer. The MKT and HH data show no significant difference between them, strongly suggesting that the entire set of random trajectories started in a 4 × 4 × 4 × 4 hypercube are part of the same chaotic sea, filling all of space in a Gaussian manner for the Martyna-Klein-Tuckerman and HooverHolian oscillators.
## IX. FURTHER TESTS
Patra and Bhattacharya also determined phase-space density numerically, in one- or twodimensional cross sections 15 . They also computed mean deviations of long-time-averaged distributions from the canonical one as averages. Their final conclusion is that the MartynaKlein-Tuckerman algorithm is not ergodic. Our own evidence for that same problem points in
## Exponents for Oscillator Models
FIG. 4: Time-averaged Lyapunov exponents for two runs 1000 times longer than those in Figure 3. The initial conditions were chosen to match those of the minimum and maximum values shown at the righthand margin of Figure 3. The convergence of the two outliers toward a common value is a strong indicator of ergodicity.
the opposite direction, toward ergodicity. We thank Puneet Patra, Baidurya Bhattacharya, and Clint Sprott for an exchange of hundreds of stimulating emails on this general subject, and specially appreciated a personal visit from Professor Bhattacharya this summer.
## X. CONCLUSIONS - IAN SNOOK PRIZE FOR 2014
The disagreement between our own investigations past and present ( which agree with those of Martyna, Klein, and Tuckerman ) and those of Patra and Bhattacharya are both thought-provoking and stimulating. Last year we offered a prize to honor the memory of our Australian Colleague Ian Snook. We asked arXiv readers for a time-reversed version of a simple pseudorandom number generator with a cycle length of 2 22 . Within 24 hours Professor Federico Ricci-Tersenghi ( University of Rome ) won the prize with the elegant solution described in his arXiv.1305.1805 contribution 16 , 'The Solution to the Challenge in 'Time-Reversible Random Number Generators' by Wm. G. Hoover and Carol G. Hoover'. His generator can be used, for instance, to reverse Brownian dynamics trajectories, a demonstration problem in our forthcoming book on the management and control of nonequilibrium systems.
FIG. 5: Shuichi Nos´ ( 1951-2005 ) and Ian Snook ( 1945-2013 ) e
This year we invite the readers of Computational Methods in Science and Technology to consider thoughtfully this interesting problem in classical statistical mechanics and dynamical systems theory. Specifically the Ian Snook Prize for 2014 will be awarded to whomever submits the most convincing treatment of this problem to us prior to 1 January 2015. The question to be discussed is 'To what extent are trajectory-based solutions of the MartynaKlein-Tuckerman harmonic-oscillator motion equations ergodic?'
It is our intention to reward the most convincing entry received ( or submitted to the arXiv or to Computational Methods in Science and Technology ) prior to 1 January 2015. The 2014 Ian Snook prize of five hundred United States dollars will be presented to the winner in January 2015. We would be very grateful for your contributions. We dedicate this work to the memories of our two colleagues shown in Figure 5 .
## XI. ACKNOWLEDGMENTS
We thank Puneet Patra and Baidurya Bhattacharya for bringing this problem to our attention and additionally thank Clint Sprott for his insights into a wide variety of fascinating problems in dynamical systems theory and for his thoughts on the current prize offer.
- 1 S. Nos´, 'A Unified Formulation of the Constant Temperature Molecular Dynamics Methods', e Journal of Chemical Physics 81 , 511-519 (1984).
- 2 S. Nos´ e, 'Constant Temperature Molecular Dynamics Methods', Progress in Theoretical Physics Supplement 103 , 1-46 (1991).
- 3 Wm. G. Hoover, 'Canonical Dynamics: Equilibrium Phase-Space Distributions', Physical Review A 31 , 1695-1697 (1985).
- 4 C. P. Dettmann and G. P. Morriss, 'Hamiltonian Formulation of the Gaussian Isokinetic Thermostat', Physical Review E 54 , 2495-2500 (1996).
- 5 H. A. Posch, W. G. Hoover, and F. J. Vesely, 'Canonical Dynamics of the Nos´ e Oscillator: Stability, Order, and Chaos', Physical Review A 33 , 4253-4265 (1986).
- 6 H. A. Posch and Wm. G. Hoover, 'Time-Reversible Dissipative Attractors in Three and Four Phase-Space Dimensions', Physical Review E 55 , 6803-6810 (1997).
- 7 D. Kusnezov, A. Bulgac, and W. Bauer, 'Canonical Ensembles from Chaos', Annals of Physics 204 , 155-185 (1990).
- 8 D. Kusnezov and A. Bulgac, 'Canonical Ensembles from Chaos: Constrained Dynamical Systems', Annals of Physics 214 , 180-218 (1992).
- 9 G. J. Martyna, M. L. Klein, and M. Tuckerman, 'Nos´-Hoover Chains-the Canonical Ensemble e via Continuous Dynamics', Journal of Chemical Physics 97 , 2635-2643 (1992).
- 10 Wm. G. Hoover and B. L. Holian, 'Kinetic Moments Method for the Canonical Ensemble Distribution', Physics Letters A 211 , 253-257 (1996).
- 11 P. K. Patra and B. Bhattacharya, 'A Deterministic Thermostat for Controlling Temperature using All Degrees of Freedom', Journal of Chemical Physics 140 , 064106 (2014).
- 12 K. P. Travis and C. Braga, 'Configurational Temperature and Pressure Molecular Dynamics: Review of Current Methodology and Applications to the Shear Flow of a Simple Fluid',
Molecular Physics 104 , 3735-3749 (2006).
- 13 Wm. G. Hoover, Carol Hoover, and Dennis Isbister, 'Chaos, Ergodic Convergence, and Fractal Instability for a Thermostated Canonical Harmonic Oscillator', Physical Review E 63 , 35413546 (2000).
- 14 Wm. G. Hoover and C. G. Hoover, 'Time-Reversible Random Number Generators : Solution of Our Challenge by Federico Ricci-Tersenghi': arXiv.1305.0961.
- 15 Puneet Kumar Patra and Baidurya Bhattacharya, 'Non-Ergodicity of Nos´-Hoover Chain There mostat in Computationally Achievable Time': arXiv.1407.2353.
- 16 F. Ricci-Tersenghi, 'The Solution to the Challenge in 'Time-Reversible Random Number Generators' by Wm. G. Hoover and Carol G. Hoover': arXiv.1305.1805. | 10.12921/cmst.2014.20.03.87-92 | [
"William Graham Hoover",
"Carol Griswold Hoover"
] | 2014-08-01T18:15:52+00:00 | 2014-08-01T18:15:52+00:00 | [
"nlin.CD",
"cond-mat.stat-mech"
] | Ergodicity of a Time-Reversibly Thermostated Harmonic Oscillator and the 2014 Ian Snook Prize | Shuichi Nos\'e opened up a new world of atomistic simulation in 1984. He
formulated a Hamiltonian tailored to generate Gibbs' canonical distribution
dynamically. This clever idea bridged the gap between microcanonical molecular
dynamics and canonical statistical mechanics. Until then the canonical
distribution was explored with Monte Carlo sampling. Nos\'e's dynamical
Hamiltonian bridge requires the "ergodic" support of a space-filling structure
in order to reproduce the entire distribution. For sufficiently small systems,
such as the harmonic oscillator, Nos\'e's dynamical approach failed to agree
with Gibbs' sampling and instead showed a complex structure, partitioned into a
chaotic sea, islands, and chains of islands, that is familiar textbook fare
from investigations of Hamiltonian chaos. In trying to enhance small-system
ergodicity several more complicated "thermostated" equations of motion were
developed. All were consistent with the canonical Gaussian distribution for the
oscillator coordinate and momentum. The ergodicity of the various approaches
has undergone several investigations, with somewhat inconclusive (
contradictory ) results. Here we illustrate several ways to test ergodicity and
challenge the reader to find even more convincing algorithms or an entirely new
approach to this problem. |
1408.0257v2 | ## Efficiencies of Quantum Optical Detectors
Daniel Hogg, 1 Dominic W. Berry, 2 and A. I. Lvovsky 1, 3
1 Institute for Quantum Science and Technology, University of Calgary, Alberta T2N1N4, Canada 2 Department of Physics and Astronomy, Macquarie University, Sydney, NSW 2109, Australia 3 Russian Quantum Centre, 100 Novaya St., Skolkovo, Moscow 143025, Moscow, Russia ∗
(Dated: December 17, 2014)
We propose a definition for the efficiency that can be universally applied to all classes of quantum optical detectors. This definition is based on the maximum amount of optical loss that a physically plausible device can experience while still replicating the properties of a given detector. We prove that detector efficiency cannot be increased using linear optical processing. That is, given a set of detectors, as well as arbitrary linear optical elements and ancillary light sources, it is impossible to construct detection devices that would exhibit higher efficiencies than the initial set.
Optical detectors - devices for converting optical signals into electric ones - are paramount not only in physics, but also in many aspects of our everyday life. In quantum optics and its applications to information processing and communication, detectors are particularly diverse and subject to intense study [1-4] and development [5, 6]. This is necessary to satisfy the demands associated with various methods of quantum-optical state measurement required for different quantum technology applications.
the local oscillator [16]. However, nonlinear optical processing is typically lossy and requires sophisticated technology. In contrast, linear optical elements with very low loss are routinely manufactured. It would be therefore much more preferable to increase the efficiency of a set of detectors using only linear optical elements.
A primary performance benchmark of any optical detector is its quantum efficiency. In spite of its universality and intuitiveness, this fundamental characteristic does not have a uniform definition applicable to all classes of detectors. For example, the quantum efficiency of a photodiode is defined as a ratio of the number of photoelectrons to the number of incident photons; that of a single-photon detector is the probability to generate a 'click' given a single input photon; for a balanced homodyne detector, the efficiency is obtained by means of a relatively complex calculation that includes the efficiencies of its photodiodes, the mode matching of the signal and the local oscillator [7], the electronic noise [8] and other parameters.
A further important outstanding problem is the construction of optical detectors with high efficiency. Efficient optical detection is a primary requirement in applications ranging from quantum information processing and communications [9-11] to fundamental tests such as loophole-free locality violation [12, 13]. Although significant progress has been made in recent years in photon detector technology [5], highly efficient optical detectors remain expensive and unavailable for certain wavelengths. It would therefore be useful to develop means of increasing detector efficiency by optical means. That is, construct an all-optical device involving lower-efficiency detectors that behaves as a higher-efficiency detector.
It is possible to increase photon detection efficiency using non-linear optics. Examples include non-demolition detection [14] and the CNOT gate [15]. For quadrature detection, the efficiency can be enhanced via optical squeezing to amplify the quadrature that is in phase with
In this paper, we introduce a generalized definition of detector efficiency. Thereafter, we are able to address the principal question of this paper: Can the efficiency of a set of detectors be increased using linear optics? We show that the answer to this question is negative.
The efficiency of a photon detector can be defined in terms of equivalent optical loss [17, 18]. That is, for a given detector, we look for equivalent configurations consisting of another detector preceded by a loss channel (attenuator) of transmissivity η [Fig. 1(a)]. The detector efficiency would be the minimum (or infimum) value of η for all such equivalent representations that are theoretically allowed.
Mathematically, the generalized efficiency of a detector with a positive-operator valued measure (POVM) Σ is /vector ˆ given by
$$E ( \vec { \hat { \Sigma } } ) = \inf \left \{ \eta | \exists \vec { \hat { \Pi } }, \mathcal { F } _ { \eta } ( \vec { \hat { \Pi } } ) = \vec { \hat { \Sigma } } \right \}, \quad \quad ( 1 ) \\ \intertext { w o r } \tau / \vec { \hat { \Pi } } \vee \text{www.contest a dotoot on with } D \partial \mathcal { U M } \vec { \hat { \Pi } } \text{ with }$$
where F η ( ˆ ) /vector Π represents a detector with POVM Π with /vector ˆ an attenuator of transmissivity η placed in front of it [Fig. 1(a)]. /vector ˆ Π must be a theoretically allowed POVM, i.e. a set of non-negative self-adjoint operators that sum to unity.
For a detector whose POVM ˆ is known, the loss trans/vector Π formation F η ( ˆ ) /vector Π can be calculated as follows. For an input quantum state ˆ, ρ the probability of obtaining a specific ( /lscript th) measurement outcome is
$$p _ { \ell } ( \hat { \rho } ) & = \text{Tr} [ \hat { \rho } \mathcal { F } _ { \eta } ( \hat { \Pi } _ { \ell } ) ]. \quad \quad ( 2 ) \\. \quad. \quad. \quad. \quad. \quad. \quad.$$
On the other hand, the detection process shown in Fig. 1(a) is equivalent to that in Fig. 1(b), so we can write
$$p _ { \ell } ( \hat { \rho } ) = \text{Tr} [ \mathcal { E } _ { \eta } ( \hat { \rho } ) \hat { \Pi } _ { \ell } ],$$
FIG. 1. Equivalent loss model of detector. (a) An inefficient detector with POVM ˆ Σ and efficiency /vector η is equivalent to a detector with POVM Π preceded by an attenuator with /vector ˆ transmissivity η . (b) An equivalent model of quantum state measurement with an imperfect detector.
where E η (ˆ) is the loss transformation of state ρ ˆ, ρ which in the Fock basis takes the form of the the generalized Bernoulli transformation [19, 20]. This can then be used to derive F η ( ˆ ) /vector Π in that basis as shown in Appendix A. For a diagonal POVM, relevant for phase-insensitive detection, the map takes the form
$$\langle n | \, \mathcal { F } _ { \eta } ( \hat { \Pi } _ { \ell } ) \, | n \rangle = \sum _ { k = 0 } ^ { n } \binom { n } { k } ( 1 - \eta ) ^ { n - k } \eta ^ { k } \, \langle k | \, \hat { \Pi } _ { \ell } \, | k \rangle \,. \quad ( 4 ) \, & \text{physi} \, \\ \tau _ { 0 } \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, & \text{k} \,. \quad \text{k} \,. \quad \text{k} \,. \quad \text{k} \,.$$
To illustrate the results above, let us consider a nondiscriminating single-photon detector of efficiency η , defined as the probability for a single incident photon to generate a 'click'. This detector is described by the wellknown POVM [21, p. 118]
$$\hat { \Pi } _ { \text{off} } ( \eta ) & = \sum _ { n = 0 } ^ { \infty } ( 1 - \eta ) ^ { n } \left | n \right \rangle \left \langle n \right |, \quad \quad ( 5 a ) \quad \text{$2e^{\prime}$} \\ \hat { \rightharpoonup } \quad \cdot \quad \Xi ^ { \infty }. \quad \cdot \quad \cdot \quad \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot$$
$$\hat { \Pi } _ { \text{on} } ( \eta ) = \sum _ { n = 0 } ^ { n = \upsilon } [ 1 - ( 1 - \eta ) ^ { n } ] \left | n \right \rangle \left \langle n \right |. \quad \ \ ( 5 b ) \\. \quad. \quad. \quad. \quad. \quad. \quad. \cdots. \quad t \cdots. \quad t \cdots.$$
If we place an attenuator with transmissivity η ′ in front of that detector, transformation (4) will lead to
$$\mathcal { F } _ { \eta ^ { \prime } } ( \hat { \Pi } _ { \text{off} } ( \eta ) ) = & \sum _ { n = 0 } ^ { \infty } ( 1 - \eta \eta ^ { \prime } ) ^ { n } \left | n \right \rangle \left \langle n \right | = \hat { \Pi } _ { \text{off} } ( \eta \eta ^ { \prime } ), \quad ( 6 a ) \begin{array} { c } \detect \\ \text{specia} \\ \text{to me} \\ \text{Let} \\ \text{to} \end{array} \end{array}$$
$$\mathcal { F } _ { \eta ^ { \prime } } ( \hat { \Pi } _ { \text{on} } ( \eta ) ) = \sum _ { n = 0 } ^ { n = 0 } [ 1 - ( 1 - \eta \eta ^ { \prime } ) ^ { n } ] \left | n \right \rangle \left \langle n \right | = \hat { \Pi } _ { \text{on} } ( \eta \eta ^ { \prime } ), \quad \text{tors } u \\ ( 6 b ) \quad \text{equiv} \\ \text{tenua}$$
i.e. this setting is equivalent to a non-discriminating single-photon detector of efficiency ηη ′ .
By the same token, a detector with POVM (5) is equivalent to a non-discriminating detector of efficiency η/η ′ preceded by an attenuator with transmissivity η ′ . The POVMelements ˆ Π off,on ( η/η ′ ) are non-negative for η ′ ≥ η , and negative (unphysical) for η ′ < η . This implies that the generalized efficiency (1) of detector (5) equals η , so our new definition is consistent with the traditional one. One can use similar arguments to show this consistency for other types of detectors.
The equivalent-loss approach to quantum efficiencies has previously been applied to investigate the question of whether linear optical processing can increase the efficiency of single photon sources. Originally investigated in [22, 23], an explicit definition of source efficiency was constructed, along with a proof that linear optical processing cannot increase it [24]. Later, this proof was extended to the case of multiple sources [25]. Below, we address a similar problem for detectors.
Suppose we are given a set of single-mode detectors with efficiencies { η i } , which we call 'physical'. One may use these detectors, an arbitrary number of linear optical elements, such as beam splitters and phase shifters, as well as any ancillary light sources, to construct a set of single-mode optical state measurement devices which we call 'virtual detectors' [Fig. 2(a)]. Different virtual detectors do not share any optical elements. In particular, each virtual detector uses a different subset of the physical detectors. We show that the efficiencies η ′ i of these virtual detectors are bounded by the efficiencies of the physical detectors:
$$\eta _ { i } ^ { \prime \downarrow } \leq \eta _ { i } ^ { \downarrow }, \quad \ \ \ ( 7 )$$
where the downward arrow denotes sorting in nonincreasing order.
Consider first a single virtual detector. It can be represented by a scheme shown in Fig. 2(a). The mode ˆ a 1 to be measured, as well as the ancillary modes ˆ a 2 ,...,M , are processed by an interferometer. Such an interferometer can be represented in the Heisenberg picture as a unitary transformation W of the input and output mode annihilation operators.
$$\hat { a } _ { i } = \sum _ { j = 1 } ^ { M } W _ { i j } \hat { b } _ { j }. \quad \quad ( 8 ) \\ \cdot \ \hat { \hat { \cdot } } \ \,, \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot$$
The output modes { ˆ b j } may be incident onto the physical detectors or simply discarded. The latter case requires no special treatment because discarding a mode is equivalent to measuring it with a detector with efficiency zero.
Let η max be the highest efficiency of all physical detectors used in a given virtual detector. According to our definition of detector efficiency, the physical detectors are equivalent to higher-efficiency detectors preceded by attenuators of transmissivity η max . As our definition of efficiency uses an infimum, the transmissivity may need to be taken as η max + /epsilon1 for arbitrarily small (but positive) /epsilon1 . It is trivial to include this /epsilon1 , and it does not qualitatively change the proof. We therefore omit it in the following discussion. As demonstrated in Ref. [24], the attenuators can be commuted to precede the interferometer. We now present a simpler argument to that effect.
FIG. 2. A single-mode virtual detector. (a) A generalized model. The mode to be measured and the ancillary modes are processed by an interferometer and impinge onto the physical detectors, whose equivalent models are shown inside the dotted rectangles. (b) An equivalent model of the virtual detector. The input mode is subjected to immediate attenuation, which implies that the detector efficiency cannot exceed η max .
Using the beam splitter model of loss [26], we decompose the modes incident on them as
$$\text{one moves incident on } \text{em as } & \text{to } \text{r} \\ \hat { b } _ { j } & = \sqrt { \eta ^ { \max } } \hat { b } _ { j } ^ { \prime } + \sqrt { 1 - \eta ^ { \max } } \hat { d } _ { j }, & \text{(9)} & \text{c} \\ \hat { v } _ { j } & = \sqrt { 1 - \eta ^ { \max } } \hat { b } _ { j } ^ { \prime } - \sqrt { \eta ^ { \max } } \hat { d } _ { j }, & \text{(10)} & \text{with} \\ & \text{in } \text{f}$$
where modes ˆ d j are discarded. Putting Eqs. (8) and (9) together, we write for the incoming mode
$$o _ { j } & = \sqrt { \eta } ^ { \dots } \dots o _ { j } + \sqrt { 1 } - \eta ^ { \dots } \dots a _ { j }, \quad ( \mathfrak { Y } ) \quad \vec { \text{whi} } \\ \hat { v } _ { j } & = \sqrt { 1 - \eta ^ { \max } } \hat { b } ^ { \prime } _ { j } - \sqrt { \eta ^ { \max } } \hat { d } _ { j }, \quad ( 1 0 ) \quad \text{$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$ \quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\ \text{upsp} \\ \dots \quad \dots \quad \epsilon \quad \dots \quad \epsilon \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \digma \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \dag \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \dangle \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \dod \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \delta \quad \d }$$
with
$$\omega, \, \nu \, \ m v e \, \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ M v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v u \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v o \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v a \ m m \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v e \ m v s \ m m.$$
$$\hat { a } _ { i } ^ { \prime } \coloneqq \sum _ { j = 1 } ^ { N } W _ { i j } \hat { b } _ { j } ^ { \prime } \text{ and } \hat { d } _ { i } ^ { \prime } \coloneqq \sum _ { j = 1 } ^ { N } W _ { i j } \hat { d } _ { j }. \quad ( 1 2 ) \quad \text{ada} _ { I } \text{ have} \\ \vdots \colon \colon \dots \dots \colon \dots \colon \dots \colon \dots \colon \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \end{cases}$$
At the same time, Eq. (10) can be rewritten as
$$\hat { v } _ { i } ^ { \prime } \coloneqq \sum _ { j = 1 } ^ { N } W _ { i j } \hat { v } _ { j } = \sqrt { 1 - \eta ^ { \max } } \hat { a } _ { i } ^ { \prime } - \sqrt { \eta ^ { \max } } \hat { d } _ { j } ^ { \prime }. \ \ ( 1 3 ) \quad \text{inter}. \\ \text{These results imnlv that the setun of Fij} \ 2 ( a ) \text{ is eniv}. \quad \text{$one$}$$
These results imply that the setup of Fig. 2(a) is equivalent to that of Fig. 2(b). Indeed, Eqs. (11) and (13) signify a beam splitter transformation. Additionally, modes
1
FIG. 3. (Color online) A single-mode virtual detector with adaptive measurements for M = 7. The virtual detector is again shown inside the dotted rectangle. The dotted lines from the detectors to the interferometers indicate that the interferometers are controlled based on the results of the detections.
ˆ v ′ i can be obtained from vacuum modes ˆ v j by means of a linear optical transformation, so they must also be in the vacuum state. Modes ˆ d ′ i are related in a similar fashion to modes ˆ , and so can be treated as discarded. d j
Our virtual detector is hence equivalent to a setup in which an attenuator with transmissivity η max is placed in front of input mode ˆ a 1 . This implies that η max is an upper bound for the efficiency of this detector.
This result is readily extended to virtual detectors with adaptive measurements; that is, detectors in which the configuration of interferometer W can be modified dependent on the results of the measurements by a subset of the physical detectors. To see this, we model the adaptive virtual detector as shown in Fig. 3. We would have the interferometer W M on modes 1 , . . . , M , followed by interferometer W M -1 on modes 1 , . . . , M -1, and so forth up to interferometer W 2 on modes 1 and 2. Each interferometer W k on modes 1 , . . . , k would depend on the results of measurements of modes k +1 , . . . , M . This interferometer would then be followed by a measurement on mode k .
As discussed above, each physical detector can be modeled as having a beam splitter with transmissivity η max before it. In the same way as above, the beam splitters
after interferometer W 2 can be commuted to before W 2 , so that there are three equal reflectivity beam splitters after interferometer W 3 . These can then be commuted through W 3 , and so forth, until we have beam splitters with transmissivity η max before the first interferometer W M . Although the interferometers W 2 to W M -1 can depend on measurement results, the loss commutes independently of the interferometers. This means that the above argument holds, and the efficiency for this virtual detector cannot exceed η max .
We now proceed to proving Eq. (7) for multiple virtual detectors. We have so far shown that the efficiency η ′ i of the i th virtual detector cannot exceed the efficiency η max i of the best physical detector used in its construction; that is, η ′ i ≤ η max i . Let σ i ( ) and τ ( ) be the permui tations that define the sorting of sequences η max i and η ′ i in non-increasing order: η max ↓ σ i ( ) = η max i and η ′ ↓ i = η ′ τ ( ) i , respectively. Then we obtain
$$\eta _ { i } ^ { \prime \downarrow } = \eta _ { \tau ( i ) } ^ { \prime } \leq \eta _ { \tau ( i ) } ^ { \max } = \eta _ { \sigma ( \tau ( i ) ) } ^ { \max \downarrow } \quad \quad ( 1 4 ) \quad [ 2 1 ]$$
Now consider the case that Eq. (7) were violated; that is η ′ ↓ i > η ↓ i . That would imply that η ′ k ↓ > η ↓ i for all k ≤ i , which would in turn imply that η max ↓ σ τ ( ( k )) > η ↓ i for all k ≤ i . But, that would imply that there are i values of η ↓ k that are larger than η ↓ i , which is a contradiction, because η ↓ i is sorted in non-ascending order. This contradiction implies that Eq. (7) must hold.
Linear optics are cheap and easy to manufacture. In addition, their properties are well-understood, and linear processes in general have high efficiency. It would be extremely fortunate if we could somehow use these processes to increase the efficiency of sources or detectors, but this is not the case. Nonlinear optics appears to be the only alternative.
This work has been supported by NSERC. We thank B. C. Sanders and A. S. Prasad for helpful discussions. DWB is funded by an Australian Research Council Future Fellowship (FT100100761).
- ∗ [email protected]
- [1] J. Lundeen et al. , Nature Phys. 5 , 27 (2008).
- [2] L. Zhang et al. , Nature Photon. 6 , 364 (2012).
- [3] R. H. Hadfield, Nature Photon. 3 , 696 (2009).
- [4] M. D. Eisaman, J. Fan, A. Migdall, and S. V. Polyakov, Rev. Sci. Instrum. 82 , 071101 (2011).
- [5] G. Buller and R. Collins, Meas. Sci. Technol. 21 , 012002 (2010).
- [6] D. Achilles, C. Silberhorn, C. Sliwa, K. Banaszek, ´ and I. A. Walmsley, Opt. Lett. 28 , 2387 (2003).
- [7] A. I. Lvovsky and M. G. Raymer, Rev. Mod. Phys. 81 , 299 (2009).
- [8] J. Appel, D. Hoffman, E. Figueroa, and A. I. Lvovsky, Phys. Rev. A 75 , 035802 (2007).
- [9] M. Varnava, D. E. Browne, and T. Rudolph, Phys. Rev. Lett. 100 , 060502 (2008).
- [10] A. Scherer, B. C. Sanders, and W. Tittel, Opt. Express 19 , 3004 (2011).
- [11] M. Koashi, arXiv preprint quant-ph/0609180 (2006).
- [12] M. Giustina et al. , Nature 497 , 227 (2013).
- [13] B. G. Christensen, K. T. McCusker, J. B. Altepeter, B. Calkins, T. Gerrits, A. E. Lita, A. Miller, L. K. Shalm, Y. Zhang, S. W. Nam, N. Brunner, C. C. W. Lim, N. Gisin, and P. G. Kwiat, Phys. Rev. Lett. 111 , 130406 (2013).
- [14] W. J. Munro, K. Nemoto, R. G. Beausoleil, and T. P. Spiller, Phys. Rev. A 71 , 033819 (2005).
- [15] K. L. Brown, J. H. M. Plaskus, H. E. Broadu, and J. P. Dowling, arXiv:1308.2645 (2013).
- [16] U. Leonhardt and H. Paul, Phys. Rev. Lett. 72 , 4086 (1994).
- [17] J. J. Renema et al. , Opt. Express 20 , 2806 (2012).
- [18] B. Calkins et al. , Opt. Express 21 , 22657 (2013).
- [19] C. T. Lee, Phys. Rev. A 48 , 2285 (1993).
- [20] T. Kiss, U. Herzog, and U. Leonhardt, Phys. Rev. A 52 , 2433 (1995).
- [21] P. K. Kok and B. W. Lovett, Introduction to Optical Quantum Information Processing (Cambridge University Press, 2010).
- [22] D. W. Berry et al. , New J. Phys. 6 , 93 (2004).
- [23] D. W. Berry, S. Scheel, B. C. Sanders, and P. L. Knight, Phys. Rev. A 69 , 031806 (2004).
- [24] D. W. Berry and A. I. Lvovsky, Phys. Rev. Lett. 105 , 203601 (2010).
- [25] D. W. Berry and A. I. Lvovsky, Phys. Rev. A 84 , 042304 (2011).
- [26] U. Leonhardt, Measuring the Quantum State of Light , Cambridge Studies in Modern Optics (Cambridge University Press, 1997).
## Appendix A: POVM loss transformation
Here we determine an expression for the map F η on the detector POVM under loss. For full generality, we derive the transformation for the case where the POVM is not diagonal, so elements can be written in the Fock basis as
$$\hat { \Pi } _ { \ell } = \sum _ { m, n = 0 } ^ { \infty } ( \hat { \Pi } _ { \ell } ) _ { m n } \left | m \right \rangle \left \langle n \right |. \quad \quad ( 1 5 ) \\.. \quad \dots \quad. \dots \quad - \quad. \quad.$$
Using conditions (2) and (3) on F η , taking the trace over both sides in the Fock basis and inserting identity matrices yields
$$\L \cdots \, \cdots \\ \sum _ { m = 0 } ^ { \infty } \sum _ { n = 0 } ^ { \infty } \langle m | \, \hat { \rho } \, | n \rangle \, \langle n | \, \mathcal { F } _ { \eta } ( \hat { \Pi } _ { \ell } ) \, | m \rangle \\ = \sum _ { m = 0 } ^ { \infty } \sum _ { n = 0 } ^ { \infty } \langle m | \, \mathcal { E } _ { \eta } ( \hat { \rho } ) \, | n \rangle \, \langle n | \, \hat { \Pi } _ { \ell } \, | m \rangle \,. \quad \ ( 1 6 ) \\ \text{stituting the expression for} \, \mathcal { E } _ { n } ( \hat { \rho } ) \, \text{from the generalized}$$
Substituting the expression for E η (ˆ) from the generalized ρ Bernoulli transformation [19, 20] then gives
$$& \sum _ { m = 0 } ^ { \infty } \sum _ { n = 0 } ^ { \infty } \langle m | \hat { \rho } | n \rangle \langle n | \mathcal { F } _ { \eta } ( \hat { \Pi } _ { \ell } ) | m \rangle \\ & = \sum _ { m = 0 } ^ { \infty } \sum _ { n = 0 } ^ { \infty } \sum _ { k = 0 } ^ { \infty } \langle m + k | \hat { \rho } | n + k \rangle \sqrt { \binom { m + k } { k } } \binom { n + k } { k } ( 1 - \eta ) ^ { k } \eta ^ { \frac { k } { 2 } ( m + n ) } \langle n | \hat { \Pi } _ { \ell } | m \rangle \\ & = \sum _ { m = 0 } ^ { \infty } \sum _ { k = 0 } ^ { \min ( m, n ) } \sum _ { k = 0 } ^ { \min ( m, n ) } \langle m | \hat { \rho } | n \rangle \sqrt { \binom { m } { k } } ( 1 - \eta ) ^ { k } \eta ^ { \frac { k } { 2 } ( m + n ) - k } \langle n - k | \hat { \Pi } _ { \ell } | m - k \rangle \\ \intertext { nce the elements } \langle m | \hat { \rho } | n \rangle \, \text{are arbitrary,} \, \mathcal { F } _ { \eta } \, \text{ yields a transformation described by}$$
Since the elements 〈 m ρ n | ˆ | 〉 are arbitrary, F η yields a transformation described by
$$\langle n | \, \mathcal { F } _ { \eta } ( \hat { \Pi } _ { \ell } ) \, | m \rangle = \sum _ { k = 0 } ^ { \min ( m, n ) } \sqrt { \binom { m } { k } } \binom { n } { k } ( 1 - \eta ) ^ { k } \eta ^ { \frac { 1 } { 2 } ( m + n ) - k } \, \langle n - k | \, \hat { \Pi } _ { \ell } \, | m - k \rangle \,.$$ | 10.1103/PhysRevA.90.053846 | [
"Daniel Hogg",
"Dominic W. Berry",
"A. I. Lvovsky"
] | 2014-08-01T18:21:13+00:00 | 2014-12-15T23:44:08+00:00 | [
"quant-ph"
] | Efficiencies of Quantum Optical Detectors | We propose a definition for the efficiency that can be universally applied to
all classes of quantum optical detectors. This definition is based on the
maximum amount of optical loss that a physically plausible device can
experience while still replicating the properties of a given detector. We prove
that detector efficiency cannot be increased using linear optical processing.
That is, given a set of detectors, as well as arbitrary linear optical elements
and ancillary light sources, it is impossible to construct detection devices
that would exhibit higher efficiencies than the initial set. |
1408.0258v1 | ## The Pricing War Continues: On Competitive Multi-Item Pricing
Omer Lev , Joel Oren , Craig Boutilier 1 2 2 , and Jeffery S. Rosenschein 1
1 The Hebrew University, Israel omerl,[email protected] 2 University of Toronto, Canada oren,[email protected]
Abstract. We study a game with strategic vendors who own multiple items and a single buyer with a submodular valuation function. The goal of the vendors is to maximize their revenue via pricing of the items, given that the buyer will buy the set of items that maximizes his net payoff.
We show this game may not always have a pure Nash equilibrium, in contrast to previous results for the special case where each vendor owns a single item. We do so by relating our game to an intermediate, discrete game in which the vendors only choose the available items, and their prices are set exogenously afterwards. We further make use of the intermediate game to provide tight bounds on the price of anarchy for the subset games that have pure Nash equilibria; we find that the optimal PoA reached in the previous special cases does not hold, but only a logarithmic one.
Finally, we show that for a special case of submodular functions, efficient pure Nash equilibria always exist.
## 1 Introduction
Consider a scenario in the world of e-commerce, where a single consumer is seeking to buy a set of products through an online website with multiple vendors, such as Amazon or eBay. Given the items available for sale and their prices, the buyer will purchase some subset of them, according to his valuation of the items and their prices.
The vendors strive to maximize their profits-for ease of exposition we assume production costs are zero, hence profits can be equated with revenues, or the sum of the prices of their sold items. A vendor can both competitively tailor the set of items it offers and adjust the prices of these items to react to their competitors. 3 Indeed, automatic mechanisms for rapid online price optimization exist in a variety of markets and industries [1]. This practice, sometimes called competitive price intelligence [2], is a growing phenomenon within online retail. The specific question that it addresses is how a company should price its products in this competitive environment.
Furthermore, as argued by Babaioff et al. [3], such a setting introduces subtle algorithmic questions, since changing the prices of the products may affect the resulting revenues in a complex fashion, which may induce responses by one's competitors. Therefore, studying the convergence properties of such pricing dynamics is of interest.
3 Note that pricing an item sufficiently high can be regarded as not offering it.
In this paper, we take a game-theoretic approach to price competition among multiple sellers, each with a set of multiple items. As in Babaioff et al. [3], we study a setting with a single buyer with a (combinatorial) valuation function, taken to be a monotone and submodular set function over the set of items, which is fully known to the vendors. However, unlike that earlier work, we examine the more general case where each of the k vendors controls a disjoint set of items A i , rather than a single item. Given the prices of all of the items, the buyer will buy the set with the highest net-payoff (valuation minus the total price). Our model induces a game in which each of the vendors' strategy is a pricing of their items.
Contributions We begin by discussing a related two-phase game, that serves as a waystation in our study of the main game. In this intermediate game, vendors can only modify the sets of items being offered, whereas their prices for these items are subsequently set by a specific pricing mechanism. We show that this game, resulting only from this modification of game dynamics (without changing its parameters), has two critical properties. First, any strategy profile in the original game (a price vector) has a corresponding strategy profile that results in at least as much revenue for each of the vendors (Proposition 4.2).
More importantly, we show that for any pure Nash equilibrium in the original game, there is a corresponding pure Nash equilibrium in the intermediate game, in which each vendor sells the same items at the same prices as in the original equilibrium (Theorem 4.3). Hence, we reduce the pricing decision to a decision over what items to sell, allowing a significant simplification of the problem.
We next study basic game theoretic properties of our game. We first show that there are games which admit no pure Nash equilibrium. To do so, we show that our twophase game admits no pure Nash equilibria, which then implies the nonexistence of a pure Nash equilibrium in the original game, using Theorem 4.3 (see Proposition 5.1). This result suggests the following question: suppose we restrict attention to instances of the game that have some pure Nash equilibria-can we then say anything about their value? To accomplish this, we analyze the price of anarchy (PoA) of this subset of games, where the objective function in question is the social welfare value, taken to be simply the buyer's valuation of the set of items that he purchased. We provide a tight bound of Θ (log m ) , where m is the maximal number of items controlled by any of the vendors.
Finally, as an additional way of dealing with the consequences of Proposition 5.1, we investigate a special class of valuation functions, which we call category-substitutable , that, informally, partition products into 'equivalence classes' or categories, such that only a single item will be chosen within a specific category, while different categories do not influence one another. We show not only that efficient pure Nash equilibria always exist given such buyer valuations, but also provide a precise characterization of such equilibria.
## 2 Previous Work
Multi-item pricing has been a significant topic of research for many years [4, 5], including analyses of the price of anarchy (see [6] and follow-up papers). The work of
Babaioff et al. [3] is the most directly related to the model developed here; indeed, the game that they study is a special case of our game, in which each vendor sells only a single item. They show that for a buyer with a general valuation function, a pure Nash equilibria may not exist, though they prove several properties of the equilibria for games where one does. Furthermore, for submodular valuation functions (the focus of our paper), they show not only that pure Nash equilibria exist, but that they are unique 4 and efficient-i.e., have a price of anarchy (PoA) of one. In contrast to their setting, we show that in our more general case there exist games with no pure Nash equilibria. In cases where they do, we provide a characterization similar to theirs, though in our more general case, the PoA is significantly higher.
Non-competitive (i.e., single-vendor) optimal-pricing problems have been studied in the theoretical computer science community. [7] study a number of settings with multiple buyers possessing various valuation functions. They show that even with unitdemand buyers and an unlimited supply of each item, selecting the optimal price vector is APX-hard; they then provide a logarithmic approximation algorithm for the same case. It should be noted that Babaioff et al. also provide a log n approximation algorithm for the case of a single vendor, and for that of a single buyer with a submodular valuation function.
In a recent paper, Oren et al. [8] analyze a model in which fixed prices are given exogenously, and there are multiple unit-demand buyers. As above, their model assumes an unlimited supply of each item. The strategies of the vendors are which sets of items to sell. Having the vendor make decisions only about the set of items to sell has traditionally been studied in the field of operation research. In particular, assortment optimization [9] deals with optimizing a seller's 'assortment' (e.g., his catalog, or shelf), under various circumstances. Although our game does not fall directly in this category, we do define a discrete game in which the vendors' decisions are similar to those in assortment optimization (although the pricing procedure differs).
## 3 Preliminaries - The Vendor Competition (VC) Game
We consider the following setting: there is a set of k vendors, with a corresponding vector of pairwise-disjoint sets of items A = ( A , . . . , A 1 k ) , such that | A i | = n i , and n = ∑ k i =1 n i . We let A ∗ = ⋃ k i =1 A i .
A strategy profile of the vendors is a price vector p ∈ R n + , where p a ( ) denotes the price of item a according to p . For a set S ⊆ A ∗ , we let p S ( ) = ∑ a ∈ S p a ( ) . For a vendor i ∈ [ k ] , we let p i ∈ R n i + denote vendor i 's price vector for the items in A i , and as before, for an item a ∈ A i , p i ( a ) denotes vendor i 's price for item a according to p i . For convenience, we will let p -i denote the price vector corresponding to the items not in A i (of all other vendors).
The buyer's valuation function We assume that there is a single buyer with a valuation function v : 2 A ∗ → R + ; that is, the function v ( ) · assigns a non-negative value to every
4 More precisely, they give a closed-form characterization of the prices of the items that are sold .
bundle (or subset) of items. We let m S a ( ) = v S ( ∪ { a } ) -v S ( ) , where S ⊆ A ∗ , denote the marginal contribution of item a to the set S . Following [3], we assume that v ( ) · is non-decreasing : for S ⊆ T ⊆ A ∗ , v S ( ) ≤ v T ( ) (implying that v ( ) · is maximized at A ∗ ). Furthermore, we assume that the valuation function is submodular : for S ⊆ T ⊆ A ∗ and a ∈ A ∗ \ T , we have that m S a ( ) ≥ m T a ( ) . Both of these assumptions are central in the model proposed by Babaioff et al. [3] (although additional discussion and results are provided for non-submodular functions as well). Note that v ( ) · is said to be submodular if the following, equivalent property holds:
$$v ( S ) + v ( T ) \geq v ( S \cup T ) + v ( S \cap T ), \text{ \ for all } S \subseteq T \subseteq A ^ { * }.$$
Finally, slightly abusing notation, let the valuation be defined over vectors of item sets as follows: for S ∗ ⊆ A ∗ , define S = ( S , . . . , S 1 k ) where S i = S ∗ ∩ A i , for i = 1 , . . . , k . Then v ( S ) = v S ( ∗ ) . We adapt the rest of our function definitions in an analogous fashion.
The buyer is assumed to have a quasi-linear utility function: given a vector of prices p ∈ R n + , the buyer's utility for a bundle S ⊆ A ∗ is u b ( S, p ) = v b ( S ) -∑ a ∈ S p a ( ) . The demand correspondence of the buyer is the family of sets that maximizes his utility:
$$D ( v ; \mathbf p ) = \{ S \subseteq A ^ { * } \, \colon u _ { b } ( S ) \geq u _ { b } ( S ^ { \prime } ), \forall S ^ { \prime } \subseteq A ^ { * } \}.$$
The buyer's decision function X v ( ; p ) ⊆ 2 A ∗ must satisfy X v ( ; p ) ∈ D v ( ; p ) . That is, given the price vector p , the buyer buys the bundle X v ( ; p ) . The buyer's decision is said to be maximal (or simply, the buyer is maximal), if there does not exist a set ˜ X ∈ D v ( ; p ) such that X v ( ; p ) /subsetnoteql ˜ X . Babaioff et al. [3] showed this property is critical to ensure the existence of a pure Nash equilibrium in their setting. In our work, we will explicitly state where this property is required.
Vendor payoffs Given the buyer's decision function X , and a (fixed) price vector p = ( p , i p -i ) , vendor i 's utility is u X i ( p ) = ∑ a ∈ X v ( ; p ) ∩ A i p a ( ) . If the vendors select mixed strategies, then a vendor's utility is defined to be his expected utility. Vendor j 's best response to the other agents' mixed strategies is a distribution over prices for A j that maximizes his expected utility.
This setup defines a game, parameterized by the vector A and the valuation function v , in which each of the vendors prices his items to maximize his utility. We will refer to such a game as a vendor competition game , or simply a VC game .
When discussing our special case, in Section 6, we will also make use of the following theorem, which was proved by Babaioff et al.:
Theorem 3.1 ([3]). Consider the case where each vendor owns a single item, that is n i = 1 for every i = 1 , . . . , k ; that is, A i = { a i } , for i = 1 , . . . , k . Then if the buyer's valuation function v ( ) · is non-decreasing and submodular, then there exists a pure Nash equilibrium, p ∈ R k + , of the following form: for every vendor i , such that m a i ( A ∗ \ a i ) > 0 , p a ( i ) = m a i ( A ∗ \ a i ) , and a i ∈ X v ( ; p ) . Also, the payoff of each vendor i is precisely m a i ( A ∗ \ a i ) .
The objective function Given a VC game G = ( v, S ) and pricing vector p ∈ R n + , we use the standard definition of social welfare, namely, the total payoff of all the parties in the game, including the (non-strategic) buyer. Notice that by this definition, social welfare is simply the valuation of the set bought by the buyer, v X v ( ( ; p )) , since all payments are simply transferred from the buyer to the vendors. We let f ( p ) denote the social welfare resulting from a price vector p .
## 4 A Related Discrete Game
The game in its current formulation may seem somewhat hard to reason about, due to large (continuous) strategy spaces. 5 To simplify our analysis, we use the following discrete game, which can be thought of as imposing a specific pricing mechanism given the vendors' selection of items (the design of the pricing scheme is influenced by the results of [3]).
Definition 4.1 (The price-moderated VC game). Given a buyer valuation function over the vendors' items, consider the following two-round process:
- 1. Each vendor i ∈ [ k ] commits to offering a subset of S i ⊆ A i of items; this is its (discrete) strategy;
- 2. Given the strategy vector S = ( S , . . . , S 1 k ) , item prices are set to be their marginal values. That is, if we set S ∗ = ⋃ k i =1 S i , then for each a ∈ S ∗ , the mechanism will set ˜( p a ) = m S a ( ∗ \ { a } ) . For each item a ′ / ∈ S ∗ the mechanism sets ˜( p a ′ ) = v A ( ∗ ) + 1 . Let ˜ p be the resulting price vector.
The consumer then buys the set X v ( ; ˜ p ) , as before. We call the resulting game a pricemoderated vendor competition game, or more succinctly, a PMVC game.
By analogy to our definitions for the original game, let X v ′ ( ; S ) denote the set of items sold, given the strategy profile S . That is, given the price vector ˜ p imposed by the pricing mechanism in the second round, X v ′ ( ; S ) = X v ( ; ˜ p ) . We similarly define a vendor's utility to be u ′ i ( S , i S -i ) , for i ∈ [ k ] .
Note that the specified pricing ( v A ( ∗ + 1) ) of items not offered (i.e., not in S ∗ ) ensures that the consumer will never buy them (i.e., X v ( ; ˜ p ) ⊆ S ∗ ). Further observe that the set of price vectors ˜ p that correspond to the discrete strategy profiles S in the PMVC game is a strict subset of the strategy space in the original VC game. We justify our use of this game in our analysis by establishing the relationship between the original VC game and the proposed PMVC game, using a number of straightforward results.
Assumption For ease of exposition, we assume that the buyer is maximal. As we shall see, this implies that X v ′ ( ; S ) = X v ( ; p ) . However, we can adapt the pricing mechanism by judiciously setting the prices to be slightly below the marginal contributions to ensure maximality (we leave the details of such a modification to an expanded version
5 In particular, the game is clearly not in normal form. As such, we cannot directly apply Nash's theorem about the existence of a mixed equilibrium. We defer the treatment of such equilibria to future study.
of the paper). We now describe an important relationship between the VC and PMVC game that simplifies our subsequent analysis by relating our original model to to a simpler discrete game:
Proposition 4.2. For every strategy profile p in the VC game and valuation v , there is a strategy profile S in the PMVC game such that X v ′ ( ; S ) = X v ( ; p ) , and u ′ i ( S , i S -i ) ≥ u i ( p , i p -i ) for each vendor i .
Proof. Let p be a strategy profile in the VC game, and let T = X v ( ; p ) . Consider the strategy profile S where S i = X v ( ; p ) ∩ A i , for i = 1 , . . . , k , and let ˜ p be the resulting price vector imposed by the pricing mechanism. Furthermore, we let ˜ = T X v ′ ( ; S ) . We begin by showing that T = ˜ T . First, notice that, as for all a / T ∈ , ˜( p a ) = v A ( ∗ ) + 1 , and hence item a is not sold, and ˜ T ⊆ T . Next, suppose for the sake of contradiction that ˜ T /subsetnoteql T , and let a ∈ T \ ˜ T . By the submodularity of the function v ( ) · , we have that m T a ( ˜ ) ≥ m T a ( ) ≥ 0 . This implies that
$$- \quad \Phi _ { \Phi } ^ { \prime } - \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
\quad u _ { b } ( \tilde { T } \cup a, \tilde { \mathbf p } ) & = v ( \tilde { T } \cup a ) - \sum _ { a ^ { \prime } \in \tilde { T } } \tilde { \mathbf p } ( a ^ { \prime } ) - m _ { a } ( \tilde { T } ) \\ & = u _ { b } ( \tilde { T }, \tilde { \mathbf p } ). \\ \cdot \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Philon } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Psi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phicon } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phphi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad &\quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } & \quad \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phid } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & = \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Ph i } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad & \quad \Phi _ { \Phi } ^ { \prime } \quad &.$$
By maximality, the buyer would rather buy item a as well, resulting in a contradiction.
We now claim that u ′ i ( S , i S -i ) ≥ u i ( p , i p -i ) . This follows from the fact that marginal contributions are the maximal prices at which the buyer still buys X v ′ ( ; ˜ p ) . That is, any increase in the price would result in the buyer not buying the product:
$$u _ { b } ( \tilde { T }, \tilde { \mathbf p } ) & = v ( \tilde { T } ) - \sum _ { a ^ { \prime } \in \tilde { T } \searrow a } \tilde { p } ( a ^ { \prime } ) - m _ { a } ( T \, \bar { \ } a ) \\ & = v ( \tilde { T } \, \bar { \ } a ) - \sum _ { a ^ { \prime } \in \tilde { T } \searrow a } \tilde { p } ( a ^ { \prime } ) = u _ { b } ( \tilde { T } \, \bar { \ } a, \tilde { \mathbf p } )$$
/intersectionsq
/unionsq
Similarly to the previous proposition, which offered a mapping of strategy profiles in a way that does not cause the vendor's utilities to deteriorate, we now show that the same mapping also preserves Nash equilibria in cases where such equilibria exist. We note that the following result uses similar arguments to those given by Babaioff et al. [3], for proving a related characterization of pure Nash equilibria.
Theorem 4.3. For every pure Nash equilibrium p of a VC game there is a pure Nash equilibrium S = ( S , . . . , S 1 k ) in the corresponding PMVC game, such that: (1) X v ′ ( ; S ) = X v ( ; p ) ; and (2) for all a ∈ X v ( ; p ) , ˜( p a ) = p a ( ) , where ˜ p is the induced price vector for S .
Proof. For convenience, let B = X v ( ; p ) . As before, we let the strategy profile in the corresponding PMVC game be S = ( S , . . . , S 1 k ) , where S i = B ∩ A i , for i = 1 , . . . , k .
/negationslash
We begin by proving part (2) of the theorem. Suppose that there is an item a ∈ B such that p a ( ) = m B a ( \ a ) . If p a ( ) > m a ( B ) = v B ( ) -v B ( \ a ) , then v B ( \ a ) -p B ( \ a ) > v B ( ) -p B ( ) , implying that the buyer would not buy item a , contradicting our assumption that a ∈ B .
Assume now that p a ( ) < m a ( B ) . Letting p ′ denote the vector resulting by replacing p a ( ) in p with m B a ( \ a ) , we clearly have that u b ( B, p ′ ) = u b ( B \ a, p ′ ) . We now prove the following claim:
$$C l a i m. \ u _ { b } ( B \, \sum a, \mathbf p ^ { \prime } ) \geq u _ { b } ( T, \mathbf p ^ { \prime } ), \, \text{for all $T\subseteq B\,\sum a$.}$$
Proof. Let ( B \ a ) \ T = { c 1 , . . . , c m } , and let P t = ( B \ a ) \ { c 1 , . . . , c t } for all t = 1 , . . . , m . Then u b ( B \ a, p ′ ) ≥ u b ( P , t p ′ ) . We prove the claim inductively. Suppose the claim is true for t < m : u b ( B \ a, p ′ ) ≥ u b ( P , t p ′ ) . We now show that the same inequality holds for t +1 as well:
$$u _ { b } ( P _ { t + 1 }, \mathbf p ^ { \prime } ) & = v ( P _ { t + 1 } ) - p ^ { \prime } ( P _ { t + 1 } ) = v ( P _ { t } ) - p ^ { \prime } ( P _ { t } ) - ( m _ { c _ { t + 1 } } ( P _ { t + 1 } ) - p ( c _ { t + 1 } ) ) \\ & \leq u _ { b } ( P _ { t }, \mathbf p ^ { \prime } ) - ( m _ { c _ { t + 1 } } ( B ) - p ( c _ { t + 1 } ) ) \leq u _ { b } ( P _ { t }, \mathbf p ^ { \prime } )$$
where the first inequality follows from submodularity, and the second inequality follows from the fact that p c ( t +1 ) ≤ m c t +1 ( B ) as we have previously shown. /intersectionsq /unionsq
Therefore, the vendor who owns a can increase his payoff by setting the price of item a to any value between p a ( ) and m B a ( ) , contradicting the equilibrium state.
What is left to prove is that S is a Nash equilibrium in the PMVC game. Note that we can assume w.l.o.g. the price of all products which are not sold is v A ( ∗ ) + 1 , as they remain unsold and continue to contribute nothing to the buyer or seller. Now, suppose S is not a Nash equilibrium, and that there is a player i , which can benefit from changing his set of sold items from S i to S ′ i , which would result in a different vector of induced prices ˜ p ′ = (˜ p , ′ i ˜ p ′ -i ) . We now argue that vendor i can make an identical improvement in his revenue by changing his price vector from p i to ˜ p ′ i , contradicting p being a Nash equilibrium. For convenience, we let B ′ = ( B \ S i ) ∪ S ′ i , and B ′′ = X v p , ( ; ˜ ′ i p -i ) .
To show this, first notice no other vendor would sell any previously unsold items as a result; that is, X v p , ( ; ˜ ′ i p -i ) \ A i ⊆ X v p , ( ; i p -i ) \ A i (since prices of items in ( A ∗ \ A i ) ∩ X v p , ( ; i p -i ) are still v A ( ∗ ) + 1 ). So B ′′ = X v p , ( ; ˜ ′ i p -i ) ⊆ X v S , ′ ( ; ′ i S ′ -i ) = B ′ . Thanks to submodularity, we have that for every a ∈ S ′ i , m B a ( ′ ) < m a ( B ′′ ) . Arguments similar to the ones given above (on p a ( ) = m B a ( ) ) imply that player i would sell all the items in S ′ i , and as the prices are unchanged from the PMVC game, will make the same profit as in the PMVC game. As this increases the player's profit in the PMVC game, it would increase its profit in the VC game as well, in contradiction to p being a Nash equilibrium. /intersectionsq /unionsq
Discussion Note that we have not shown an exact equivalence between the two games: the set of Nash equilibria in the VC game is a subset of the equilibria in the PMVC game. However, Proposition 4.2 and Theorem 4.3 allow us to reason about our original game to a considerable extent.
In contrast to the original model of Babioff et al. in which n i = 1 for all i = 1 , . . . , k , we can show that in our more general game, there may not always be a pure
Nash equilibrium. In order to do so, we provide an example of a VC game in the next section with two vendors who each control two items. We show that this game does not admit any pure Nash equilibrium by relating to its corresponding PMVC game, using Theorem 4.3. Moreover, if we restrict ourselves to VC games that do admit pure Nash equilibria, we can provide quantitative bounds on their quality. Specifically, when restricting ourselves to VC games that have pure Nash equilibria, we provide a lower bound on the price of stability of the PMVC game by analyzing an instance of the game. As the optimal objective value (the valuation of the set that is bought by the buyer) is always v N ( ) , Theorem 4.3 immediately implies that the same lower bound applies to the VC game. To complement our lower bound, we also provide an upper bound for the price of anarchy, also ensuring tightness of bounds.
## 5 Equilibrium Analysis
In Section 4, we outlined several properties of the discrete PMVC game. We now describe how the PMVC game can serve as a surrogate to help analyze the stability of the VC game, and the quality of equilibria in the VC game in cases when they exist.
## 5.1 Existence of pure Nash equilibria
We begin by showing that, as opposed to the special case where each vendor owns a single item, some instances of our game may not actually admit pure Nash equilibria.
Proposition 5.1. There exists an instance of the VC game with two vendors, where n 1 = n 2 = 2 , that does not admit a pure Nash equilibrium.
Proof. Let A 1 = { a, b } and A 2 = { c, d } . We define the buyer's valuation function v according to Table 1 (the value in each cell is the valuation of the union of the sets given at the head of the entry's row and column).
| ∅ | { c } | { d } | { c, d } |
|-----------------|---------|---------|------------|
| ∅ 0 | 2.803 | 2.703 | 4.1045 |
| { a } 3.203 | 5.404 | 5.304 | 6.5045 |
| { b } 2.503 | 5.304 | 5.204 | 6.6045 |
| { a, b } 4.4045 | 6.6045 | 6.5045 | 7.6045 |
Table 1. The buyer's valua-
| ∅ | { c } | { d } | { c, d } |
|----------|-------------------------|---------------|-------------|
| ∅ (0,0) | (0,2.803) | (0,2.7030) | (0,2.703) |
| { a } | (3.203,0) (2.601,2.201) | (2.601,2.101) | (2.4,2.301) |
| { b } | (2.503,0) (2.501,2.801) | (2.501,2.701) | (2.5,2.701) |
| { a, b } | (3.103,0) (2.501,2.2) | (2.501,2.1) | (2.1,2.1) |
tion function
Table 2. The payoffs for each vendor.
It is easy to verify that v is (strictly) non-decreasing and submodular. Now, consider the PMVC game with the same item sets and valuation function v . For each strategy profile ( S , S 1 2 ) , the mechanism prices the items according to their marginal contributions (Definition 4.1). Therefore, vendor payoffs are the sum of the prices of the items offered. The vendors' payoffs for each strategy profile are given in Table 2 (the first
entry corresponds to the row player, Vendor 1, and the second to the column player, Vendor 2). As is evident from Table 2, there is no pure Nash equilibrium in the PMVC game. Theorem 4.3 then implies our proposition.
## 5.2 How bad can equilibria be?
Given the negative nature of Proposition 5.1, we now restrict attention to the subclass of VC games that do admit pure Nash equilibria, and ask whether reasonable guarantees on social welfare in such equilibria can be derived
More formally, let G = { G = ( v, ( A , . . . , A 1 k )) : ∃ a pure Nash equilibrium in G } be the set of VC games which admit a pure Nash equilibrium. Define the price of anarchy (PoA) as follows:
$$P o A _ { \mathcal { G } } = \max _ { G \in \mathcal { G } } \frac { \max _ { \mathbf p ^ { * } } f ( \mathbf p ^ { * } ) } { \min _ { \mathbf p \colon \mathbf p } i s a \, \text{pure} \, \text{Hash equilibrium} \, f ( \mathbf p ) }$$
PoA is a commonly used worst-case measure of the efficiency of the equilibria, and in our case reflects the efficiency loss in G resulting from the introduction of strategic pricing, as opposed to using a 'centrally coordinated' pricing policy.
Theorem 5.2. Define the set of VC games G m , such that G ∈ G m iff (1) G has a pure Nash equilibrium, and (2) max k i =1 | A i | = m . Then the PoA of G m is at most H m +1 , where H m is the m 'th harmonic number.
Proof. Consider a game G = ( v, A = ( A , . . . , A 1 k ) } in G m . It is enough to provide a lower bound on the minimal social welfare of a pure Nash equilibrium in the corresponding PMVC game: by Theorem 4.3, this will establish a lower bound on the social welfare of a pure Nash equilibrium in G as well. So let S = ( S , . . . , S 1 k ) be a pure Nash equilibrium of the PMVC game. As v ( ) · is non-decreasing, we can assume without loss of generality that | S i | = { a i } , for some a i ∈ A i .
Again by the assumption that v ( ) · is non-decreasing, we know that the optimal social welfare is obtained when all of A ∗ is sold, so it is enough to upper bound v ( A ) in terms of v ( S ) .
We now show the following straightforward bound on the social welfare resulting from switching from S i to A i :
$$\underset { 1, \dots, k. } { \text{Lemma 5.3. } } \, \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \end{array} \, \| \begin{array} { c } \text{Lemma} \, 5. 3. \, \text{$v} \| \, ( A _ { i }, \text{$S_{i}$} ) \, - \, \text{$v} \| \, ( \text{$S_{i}$} ) \| \,, \text{for all $i$} \, = \, \\ 1, \dots, k. \, \text{$\text{$1}$} ) \end{array}$$
Proof. As S is a Nash equilibrium, the profit from selling a i is higher than selling any set B ⊆ A i . Using the definition of the pricing mechanism of the PMVC game, we know
$$v ( S _ { i }, \mathbf S _ { - 1 } ) - v ( \emptyset, \mathbf S _ { - 1 } ) \geq \sum _ { b \in B } m _ { b } ( B \, \sum b, \mathbf S _ { - 1 } ), \quad \text{ for all } B \subseteq A _ { i }$$
By an averaging argument, this means that for all B ⊆ A i , there exists an item b ∈ B such that
$$\frac { 1 } { | B | } ( v ( S _ { i }, \mathbf S _ { - i } ) - v ( \emptyset, \mathbf S _ { - i } ) ) \geq m _ { b } ( B \, \langle \, b, \mathbf S _ { - i } ) \, \quad \quad ( 1 )$$
The above implies that there is a relabelling of the items in A i , so that: (1) A i = { b 1 , . . . , b n i } , (2) b 1 = a 1 , and (3) if set P t = { b 1 , . . . , b t } ∪ S -i and P 0 = S -i , the following holds:
$$v ( A _ { i }, \mathbf S _ { - i } ) & = v ( \emptyset, \mathbf S _ { - i } ) + \sum _ { t = 1 } ^ { n _ { i } } m _ { b _ { i } } ( P _ { t - 1 } ) \leq v ( \emptyset, \mathbf S _ { - i } ) + \sum _ { i = 1 } ^ { n _ { i } } \frac { 1 } { t } ( v ( S _ { i }, \mathbf S _ { - i } ) - v ( \emptyset, \mathbf S _ { - i } ) ) \\ & = v ( \emptyset, \mathbf S _ { - i } ) + H _ { n _ { i } } ( v ( S _ { i }, \mathbf S _ { - i } ) - v ( \emptyset, \mathbf S _ { - i } ) )$$
where the first equality follows from a simple telescopic series, and the first inequality follows from Eq. 1. /intersectionsq /unionsq
Next, we show the following useful bound:
$$L e m m a { 5. 4. } \sum _ { i = 1 } ^ { k } v ( A _ { i }, S _ { - 1 } ) \geq v ( A _ { i }, A _ { - 1 } ) + ( k - 1 ) v ( S _ { i }, S _ { - 1 } )$$
Proof. L ( ) t = ( A , . . . , A 1 t , S t +1 , . . . , S k ) , for t = 1 , . . . , k , and L (0 ) = S . That is, L ( ) t is the strategy profile resulting from replacing the lengtht prefix of S with that of A . We prove by induction that
$$\sum _ { i = 1 } ^ { t } v ( A _ { i }, \mathbf S _ { - \mathbf i } ) \geq v ( L ^ { ( t ) } ) + ( t - 1 ) v ( \mathbf S )$$
and the lemma would follow by setting t = k .
The inequality clearly holds for t = 1 , due to the monotonicity of v ( ) · . Assume that the inequality holds for t < k . Thus, for t +1 , we have:
$$\sum _ { i = 1 } ^ { t + 1 } v ( A _ { i }, S _ { - 1 } ) \geq v ( L ^ { ( t ) } ) + ( t - 1 ) v ( \mathbf S ) + v ( A _ { t + 1 }, \mathbf S _ { - ( t + 1 ) } )$$
By the second definition of submodularity, we know v L ( ( ) t ) + v A ( t +1 , S -( t + ) 1 ) ≥ v L ( ( +1) t ) + v ( S ) . Putting this in the preceding inequality concludes the proof. /intersectionsq /unionsq
We can also prove an upper bound on the optimal social welfare in terms of the social welfare of S . By the above two lemmas, we get:
$$\text{welfare of $S$}. By the above two lemmas, we get: \\ v ( A ) & \leq \sum _ { i = 1 } ^ { k } v ( A _ { i }, S _ { - i } ) - ( k - 1 ) v ( S ) \\ & \leq \sum _ { i = 1 } ^ { k } v ( \emptyset, S _ { - i } ) + \sum _ { i = 1 } ^ { k } H _ { n _ { i } } ( v ( S _ { i }, S _ { - i } ) - v ( \emptyset, S _ { - i } ) ) - ( k - 1 ) v ( S ) \\ & \leq \sum _ { i = 1 } ^ { k } v ( S _ { i }, S _ { - i } ) + \sum _ { i = 1 } ^ { k } H _ { n _ { i } } ( v ( S _ { i }, S _ { - i } ) - v ( \emptyset, S _ { - i } ) ) - ( k - 1 ) v ( S ) \\ & = v ( S ) + \sum _ { i = 1 } ^ { k } H _ { n _ { i } } ( v ( S _ { i }, S _ { - i } ) - v ( \emptyset, S _ { - i } ) ),$$
where the third inequality follows from the monotonicity of
By submodularity and that n m for i = 1 , . . . , k , v ( ) · . i ≤
$$v ( \mathbf A ) & \leq v ( \mathbf S ) + H _ { m } \sum _ { i = 1 } ^ { k } ( v ( S _ { i }, \mathbf S _ { - i } ) - v ( \emptyset, \mathbf S _ { - i } ) ) \\ & \leq v ( \mathbf S ) + H _ { m } v ( \mathbf S ) = v ( \mathbf S ) ( H _ { m } + 1 ),$$
which establishes our upper bound on the PoA.
We also give an example of a game with a pure Nash equilibrium that matches the above bound.
Theorem 5.5. There exists a game in G m with a price of anarchy of H m .
Proof. Our counter-example is obtained by making the bound of Lemma 5.3 tight. Consider a game G = ( v, A = ( A , . . . , A 1 k )) , in which | A i | = m , for i = 1 , . . . , k .
Wedefine the valuation function as follows. For a strategy profile T = ( T , . . . , T 1 k ) , we set v ( T ) = ∑ k i =1 /lscript ( T i ) , where /lscript ( T i ) = 0 if | T i | = 0 , and otherwise we set /lscript ( T i ) = H | T i | . Observe that the vendors are all symmetric, and that furthermore, the payoffs only depend on their own prices.
We now consider the following strategy profile p . Pick an arbitrary item a i from each A i , for i = 1 , . . . , k , and set p a ( i ) = 1 . Price the remaining items at v A ( ∗ ) + 1 . Note that the payoff of each vendor is precisely 1 . 6
/negationslash
It is easy to see that p is a pure Nash equilibrium. Indeed, suppose that it is not, and let i be an arbitrary vendor. Then he has an alternative pricing p ′ i = p i , such that deviating to it would improve his payoff of 1 . Suppose that the set of items being bought under a deviation to p ′ i is B = X v p , ( ; ′ i p ′ -i ) , such that ∑ a ∈ B p a ( ) > 1 . Then there exists an item b ∈ B , such that p b ( ) > 1 / B | | . But then by the definition of the valuation function we have:
$$\ell ( B ) - p ( B ) = H _ { | B | } - p ( B \, \| \, b ) - p ( b ) < H _ { | B | - 1 } - p ( B \, \| \, b )$$
contradicting the assumption that the buyer ends up buying set B .
/intersectionsq
/unionsq
Note that the above construction can be extended to show that the Price of Stability (PoS) is identical:
Corollary 5.6. The Price of Stability of the class of games G m is Ω H ( m ) .
Sketch of Proof Use the construction in the proof for Proposition 5.5, but set /lscript ( T i ) = 1 if | T i | = 1 , and /lscript ( T i ) = H | T i | -/epsilon1 , if | T i | > 1 , for sufficiently small /epsilon1 . It is not hard to show that the aforementioned pure Nash equilibrium is the only Nash equilibrium. A similar bound follows. /intersectionsq /unionsq
6 If the buyer is not maximal, we can decrease the prices of the a i 's by some small /epsilon1 .
## 6 Special Case: Product Categories
A particular VC gane of interest is one in which have classes of items that are roughly equivalent; as such the buyer is interested in at most one item from each class (e.g., TV sets of a certain size, with different manufacturers and sets of feature). Items is different classes however are 'unrelated' so the buyer's valuation for any set of items is additive across these classes. This scenario reflects the common case of shops selling very similar products, of which the buyer only needs one, and tries to understand how the model's pricing behaviour.
Definition 6.1. A Category-Divided Substitutable-Product Vendor Competition game (CDSP-VC) is a VC game with a buyer that has a category-product-substitutable valuation function:
/negationslash
- -For S ⊆ A ∗ , v S ( ) = ∑ r i =1 v S ( ∩ T ( ) i ) .
- -A ∗ is partitioned into r pairwise-disjoint sets, T (1) , . . . , T ( r ) . That is, T ( ) i ∩ T ( j ) = ∅ for i = j , and ⋃ r i =1 T ( ) i = A ∗ . We refer to each set T ( j ) for j = 1 , . . . , r , as a category.
- -For S ⊆ A ∗ and 1 ≤ i ≤ r , v S ( ∩ T ( ) i ) = max a ∈ ∩ S T ( ) i v a ( ) .
Due to the additivity of different item classes, we can focus on the pricing dynamic within a specific category and easily generalize the results.
Observation 6.2 For a category T ( j ) , regardless of the other vendors' strategies, no vendor can profit by selling any items other than his most valuable one in category T ( j ) .
/negationslash
Proof. As the buyer only buys a single item from each category, vendors can make positive revenue from at most one item. Consider a vendor i and category T ( j ) such that A i ∩ T ( j ) = ∅ . Suppose that a ∈ arg max b ∈ T ( j ) ∩ A i v b ( ) , and consider a strategy profile p . For any item b ∈ ( A i \ a ) ∩ T ( j ) , if b ∈ X v p , ( ; i p -i ) , then if vendor i switches to a price vector p ′ i , which prices a the same as b and prices b at v A ( ∗ ) + 1 , then surely a ∈ X v p , ( ; ′ i p -i ) , and the player's profit does not decrease.
/negationslash
The above observation implies that within every category, each of the vendors is better off effectively trying to sell his highest valued item. In other words, we can assume without loss of generality, that for every vendor i and category T ( j ) , the vendor can pick an item a ( j ) i ∈ arg max a ∈ T ( j ) ∩ A i v a ( ) (if such item exists) and set p b ( ) = v A ( ∗ ) + 1 , for all b = a ( j ) i , without incurring a loss as a result. Therefore, this reduces our game to r independent special cases of the VC game, in which each vendor owns a single item.
We turn to the result given by Babaioff et al. [3] (Theorem 3.1 in the preliminaries). Their result implies the following characterization of the prices in a pure Nash equilibrium.
/negationslash
Corollary 6.3. Every CDSP-VC game has a pure Nash equilibrium of the following form. For every category T ( j ) , let c ( j ) i = argmax a ∈ ( T ( j ) ∩ A i ) v a ( ) , and w = argmax i c ( j ) i . Let b ( j ) = argmax a ∈ ( T ( j ) \ A w ) v a ( ) or b ( j ) = 0 if | T ( j ) | = 1 . Then p c ( ( j ) w ) = v c ( ( j ) w ) -b ( j ) , and for every player i = w p c , ( ( j ) i ) = 0 . For all other items a ∈ T ( j ) , p a ( ) = v A ( ∗ ) + 1 .
Proof. Once we know that each player sells only a single product, products that are not sold need to be priced high, and the rest of the result stems from Theorem 3.1, as the above difference constitutes the marginal contribution of the item c ( j ) .
## 7 Conclusions and Future Work
In analyzing the multi-item, multi-vendor problem, we began by defining a discrete game, which allowed us to consider a related case on the road to analyzing the scenario of a 'multiple item per vendor' game. The main property of the discrete game was to transform player strategies from pricing, to selecting what items to sell. To paraphrase Clausewitz's famous dictum, displaying (what to sell) became pricing by other means.
Utilizing this discrete game, we were able to prove that a multi-item, multi-vendor game with submodular buyer valuations does not necessarily have a Nash equilibrium (unlike the 'single item per vendor' model), and furthermore, even when an equilibrium exists, the equilibrium may provide only a logarithmic price of anarchy (again, unlike the single item per vendor model). Building on these results, we showed that in a particular (yet, in our view, common) category-substitute model, there will always be an efficient pure Nash equilibrium.
There are many open problems in this area, even before the 'holy grail' of pricing multi-item multi-buyer scenarios. We believe that there is a need to establish the characteristics of valuation functions that guarantee the existence of a Nash equilibrium. This is true both in our scenario, where even when assuming submodular functions a Nash equilibrium is not assured, as well as in simpler scenarios, such as the one item per vendor model, where, while equilibrium is assured for submodular valuations, it is not known if that is a characterization of the properties required for an equilibrium.
Adding more buyer valuations changes the model significantly, as vendors do not simply construct some 'buyer in expectation' and act according to it, but rather have a wider range of options to pursue (primarily bundling). Perhaps using a metric to define a set of similar, yet not identical, buyers, it might be possible to build on our results (and those of others), and construct extensions to the current model incorporating multiple buyers.
## Bibliography
- [1] Angwin, J., Mattioli, D.: Coming soon: Toilet paper priced like airline tickets. The Wall Street Journal (2012)
- [2] Skorupa, J.: Pricing intelligence goes to war. Retail Infor Systems News (2014)
- [3] Babaioff, M., Nisan, N., Leme, R.P.: Price competition in online combinatorial markets. CoRR abs/1401.1559 (2014)
- [4] Hart, S., Nisan, N.: Approximate revenue maximization with multiple items. In: Proceedings of the 13th ACM Conference on Electronic Commerce (EC), Valencia, Spain (June 2012) 656
- [5] Hart, S., Reny, P.J.: Maximizing revenue with multiple goods: Nonmonotonicity and other observations. DP 630, Hebrew University of Jerusalem, Center for Rationality (2012)
- [6] Christodoulou, G., Kov´ acs, A., Schapira, M.: Bayesian combinatorial auctions. In: Proceedings of the 35th International Colloquium on Automata, Languages and Programming (ICALP), Reykjavik, Iceland (July 2008) 820-832
- [7] Guruswami, V., Hartline, J.D., Karlin, A.R., Kempe, D., Kenyon, C., McSherry, F.: On profit-maximizing envy-free pricing. In: Proceedings of the Sixteenth Annual ACM-SIAM Symposium on Discrete Algorithms. SODA '05, Philadelphia, PA, USA, Society for Industrial and Applied Mathematics (2005) 1164-1173
- [8] Oren, J., Narodytska, N., Boutilier, C.: A game-theoretic analysis of catalog optimization. In: AAAI. (2014) . Forthcoming
- [9] Sch¨ on, C.: On the product line selection problem under a ttraction choice models of consumer behavior. European Journal of Operational Research 206 (1) (October 2010) 260-264 | null | [
"Omer Lev",
"Joel Oren",
"Craig Boutilier",
"Jeffery S. Rosenschein"
] | 2014-08-01T18:30:00+00:00 | 2014-08-01T18:30:00+00:00 | [
"cs.GT",
"cs.MA"
] | The Pricing War Continues: On Competitive Multi-Item Pricing | We study a game with \emph{strategic} vendors who own multiple items and a
single buyer with a submodular valuation function. The goal of the vendors is
to maximize their revenue via pricing of the items, given that the buyer will
buy the set of items that maximizes his net payoff.
We show this game may not always have a pure Nash equilibrium, in contrast to
previous results for the special case where each vendor owns a single item. We
do so by relating our game to an intermediate, discrete game in which the
vendors only choose the available items, and their prices are set exogenously
afterwards.
We further make use of the intermediate game to provide tight bounds on the
price of anarchy for the subset games that have pure Nash equilibria; we find
that the optimal PoA reached in the previous special cases does not hold, but
only a logarithmic one.
Finally, we show that for a special case of submodular functions, efficient
pure Nash equilibria always exist. |
1408.0259v2 | ## Permutation Trellis Coded Multi-level FSK Signaling to Mitigate Primary User Interference in Cognitive Radio Networks
Raghed El-Bardan, Student Member, IEEE, Engin Masazade, Member, IEEE, Onur Ozdemir, Member, IEEE, Yunghsiang S. Han, Fellow, IEEE, and Pramod K. Varshney, Fellow, IEEE
## Abstract
We employ Permutation Trellis Code (PTC) based multi-level Frequency Shift Keying signaling to mitigate the impact of Primary Users (PUs) on the performance of Secondary Users (SUs) in Cognitive Radio Networks (CRNs). The PUs are assumed to be dynamic in that they appear intermittently and stay active for an unknown duration. Our approach is based on the use of PTC combined with multi-level FSK modulation to make the SU transmissions robust against PU interference and noise disturbances and help an SU improve its data rate by increasing its transmission bandwidth while operating at low power and not creating destructive interference for PUs. We evaluate system performance by obtaining an approximation for the actual Bit Error Rate (BER) using properties of the Viterbi decoder and carry out a thorough performance analysis in terms of BER and throughput. The results show that the proposed coded system achieves i) robustness by ensuring that SUs have stable throughput in the presence of heavy PU interference and ii) improved resiliency of SU links to interference in the presence of multiple dynamic PUs.
R. El-Bardan and P. K. Varshney are with the Department of Electrical Engineering and Computer Science, Syracuse University, Syracuse, NY 13244 USA.
- E. Masazade is with the Department of Electrical and Electronics Engineering, Yeditepe University, Turkey.
- O. Ozdemir is with Boston Fusion Corp., 1 Van de Graaff Drive, Suite 107, Burlington, MA 01803 USA.
- Y. S. Han is with the Department of Electrical Engineering at National Taiwan University of Science and Technology, Taipei, Taiwan. He visited Syracuse University during 2012-13.
Email: { raelbard, oozdemir, varshney } @syr.edu, [email protected], [email protected]
## Index Terms
Cognitive Radio Networks, Primary User Interference, Permutation Trellis Codes, FSK Systems
## I. INTRODUCTION
Today a large portion of the frequency spectrum assigned by the Federal Communications Commission (FCC) is used intermittently resulting in inefficient usage of the assigned frequency bands. In order to accommodate the increased demand for spectrum in wireless applications, Dynamic Spectrum Access (DSA) has been proposed as the new communication paradigm for wireless systems where the existing spectrum is utilized opportunistically [1], [2]. The radios using this technique are called Cognitive Radios (CRs). In a Cognitive Radio Network (CRN), Primary Users (PUs) are the licensed users who have the right to access their spectrum at any time. When Secondary Users (SUs) have data to transmit, they look for spectrum holes which are the frequency bands not being used by primary users and transmit their data over them [3], [4]. In order to ensure a non-degraded PU operation in the network, existing literature has considered several effective protection techniques, e.g., beamforming of SU signals and putting interference temperature limits to satisfy PU signal-to-interference plus noise constraint [5], [6]. In addition, there are scenarios in which SUs can co-exist with the PUs as long as they do not violate the quality of service requirements of the PUs [7], [8].
Orthogonal frequency division multiplexing (OFDM) has been suggested as a multi-carrier communication candidate for CR systems where the available spectrum is divided into subcarriers each of which carries a low rate data stream [9]-[13]. A typical approach for a CR using OFDM is to sense the PU activity over the sub-carriers and then adjust its communication parameters accordingly. The goal is to protect PUs as well as intended SU receivers from possible collisions resulting from the use of the same sub-carriers. Continuous spectrum sensing and reformation of wireless links may result in substantial performance degradation for SUs [14]. For instance, the throughput of the secondary system is affected by the time spent for channel sensing. When an SU spends more time on spectrum sensing, a smaller number of information bits will be transmitted over a smaller interval of time resulting in reduced system throughput. On the other hand, decreasing sensing duration may result in a larger probability of making incorrect decisions, thereby decreasing the throughput of both SU and PU. In this regard, the
authors in [15] numerically analyze the trade-offs between throughput and sensing duration in addition to coding blocklength and buffer constraints. In [16], the authors study the problem of designing a sensing slot duration to maximize the achievable throughput for the SUs under the constraint that the PUs are sufficiently protected.
Error correction coding (ECC) has been proposed for CRNs as one possible technique to better protect SU links and prevent degradation in their performance. In [10], it is assumed that an SU transmitter vacates the band once a PU is detected. Due to the sudden appearance of a PU, rateless codes have been considered to compensate for the packet loss in SU data which is transmitted through parallel subchannels. The authors in [11] consider the design of two efficient anti-jamming coding techniques for the recovery of lost transmitted packets via parallel channels, namely rateless and piecewise coding. Similar to the spectrum model defined in [10], a system's throughput and goodput are analyzed and a performance comparison is carried out between these two coding techniques. For an OFDM-based CRN presented in [12], SU transmitters and receivers continuously sense the spectrum, exchange information and decide on the available and unavailable portions of the frequency spectrum. Depending on frequency availability, an appropriate Reed-Solomon coding scheme is used to retrieve the bits transmitted over the unavailable portions of the frequency spectrum. The authors in [13] further explore LowDensity-Parity-Check (LDPC) codes in an OFDM scheme where a switching model is considered for dynamic and distributed spectrum allocation. They also analyze the effects of errors during PU detection on channel capacity and system performance. The switch is assumed to be open for each SU detecting a PU. When the switch is open, the channel is modeled as a binary erasure channel (BEC) and the cognitive transmitter continues to transmit its message allowing bits to be erroneous when received. Another major use of the ECC schemes is presented in [17] where the authors study the performance of cooperative relaying in cognitive radio networks using a rateless coding error-control mechanism. They assume that an SU transmitter participates in PU's transmission as a relay instead of vacating the band in order to reduce the channel access time by a PU. Since the use of rateless codes allows an SU receiver to decode data regardless of which packets it has received as long as enough encoded packets are received, these codes are very suitable for cooperative schemes. The authors in [18] propose an end-to-end hybrid ARQ scheme in CRNs consisting of unidirectional opportunistic links to reduce the number of retransmissions with a fixed throughput offset. Their error control approach is based on coded
cooperation among paths and amplify-and-forward relaying of packets within a path such that this hybrid ARQ works for CRNs even if some coded data are missing. The authors implement their scheme using convolutional codes combined with BPSK modulation.
In this paper, we investigate the error correction performance of Permutation Trellis Codes (PTCs) [19], [20] combined with multi-level Frequency Shift Keying modulation systems in CRNs. The proposed PTC based framework is quite general and is applicable to many systems beyond CRNs where interference is an issue. For example, the framework has been applied to power line communications [19], [20] where strong interferers are assumed to be always present as an extreme case. In this paper, we show a much wider applicability of the PTC based framework as an interference mitigation approach by applying it to CRNs where the special challenge is to guarantee reliable communication by SUs in the presence of intermittent PUs that stay active for an unknown duration without the need of detecting them. These dynamic PUs serve as the source of strong interference for SU communications. Another challenge lies in the susceptibility of SU transmissions in a CRN to intentional interference like jamming attacks. In such a case, it is extremely important to guarantee a minimum degree of operability of the network. In other words, additive white Gaussian noise (AWGN) is not the critical limitation here, but rather an unpredictable and/or intermittent interference, e.g., PU transmissions and jamming attacks. Therefore, the emphasis of the proposed PTC based framework is on robustness, rather than on data rate or bandwidth use. To the best of our knowledge, our work is the first to consider the use of PTCs in CRNs. The motivation for this proposed approach is to overcome the tremendous degrading effect of PU interference on an SU transmission and to provide a stable level of reliable information reception regardless of how severe or prolonged a dynamic PU activity is and without requiring an accurate detection of it. In our model, an SU transmits its own information using low power concurrently with PU transmissions without the need to relay PU's traffic [17] or to vacate the band [10]. Different from [12], we assume that no information exchange or spectrum negotiation takes place between the SU transmitter-receiver pair. Our proposed scheme is different from existing work where communication sessions are carried over parallel subchannels [10]-[13]. This is due to the fact that our scheme is based on multi-level FSK modulation with PTC using a single subchannel at a time. By using PTC, continuous channel sensing by the SU transmitter-receiver pair is no longer required, since an appropriate PTC can cope with high levels of PU interference on a given SU link. Thus, SUs
no longer suffer from the huge overhead created by continuous sensing of the spectrum.
We consider a similar system model as devised in [19], [20] and carry out a thorough performance analysis in terms of BERs and throughput for dynamically varying interference. In [19], [20], the authors were interested in the design of PTCs that mitigate the impact of permanent interferences that exist in power line communications. In their work, they conducted a simulation study and presented results on the performance of different PTCs with respect to the probability of an element of the code matrix being in error in the presence of different noises on one or more channels. They also investigated the choice of distance increasing/conservative/reducing mappings for a given PTC with respect to narrowband interference that always exists on one band in the presence of background noise in terms of BER. The authors in [19], [20] provide a BER performance analysis via simulations and not analytically.
The analytical BER analysis of a PTC coded M-FSK system is imperative to determine its link quality. In other words, the BER estimate is a useful tool for cross-layer design. Given that the PU stays active once it starts transmitting, the exact bit error rate (BER) is derived using an exhaustive search [21]. Hence, the analytical evaluation of BER becomes computationally prohibitive for large codes. Rather than using an exhaustive search, in this paper, we develop an approximation of BER using the properties of the Viterbi decoder. Also, in our previous work [21], a very special case is considered where a PU stays active once it starts transmission. In this paper, we consider a more practical scenario where a 2-state Markov chain is employed to model the dynamic PU activity in terms of alternating On-Off periods. It should be noted that the work on [19] and [20] also considered an extreme case where the authors assumed that the interference was always present and did not consider dynamically varying interference. In the more general framework presented in this paper, the performance results provided, e.g., BER, are intuitive and more useful in comparison to the worst-case guaranteed performance presented in [21].
Under the same resources such as energy per bit, transmission bandwidth, and transmission time, we compare the throughput performance of the proposed communication system with a simple uncoded opportunistic M-FSK system and a convolutionally coded OFDM system using BPSK modulation. The simulation results show that the proposed system outperforms the aforementioned systems in the presence of heavy PU interference. We also compare the BER performance of the proposed communication system with an LDPC code coupled with an M -
FSK modulation system for static and dynamic activities of a PU. It is shown that the proposed scheme outperforms the other one. Moreover, we present results on the degree of resiliency to interference that a PTC can guarantee for the SU link in the presence of interference by multiple dynamic PUs.
To summarize, the main contributions of the paper are:
- · We consider a more practical scenario in which we represent the cognitive radio network using a dynamic PU channel occupancy model including the case of multiple PUs.
- · To mitigate interference created by these dynamic PUs, we propose the use of PTC-based multi-level FSK signaling that incurs no overhead costs as no channel sensing is required before SU transmissions.
- · Using the proposed PTC based framework, the SU transmissions achieves robustness against the dynamic PUs' interference and noise disturbances and, at the same time, the resiliency of SU links improves.
- · A thorough analytical performance analysis of the proposed scheme is presented in terms of BER and throughput for static and dynamic activities of multiple PUs.
The rest of the paper is organized as follows. In Section II, we provide the system model and introduce the concept of PTC. In Section III, we introduce hard-decision Viterbi decoding of the system. We also approximate the actual BER through the use of a specified finite number of paths forming the Viterbi trellis. Section V presents numerical results demonstrating the improvement in the performance of the PTC-based multi-level FSK system, validating the suggested BER approximation, supporting the effectiveness of the proposed communication scheme, as well as demonstrating the efficiency of PTC in the presence of multiple dynamic PU interference. Finally, we conclude the paper and address some future research directions in Section VI.
## II. SYSTEM MODEL
We consider a simple cognitive radio network as shown in Fig. 1, which consists of a PU transmitter-receiver pair and an SU transmitter-receiver pair. The SU pair is assumed to be located within the transmission range of the PU. The transmission range shown in Fig. 1 is the maximum distance covered by a PU transmission such that the signal to interference plus noise ratio (SINR) at the PU receiver equals a minimum threshold value, SINR ∗ PU . In this model, we assume a free space path loss model and a Line-Of-Sight (LOS) AWGN channel. The power in
Fig. 1: A simple cognitive radio network.
the transmitted signal is P t , so the received power is given in [22] by
$$\Phi _ { \Phi } = \ P _ { T } \cdot \left ( \frac { \sqrt { G _ { l } } \lambda } { 4 \pi d } \right ) ^ { 2 },$$
where √ G l is the product of the transmit and receive antenna field gains in the LOS direction and λ is the signal wavelength defined as the ratio of the speed of light to the center frequency of the band in operation, C f j . In this model, we assume omnidirectional antennas so √ G l = 1 . Both licensed and unlicensed users coexist and they can simultaneously operate over the same band. One further assumption made here is that interference, created by an SU, that deteriorates the QoS of the PU is negligible compared to the received PU power. This is due to the fact that the transmission power of PU is much larger than the transmission power of SU, i.e., P SU T glyph[lessmuch] P PU T . In this approach, the SINR at the PU receiver [23] given in (2) satisfies SINR ≥ SINR ∗ PU , that is,
$$S I N R \ = \ \frac { P _ { R } ^ { P U } } { N _ { 0 } + P _ { I } ^ { S U } } \geq S I N R _ { P U } ^ { * },$$
where P SU I is the SU interference at the PU receiver and SINR ∗ PU is the minimum threshold above which a PU transmission is received successfully. It is assumed that the SU transmitter knows the location of its corresponding receiver possibly by means of extra signaling. An overview of the signal processing model that combines the PTC scheme with the multi-level FSK communication system is provided in Fig. 2. The information bits, m , are loaded into a rate R = m n convolutional encoder in parallel, e.g., R = 1 2 with a two-stage shift register and generator 7, 5 (octal) [19]. It is well-known that due to the incorporation of a channel coding scheme in a
Fig. 2: Block diagram of the Coded multi-level FSK System
communication system, the overall transmission energy per bit is less than that without a coding scheme. However, due to its error correction capabilities, the purpose of convolutional encoding is to improve the performance of the system.
For each of the convolutional coded output symbols, the PTC encoder assigns a certain code matrix which is transmitted over both time and frequency. We present an example for illustration, in which the coded symbol '01' is mapped onto '213'. This permutation code matrix is to be transmitted in both time and frequency domains resulting in a 3 × 3 ( H × H ) binary code matrix as given below.
$$\mathbf T _ { i } = \left [ \begin{array} { c c c } 0 & 1 & 0 \\ 1 & 0 & 0 \\ 0 & 0 & 1 \end{array} \right ],$$
where 1 ≤ i ≤ M M . = 2 m denotes the number of symbols and H defines the number of frequency bands as well as the number of time steps used in transmitting the outputs of the encoder. In this paper, we define the PTC-based multi-level FSK system as the H -FSK system since the PTC-based multi-level FSK modulated system leads to an H x H binary code matrix. According to (3), transmission takes place on f 2 , f 1 , and f 3 (first, second and third columns of T i ) corresponding to the time steps T s , 2 T s , and 3 T s respectively. Tables I and II present the symbol (with 1 bit and 2 bits) mappings onto the corresponding unique permutation code matrices respectively [19]. For mapping tables for larger values of H , we refer the reader to [19]. For M different symbols, T i denotes the set of transmitted code matrices such that T i ∈ { T 1 , . . . , T M }
TABLE I: Mapping of symbols into permutation code matrices ( M = H = 2)
| Label | Symbol | Permutation code matrix |
|---------|----------|---------------------------|
| T 1 | 0 | 12 |
| T 2 | 1 | 21 |
TABLE II: Mapping of symbols into permutation code matrices ( M = 4, H = 3)
| Label | Symbol | Permutation code matrix |
|---------|----------|---------------------------|
| T 1 | 00 | 231 |
| T 2 | 01 | 213 |
| T 3 | 10 | 132 |
| T 4 | 11 | 123 |
has the following general form:
$$\mathbf T _ { i } = \left [ \begin{array} { c c c c } q _ { 1, 1 } & q _ { 1, 2 } & \dots & q _ { 1, H } \\ \colon & \ddots & \dots & \colon \\ q _ { H, 1 } & \dots & \dots & q _ { H, H } \end{array} \right ],$$
where q j,k ∈ { 0 1 , } denotes the ( j, k ) binary element in the i -th transmitted code matrix, j indicates the output of the detector for frequency f j at time step k in the code matrix. At each time step, the element where 1 is present in the code matrix is modulated and transmitted. At a given time, only one frequency band is used for transmission. Using the H -FSK scheme, the transmitted signal, assumed to be sufficiently narrowband, over j -th frequency and k -th time step where j, k ∈ { 1 2 , , . . . , H } , is given by:
$$\dots \dots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots.$$
where E s = P SU T · T s is the transmitted signal energy per information symbol and T s is the symbol duration. In this model, we assume a strong LOS path between a stationary SU transmitter and a stationary receiver. Therefore, channel noise is modeled as Additive White Gaussian
Noise (AWGN) with zero-mean and variance N 0 2 . Since path loss is a function of the operating frequency, we need to ensure that the received signal power over the different frequency channels is the same, P SU R ( f 1 ) = . . . = P SU R ( f H ) = P SU R , in order to satisfy a fixed signal-to-noise ratio, E /N s 0 at the receiver. Therefore, we adjust each SU's transmitting power over each frequency, P SU T ( f j ) , according to the free space path loss model given in (1). Therefore, the SU transmitter needs to know the location of its corresponding receiver. At the input of the SU demodulator, the received signal is given by:
$$x _ { j } ( t ) = \begin{cases} \ s _ { j } ^ { r } ( t ) + i _ { P U } ^ { r } ( t ) + w ( t ), & \text{if $\exists\,PU$ on $f_{j}$,} \\ \ s _ { j } ^ { r } ( t ) + w ( t ), & \text{otherwise,} \end{cases}$$
where s r j ( ) = t √ 2 E /H r s T s cos(2 πf t j + ) θ , i r PU ( ) = t √ 2 I PU /H T s cos(2 πf t j + φ ) , and E r s glyph[defines] P SU R T s is defined as the symbol energy at the receiver. I PU glyph[defines] P PU I T s and P PU I are defined as the interference energy per coded symbol and the interference power due to the PU transmitter at the SU receiver, respectively, and w t ( ) represents the channel noise at the receiver. It should be mentioned that a PU transmitter can modulate the transmitted signal using any scheme and is not limited to the use of FSK-modulated signals. The fact that the PU power is very high compared to that of the SU implies that the received signal is dominated by the PU signal which, in turn, shapes the behavior of the system. Without loss of generality, we assume that the PU uses an FSK-modulated signal to transmit its information similar to the SU.
At the receiver side, non-coherent detection is employed using a bank of H quadrature receivers so that each consists of two correlation receivers corresponding to the in-phase and quadrature components of the signal. The in-phase component of the signal received, x I j,k , is given by:
$$x _ { I _ { j, k } } & = \\ & \begin{cases} \frac { E _ { s } ^ { r } } { H } \cos \theta + \frac { I _ { P W } } { H } \cos \phi + w, & \text{if PU exists on $f_{j}$ at time $k$,} \\ \frac { E _ { s } ^ { r } } { H } \cos \theta + w, & \text{otherwise,} \end{cases}$$
where θ and φ are uniformly distributed over [0, 2 π ], i.e., θ , φ ∼ U (0 , 2 π ) , and denote the random phase components of the SU and the PU signals, respectively. The noise term w in (6) is modeled as AWGN, i.e., w ∼ N (0 , N 0 2 ) . E r s is the received symbol energy over a given signaling interval.
Similarly, x Q j,k is the received signal's quadrature component defined as follows:
$$& x _ { Q _ { j, k } } = \\ & \left \{ \begin{array} { c c c c c } \frac { E ^ { r } } { H } \sin \theta + \frac { I _ { P W } } { H } \sin \phi + w, & \text{if} \, PU \, \text{exists on} \, f _ { j } \, \text{at} \, \text{time} \, k, \\ & \frac { E ^ { r } } { H } \sin \theta + w, & \text{otherwise}. \\. &. &. &. &. &. &. &. &. &. &. &. &. &. &. \end{array} \right \}$$
The envelope of each quadrature receiver over frequency j and time step k , l j,k , is defined as the square root of the sum of the squared in-phase and quadrature components of the correlator output as
$$l _ { j, k } = \sqrt { x _ { I _ { j, k } } ^ { 2 } + x _ { Q _ { j, k } } ^ { 2 } }.$$
At the receiver, a hard decision decoding scheme is used where the envelope value of each of the H quadrature receivers is compared to a threshold value, l th . The threshold value we use in this paper is the same as that used by the authors in [19], namely l th = 0 6 . √ E r s .
The received code matrix R i is of the form,
$$\cdots \dots \dots \dots \dots \dots, \\ \mathbf R _ { i } = \left [ \begin{array} { c c c c } b _ { 1, 1 } & b _ { 1, 2 } & \dots & b _ { 1, H } \\ \colon & \ddots & \dots & \colon \\ b _ { H, 1 } & \dots & \dots & b _ { H, H } \end{array} \right ],$$
where b j,k ∈ { 0 1 , } and can be determined from,
$$b _ { j, k } \ = \ \begin{cases} \ 1, & l _ { j, k } \geq l _ { t h }, \\ 0, & \text{otherwise}. \end{cases}$$
## III. BIT ERROR RATE ANALYSIS
The link quality between two SUs is determined by how much PU interference it can tolerate. We employ BER as the QoS metric to characterize successful communication for an SU. We consider the system model presented in Fig. 2 and obtain an approximation for BER. This system model is the same as the one considered in [19], [20] where the PTC encoder includes a convolutional encoder and the corresponding decoder uses Viterbi algorithm for decoding.
Here, we assume that the PTC encoder uses a rate 1 2 convolutional code whose output is converted to a symbol which, in turn, is mapped onto a permutation matrix to be transmitted. It is important to point out that the exact BER analysis of the proposed system is computationally prohibitive. Hence, we approximate BER using some properties of the Viterbi decoder.
Fig. 3: Finite State Machine of a ( R = 1 2 ) convolutional coded PTC system
Fig. 4: A ( R = 1 2 ) convolutional code trellis for our PTC-based system.
The encoder of the binary convolutional code with a given rate is viewed as a finite state machine as shown in Fig. 3. This state diagram, corresponding to the PTC shown in Table II, results in the trellis presented in Fig. 4. In order to optimally decode the received information, we employ the Viterbi algorithm which reconstructs the maximum-likelihood path given the input information sequence. The Viterbi decoder gives as an output the path in the trellis that has the minimum overall Hamming distance with respect to the received sequence. In this regard, an error event may occur if for a transmitted path in the trellis the overall number of differences with the demodulator outputs is larger than or equal to that of a competing path.
In order to compute an upper-bound on the probability of bit error, all error events have to be taken into account. It is important to note that the number of error events is proportional to the number of input bits to be transmitted. In other words, the number of competing paths and the number of error events increase as more input bits are serially loaded into the convolutional encoder. Then, the probability of bit error of a convolutional code is upper-bounded according to [24], [25] as
$$P _ { e } \leq \sum _ { d = d _ { f r e e } } ^ { \infty } a _ { d } P _ { 2 } ( d ),$$
where d free is the free distance of the convolutional code, a d is the number of paths that differ by d bits from the transmitter codeword, and P 2 ( d ) is the probability that the decoded path differs by d bits from the transmitted codeword. In particular, a d and P 2 ( d ) are independent of the transmitted codeword. In fact, they can be calculated assuming that an all-zero codeword is transmitted. Knowing that all the code matrices to be sent include H non-zero elements, we next prove that there is a similar upper-bound on the system given in Fig. 2.
Now, let c be a codeword of the convolutional code and P ( c ) be the corresponding expanded codeword comprising the permutation code matrices mapped from c . The length of P ( c ) is ( L + m H ) 2 , where L denotes the size of packet to be transmitted and L + m the number of branches in the trellis. In order to apply (11), we need to prove that the upper-bound given in (11) is independent of the transmitted expanded codeword. Let P e ( P ( c )) denote the probability of bit error when P ( c ) is the transmitted codeword. Knowing that P ( c ) is a one-to-one mapping from the codeword c , by the argument of (11) given in [24], [25], we have
$$P _ { e } ( \mathcal { P } ( \mathbf c ) ) \quad \leq \sum _ { d = d _ { f r e e } ^ { * } } ^ { \infty } a _ { d } ^ { \mathcal { P } ( \mathbf c ) } P _ { 2 } ^ { \mathcal { P } ( \mathbf c ) } ( d ),$$
where d ∗ free is the free distance of the expanded code, and a P ( c ) d and P P ( c ) 2 ( d ) are the number of paths and the probability that the paths differ by d bits from the transmitted codeword P ( c ) , respectively.
Proposition 1: The bound in (12) is independent of the transmitted codeword if the noise caused by the PU in a subchannel will affect all bits in this channel that are generated by one or more trellis branches.
Proof: See proof in Appendix.
The fact that we have no information regarding the length of the transmitted bits sequence means that the calculation of the upper bound, in (11), involves an infinite sum over all possible error events. In other words, an infinite number of trellis' paths have to be taken into consideration which will render the process of evaluating the error bound computationally prohibitive for applications such as link adaptation. Therefore, it is important to approximate BER using a suitably selected finite number of trellis' paths.
For every convolutional code, there exists a transfer function that uniquely determines the number of paths by which a received sequence is d bits different from an assumed all-zero transmitted bit sequence. For instance, the transfer function of the mapped permutation code in the proposed H -FSK scheme based on Fig. 3, i.e. H = 3, is given by T D ( ) where
$$T ( D ) = D ^ { 1 6 } + 2 \cdot D ^ { 2 0 } + 4 \cdot D ^ { 2 4 } + 8 \cdot D ^ { 2 8 } + \cdots$$
Equivalently, we are saying that there is only one path, { a → b, b → c, c → a } , with an input sequence { 1 0 0 , , } and permutation code matrix sequence { 123 132 123 , , } that has 16 bits different from the assumed all-zero transmitted bit sequence ( d H (231,123) + d H (231,132) + d H (231,123) = 6 + 4 + 6), 2 paths that have 20 bits difference, and so on. It is most likely that the first few error events (paths) in the trellis dominate all the remaining ones. In this manner, the approximate BER can be written as
$$\hat { P } _ { e } \quad \approx \sum _ { d = d _ { f r e e } ^ { * } } ^ { d _ { f r e e } ^ { * } + z } a _ { d } ^ { \mathcal { P } ( \mathbf c ) } P _ { 2 } ^ { \mathcal { P } ( \mathbf c ) } ( d ),$$
where z ∈ N and z + 1 denotes the number of paths involved in the BER approximation for which numerical results are presented in Section V.
In order to compute P P ( c ) 2 ( d ) we denote by T u and R u the transmitted and received code matrices at stage u in a trellis composed of V stages. Then, P P ( c ) 2 ( d ) can be expressed as,
$$\text{ matrices at stage } u \text{ in a trellis composed of } \mathcal { V } \text{ stages. Then, } P _ { 2 } ^ { \rho } \stackrel { \nu } { \langle } ( d ) \text{ can be expressed as, } \\ P _ { 2 } ^ { \mathcal { P } ( e ) } ( d ) \text{ } & = \prod _ { u = 1 } ^ { \mathcal { V } } P \left ( \mathbf R ^ { u } | \mathbf T ^ { u } \right ) \\ & = \prod _ { u = 1 } ^ { \mathcal { V } } j = 1 \, k = 1 } ^ { \mathbf V } P \left ( b _ { j, k } ^ { u } | q _ { j,$$
where q u j,k and b u j,k are the ( j, k ) binary elements given by (4) and (9) respectively in the u -th stage of the trellis and PU j and ¯ PU j denote the presence and absence of PU activity over channel f j respectively. Here we note that P PU ( j ) and P PU ( ¯ j ) depend on channel f j 's occupancy model which, in turn, is discussed in Section IV-1.
In each stage of the trellis, the absence and presence of PU need to be taken into consideration when computing the likelihoods in (15) that also depend on the code matrix to be transmitted.
## A. Computation of likelihoods in the absence of PU
Suppose there is no PU transmission over frequency band j at time step k . If the SU is active, i.e. transmitting a bit '1', then the demodulator in-phase component x I j,k ∼ N ( √ E r s H cos θ, N 0 2 ) and the demodulator quadrature component x Q j,k ∼ N ( √ E r s H sin θ, N 0 2 ) . The fact that x I j,k and x Q j,k are statistically independent random variables with non-zero means, then implies l j,k ∼ Rice ( √ E r s H , √ N 0 2 ) . Accordingly, P ( b j,k = 0 | q j,k = 1 , P U ¯ j ) can be computed as
$$\cdot \quad \cdot \quad & \cdot \\ P \left ( b _ { j, k } = 0 | q _ { j, k } = 1, \bar { P } U _ { j } \right ) \ & = \ \ P ( l _ { j, k } < l _ { t h } | q _ { j, k } = 1, \bar { P } U _ { j } ) \\ & = \ \ F _ { L _ { j, k } } ( l _ { t h } ) \\ & = \ 1 - Q _ { 1 } \left ( \sqrt { 2 \frac { E ^ { r } s / H } { N _ { 0 } } }, 0. 6 \sqrt { 2 \frac { E ^ { r } } { N _ { 0 } } } \right ), \quad \quad \quad$$
where F L j,k ( l th ) is the cumulative distribution function of l j,k evaluated at l th , Q 1 ( v, w ) is the Marcum's Q-function defined as [26],
$$Q _ { 1 } ( v, w ) = \int _ { w } ^ { \infty } x \exp \left \{ - \frac { x ^ { 2 } + v ^ { 2 } } { 2 } \right \} I _ { 0 } ( v x ) \, d x$$
and I 0 ( vx ) is the zeroth order modified Bessel function. P ( b j,k = 1 | q j,k = 1 , P U ¯ j ) , in this case, is nothing but the complement of (16) and is given by
$$P \left ( b _ { j, k } = 1 | q _ { j, k } = 1, \, \bar { P } \bar { U } _ { j } \right ) \ & = \ \, P ( l _ { j, k } \geq l _ { t h } | q _ { j, k } = 1, \, \bar { P } \bar { U } _ { j } ) \\ & = \ 1 - F _ { L _ { j, k } } ( l _ { t h } ) \\ & = \ Q _ { 1 } \left ( \sqrt { 2 \frac { E _ { s } ^ { r } / H } { N _ { 0 } } }, 0. 6 \sqrt { 2 \frac { E _ { s } ^ { r } } { N _ { 0 } } } \right ). \\.$$
If the SU is not transmitting any information on subchannel j at time step k , then, x I j,k ∼ N (0 , N 0 2 ) and x Q j,k ∼ N (0 , N 0 2 ) , and we have l j,k ∼ Rayleigh ( √ N 0 2 ) . Then, P b ( j,k = 1 | q j,k =
0 , P U ¯ j ) and P b ( j,k = 0 | q j,k = 0 , P U ¯ j ) can be computed as,
$$P ( b _ { j, k } = 1 | q _ { j, k } = 0, \bar { P U } _ { j } ) \ & = \ P ( l _ { j, k } \geq l _ { t h } | q _ { j, k } = 0, \bar { P U } _ { j } ) \\ & = \ 1 - F _ { L _ { j, k } } ( l _ { t h } ) \\ & = \ \exp \left ( \ - 0. 3 6 \frac { E ^ { r } } { N _ { 0 } } \right )$$
and
$$P ( b _ { j, k } = 0 | q _ { j, k } = 0, \bar { P U } _ { j } ) \ & = \ P ( l _ { j, k } < l _ { t h } | q _ { j, k } = 0, \bar { P U } _ { j } ) \\ & = \ F _ { L _ { j, k } } ( l _ { t h } ) \\ & = \ 1 - \exp \left ( - \ 0. 3 6 \frac { E _ { r } ^ { r } } { N _ { 0 } } \right ).$$
## B. Computation of likelihoods in the presence of PU
In the presence of PU activity and an active SU, i.e., the SU is transmitting a '1', the signal received at the input of the demodulator has two significant terms besides noise. One corresponds to the actual signal energy of the transmitted symbol and the other is created by the PU activity. It is assumed that the SU transmitting power is sufficiently small so that QoS of the PU session is maintained. Therefore, the SU signal is negligible as compared to that of the PU. In this case, x I j,k ∼ N ( √ I PU H cos φ, N 0 2 ) and x Q j,k ∼ N ( √ I PU H sin φ, N 0 2 ) . Thus, l j,k ∼ Rice ( √ I PU H , √ N 0 2 ) . P b ( j,k = 0 | q j,k = 1 , P U j ) and P b ( j,k = 1 | q j,k = 1 , P U j ) , in this scenario, are calculated according to,
$$P ( b _ { j, k } = 0 | q _ { j, k } = 1, P U _ { j } ) & \ = \ \ P ( l _ { j, k } < l _ { t h } | q _ { j, k } = 1, P U _ { j } ) \\ & \ = \ \ F _ { L _ { j, k } } ( l _ { t h } ) \\ & \ = \ \ 1 - Q _ { 1 } \Big ( \sqrt { 2 \frac { I _ { P U } / H } { N _ { 0 } } }, 0. 6 \sqrt { 2 \frac { E _ { s } ^ { r } } { N _ { 0 } } } \Big ) \quad \quad ( 2 1 ) \\. \, \text{a}$$
and
$$P ( b _ { j, k } = 1 | q _ { j, k } = 1, P U _ { j } ) \ & = \ \ P ( l _ { j, k } \geq l _ { t h } | q _ { j, k } = 1, P U _ { j } ) \\ & = \ 1 - F _ { L _ { j, k } } ( l _ { t h } ) \\ & = \ Q _ { 1 } \left ( \sqrt { 2 \frac { I _ { P U } / H } { N _ { 0 } } }, 0. 6 \sqrt { 2 \frac { E _ { s } ^ { r } } { N _ { 0 } } } \right ).$$
If the SU does not transmit on subchannel j at time step k , the output of the demodulator has the noisy signal received from the PU. In that case, and similar to the analysis approach followed earlier, x I j,k ∼ N ( √ I PU H cos φ, N 0 2 ) and x Q j,k ∼ N ( √ I PU H sin φ, N 0 2 ) . Consequently, l j,k ∼ Rice ( √ I PU H , √ N 0 2 ) . So, the following probabilities P b ( j,k = 0 | q j,k = 0 , P U j ) and P b ( j,k = 1 | q j,k = 0 , P U j ) are the same as P b ( j,k = 0 | q j,k = 1 , P U j ) and P b ( j,k = 1 | q j,k = 1 , P U j ) , respectively.
Once the probabilities presented above are evaluated, we can calculate P ( R T i ′ | i ) for each stage of the V stages in the trellis and, eventually, approximate the BER of the proposed scheme as given in (14).
## IV. THROUGHPUT ANALYSIS
So far, we have computed the BER approximately and evaluated the performance of the H -FSK system in the presence of PU interference. We have also proposed the adoption of the PTC-based multi-level FSK in the transmission strategy of SU in order to provide resilience to spontaneous narrowband PU interference. In other words, heavy redundancy introduced by the PTC ensures successful communication sessions of the SUs in the presence of PU interference. In order to further explore the performance of the proposed H -FSK communication system and validate its effectiveness, we perform a throughput analysis based on the approximate BER provided in Section III. Accordingly, an SU, adopting the proposed H -FSK communication scheme, is expected to provide a level of reliable information reception which is better than what the SU could achieve otherwise under heavy PU interference.
1) Licensed Channel Dynamic Occupancy Model: Different from [21] where the licensed channel was assumed to be always occupied by the PU, we model the channel available for PU active transmissions and its occupancy as a 2-state Markov chain as shown in Fig. 5. In this model, the channel has two alternative states denoted by On (Busy) and Off (Idle). This assumption is practical in the sense that the licensed spectrum occupancy will experience alternating On-Off periods rather than the assumption of the channel being always occupied (PU being always present once it starts transmission during an SU transmission). The periods during which the licensed spectrum's state is idle or busy, also known as holding times, are geometrically distributed with known independent parameters r and p as shown in Fig. 5. An On state represents the state in which the licensed band is occupied by a PU resulting in degraded
Fig. 5: Two-state Markov chain model for PU activity over licensed spectrum
BER performance for SU transmissions, while an Off (idle) state represents the state in which the primary band is idle. Accordingly, the steady state probabilities of the primary channel's state being Off or On are given respectively as
$$P _ { c h } ^ { O f f } = \frac { r } { r + p }$$
and
$$P _ { c h } ^ { Q n } = \frac { p } { r + p }.$$
2) Throughput Analysis: In order to examine the effectiveness of the proposed H -FSK approach, we conduct a throughput analysis that monitors the average throughput of the SU communication session in the presence of PU activity modeled as a 2-state Markov chain. Knowing that the SU transmitter has no information regarding the presence or absence of the PU, we determine the unlicensed throughput for SU communications as a function of the probability of a licensed channel being occupied by a PU. Such analysis illustrates the efficiency of the proposed scheme in ensuring a stable throughput regardless of any PU activity over the licensed spectrum. Accordingly, we present the approximate expected throughput ( T e ) of the H -FSK communication scheme based on the approximate BER given as
$$T _ { e } \approx \left ( 1 - P \hat { E } R \right ) \times R _ { p } \times L,$$
where ˆ PER = 1 -(1 -ˆ ) P e L is the approximate packet error rate (PER), ˆ P e is the approximate BER computed with z +1 paths knowing the presence of dynamic PU interference as described in (14) and (15), and R p is the rate at which information packets are sent. Throughput, in our
model, is defined as the number of correctly received packets per unit time. For simplicity, we do not consider header and overhead bits when computing the link throughput.
## V. NUMERICAL RESULTS AND DISCUSSION
## A. Approximate BER with Single Permanent PU Interference
For the coded H -FSK communication system shown in Fig. 2, the following values of the physical parameters are assumed. The transmit power of the SU is varied between [25 µW, 4 mW ] and the noise power density is selected as N 0 = 2 5 . × 10 -14 . These values are chosen in such a way so as to not create destructive interference to the PU. We further assume that the lowest frequency band in the available frequency spectrum, f 1 , is selected as 56 MHz, and the bandwidth spacing between any two subchannels is selected as 6 MHz. The PU is assumed to be operating at f 2 and has a total transmitting power of 1 MW. The parameters' values used in the simulation are chosen in accordance with the IEEE 802.22 standard [27]. The distances between the PU transmitter and SU receiver and that between the SU transmitter-receiver pair are selected as 10 m, in order to analyze the quality of the SU link under a high interference scenario.
In Fig. 6, we fix H = 3 and compare the simulated BER with its approximations using different number of paths used in the approximation, i.e., z +1 . It is clear from the figure that z = 3 approximates BER well for H = 3 . In Fig. 7, we fix SNR = 7 dB and compare absolute error of the BER approximation for different values of H . It is again clear that z = 3 , i.e., using only z +1 = 4 paths of the convolutional code's trellis, provides good BER approximation for different sizes of PTC. This is due to the fact that the first few error events with a few errors in the trellis dominate all the remaining transitions. This result confirms what we have mentioned earlier regarding the domination of a small number of paths in the trellis over all the remaining paths that may evolve due to incoming bits. The approximate BER we provide in the section could be used as a QoS metric in an adaptive scenario where certain parameters of the system such as H could be adapted in an online manner using this approximation. In Table III, we present results on the performance of exhaustive search 1 compared to the proposed
1 Using this technique, decisions on the transmitted code matrices are made by the permutation trellis decoder which decides in favor of the transmitted code matrix, i.e., among 2 m possible code matrices, which has the minimum Hamming distance with respect to the received one. In order to determine all the decisions for every possible code matrix, a brute force method is used to compare each possible received code matrix to every code matrix. This method needs roughly 2 H 2 + m comparisons [21].
Fig. 6: BER approximations for PTC with H = 3.
Fig. 7: | BER -BER apx | for different PTC ( H = 2, 3, and 4) and SNR = 7 dB.
approximation approach in this paper in terms of the time needed to compute BER for different PTC. As shown in Table III, it takes much shorter time to compute the BER approximately as compared to its exact calculation.
TABLE III: A comparison of the time needed to compute BER between exhaustive search [21] and approximation approach for different PTC ( H = 2,3, and 4)
| PTC | Exhaustive Search Approach | Approximate Approach |
|-------|------------------------------|------------------------|
| H = 2 | 0 . 33 s | 0 . 31 s |
| H = 3 | 42 . 21 s | 7 . 24 s |
| H = 4 | 26274 . 99 s | 245 . 32 s |
## B. Throughput Analysis of H -FSK
For comparison purposes, we first consider a competing scenario where an SU does not employ any coding technique in its spectrum interweaved transmission strategy which, in turn, is assumed to be based on M -FSK modulated signals. In this scenario, an SU makes use of the CR channel sensing feature in order to determine if there is an active PU in the network or not. Depending on the channel sensing measurements of the licensed spectrum, a decision is made on whether the SU should vacate the band or not. In the case where the SU detects a white space, it adapts the transmission parameters to utilize the full spectrum available and achieve the maximum throughput. On the other hand, when a PU is detected an SU vacates the band and adjusts its transmission parameters accordingly by employing a modified FSK modulation scheme supporting the transmission of M ′ symbols where M < M ′ . In order to gain more insight regarding the throughput performance comparison of the PTC-based multi-level FSK with the uncoded-opportunistic (adaptive) M -FSK system, we assume that packet transmissions take place over an observation window of length T seconds divided into x time slots. For each time slot, the steady state probability of a channel being occupied by an active PU, given in (24), is evaluated. Knowing each interval's time and the time required for transmitting a single packet, the incorrectly received packets can be determined. As a result, the unlicensed network's throughput is computed and analyzed. We further assume that the available spectrum consists of four frequency bands. With the use of the PTC-based multi-level FSK approach, an SU does not need to sense the channel or adapt its wireless links. In the uncoded opportunistic system, an SU senses the licensed frequency band at the beginning of a time slot. Assuming zerocost perfect detection of a licensed user's activity, an SU decides on which M -FSK modulated
P Ch
Fig. 8: Throughput Comparison of H -FSK and uncoded opportunistic M -FSK as well as convolutionally coded BPSK schemes for SNR = 4 dB, packet size = 256 bits, and R p = 100 packets/sec.
signals are to be transmitted. In the case where an SU detects a spectrum hole, it utilizes the full available spectrum (4 bands) using a 4-FSK modulation scheme. When a PU is detected, an SU vacates the licensed band and utilizes the remaining spectrum (2 frequency bands out of the 4 bands assumed to comprise the spectrum) using BFSK modulated signals. Different from [28], the licensed channel can change its state at any instant in a time slot. In Fig. 8, we present the average throughput of H -FSK and that of an uncoded opportunistic M -FSK system as a function of P On ch for H = M = 4 where the packet size L = 256 bits, R p = 100 packets/sec, and SNR is 4 dB. Fig. 8 illustrates that the average throughput in H -FSK is constant regardless of P On ch values while the average throughput of the uncoded opportunistic M -FSK decreases as P On ch increases and it matches the theoretical approximation given in (25). As shown, there exists a probability p ∗ 1 ≈ 0 7 . such that for P On ch ≥ p ∗ 1 the H -FSK approach outperforms the uncoded opportunistic scheme in terms of throughput performance besides BER. Thus, the use of the proposed H -FSK communication scheme for values of P On ch less than p ∗ 1 is not beneficial because higher throughput values are obtained using the uncoded opportunistic scheme. In other
words, the H -FSK system is effective for situations where the PU activity is relatively high. In practice, the throughput evaluated for the uncoded M -FSK system is expected to be lower than what is shown in Fig. 8 due to channel sensing intervals. It should be noted that the proposed H -FSK communication scheme does not need the channel sensing mechanism that might not be available for some applications.
Along these lines, we compare the average throughput of the proposed H -FSK scheme to a rate1 2 convolutionally coded BPSK modulated OFDM system. We assume that the latter transmits a BPSK modulated signal in parallel over the available frequency bands in a given time window similar to what is considered in the previous comparison setting. For fair comparison, neither scheme carries out sensing, i.e., they transmit at all times. It is worth noting that the throughput performance of this BPSK OFDM system is investigated assuming the same transmission bandwidth, power, and time frame as with the H -FSK system. As shown in Fig. 8, the average throughput of the convolutionally coded BPSK system also decreases as P On ch increases. Similar to the previous throughput comparison, there exists a probability p ∗ 2 ≈ 0 55 . such that for P On ch ≥ p ∗ 2 the H -FSK system outperforms the convolutionally coded BPSK scheme in terms of average SU's throughput. Therefore, the use of the proposed H -FSK communication scheme for values of P On ch less than p ∗ 2 is not beneficial because higher throughput values would be obtained using the other coded scheme.
## C. SU Link's Resiliency in the Presence of Multiple PUs
In this section, we further carry out our performance analysis and examine the resiliency of the network (SU's link) to interference provided by the proposed PTC based framework in the presence of multiple PUs when i) PUs always transmit and ii) PUs are dynamic and their activities vary dynamically. It is worth noting here that the PUs are located 10 m away from the SU receiver in order to analyze the quality of the SU link under a high interference scenario.
1) Static Channel Occupancy: First, we consider multiple PUs that are always active in the network during SU communication. In Fig. 9, we plot the approximate BER performance of the SU link for different PTCs in the presence of multiple PUs. The results obtained in the presence of multiple PUs that are always ON are similar to those presented in [19], [20]. These results show that the use of larger PTC (higher value of H ) adds more robustness to SU communications, i.e., BER decreases as the value of H increases. It is also shown in these results that the approximate
Fig. 9: Approximate BER performance of different PTCs with respect to PUs that are always ON and transmitting.
Fig. 10: Approximate BER versus Simulated BER for a PTC with H = 2 , 3 , and 4 and P ON ch = 0 35 . .
BER performance of an SU link degrades as the number of PUs joining the network increases. Of course, this is achieved at the expense of larger overhead when larger values of H are used.
2) Dynamic Channel Occupancy: The case of static PUs that always transmit is more of a pessimistic scenario as presented above. In this paper, we extend the work in [19], [20] and consider a more practical scenario to model the intermittent dynamic activities of PUs in the network. Using a 2-state Markov chain to model channels' occupancy by PUs, we provide a more
Fig. 11: Approximate BER performance of a PTC ( H = 4) with respect to dynamically varying PUs activities and PUs that are always ON.
accurate characterization of system performance and verify the approximate BER in the presence of a dynamically transmitting PU with simulated BER for a given PTC and known parameters of the channel occupancy model. In Fig. 10, we present the approximate BER performance of the proposed framework as a function of E s N 0 for different PTC schemes, i.e., H = 2 3 , , and 4 , and P ON ch = 0 35 . . Both analytical and simulation results are presented which match each other quite well.
In addition, in Fig. 11, we further examine the approximate BER performance of SU communications for a given PTC, i.e., H = 4 , in the presence of one, two, and three PUs that i) are static in nature and always transmit and ii) dynamic in nature. The results obtained in Fig. 11 clearly show that there exists a gap between the performance of the systems in both cases. Our analysis based on a more practical occupancy model is able to predict SU communication performance more accurately and which is better than the pessimistic case of PUs being always ON. For example, the approximate BER value in the presence of one dynamic PU is lower than when a PU is always transmitting. This is intuitive because only a fraction of the transmitted information packets are affected by interference from the dynamic PU when
Fig. 12: A BER comparison between the proposed PTC scheme with H = 4 and a 1 / 2 rate ( 64800 32400 , ) LDPC encoding with a 4 -FSK modulation system with respect to static ( P On ch = 1) and dynamic ( P On ch = 0.75) activities of a PU.
it is transmitting compared to the static PU case where all the transmitted information packets experience permanent interference. In fact, this is also observed for the case of 2 and 3 PUs being active in the network.
Moreover, we compare the BER performance of the proposed communication system when H = 4 with another system that is based on a 1 / 2 rate ( 64800 32400 , ) LDPC encoding given in Matlab toolbox with a 4 -FSK modulation system. Fig. 12 illustrates how the proposed communication system outperforms the aforementioned system for both static and dynamic activities of a PU.
## VI. CONCLUSION
In this paper, we employed a PTC based framework to mitigate the impact of PUs modeled using a practical dynamic channel occupancy model in CRNs. We, further, computed the SU link's BER approximately which was shown to be quite accurate. This approximation allows one to use BER as a QoS metric to determine the link quality of an SU link for applications such as link adaptation. Furthermore, in order to assess the effectiveness of the proposed PTC based
multi-level FSK communication scheme, we compared the performance to that of an uncoded opportunistic M -FSK system, a coded M -FSK system, and a coded BPSK modulated system deploying parallel transmissions. Based on the comparisons described in Section V, we showed that the proposed scheme outperforms the latter two under relatively heavy PU interference. We also presented results that exhibit the resiliency of an SU link to interference for PTCs in the presence of multiple dynamic PUs activities.
In this work, we evaluated the performance of PTC codes for a single SU link. Optimal code assignments in a multiple SU scenario will be considered as a future research direction. We also plan to consider 1) more realistic channel models that account for shadowing and fading effects; and 2) more effective decoding techniques, e.g., soft-decision decoding. Based on the promising results we have obtained in a CRN, we intend to generalize the proposed scheme to tackle wireless networks subject to jamming attacks and malicious nodes' activities.
## APPENDIX
## PROOF OF PROPOSITION 1
Let P ( c 1 ) be the transmitted codeword and R ( ) i be an arbitrary received vector. Assume that P ( c 2 ) is another possible codeword. If there exists a received vector R ( j ′ ) such that d H ( R ( j ′ ) , P ( c 2 )) = d H ( R ( ) i , P ( c 1 )) and P ( R ( j ′ ) |P ( c 2 )) = P ( R ( ) i |P ( c 1 )) , then the upper-bound given in (12) is the same no matter what the transmitted codeword is ( P ( c 2 ) or P ( c 1 ) ). Let s 1 = ( s 1 ,glyph[lscript]H 2 , s 1 ,glyph[lscript]H 2 +1 , . . . , s 1 ( , glyph[lscript] +1) H 2 -1 ) be a portion of P ( c 1 ) generated from the corresponding ( glyph[lscript] +1) th output branch of the convolutional code and s 2 be that of P ( c 2 ) for the same bit positions. Since the number of non-zero elements in s 1 and s 2 is H , we can assume that they occupy the following bit indices o 1 0 , , o 1 1 , , . . . , o 1 ,H -1 in s 1 and o 2 0 , , o 2 1 , , . . . , o 2 ,H -1 in s 2 , respectively. Let r 1 = ( r 1 ,glyph[lscript]H 2 , r 1 ,glyph[lscript]H 2 +1 , . . . , r 1 ( , glyph[lscript] +1) H 2 -1 ) be the corresponding received vector of s 1 . Note that r 1 here is in the form of an arbitrary received code matrix for each output branch of the convolutional code and is mapped to a permutation trellis code matrix. Then, R ( j ′ ) can be constructed from R ( ) i as provided below.
Define a one-to-one mapping π glyph[lscript] from indices { glyph[lscript]H 2 , glyph[lscript]H 2 +1 , . . . , ( glyph[lscript] +1) H 2 - } 1 to itself as
$$\mu _ { \ell } ( q ) = \begin{cases} \lambda _ { \ell } \\ c \end{cases}$$
$$\pi _ { \ell } ( q ) = \begin{cases} \ o _ { 2, i }, & \text{if $q=0_{1,i}$ for $0\leq i\leq H-1$,} \\ \ o _ { 1, i }, & \text{if $q=0_{2,i}$ for $0\leq i\leq H-1$,} \\ \ q, & \text{otherwise.} \end{cases}$$
It is easy to see that the above mapping rule switches s 1 to s 2 and vice versa. We can write π ( P ( c 1 )) = P ( c 2 ) and π ( P ( c 2 )) = P ( c 1 ) . At the same time, the mapping rule also switches r 1 ,o 1 ,i with r 1 ,o 2 ,i for 0 ≤ i ≤ H -1 when applied to r 1 . Then, the combination of glyph[lscript] mappings in π with respect to R i , results in
$$\mathbf R ^ { ( j ^ { \prime } ) } = \pi ( \mathbf R ^ { ( i ) } ) = ( r _ { 1, \pi _ { 0 } ( 0 ) }, r _ { 1, \pi _ { 0 } ( 1 ) }, \dots, r _ { 1, \pi _ { 0 } ( H ^ { 2 } - 1 ) }, r _ { 1, \pi _ { 1 } ( H ^ { 2 } ) }, \dots, r _ { 1, \pi _ { L + m - 1 } ( ( L + m - 1 ) H ^ { 2 } - 1 ) } ).$$
## Accordingly,
$$d _ { H } ( \mathbf R ^ { ( i ) }, \mathcal { P } ( \mathbf c _ { 1 } ) ) = d _ { H } ( \pi ( \mathbf R ^ { ( i ) } ) ), \pi ( \mathcal { P } ( \mathbf c _ { 1 } ) ) ) = d _ { H } ( \mathbf R ^ { ( j ^ { \prime } ) } ), \mathcal { P } ( \mathbf c _ { 2 } ) )$$
for the one-to-one mapping conserves the Hamming distance of the convolutional code. In the case where noise for each received bit is independent, we have
$$P ( { \mathbf R } ^ { ( i ) } | { \mathcal { P } } ( { \mathbf c } _ { 1 } ) ) = P ( \pi ( { \mathbf R } ^ { ( i ) } ) | \pi ( { \mathcal { P } } ( { \mathbf c } _ { 1 } ) ) ) = P ( { \mathbf R } ^ { ( j ^ { \prime } ) } | { \mathcal { P } } ( { \mathbf c } _ { 2 } ) )$$
and since noise of the received bits in a subchannel occupied by the PU is not independent, P ( R ( ) i |P ( c 1 )) is not always equal to P ( R ( j ′ ) |P ( c 2 )) . However, if we assume that the noise caused by the PU in a subchannel will affect all bits in this channel that are generated by one or more trellis branches, then
$$P ( { \mathbf R } ^ { ( i ) } | { \mathcal { P } } ( { \mathbf c } _ { 1 } ) ) = P ( \pi ( { \mathbf R } ^ { ( i ) } ) | \pi ( { \mathcal { P } } ( { \mathbf c } _ { 1 } ) ) ) = P ( { \mathbf R } ^ { ( j ^ { \prime } ) } | { \mathcal { P } } ( { \mathbf c } _ { 2 } ) ).$$
We can reason this from the fact that bit-switching by π on R ( ) i does not move any bit out of the row it is located in.
## ACKNOWLEDGMENT
This material is based on research sponsored in part by CASE: The Center for Advanced Systems and Engineering, an NYSTAR center for advanced technology at Syracuse University. Part of Han's work is supported by National Science Council, Taiwan, R.O.C., under grants NSC 101-2221-E-011 -069 -MY3 and NSC 99-2221-E-011 -158 -MY3.
## REFERENCES
- [1] I. F. Akyildiz, W.-Y. Lee, M. C. Vuran, and S. Mohanty, 'Next generation/dynamic spectrum access/cognitive radio wireless networks: A survey,' Computer Networks , vol. 50, no. 13, pp. 2127-2159, 2006.
- [2] Q. Zhao and A. Swami, 'A decision-theoretic framework for opportunistic spectrum access,' IEEE Trans. Wireless Comm. , vol. 14, no. 4, pp. 14-20, Aug. 2007.
- [3] S. Haykin, 'Cognitive radio: brain-empowered wireless communications,' IEEE J. Sel. Areas in Comm. , vol. 23, no. 2, pp. 201-220, Feb. 2005.
- [4] J. Mitola and G. Q. Maguire, 'Cognitive radio: making software radios more personal,' IEEE Personal Comm. , vol. 6, no. 4, pp. 13-18, Aug. 1999.
- [5] Y.-S. Kwon, H.-S. Kim, J.-H. Yoo, and J.-H. Chung, 'A non-interfering cognitive radio system for spectrum sharing,' in Proc. Int. Symp. on Comm. and Information Technology , Sep. 2009, pp. 664-665.
- [6] Y. Wu and Z. Yang, 'Spectrum sharing under interference temperature and SINR constraints,' in Proc. Int. Conf. on Wireless Comm., Networking and Mobile Computing , Oct. 2008, pp. 1-4.
- [7] Federal Communications Commission, 'Establishment of interference temperature metric to quantify and manage interference and to expand available unlicensed operation in certain fixed mobile and satellite frequency bands,' Et Docket 03-289, Notice of Inquiry and Proposed Rulemaking , 2003.
- [8] T. C. Clancy, 'Formalizing the interference temperature model,' Wiley J. on Wireless Comm. and Mobile Computing , vol. 7, no. 9, pp. 1077-1086, Nov. 2007.
- [9] H. Mahmoud, T. Yucek, and H. Arslan, 'OFDM for cognitive radio: merits and challenges,' IEEE Trans. Wireless Comm. , vol. 16, no. 2, pp. 6-15, Apr. 2009.
- [10] F. Shayegh and M. R. Soleymani, 'Rateless codes for cognitive radio in a virtual unlicensed spectrum,' in Proc. IEEE Sarnoff Symp. , May 2011, pp. 1-5.
- [11] G. Yue and X. Wang, 'Anti-jamming coding techniques with application to cognitive radio,' IEEE Trans. Wireless Comm. , vol. 8, no. 12, pp. 5996-6007, Dec. 2009.
- [12] J. Ko and C. Kim, 'Communication method between spectrum heterogeneous secondary users in OFDM-based cognitive radio,' Electronics Letters , vol. 47, no. 14, pp. 827-829, Jul. 2011.
- [13] C. Nicola, H. Mercier, and V. K. Bhargava, 'Error-correcting codes for dynamic spectrum allocation in cognitive radio systems,' in Proc. Int. Symp. on Signals, Syst. and Electron. , Aug. 2007, pp. 247-250.
- [14] S. Wang, Y. Wang, J. P. Coon, and A. Doufexi, 'Energy-efficient spectrum sensing and access for cognitive radio networks,' IEEE Trans. Veh. Technol. , vol. 61, no. 2, pp. 906-912, Feb. 2012.
- [15] G. Ozcan and M. C. Gursoy, 'Throughput of cognitive radio systems with finite blocklength codes,' in Proc. IEEE Annual Conf. on Inf. Sciences and Systems (CISS) , Mar. 2012, pp. 1-6.
- [16] Y.-C. Liang, Y. Zeng, E. Peh, and A. T. Hoang, 'Sensing-throughput tradeoff for cognitive radio networks,' IEEE Trans. Wireless Comm. , vol. 7, no. 4, pp. 1326-1337, Apr. 2008.
- [17] B. Shahrasbi and N. Rahnavard, 'Rateless-coding-based cooperative cognitive radio networks: Design and analysis,' in Proc. Annual IEEE on Comm. Society Conf. on Sensor, Mesh and Ad Hoc Comm. and Networking (SECON) , June 2011, pp. 224-232.
- [18] W.-C. Ao and K.-C. Chen, 'End-to-end HARQ in cognitive radio networks,' in Proc. IEEE Wireless Comm. and Networking Conf. (WCNC) , Apr. 2010, pp. 1-6.
- [19] H. C. Ferreira, A. J. H. Vinck, T. G. Swart, and I. De Beer, 'Permutation trellis codes,' IEEE Trans. Communications , vol. 53, no. 11, pp. 1782-1789, Nov. 2005.
- [20] T. G. Swart, I. De Beer, H. C. Ferreira, and A. J. H. Vinck, 'Simulation results for permutation trellis codes using M-ary FSK,' in Proc. Int. Symp. on Power Line Comm. and its Applications , Apr. 2005, pp. 317-321.
- [21] R. El Bardan, E. Masazade, O. Ozdemir, and P. K. Varshney, 'Performance of permutation trellis codes in cognitive radio networks,' in Proc. IEEE Sarnoff Symp. , May 2012, pp. 1-6.
- [22] A. Goldsmith, Wireless Communications . New York, NY, USA: Cambridge University Press, 2005.
- [23] M. Elalem, L. Zhao, and Z. Liao, 'Interference mitigation using power control in cognitive radio networks,' in Proc. IEEE Vehicular Tech. Conf. , May 2010.
- [24] J. G. Proakis, Digital Communications . McGraw -Hill, 2000.
- [25] S. Lin and D. J. Costello, Jr., Error Control Coding: Fundamentals and Applications . Prentice-Hall, 1983.
- [26] M. K. Simon, Probability Distributions Involving Gaussian Random Variables: A Handbook for Engineers, Scientists and Mathematicians . Springer-Verlag New York, 2006.
- [27] C. Stevenson, G. Chouinard, Z. Lei, W. Hu, S. J. Shellhammer, and W. Caldwell, 'IEEE 802.22: The first cognitive radio wireless regional area network standard,' IEEE Comm. Mag. , vol. 47, no. 1, pp. 130-138, Jan. 2009.
- [28] Q. Zhao, B. Krishnamachari, and K. Liu, 'On myopic sensing for multi-channel opportunistic access: structure, optimality, and performance,' IEEE Trans. Wireless Comm. , vol. 7, no. 12, pp. 5431-5440, Dec. 2008. | null | [
"Raghed El-Bardan",
"Engin Masazade",
"Onur Ozdemir",
"Yunghsiang S. Han",
"Pramod K. Varshney"
] | 2014-07-11T19:02:23+00:00 | 2014-12-12T16:17:05+00:00 | [
"cs.IT",
"math.IT"
] | Permutation Trellis Coded Multi-level FSK Signaling to Mitigate Primary User Interference in Cognitive Radio Networks | We employ Permutation Trellis Code (PTC) based multi-level Frequency Shift
Keying signaling to mitigate the impact of Primary Users (PUs) on the
performance of Secondary Users (SUs) in Cognitive Radio Networks (CRNs). The
PUs are assumed to be dynamic in that they appear intermittently and stay
active for an unknown duration. Our approach is based on the use of PTC
combined with multi-level FSK modulation so that an SU can improve its data
rate by increasing its transmission bandwidth while operating at low power and
not creating destructive interference for PUs. We evaluate system performance
by obtaining an approximation for the actual Bit Error Rate (BER) using
properties of the Viterbi decoder and carry out a thorough performance analysis
in terms of BER and throughput. The results show that the proposed coded system
achieves i) robustness by ensuring that SUs have stable throughput in the
presence of heavy PU interference and ii) improved resiliency of SU links to
interference in the presence of multiple dynamic PUs. |
1408.0261v1 | ## Analysis of the High Water Mark Convergents of Champernowne's Constant in Various Bases
John K. Sikora [email protected]
August 4, 2014
## Abstract
In this paper, we show that patterns exist in the properties of the High Water Mark (HWM) convergents of Champernowne's Constant in various bases ( C b ), specifically in bases 2 through 124. The convergents are formed by truncating the Continued Fraction Expansion (CFE) of C b immediately before the CFE HWMs. These patterns have been extended from the known patterns in the CFE HWMs of Champernowne's Constant in base ten ( C 10 ). We show that the patterns may be used to efficiently calculate the CFE coefficients of C b , and to calculate and predict the lengths of the HWM coefficients, the number of correct digits of C b as calculated by the convergent, and the convergent error. We have discovered that minor corrections to the pattern formulations in base 10 are required for some bases, and these corrections are presented and discussed. The resulting formulations may be used to make the calculations for C b in any base.
## 1 Introduction
Champernowne's Constant, C 10 , was formulated in 1933 by British mathematician and economist D. G. Champernowne as an example of a normal number [1]. It is formed by concatenating the positive integers to the right of a decimal point, i.e., 0 123456789101112 . . . . , without end. In 1937, Kurt Mahler proved that C 10 is transcendental [2].
It is well known that the CFE of C 10 consists mostly of coefficients with a reasonably small number of digits, interspersed with coefficients with a very large number of digits. A coefficient that has a value that is larger than any previous coefficient is called a HWM. As one progresses along the CFE, the pattern of relatively small coefficients sprinkled with very large coefficients continues between the HWMs, even though the other large coefficients are not themselves HWMs.
There are many patterns in the properties of the convergents formed by truncating the CFE of C 10 immediately before the HWMs for HWM numbers greater than 3. These properties and their patterns have been formulated and discussed [3, Conjectures 1-7]. The purpose of this paper is to extend the formulations and the conjectures from base 10 to bases 2 through 124, with exceptions noted and the pertinent formulae provided.
In addition, another conjecture which describes the lengths of the largest coefficients between the HWMs [4, Conjecture 8] is generalized from C 10 to C b . We show that deviations from the expected behavior occur in some bases and these deviations are noted and briefly discussed.
## 2 Definitions
Definition 1: C 10 - Champernowne's Constant in base 10.
Definition 2: C b - Champernowne's Constant in base b . For example C 3 is 0 12101112202122100101 . . . . 3
Definition 3: C b position , or position - For an integer, it is the position of the first digit of the integer in the consecutive sequence used to generate C b , starting at position 0 for the 0 to the left of the radix point.
The radix point is not counted. Therefore, for C 10 , the integer 1 is in position 1 and the integer 12 is in position 14. For C 3 , the integer 11 3 is in position 5.
Definition 4: CFE coefficient number , or coefficient number - The number of the CFE coefficient, starting at coefficient number 0. Coefficient number 0 is always 0 for C b .
Definition 5: HWM number , or HWM # - The number of the HWM. The first HWM is 0 and it is HWM number 1.
Definition 6: HWM length -The number of digits of the HWM of C b as represented in base b .
Definition 7: 10 X b -This is equal to b X and it is read one-zero, or 'ten', to the power of X in base b . This notation is used instead of b X primarily because the computer programs that were used in the analysis calculate the HWM convergents with two primary inputs: base b , and the power of 'ten', X . This terminology was chosen so that there would be consistency between this paper and the programs.
Definition 8: P 10 b X , , or C b position of 10 X b - The formula for the C b position of the integer 10 X b , X ≥ 0, is:
$$C _ { b } \text{ position of } 1 0 _ { b } ^ { X } & = P 1 0 _ { b, X } = 1 + \sum _ { x = 0 } ^ { X } ( b - 1 ) \cdot x \cdot b ^ { ( x - 1 ) } \\ \swarrow \quad \dots \quad \tilde { \lambda } \wedge \wedge \dots \quad \dots \quad \tilde { \lambda } \wedge \tilde { \lambda } \tilde { \lambda } \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \end{array}$$
Thus, the C b position of 10 0 b is 1, the position of 10 1 b is b , and the position of 10 2 3 is 15.
Definition 9: 10 X b HWMconvergent , or 10 X b convergent - The convergent formed from the numerator, N b X , , and the denominator, D b X , , as calculated using Conjecture 7 and Conjecture 6, respectively. The convergent is named in this manner because the C b position 10 X b is used in the calculation of the convergent. The CFE derived from the convergent does not calculate the 10 X b HWM coefficient as defined in Definition 10. The HWM coefficient is, however, the next term in the CFE.
Definition 10: 10 X b HWMcoefficient , or 10 X b coefficient - The CFE coefficient immediately following the last coefficient as calculated by the 10 X b convergent. With a few exceptions as noted later in this paper, these coefficients constitute the HWMs of the CFE of C b .
Definition 11: The denominator of the 10 X b HWM convergent , or D b X , -The denominator for the 10 X b HWM convergent as constructed using Conjecture 6.
Definition 12: The numerator of the 10 X b HWM convergent , or N b X , -The numerator for the 10 X b HWM convergent as calculated using Conjecture 7.
Definition 13: Number of correct digits , or NCD , or NCD b,X ( ) -The number of correct consecutive digits of C b as calculated by a convergent, starting at, and counting the 0 to the left of the radix point. The radix point itself is not counted. Therefore, if the position of the last correct digit, p c , is known, the number of correct digits is p c + 1 since the 0 is counted as the first correct digit, but it is in position 0. When given with arguments, base b , and power of 'ten' X NCD b,X , ( ) is the NCD of the 10 X b HWM convergent as defined in Definition 9.
In general, in this paper we refer to the HWM convergents in a manner different from that in our previous paper which detailed the HWM convergent properties of C 10 [3]. In the previous paper, a HWM convergent is referred to as 'the convergent before HWM #N'. In this paper a HWM convergent is referred to as 'the 10 X b HWM convergent', or as the ' 10 X b convergent' as defined in Definition 9. The reason is that in this paper, bases 2 through 124 are discussed, rather than base 10 exclusively, and the HWM numbers are different for the various bases. 1
1 The HWM numbers match for bases 6 through 124, but not for bases 2 through 5. See Conjecture 1 and Conjecture 5 for details.
## 3 Method
A program was written in Ruby to calculate C b in the desired base by taking the digits of each of the integers used to form C b , dividing by the appropriate power of the base for each digit, and summing. The Euclidean Algorithm applied to real numbers was then used to calculate the CFE coefficients of C b . The use of this method was difficult and care was taken to calculate C b to sufficient accuracy so that the desired number of coefficients could be accurately calculated. Bases 2, 7, 8, 9, and 10 were used in the initial investigation.
It was noticed that large coefficients (HWMs) in the CFE of C b appeared in other bases, similar to the behavior of the CFE of C 10 . The CFE coefficients of C b , terminated before the HWMs were then used to calculate the numerator and denominator of each convergent. Patterns were observed in the denominators of the convergents as given in Conjecture 6. The means of calculating the numerators of the convergents was determined as given in Conjecture 7.
Programs were then written in C to calculate the numerators and denominators using these conjectures and the resulting calculations led to the formulation of Conjecture 1 through Conjecture 5. These calculations were checked for each 10 X b convergent and HWM as described in Section 5.1.
## 4 Conjectures
In this section we extend the C 10 conjectures [3, Conjectures 1-7] to C b . The order of the conjectures in the present paper matches the order of the conjectures in the previous paper for ease of comparison. However, Conjecture 1 through Conjecture 5 follow from analyzing the convergents formed using Conjecture 6 and Conjecture 7.
Conjecture 1: The 10 X b convergent, for X ≥ 0 , calculates the value of C b correctly until the integer 10 X + 1 b -2 in the integer sequence used to form C b is reached. Furthermore, this integer is calculated as 10 X + 1 b -1 instead of 10 X + 1 b -2.
Exception 1.1 For base 2, the convergents that calculate C 2 per Conjecture 1 start with the 10 X 2 convergent, X ≥ 2, rather than X ≥ 0.
Exception 1.2 For base 3 and base 4, the HWM convergents that calculate C 3 and C 4 per Conjecture 1 start with the 10 X 3 and 10 X 4 convergent, X ≥ 1, rather than X ≥ 0.
The following are examples of the conjecture: the 10 1 10 HWM convergent calculates C 10 as '0 123456 . . . . 95969799 . . . ' instead of '0 123456 . . . . 95969798 . . . '. Similarly, the 10 2 5 HWM convergent calculates C 5 as '0 12341011 . . . . 440441442444 . . . 5 ' instead of '0 12341011 . . . . 440441442443 . . . 5 '.
Conjecture 2: The NCD of the 10 X b HWM convergent for base b, X ≥ 0 is given by:
$$N C D ( b, X ) = C _ { b } \text{ position of } 1 0 _ { b } ^ { X + 1 } - X - 2$$
Exception 2.1 For b = 2, Equation (4.1) is valid for X ≥ 2.
Exception 2.2 For b = 3 and b = 4, Equation (4.1) is valid for X ≥ 1.
The C b position of 10 X b is given in Equation (2.1). Thus, the NCD(10,0) is 8, the NCD(5,3) is 2340, the NCD(16,5) is 99544809, and the NCD(10,7) is 788888881.
Conjecture 3: The 10 X b HWM coefficient numbers are even. The coefficient numbers start at 0 per Definition 4.
The 10 X b HWM convergents calculate C b slightly larger than C b . Per Definition 4, this means that the 10 X b HWM coefficient numbers will be even.
This conjecture is very useful at the end of the calculation of the CFE coefficients as the algorithm cannot distinguish the degenerate case where the last coefficient is 1 from the case where the last coefficient is not 1. In the degenerate case where the last coefficient equals 1, the algorithm calculates the last term as Y +1 instead of Y , and the calculation ends on an even numbered coefficient. However, if the calculation ends on an even numbered coefficient, it is taken from Conjecture 3 that this is the degenerate case, the even numbered coefficient is Y , and the last coefficient (the one before the HWM) is 1.
## Conjecture 4: The error of the 10 X b HWM convergent is given by:
$$\text{Concrete} 4 \colon \ 1 n e \text{ error of the } 1 U _ { b } ^ { \prime } \ 1 W M \text{ convergent as given } \theta \text{.} \\ \text{error} = \begin{cases} 1. 0 \times b ^ { - ( \text{exponent} - 1 ) } & \text{base} \, b \geq 5, X = 0 \\ ( b - 1 ). 1 _ { b } \times b ^ { - \text{exponent} } & \text{base} \, b \geq 3, X = 1 \\ 1. 0 1 _ { 2 } \times b ^ { - \text{exponent} } & \text{base} \, b = 2, X = 2 \\ ( b - 1 ). \underbrace { 0 0 \dots 0 0 ( b - 1 ) ( b - 1 ) _ { b } \times b ^ { - \text{exponent} } & b = 2, X \geq 3 ; \, b \geq 3, X \geq 2 \end{cases}$$
$$\text{number of zeroes} & = X \\ \text{exponent} & = N C D ( b, X ) + X + 2$$
For example, the 10 0 8 HWM has an error of 1 0 . × 8 -7 , the 10 1 3 HWM has an error of 2 1 . 3 × 3 -15 10 , the 10 2 2 HWM has an error of 1 01 . 2 × 2 -18 10 , the 10 2 10 HWM has an error of 9 0099 . 10 × 10 -2890 10 , and the 10 5 15 HWM has an error of e.00000ee 15 × 15 -67530135 10 . Note that the mantissa is given in base b while the exponent is given in base ten, similar to the output from the GNU Multiple Precision Arithmetic (GMP) library [5] function mpf out str . This library was utilized by the programs used in the analysis.
The value of exponent is also equal to the C b position of 10 X + 1 b . Note that for the base b = 2 , X = 2 case we could have explicitly provided the value of exponent , but we wanted to show that the value of exponent was consistent for all of the given cases as formulated.
## Conjecture 5: Let:
/negationslash
$$( b = = 2 ) = \begin{cases} 0 & b \ne 2 \\ 1 & b = 2 \end{cases}$$
/negationslash
$$( b = = 4 ) = \begin{cases} 0 & b \neq 4 \\ 1 & b = 4 \end{cases}$$
The length, as represented in base b , of the HWM coefficient that occurs after the last coefficient of the CFE calculated by the 10 X b HWM convergent for X ≥ 0 is given by:
$$l e n g t h = N C D ( b, X ) - 2 \cdot N C D ( b, X - 1 ) - 3 \cdot ( X - ( b = = 2 ) ) - 2 + ( b = = 4 )$$
Exception 5.1 For base 2, Equation (4.3) is valid for X ≥ 2, rather than X ≥ 0. In addition, the 10 2 2 coefficient is coefficient number 6 and it is of length 3, per Conjecture 5. It is not a HWM but the convergent does calculate C 2 per Conjecture 1. The coefficient number of the next HWM, which is the 10 3 2 HWM coefficient, is 14. 2 All subsequent HWMs are 10 X 2 HWM coefficients.
Exception 5.2 For base 3 and base 4, Equation (4.3) is valid for X ≥ 1, rather than X ≥ 0. For base 3, the 10 1 3 HWM coefficient is coefficient number 6, and for base 4, the 10 1 4 HWM coefficient is coefficient number 12. Both are HWMs, and subsequent HWMs are 10 X b HWM coefficients.
2 The 10 2 2 coefficient has a value of 5, one less than the value of coefficient numbers 2 and 5, which are the largest terms along with coefficient number 10 until the 10 3 2 HWM coefficient at coefficient number 14.
Exception 5.3 For base 5, the 10 0 5 coefficient is coefficient number 4 and it is of length 1, per Conjecture 5. 3 It is not a HWM but the convergent does calculate C 5 per Conjecture 1. The coefficient number of the 10 1 5 HWM coefficient is 10. The HWMs at coefficient number 7 and coefficient number 9 are not 10 X 5 HWM coefficients.
For example, the 10 3 2 HWM as represented in binary has a length of 9, the 10 2 10 HWM as represented in decimal has a length of 2504, and the 10 5 16 HWM as represented in hexadecimal has a length of 89198852.
For the 10 0 b coefficient, b ≥ 5, the coefficient length is b -4. For bases greater than 5, the 10 0 b coefficient is a HWM, and it will be HWM #4 at coefficient number 4, per Definition 4 and Definition 5. All subsequent HWMs are those generated by a 10 X b convergent, X ≥ 1. 4
Furthermore, it appears that for b ≥ 5, the first 4 CFE terms are [0; b -2, b -1, 1] = [0; b -2, b ]. This gives a 10 0 b convergent of
$$\Phi ^ { \Phi } _ { b } ^ { 0 } \text{ convergent} = \frac { b } { ( b - 1 ) ^ { 2 } }, \ b \geq 5$$
As stated above, the 10 0 b convergent is a HWM for b ≥ 6. For the numerators and denominators for other convergents, see Equation (4.6) and Equation (4.5), respectively.
It should be noted that for the lowest value of X allowed in any base, the NCD b,X ( -1) term is not the true NCD as the 10 X -1 b convergent is not a valid HWM convergent. However, the calculated value as used in Equation (4.3) gives the correct HWM length. It should also be reiterated that the HWM following the 10 X b HWM convergent is not calculated by the convergent. It is calculated by the 10 X + 1 b HWM convergent, or by other means [6].
It is unknown why the HWM lengths in base 4, as formulated, are longer by 1 digit than the HWM lengths in other bases (other than base 2). It is also unknown why the HWM lengths in base 2, as formulated, are longer by 3 digits than the HWM lengths in other bases (other than base 4). However, in the base 2 case, we feel that since there are no 10 X b HWM convergents for X < 2, this may increase the lengths of the HWM coefficients.
Conjecture 6: The denominator, D b 0 , , used to calculate the 10 0 b HWMconvergent for base, b ≥ 5 , X = 0 , is given in Equation (4.4) . The denominator used to calculate the 10 X b HWM convergent for base, b, X ≥ 1 , is given by:
$$\text{Convecue } \colon \text{$1$ line} \text{in} \text{in}, \text{$1$b,0$, use } \colon \text{$2$} \colon \text{$3$} \colon \text{$4$} \colon \text{$5$} \colon \text{$6$} \colon \text{$7$} \colon \text{$8$} \colon \text{$9$} \colon \text{$10$} \colon \text{$11$} \colon \text{$12$} \colon \text{$13$} \colon \text{$14$} \colon \text{$15$} \colon \text{$16$} \colon \text{$17$} \colon \text{$18$} \colon \text{$19$} \colon \text{$20$} \colon \text{$21$} \colon \text{$22$} \colon \text{$23$} \colon \text{$24$} \colon \text{$25$} \colon \text{$26$} \colon \text{$27$} \colon \text{$28$} \colon \text{$29$} \colon \text{$30$} \colon \text{$31$} \colon \text{$32$} \colon \text{$32$} \colon \text{$32$} \colon \text{$32$} \colon \text{$32$} \colon \text{$32$} \colon \text{$32$} \colon \text{$32$} \colon \text{$32$} \colon \text{$32$} \colon \text{$32$} \colon \text{$32$} \colon \text{$32$} \colon \text{$32$} \colon \text{$32$} \colon \text{$32$} \colon \text{$32$} \colon \text{$32$} \colon \text{$32$} \colon \text{$32$} \colon \text{$64$} \colon \text{$65$} \colon \text{$66$} \colon \text{$67$} \colon \text{$68$} \colon \text{$69$} \colon \text{$70$} \colon \text{$71$} \colon \text{$72$} \colon \text{$72$} \colon \text{$72$} \colon \text{$72$} \colon \text{$72$} \colon \text{$72$} \colon \text{$72$} \colon \text{$72$} \colon \text{$72$} \colon \text{$72$} \colon \text{$72$} \colon \text{$72$} \colon \text{$72$} \colon \text{$72$} \colon \text{$72$} \colon \text{$72$} \colon \text{$72$} \colon \text{$72$} \colon \text{$72$} \colon \text{$72$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$72$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$72$} \colon \text{$72$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$72$} \colon \text{$73$} \colon \text{$72$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$72$} \colon \text{$72$} \colon \text{$72$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$72$} \colon \text{$73$} \colon \text{$73$} \colon \text{$72$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$72$} \colon \text{$73$} \colon \text{$72$} \colon \text{$72$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$72$} \colon \text{$72$} \colon \text{$73$} \colon \text{$72$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$72$} \colon \text{$72$} \colon \text{$72$} \colon \text{$72$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$72$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$72$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$72$} \colon \text{$73$} \colon \text{$73$} \colon \text{$72$} \colon \text{$72$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$72$} \colon \text{$73$} \colon \text{$72$} \colon \text{$73$} \colon \text{$72$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$72$} \colon \text{$73$} \colon \text{$72$} \colon \text{$72$} \colon \text{$72$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$72$} \colon \text{$72$} \colon \text{$73$} \colon \text{$73$} \colon \text{$72$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$72$} \colon \text{$72$} \colon \text{$73$} \colon \text{$72$} \colon \text{$72$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$73$} \colon \text{$72$} \$$
3 The coefficient is 1.
4 There is a close call for base 6 as coefficient number 8 is not a 10 X 6 coefficient and it is only one less than coefficient number 4, the 10 0 6 coefficient.
5 Although the 10 1 2 HWM convergent is not valid, NCD (2 , 1) = 3 which gives a valid exponent of 8 for the 10 2 2 HWM convergent denominator, 0 110001 . 2 × 2 8 10 .
For example, D 2 4 , , the denominator for the 10 4 2 HWM convergent is 0 1111000001 . 2 × 2 54 10 , D 5 3 , is 4 4430001 . 5 × 5 348 10 , D 10 8 , is 4 999999990000000005 . 10 × 10 788888898 10 , and D 15 5 , is e.eeeed000001 15 × 15 3742640 10 .
If X = 1 for an odd base, then the ( b -1) term is to the left of the radix point and the ( b -2) term is immediately to the right of the radix point. For example, the denominator for the 10 1 7 HWM convergent is 6 501 . 7 × 7 8 10 .
Conjecture 7: The numerator, N b 0 , , used to calculate the 10 0 b HWM convergent for base, b ≥ 5 , X = 0 , is given in Equation (4.4) . For other b and X N , b X , is calculated as described below.
## Given:
- 1. The denominator, D b X , , of a 10 X b convergent as calculated from Conjecture 6
- 2. The C b position of the integer 10 X b , P 10 b X , , as given by Equation (2.1)
- 3. C b [0 1 . . . . P 10 b X , ] denotes C b truncated to position P 10 b X , 6
- 4. Ceil ( f ) denotes the integer immediately above non-integer f .
Then the numerator, N b X , , X ≥ 1 , of the convergent is calculated as:
$$N _ { b, X } = \begin{cases} C e i l ( D _ { b, X } \cdot C _ { b } [ 0. 1 \dots P 1 0 _ { b, X } ] ) + 1 & b \text{ is odd} \\ C e i l ( D _ { b, X } \cdot C _ { b } [ 0. 1 \dots P 1 0 _ { b, X } ] ) & b \text{ is even} \end{cases}$$
Exception 7.1 For base 2, Equation (4.6) is valid for X ≥ 2, rather than X ≥ 1.
For example, the calculation for N 2 3 , , the numerator for the 10 3 2 HWM convergent is: Ceil(0 11100001 . 2 × 2 21 10 · 0 110111001011101111 ) = 110000100000000100001 . . 2 2
The calculation for N 3 2 , is:
Ceil(2 21001 . 3 × 3 17 10 · 0 121011122021221 ) + 1 = 112222220011222111 . . 3 3
The calculation for N 10 1 , is:
Ceil(4 9005 . 10 × 10 11 10 · 0 1234567891 . 10 ) = 60499999499 10 .
And the calculation for N 15 1 , is:
Ceil(e.d01 15 × 15 16 10 · 0 123456789abcde1 . 15 ) + 1 = 120eeeeeeeeeedeed 15 .
## 5 Results, Verification, and Data Location
## 5.1 Results and Verification
The C b CFE coefficients may be calculated very efficiently from the 10 X b HWM convergents. For X ≥ 1, the denominator of the convergent is given by Conjecture 6 and the numerator of the convergent may be calculated by multiplying the denominator by a sufficient number of digits of C b . The required number of digits is P 10 b X , + 1 as given in Conjecture 7, resulting in a very efficient calculation for the C b CFE coefficients. 7 The Euclidean Algorithm is then used to calculate the CFE coefficients from the numerator and the denominator. Conjecture 3 is used to determine if the last coefficient is 1. For b ≥ 5 , X = 0, Equation (4.4) is used to calculate the numerator and the denominator.
The properties of the convergents as given in Conjecture 1 through Conjecture 4 may be observed by dividing the numerator of the convergent by the denominator of the convergent and comparing the result with C b for each base and power. The property of the convergents as given in Conjecture 5 may be checked by examination of the C b CFE coefficients.
In the current study, the calculations and comparisons were performed for bases 2 to 124. Table 1 gives the base, b , and the powers of 'ten', X , in base b that were computed. The calculated results for each base and each power were checked against the predictions made in Conjecture 1 through Conjecture 5.
6 The last digit is a 1 as it is the first digit of 10 X b in the integer sequence used to form C b .
7 The required number of digits is P 10 b X , +1 because the zero to the left of the radix point has been included in the count.
The maximum power for each base was usually dictated by memory size. The amount of memory required to calculate C b , the C b CFE coefficients, and the error for the 10 X -1 b HWMconvergent is approximately the same as the amount of memory required to calculate the C b CFE coefficients for the 10 X b HWM convergent, without calculating C b and the error. There are separate columns for each type of calculation in Table 1. For more details, see Section 5.2 and Section 7.2.
Sequences A244330 [8], A244331 [9] for base 2, and A244332 [10], and A244333 [11], for base 3 have been added to the OEIS which correspond to sequences A038705 [12] and A143534 [13] in base 10, respectively. The first member of each pair of sequences gives the CFE coefficient number of each HWM coefficient and the second gives the length of each HWM coefficient as represented in the specified base. 8
In addition, sequences A244758 [14] for base 2, and A244759 [15] for base 3 have been added to the OEIS which correspond to sequence A143532 [16] in base 10. These sequences give the number of digits as represented in the specified base for all of the calculated CFE coefficients.
We are considering adding similar sequences for base 4 through base 9. For these bases, there are existing OEIS sequences for the digits of C b , but not for the CFE coefficients of C b . Therefore, sequences for the CFE coefficients of C b would be added as well as sequences for the HWM coefficients, the coefficient number of the HWMs, the length of all of the calculated coefficients, and the length of the HWM coefficients.
It should be noted that the maximum numbers of CFE coefficients as calculated and presented in Table 1 are not necessarily the maximum number of coefficients that have been calculated in a particular base. For example, at present for C 10 , 82328 coefficients have been calculated [17].
Table 1: Calculated Bases, Powers, and Number of CFE Coefficients
| base b | Min X | Max X w/ error | Max X w/o error | Number of CFE Coefficients (for max X w/o error) |
|----------|---------|------------------|-------------------|----------------------------------------------------|
| 2 | 2 | 25 | 24 | 98093504 |
| 3 | 1 | 15 | 16 | 2982556 |
| 4 | 1 | 12 | 13 | 629420 |
| 5 | 0 | 10 | 11 | 195554 |
| 6 | 0 | 9 | 10 | 105806 |
| 7 | 0 | 8 | 9 | 53596 |
| 8-10 | 0 | 7 | 8 | base 8: 26362 base 10: 34062 |
| 11-14 | 0 | 6 | 7 | base 11: 14424 base 14: 16386 |
| 15-16 | 0 | 5 | 6 | base 15: 6080 base 16: 6258 |
| 17-27 | 0 | 4 | 5 | base 17: 2382 base 27: 3184 |
| 28-62 | 0 | 3 | 4 | base 28: 1104 base 62: 1830 |
| 63-90 | 0 | 2 | 3 | base 63: 540 base 90: 574 |
| 91-124 | 0 | 1 | 2 | base 91: 136 base 124: 128 |
A tabular view of the properties for a specific base, 9, appears in Table 2. To the best of our knowledge, the values in the shaded cells are predictions, and they have not been confirmed. A similar table, for base 10, is given in [3, Table 1].
## 5.2 Data Location and Digit Symbols
The computed data may be downloaded [19] and found under the 'data' folder. The data is available under the Open Data Commons Public Domain Dedication and License (ODC PDDL) [20].
8 The coefficient numbers start at 1 per the precedent set in A038705 [12].
Table 2: C 9 HWM Convergents Summary
| Power of 10 9 , X | HWM Coefficient Number (next CFE term) | Integer that Fails ; Fails As, in Convergent Calculation | NCD ( 9 , X ) | Convergent Error | Convergent Denominator | Length of HWM (next CFE term) |
|---------------------|------------------------------------------|------------------------------------------------------------|----------------------------|--------------------------------------|-------------------------------------------|---------------------------------|
| 0 | 4 | 7 9 ; 8 9 | 7 | 1 . 0 × 9 - 8 10 | 64 10 | 5 |
| 1 | 16 | 87 9 ; 88 9 | 150 | 8 . 1 9 × 9 - 153 10 | 8 . 701 9 × 9 10 10 | 131 |
| 2 | 52 | 887 9 ; 888 9 | 2093 | 8 . 0088 9 × 9 - 2097 10 | 8 . 87001 9 × 9 155 10 | 1785 |
| 3 | 152 | 8887 9 ; 8888 9 | 25420 | 8 . 00088 9 × 9 - 25425 10 | 8 . 8870001 9 × 9 2100 10 | 21223 |
| 4 | 492 | 88887 9 ; 88888 9 | 287859 | 8 . 000088 9 × 9 - 287865 10 | 8 . 888700001 9 × 9 25429 10 | 237005 |
| 5 | 1598 | 888887 9 ; 888888 9 | 3122210 | 8 . 0000088 9 × 9 - 3122217 10 | 8 . 88887000001 9 × 9 287870 10 | 2546475 |
| 6 | 4512 | 8888887 9 ; 8888888 9 | 32882905 | 8 . 00000088 9 × 9 - 32882913 10 | 8 . 888887 0000001 9 × 9 3122223 10 | 26638465 |
| 7 | 12164 | 88888887 9 ; 88888888 9 | 338992920 | 8 . 000000088 9 × 9 - 338992929 10 | 8 . 8888887 00000001 9 × 9 32882920 10 | 273227087 |
| 8 | 30410 | 888888887 9 ; 888888888 9 | 3438356831 | 8 . 0000000088 9 × 9 - 3438356841 10 | 8 . 88888887 000000001 9 × 9 338992937 10 | 2760370965 |
| | Conjecture 3 (even numbered) | Conjecture 1 | Conjecture 1, Conjecture 2 | Conjecture 1, Conjecture 4 | Conjecture 6, Equation (4.4) | Conjecture 5 |
The downloaded data must be unzipped and extracted. The data includes the numerator of the convergent and the CFE coefficients as calculated from the convergent for each base, b , and power of 'ten', X , as given in Table 1 under the 'Max X w/o error' column. For the bases and powers under the 'Max X w/ error' column, the data also includes C b as calculated by the convergent as well as the convergent error. However, for b = 2, X = 25, the CFE coefficients are not given due to the lengthy calculation time. 9
The base, b , and power, X , appear in each file name. The following prefixes identify the types of files:
- · cn calc
- -C b as calculated by the convergent
- · cn cfe coeffs
- · cn error
- -the CFE coefficients calculated from the convergent
- -C b as calculated by the convergent minus C b (approximately 30 digits)
- · cn numer
- -the numerator of the convergent
It should be noted that there are various suffixes, after the base and power, that indicate the specific program that was used to generate the data. These suffixes include, for example, whether or not the error was calculated.
The symbols used for the digits in various bases are summarized in Table 3. Of course, for any base, b , the symbol range is valid only for digits 0 through b -1. Referring to the table, it may be seen, for example, that when examining a data file for a base 36 calculation, the symbol for the digit 35 10 is 'z'. However, when examining a data file for a base 37 calculation, the symbol for the digit 35 10 is 'Z' and the symbol for the digit 36 10 is 'a'. When examining a data file for a base 64 calculation, the symbol for the digit 63 10 is ' 1', with the symbol representing a single character. |
Table 3: Digit Symbols for the Various Bases
| | Digit Symbols | Digit Symbols | Digit Symbols | Digit Symbols | Digit Symbols | Digit Symbols |
|--------|-----------------|-----------------|-----------------|-----------------|-----------------|-----------------|
| Base b | 0-9 | 10-35 | 36-61 | 62-71 | 72-97 | 98-123 |
| 2-10 | 0-9 | | | | | |
| 11-36 | 0-9 | a-z | | | | |
| 37-62 | 0-9 | A-Z | a-z | | | |
| 63-72 | 0-9 | A-Z | a-z | | 0- | 9 | | |
| 73-98 | 0-9 | A-Z | a-z | | 0- | 9 | | a- | z | |
| 99-124 | 0-9 | A-Z | a-z | | 0- | 9 | | A- | Z | | a- | z |
Discussions of the results of the processing of the data and compilations of the data may be found in Section 7.4.
## 6 2 nd Generation HWM Length Extension to Bases 2 through 62
It is well known that in addition to the C 10 HWM CFE coefficients, there are other coefficients that have a large number of digits that are not themselves HWMs. For example, coefficient 101 (starting from 0) is 140 digits long, but it is not a HWM since coefficient number 40 is 2504 digits long and coefficient number 18 is 166 digits long. Coefficient 101 is an example of what we have designated a 2 nd Generation HWM [4].
It is known that for C 10 , the lengths of the 2 nd Generation HWMs appear to be related to the lengths of the HWMs [4, Conjecture 8]. The differences in the lengths of a HWM and its corresponding 2 nd Generation HWM show a regular pattern, increasing by 10 for each HWM number increase. We have found that, in general, the same holds true for bases 2 through 62: the difference in the lengths of a HWM and its corresponding 2 nd Generation HWM increases by 10 for each HWM number increase. 10,11 However, there were irregularities in some bases as can be seen by the shaded cells in Table 4, which is a compilation of the data. Note, for example, the entries for bases 4, 5, 9, 22, and 36.
It should also be noted that starting at base 24, the first 2 nd Generation HWM starts after HWM #5 rather than after HWM #6. Further analysis may show that this is true for base 23 and possibly base 22,
9 The calculation is in progress.
10 The length is the number of digits as represented in the base of C b .
11 No analysis was performed for bases 63 through 124.
but this would most likely involve looking at the error of the potential 2 nd Generation HWM convergents [4, Conjecture 9].
Table 4: Bases, Difference Between HWM Lengths and 2 nd Generation HWM Lengths
| base | | HWM Length - 2 nd Generation HWM Length | HWM Length - 2 nd Generation HWM Length | HWM Length - 2 nd Generation HWM Length | HWM Length - 2 nd Generation HWM Length | HWM Length - 2 nd Generation HWM Length | HWM Length - 2 nd Generation HWM Length | HWM Length - 2 nd Generation HWM Length | HWM Length - 2 nd Generation HWM Length | HWM Length - 2 nd Generation HWM Length | HWM Length - 2 nd Generation HWM Length | HWM Length - 2 nd Generation HWM Length | HWM Length - 2 nd Generation HWM Length | HWM Length - 2 nd Generation HWM Length |
|--------|----|-------------------------------------------|-------------------------------------------|-------------------------------------------|-------------------------------------------|-------------------------------------------|-------------------------------------------|-------------------------------------------|-------------------------------------------|-------------------------------------------|-------------------------------------------|-------------------------------------------|-------------------------------------------|-------------------------------------------|
| 2 | 80 | 90 | 100 | 110 | 120 | 130 | 140 | 150 | 160 | 170 | 180 | 190 | 200 | 210 |
| 3 | 48 | 58 | 68 | 78 | 88 | 98 | 108 | 118 | 128 | 138 | 148 | | | |
| 4 | 36 | 47 | 57 | 67 | 77 | 87 | 97 | 107 | 117 | | | | | |
| 5 | 25 | 37 | 47 | 57 | 67 | 77 | 87 | 97 | | | | | | |
| 6 | 25 | 35 | 45 | 55 | 65 | 75 | 85 | | | | | | | |
| 7 | 27 | 37 | 47 | 57 | 67 | 77 | | | | | | | | |
| 8 | 25 | 35 | 42 | 55 | 65 | 75 | | | | | | | | |
| 9 | 24 | 37 | 47 | 57 | 67 | | | | | | | | | |
| 10 | 26 | 36 | 46 | 56 | 66 | | | | | | | | | |
| 11 | 27 | 37 | 47 | 57 | 67 | | | | | | | | | |
| 12 | 26 | 36 | 46 | 56 | | | | | | | | | | |
| 13 | 27 | 37 | 47 | 57 | | | | | | | | | | |
| 14 | 26 | 36 | 46 | 56 | | | | | | | | | | |
| 15 | 27 | 37 | 44 | 57 | | | | | | | | | | |
| 16 | 26 | 36 | 46 | 56 | | | | | | | | | | |
| 17 | 27 | 37 | 47 | | | | | | | | | | | |
| 18 | 24 | 36 | 46 | | | | | | | | | | | |
| 19 | 27 | 37 | 47 | | | | | | | | | | | |
| 20 | 26 | 36 | 46 | | | | | | | | | | | |
| 21 | 27 | 37 | 47 | | | | | | | | | | | |
| 22 | 26 | 36 | 43 | | | | | | | | | | | |
| 23 | 27 | 37 | 47 | | | | | | | | | | | |
| 24 | 16 | 26 | 36 | 46 | | | | | | | | | | |
| 25 | 17 | 27 | 37 | 47 | | | | | | | | | | |
| 26 | 14 | 24 | 36 | 46 | | | | | | | | | | |
| 27 | 17 | 27 | 37 | 47 | | | | | | | | | | |
| 28 | 16 | 26 | 36 | | | | | | | | | | | |
| 29 | 15 | 27 | 37 | | | | | | | | | | | |
| 30 | 16 | 26 | 36 | | | | | | | | | | | |
| 31 | 17 | 27 | 37 | | | | | | | | | | | |
| 32 | 16 | 26 | 34 | | | | | | | | | | | |
| 33 | 17 | 27 | 37 | | | | | | | | | | | |
| 34 | 16 | 26 | 36 | | | | | | | | | | | |
| 35 | 17 | 25 | 37 | | | | | | | | | | | |
| 36 | 15 | 26 | 36 | | | | | | | | | | | |
| 37 | 17 | 27 | 34 | | | | | | | | | | | |
| 38 | 16 | 26 | 36 | | | | | | | | | | | |
| 39 | 17 | 27 | 37 | | | | | | | | | | | |
| 40 | 16 | 26 | 36 | | | | | | | | | | | |
| 41 | 17 | 27 | 37 | | | | | | | | | | | |
| 42 | 16 | 26 | 36 | | | | | | | | | | | |
| 43 | 14 | 25 | 37 | | | | | | | | | | | |
| 44 | 16 | 26 27 | 36 37 | | | | | | | | | | | |
| 45 | 17 | | | | | | | | | | | | | |
continued on next page
Table 4 - continued from previous page
| base | HWM Length - 2 nd Generation HWM Length | HWM Length - 2 nd Generation HWM Length | HWM Length - 2 nd Generation HWM Length | HWM Length - 2 nd Generation HWM Length |
|--------|-------------------------------------------|-------------------------------------------|-------------------------------------------|-------------------------------------------|
| 46 | 16 | 26 | 36 | |
| 47 | 17 | 27 | 37 | |
| 48 | 16 | 26 | 36 | |
| 49 | 17 | 27 | 37 | |
| 50 | 15 | 26 | 36 | |
| 51 | 17 | 27 | 37 | |
| 52 | 16 | 24 | 36 | |
| 53 | 17 | 27 | 37 | |
| 54 | 16 | 26 | 36 | |
| 55 | 17 | 27 | 37 | |
| 56 | 16 | 26 | 36 | |
| 57 | 16 | 27 | 37 | |
| 58 | 16 | 26 | 36 | |
| 59 | 17 | 27 | 37 | |
| 60 | 15 | 24 | 36 | |
| 61 | 17 | 27 | 37 | |
| 62 | 16 | 26 | 36 | |
## 7 Notes on Computing
## 7.1 Ruby Programs
As mentioned in Section 3, a program was written in Ruby to look for the patterns in the CFE coefficients of C b in various bases. In particular, the HWM coefficients were noted and another Ruby program was written to calculate the numerators and the denominators, as represented in the particular base, of the convergents terminated immediately before the HWM coefficients. The patterns in the denominators and the methods of calculating the numerators as given in Conjecture 6 and Conjecture 7, respectively, were determined. The main CFE calculating programs were then written in C due to speed considerations. Other programs were written in Ruby and C to compile the statistics from the C program outputs and to formulate and check the conjectures.
## 7.2 C Programs
The main C b CFE coefficient calculation programs were written in C. Both the MPIR 32 bit [7] and GMP 64 bit multiple precision libraries were used. In general, the GMP library was used for the highest power of 'ten', X , for each base due to the speed advantage of GMP on Linux versus MPIR on Windows and also due to the larger data handling capability of the 64 bit GMP library versus the 32 bit MPIR library.
There are three main types of programs. The first two main types of programs each have two subtypes since the limit on the input and output base range of the MPIR and GMP libraries is 2 to 62, while the study was conducted to base 124. The last main type finds use only for low bases and large values of X , so the extension to higher bases is unnecessary. The following list shows the main types and subtypes of programs:
- · Programs that calculate the convergent, the CFE coefficients of the convergent, C b as calculated by the convergent, and the convergent error
- -Programs for bases 2 through 62
- -Programs for bases 63 through 124
- · Programs that calculate the convergent and the CFE coefficients of the convergent
- -Programs for bases 2 through 62
- -Programs for bases 63 through 124
- · A program that calculates the convergent, C b as calculated by the convergent, and the convergent error
Table 5 gives the calculation times for the various bases and powers of 'ten', X . See Section 7.5 for more details on the calculations. It may be observed that the base 63 through base 124 calculation times increase rapidly with X as the base conversion processing is not as efficient as the MPIR and GMP input and output functions.
Currently, for bases from approximately 4 to 62, the limitation in these calculations for the maximum power, X , for a given base, b , is usually memory. However, for the lower bases, particularly base 2 and base 3, and higher powers, the number of coefficients that are calculated becomes very high. In these cases, the calculation time of the CFE coefficients becomes a major factor. As more memory becomes available, the CFE coefficient calculation time may become more of a factor for bases higher than base 3 at the higher powers.
The program that calculates the convergent, C b as calculated by the convergent, and the convergent error, but not the CFE coefficients was written to further check the validity of the applicable conjectures for the highest power of ' ten' in the lower bases without the very time consuming calculation of the coefficients. This program may also be used for higher bases but the calculation time savings is not nearly as beneficial.
For bases 63 to 124, the limitation is calculation time rather than memory due to the inefficient base conversion. However, as the base increases, the difference in the number of digits between powers of 'ten' increases, and this factors into the increase in calculation time as well.
## 7.3 Program Location and Preferred Programs
All of the programs that were used in the study, including ones that are not mentioned, are available under the GNU LGPL [18]. To download the programs, see reference [19], under the 'programs' folder. As there are many programs, some of which were used to check results, the preferred program for each main type used in the HWM convergent calculations is indicated in Table 6. The preferred programs for MPIR are Microsoft Visual Studio C++ projects, while the GMP programs consist of a C program and a C header file. For instructions on how to run the programs, see the readme mpir.txt and readme gmp.txt files that may be found in the appropriate folders.
## 7.4 Other Programs and Data of Interest
As already mentioned, several programs were written to process the data and compare the output of the main programs with the conjectures. One processed data set that we feel may be of general interest is a list of the CFE coefficient lengths for all coefficient numbers, not just the HWMs. A separate file has been created for each base, b , and the maximum power of 'ten', X , in which the coefficients were calculated. In some of the lower base cases, where the files may be very large (over 1 GB), compilations for lower values of X are included.
Once the vcpp.tar.gz file has been unzipped and extracted, these files may be found in the following folders:
- · get hwm lengths generic base
- · get hwm lengths generic base auto
- · get hwm lengths generic base v hi base auto
The name of the file includes b and X . An example name is cn cfe coeffs base 2 pow 10 24 no err all coeffs lengths.txt .
There are other files in each of these directories for each b and X . These files have been processed from the files containing all of the coefficient lengths. The files with powers at the end of the name give the HWM lengths for each b and X .
Table 5: Calculation Times
| | | calculation time in seconds | calculation time in seconds | calculation time in seconds |
|--------|----------|-------------------------------|-------------------------------|-------------------------------|
| base b | X | CFE Calculation with Error | CFE Calculation without Error | With Error, No Coefficients |
| 2 | 25 | | | 219 |
| 3 | 24 15 16 | 21551 | 2041980 176523 | 105 |
| 4 | 12 13 | 6048 | | |
| | 10 | | 59031 | |
| 5 | | 1206 | | |
| | 11 | | 13169 | |
| | 9 | 869 | | |
| 6 | 10 | | 8908 | |
| 7 | 8 | 397 | | |
| | 9 | | 3161 | |
| 8-10 | 7 | base 8: 56 base 10: 928 | | |
| | 8 | | base 8: 569 base 10: 5524 | |
| 11-14 | 6 | base 11: 130 base 14: 903 | | |
| | 7 | | base 11: 476 base 14: 3330 | |
| 15-16 | 5 | base 15: 60 base 16: 36 | | |
| | 6 | | base 15: 139 base 16: 152 | |
| 17-27 | 4 | base 17: 5 base 27: 88 | | |
| | 5 | | base 17: 9 base 27: 172 | |
| | 3 | base 28: 1 | | |
| 28-62 | | base 62: 82 | | |
| | 4 | | base 28: 3 base 62: 144 | |
| 63-90 | 2 | base 63: 5271 base 90: 51821 | | |
| | 3 | | base 63: 14275 base 90: 53217 | |
| 91-124 | 1 | base 91: 1 base 124: 3 | | |
| | 2 | | base 91: 1 base 124: 3 | |
Table 6: Preferred Programs
| Type of Program | Applicable Bases | Preferred Program |
|---------------------|--------------------|--------------------------------------------------------------------------------------------------|
| w/ error | 2-62 | MPIR: cn generic base w err no itoa auto GMP: cn generic base gmp w err auto |
| w/ error | 63-124 | MPIR: cn generic base w err very hi base auto GMP: cn generic base gmp w err very hi base auto |
| w/o error | 2-62 | MPIR: cn generic base no err no itoa auto GMP: cn generic base gmp no err |
| w/o error | 63-124 | MPIR: cn generic base no err very hi base auto GMP: cn generic base gmp no err very hi base auto |
| w/ error, no coeffs | 2-62 | GMP: cn generic base gmp w err auto no coeffs calc |
## 7.5 Hardware
The results given in Table 5 were obtained on a desktop computer with a 3.40 GHz i7 -2600 CPU, running a single thread, with 8 GB of 1333 MHz DDR3 SDRAM. The programs were run with GMP on Linux, and GMP was built via the command line using --build=x86 64 -intel -linux -gnu to optimize GMP to 64 bit operation on an Intel x86 processor. Additional tuning per GMP instructions was attempted, but no performance increase was observed. Building GMP in this manner resulted in code that ran approximately twice as fast as the generic C code.
The MPIR programs were run in 32 bit mode on Windows and the build mode was for generic C code. The GMP programs ran about 8 times faster than the MPIR programs.
## 8 Suggestions for Future Analysis
## 8.1 Extending the powers of 'ten'
Naturally, extending the maximum power of 'ten', X , for each base and checking the calculations against the conjectures is desirable. Generally, the amount of memory required increases by a factor of approximately 1 25 . · base for each increment in power. When extending the calculations to a higher power of X , we feel that it is very important to make the calculations with the program that calculates C b as calculated by the convergent as well as with the program that calculates the CFE coefficients without calculating C b , with no more than one power difference between the two. We believe that ensuring that the calculation of C b conforms to Conjecture 1 for power X -1 increases the credibility of the calculations of the CFE coefficients for power X .
As explained in Section 7.2, for the lower bases, computing time becomes problematic. In general, for bases 62 and lower, each increment in power increases the time by a factor of approximately 2 5 . · base , with a minimum of approximately a factor of 8 for base 2 and high powers of X .
## 8.2 Analysis of the ratio of adjacent HWM coefficient numbers
We feel that it would be interesting to compare the ratio of the coefficient numbers of successive HWMs in the various bases. A base dependent formula may exist for the limit of this ratio as the power of 'ten', X increases without bound.
## 8.3 More efficient algorithm for calculating the CFE coefficients
Implementation of a faster, more efficient algorithm to calculate the CFE coefficients would be of great benefit for the lower bases and higher powers of 'ten'.
## 8.4 Further analysis of the 2 nd Generation HWM coefficients
Further analysis of the 2 nd Generation HWMs is suggested to see if the difference in HWM and 2 nd Generation HWM lengths stabilizes at 10 as the powers of 'ten' go higher. It would also be of interest to see if the 2 nd
Generation HWM convergent error, and denominators in other bases have similar patterns to those in base 10 [4].
## 8.5 More efficient base conversion for bases greater than base 62
The current programs for bases greater than base 62 have inefficient methods for calculating C b and for converting between bases. This results in greatly increased calculation times for higher powers of 'ten'. Therefore, it is desired that more efficient algorithms be employed.
## 9 Conclusions
We have determined that patterns exist in the CFE HWM convergents for Champernowne's Constant in bases 2 to 124, similar to the patterns that exist in the CFE HWM convergents for Champernowne's Constant in base 10. These patterns have been presented in the form of formulae in conjectures. In some bases, slight changes in the formulae are required. There are also a few exceptions to the conjectures, depending upon the base and the power of 'ten'. These differences and exceptions have been presented and discussed.
A cursory analysis of the 2 nd Generation HWM lengths in base 2 to base 62 has been presented and the results and exceptions to the expected results have been discussed. It was suggested that further study of the 2 nd Generation HWM convergents would be worthwhile.
The programs used in the analysis, and the data calculated by the programs, were discussed. The location of the programs and data was given. The programs are available under the GNU LGPL. The data is available under the ODC PDDL.
## References
- [1] D.G. Champernowne, 1933: The construction of decimals normal in the scale of ten , J. London Math. Soc. 8, p. 254-260.
- [2] K. Mahler, 1937: Arithmetische eigenschaften einer klasse von dezimalbrchen , Proc. Konin. Neder. Akad. Wet. Ser. A. 40, p. 421-428.
- [3] John K. Sikora, 2012: On the high water mark convergents of Champernowne's Constant in base ten , http://arxiv.org/abs/1210.1263 [math.NT], p. 4-10.
- [4] Ibid., p. 12-18.
- [5] T. Granlund, 2012: GNU multiple precision arithmetic library, version 5.0.5 , http://gmplib.org/
- [6] E. Weisstein, 2013: Champernowne Constant , http://mathworld.wolfram.com/notebooks/Constants/ChampernowneConstant.nb
- [7] William Hart, 2010: MPIR multiple precision integers and rationals library, version 2.5.1 , http://www.mpir.org
- [8] OEIS Foundation Inc, 2011: The On-Line Encyclopedia of Integer Sequences , http://oeis.org/A244330 .
[9]
OEIS Foundation Inc, 2011:
The On-Line Encyclopedia of Integer Sequences
,
http://oeis.org/A244331
.
- [10] OEIS Foundation Inc, 2011: The On-Line Encyclopedia of Integer Sequences , http://oeis.org/A244332 .
- [11] OEIS Foundation Inc, 2011: The On-Line Encyclopedia of Integer Sequences , http://oeis.org/A244333 .
- [12] OEIS Foundation Inc, 2011: The On-Line Encyclopedia of Integer Sequences , http://oeis.org/A038705 .
- [13] OEIS Foundation Inc, 2011: The On-Line Encyclopedia of Integer Sequences , http://oeis.org/A143534 .
- [14] OEIS Foundation Inc, 2011: The On-Line Encyclopedia of Integer Sequences , http://oeis.org/A244758 .
- [15] OEIS Foundation Inc, 2011: The On-Line Encyclopedia of Integer Sequences , http://oeis.org/A244759 .
- [16] OEIS Foundation Inc, 2011: The On-Line Encyclopedia of Integer Sequences , http://oeis.org/A143532 .
[17]
Wolfram Mathworld, 2013:
Champernowne Constant Continued Fraction
,
http://mathworld.wolfram.com/ChampernowneConstantContinuedFraction.html
.
- [18] Free Software Foundation, 2007: GNU Lesser General Public License v3 , https://www.gnu.org/licenses/lgpl.html .
- [19] John K. Sikora, 2014: Champernowne's Constant File Repository , https://drive.google.com/folderview?id=0B2oQudZQvHfSZ1hZN2VIYmU5eTQ&usp=sharing .
- [20] Open Knowledge Foundation, 2008: Open Data Commons Public Domain Dedication and License, v1.0 , http://opendatacommons.org/licenses/pddl/1.0/ | null | [
"John K. Sikora"
] | 2014-08-01T18:40:44+00:00 | 2014-08-01T18:40:44+00:00 | [
"math.NT"
] | Analysis of the High Water Mark Convergents of Champernowne's Constant in Various Bases | In this paper, we show that patterns exist in the properties of the High
Water Mark (HWM) convergents of Champernowne's Constant in various bases (Cb),
specifically in bases 2 through 124. The convergents are formed by truncating
the Continued Fraction Expansion (CFE) of Cb immediately before the CFE HWMs.
These patterns have been extended from the known patterns in the CFE HWMs of
Champernowne's Constant in base ten. We show that the patterns may be used to
efficiently calculate the CFE coefficients of Cb, and to calculate and predict
the lengths of the HWM coefficients, the number of correct digits of Cb as
calculated by the convergent, and the convergent error. We have discovered that
minor corrections to the pattern formulations in base 10 are required for some
bases, and these corrections are presented and discussed. The resulting
formulations may be used to make the calculations for Cb in any base. |
1408.0262v4 | ## Reducing uniformity in Khot-Saket hypergraph coloring hardness reductions
Girish Varma ∗
December 13, 2014
## Abstract
In a recent result, Khot and Saket [FOCS 2014] proved the quasi-NP-hardness of coloring a 2 -colorable 12 -uniform hypergraph with 2 (log n ) Ω(1) colors. This result was proved using a novel outer PCP verifier which had a strong soundness guarantee. We reduce the arity in their result by modifying their 12-query inner verifier to an 8-query inner verifier based on the hypergraph coloring hardness reductions of Guruswami et. al. [STOC 2014]. More precisely, we prove quasi-NP-hardness of the following problems on n -vertex hypergraphs.
- · Coloring a 2 -colorable 8 -uniform hypergraph with 2 (log n ) Ω(1) colors.
- · Coloring a 4 -colorable 4 -uniform hypergraph with 2 (log n ) Ω(1) colors.
## 1 Introduction
The discovery of the low-degree long code aka short code by Barak et. al. [BGH + 12] has over the last one year led to a sequence of results improving our understanding of the hardness of constant colorable hypergraphs with as few colors as possible. A k -uniform hypergraph is a collection of vertices and hyperedges such that every hyperedge is a subset of k vertices. A hypergraph is said to be q -colorable if the vertices of the hypergraph can be colored with at most q colors such that no hyperedge is monochromatic. An independent set in a hypergraph is a collections of vertices such that no hyperedge is wholly contained within the collection. Note that a hypergraph is q -colorable iff it the set of vertices can be partitioned into at most q independent sets.
Prior to the low-degree long code, all hardness reductions for hypergraph coloring where proved using the long code [BGS98] which resulted in a huge disparity between the best known positive and negative results for hypergraph coloring: the best known approximation algorithms require at least n Ω(1) colors to color a constant colorable (hyper)graph while the inapproximability results could only rule out at best (log n ) O (1) colors. The situation was redeemed by introduction of the low-degree long code, a derandomization of the long code, which was then adapted by Dinur and Guruswami [DG14] toward proving inapproximability results. Building on the DinurGuruswami framework, Guruswami et. al. [GHH 14] showed that it is quasi-NP-hard to color a + 2-colorable 8-uniform hypergraph with 2 2 Ω( √ log log n ) colors. Both the Dinur-Guruswami and Guruswami et. al. results were obtained by modifying the innermost PCP verifier to work with the
∗ Tata Institute of Fundamental Research, India. Supported by Google India under the Google India PhD Fellowship Award. Email : [email protected]
low-degree long code. Shortly thereafter, in a remarkable improvement, Khot and Saket [KS14] showed that it is quasi-NP-hard to color a 2-colorable 12-uniform hypergraph with 2 (log n ) Ω(1) colors. They obtained this result by using an 12-query inner PCP verifier based on the quadratic code, ie., a low-degree long code with degree two. However, to use a quadratic code based inner verifier, they needed an outer PCP verifier with a significantly stronger soundness guarantee than the standard outer PCP verifier obtained from parallel repetition of the PCP Theorem. In particular, they needed an outer PCP verifier, which in the soundness case, would not be satisfied by a short list of proofs even in superposition 1 . The construction of this outer PCP verifier with this stronger soundness guarantee is the main technical ingredient in the result of Khot and Saket [KS14]. We show that this outer PCP verifier of Khot and Saket can in fact be combined with a 8-query inner PCP verifier based on the Guruswami et. al. inner PCP verifier to obtain a hardness result for 2-colorable 8-uniform hypergraphs. More precisely, we show the following.
Theorem 1.1. For every constant /epsilon1 > 0 there is a quasi-polynomial time reduction from 3 -SAT to a 8 -uniform hypergraph G on n vertices such that,
- 1. YES Case: If the 3 -SAT instance is satisfiable then G is 2 -colorable.
- 2. NO Case: If the 3 -SAT instance is unsatisfiable then G does not have an independent set of relative size 2 -(log n ) 1 20 -/epsilon1 .
Guruswami et. al. [GHH 14] also proved how to reduce the uniformity in certain reductions + from 8 to 4 at the cost of increasing the number of colors from 2 to 4. We note that a similar trick can be performed in our setting to obtain the following result.
Theorem 1.2. For every constant /epsilon1 > 0 there is a quasi-polynomial time reduction from 3 -SAT to a 4 -uniform hypergraph G on n vertices such that,
- 1. YES Case: If the 3 -SAT instance is satisfiable then G is 4 -colorable.
- 2. NO Case: If the 3 -SAT instance is unsatisfiable then G does not have an independent set of relative size 2 -(log n ) 1 20 -/epsilon1 .
We remark that the analyses of the inner verifier in both the above theorems is simpler than the analyses of the corresponding inner verifiers in Guruswami et. al. and Khot-Saket results. Furthermore, in the language of covering complexity 2 introduced by Guruswami, H˚ astad and Sudan [GHS00], (the proof of) Theorem 1.2 demonstrates a Boolean 4CSP for which it is quasi-NPhard to distinguish between covering number of 2 vs. (log n ) Ω(1) .
## 2 Preliminaries
Let F denote the field GF (2) . Let F m m × be the vector space of m × m matrices over the field F . Our inner verifier is based on the quadratic code, which is a specialization of the low degree long code to degree 2 .
1 We will not require the exact definition of satisfying in superposition for this note. See Theorem 2.2 for the details of the Khot-Saket outer PCP verifier.
2 The covering number of a CSP is the minimal number of assignments to the vertices so that each hyperedge is covered by at least one assignment.
Definition 2.1 (Quadratic Code) . The quadratic code of x ∈ F m is a function A x : F m m × → F defined as A x ( X ) := 〈 X,x ⊗ x 〉 .
Our reductions makes use of the following outer PCP verifier of Khot and Saket [KS14]. As stated in the introduction, these instances have stronger soundness conditions which make them amenable for composition with a quadratic code based inner verifier.
Theorem 2.2 ([KS14, Theorem 7.2]) . There is a quasi-polynomial time reduction from an instance of 3 -SAT to a bi-regular instance ( U, V, E, Π) of Label Cover such that
- · Vertex sets U and V are bounded in size by N .
- · The label sets are F r × r 2 , F m m × 2 for U and V respectively.
- · For e ∈ E , the map π e : F m m × 2 → F r × r 2 is a linear transformation that maps symmetric matrices to symmetric matrices . 3 For an r × r matrix X X , ◦ π e is the unique m × m matrix such that 〈 X ◦ π , Y e 〉 = 〈 X,π Y e 〉 .
- · For each vertex v ∈ V , there is a constraint C v that is a a conjunction of homogeneous linear equations on the entries of the m × m matrix label.
- · δ ≤ 2 -log 1 / 3 N and k ≥ (log N ) 1 9 / .
The reduction satisfies:
- 1. Completeness : If the 3 -SAT instance is satisfiable then there is a labeling x u ⊗ x u for u ∈ U and y v ⊗ y v for v ∈ V such that
- · for each v ∈ V , y v ∈ F m 2 has the m th coordinate 1 and y v ⊗ y v satisfies the constraint C v ,
- · for each ( u, v ) ∈ E π , u,v ( y v ⊗ y v ) = x u ⊗ x u .
- 2. Soundness : If the 3 -SAT instance is not satisfiable then the following cannot hold: There are symmetric matrices M u ∈ F r × r 2 , M v ∈ F m m × 2 for u ∈ U, v ∈ V of rank ≤ k such that
- · for each v ∈ V , M v ∈ F m m × 2 has the ( m,m ) th coordinate 1 and M v satisfies the constraint C v ,
- · for δ fraction of edges e , π e ( M v ) = M u .
- 3. Smoothness : For any v ∈ V and any symmetric non-zero matrix M v with rank ≤ k , over a random choice of an edge e incident on v ,
$$\Pr [ \pi _ { e } ( M _ { v } ) = 0 ] \leq \delta / 2.$$
## 3 2 -colorable 8 -uniform hypergraphs
In this section we prove Theorem 1.1. Our reduction starts from the label cover instances given by Theorem 2.2. Let ( U, V, E, Π) be an instance of the label cover. We will construct a hypergraph G = ( V E , ) . For v ∈ V , let H v ⊆ F m m × 2 be the dual of the subspace of the set matrices that are symmetric and which satisfies the constraint C v . The set of vertices V will be the same as
3 The property that π maps symmetric matrices to symmetric matrices is easy to see from the proof of [KS14, Theorem 7.2].
V × ( F m m × 2 / H v ) . Any 2 -coloring of G is a collection of functions A ′ v : F m m × 2 / H v → { 0 1 , } for v ∈ V . For any such function, we can uniquely extend it to get A v : F m m × 2 → { 0 1 , } which is constant over cosets of H v . This method is called folding and it ensures that A v satisfies the following: if α ∈ F m m × 2 is such that ̂ A v ( α ) is non-zero, then α is symmetric and satisfies C v .
The set of edges E will be defined by the test mentioned below, which checks whether a supposed 2 -coloring A ′ v : F m m × 2 / H →{ v 0 1 , } is valid. There is an edge in E between any set of vertices in V that are queried together by the test. The test will be querying the extended functions A v at matrices in F m m × 2 instead of A ′ v . So a query to A v at X ∈ F m m × 2 corresponds to a query to A ′ v at the coset of H v that contains X .
## 2 -Colorable 8 -Uniform Test T 2 8 ,
- 1. Choose u ∈ U uniformly at random and v, w ∈ V uniformly and independently at random from the neighbors of u . Let π, σ : F m m × 2 → F r × r 2 be the projections corresponding to the edges ( u, v ) , ( u, w ) respectively. Uniformly and independently at random choose X ,X ,Y ,Y 1 2 1 2 ∈ F m m × 2 and x, y, z, x , y , z ′ ′ ′ ∈ F m 2 and F ∈ F r × r 2 . Let e m ∈ F m 2 be the vector with only the m th entry 1 and the rest is 0 .
- 2. Accept if and only if the following 8 values are not all equal :
A v ( X 1 ) A v ( X 3 ) where X 3 := X 1 + x ⊗ y + F ◦ π A v ( X 2 ) A v ( X 4 ) where X 4 := X 2 +( x + e m ) ⊗ z + F ◦ π A w ( Y 1 ) A w ( Y 3 ) where Y 3 := Y 1 + x ′ ⊗ y ′ + F ◦ σ + e m ⊗ e m A w ( Y 2 ) A w ( Y 4 ) where Y 4 := Y 2 +( x ′ + e m ) ⊗ z ′ + F ◦ σ + e m ⊗ e m
## 3.1 YES Case
Let y v ⊗ y v for v ∈ V and x u ⊗ x u for u ∈ U be a perfectly satisfying labeling of the label cover instance. That is, for every ( u, v ) ∈ E,π u,v ( y v ⊗ y v ) = x u ⊗ x u . Such a labeling is guaranteed by the YES instance of label cover, with the additional property that the m th coordinate of y v is 1 . Consider the following 2 -coloring of G : for each v ∈ V , A v ( X ) := 〈 X,y v ⊗ y v 〉 . Note that such a function is constant over cosets of H v . Let
$$x _ { 1 } & \coloneqq \langle X _ { 1 }, \overline { y } _ { v } \otimes \overline { y } _ { v } \rangle & x _ { 2 } & \coloneqq \langle X _ { 2 }, \overline { y } _ { v } \otimes \overline { y } _ { v } \rangle \\ y _ { 1 } & \coloneqq \langle Y _ { 1 }, \overline { y } _ { w } \otimes \overline { y } _ { w } \rangle & y _ { 2 } & \coloneqq \langle Y _ { 2 }, \overline { y } _ { w } \otimes \overline { y } _ { w } \rangle$$
and f := 〈 F, x u ⊗ x u 〉 . Note that 〈 F ◦ π u,v , y v ⊗ y v 〉 = 〈 F, π u,v ( y v ⊗ y v ) 〉 = 〈 F, x u ⊗ x u 〉 , and 〈 e m ⊗ e m , y v ⊗ y v 〉 = 〈 e m , y v 〉 = 1 . Using these, the assignments to the 8 query locations are:
$$\Lambda _ { \Lambda v } \cdot \Lambda _ { \Lambda v ^ { \prime } } \quad \Lambda _ { \Lambda \Lambda v ^ { \prime } } \quad \Lambda _ { \Lambda } \quad \Lambda _ { \Lambda } \quad \Lambda _ { \Lambda } \quad \Lambda _ { \Lambda } \quad \Lambda _ { \Lambda } \quad \Lambda _ { \Lambda } \quad \Lambda _ { \Lambda } \quad \Lambda _ { \Lambda } \quad \Lambda _ { \Lambda } \quad \Lambda _ { \Lambda } \quad \Lambda _ { \Lambda } \quad \Lambda _ { \Lambda } \quad \Lambda _ { \Lambda } \quad \Lambda _ { \Lambda } \quad \Lambda _ { \Lambda } \quad \Lambda _ { \Lambda } \quad \langle \overline { y } _ { v }, \overline { x } \rangle \langle \overline { y } _ { v }, \overline { y } \rangle + f \\ x _ { 1 } & & x _ { 2 } & & x _ { 2 } + ( \langle \overline { y } _ { v }, \overline { x } \rangle + 1 ) \langle \overline { y } _ { v }, \overline { z } \rangle + f \\ y _ { 1 } & & y _ { 1 } + \langle \overline { y } _ { w }, \overline { x } ^ { \prime } \rangle \langle \overline { y } _ { w }, \overline { y } ^ { \prime } \rangle + f + 1 \\ y _ { 2 } & & y _ { 2 } + ( \langle \overline { y } _ { w }, \overline { x } ^ { \prime } \rangle + 1 ) \langle \overline { y } _ { w }, \overline { z } ^ { \prime } \rangle + f + 1 \\ \text{to see at least one of the 4 rows are always not equal. Hence $A$ is a valid 2-color}$$
It is easy to see at least one of the 4 rows are always not equal. Hence A is a valid 2 -coloring of G .
## 3.2 NOCase
Suppose the reduction was applied to a NO instance of label cover. Let k and δ be the parameters specified by Theorem 2.2.
Lemma 3.1. If there is an independent set in G of relative size s then
$$s ^ { 8 } \leq \delta + \frac { 1 } { 2 ^ { k / 2 + 1 } }.$$
Proof. The proof of the lemma is similar to Section 8.2 in Khot and Saket [KS14]. Consider any set A ⊆ V of fractional size s . For every v ∈ V , let A v : F m m × 2 →{ 0 1 , } be the indicator function that is extended such that it is constant over cosets of H v . A is an independent set if and only if
$$\Theta \coloneqq \mathbb { E } _ { u, v, w } \mathbb { E } _ { X _ { i }, Y _ { i } \in \mathcal { T } _ { 2, 8 } } \, \prod _ { i = 1 } ^ { 4 } A _ { v } ( X _ { i } ) A _ { w } ( Y _ { i } ) = 0. \\. \pi \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \ dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots.$$
Now we do the Fourier expansion and take expectations over X ,X ,Y ,Y 1 2 1 2 to obtain the following:
$$\text{Now we do the Fourier expansion and take expectations over } X _ { 1 }, X _ { 2 }, Y _ { 1 }, Y _ { 2 } \text{ to obtain the follow-} \\ \text{ing} ; \\ \Theta = \underset { \substack { \mathbb { E } _ { n } \\ \sigma _ { 1 }, \alpha _ { 2 } \in \mathbb { F } _ { 2 } ^ { m } x \cdot \overline { x } } } { \mathbb { E } _ { f _ { \mathcal { S }, \mathcal { X } } } } \left [ \underset { \substack { \mathbb { E } _ { y } \\ \sigma _ { 1 } } } { \mathbb { E } _ { y } } \left [ \underset { \substack { \mathbb { E } _ { y } \\ \sigma _ { 1 } } } { \mathbb { E } _ { z } } \left ( \overline { x } \right ) ^ { 2 } \underset { \substack { \mathbb { E } _ { z } \\ \sigma _ { 1 } } } { \mathbb { E } _ { z } } \left [ \underset { \substack { \mathbb { E } _ { y } \\ \sigma _ { 1 } } } { \mathbb { E } _ { y } } \left [ \underset { \substack { \mathbb { E } _ { y } \\ \sigma _ { 1 } } } { \mathbb { E } _ { z } } \left [ \underset { \substack { \mathbb { E } _ { z } \\ \sigma _ { 1 } } } { \mathbb { E } _ { z } } \left [ \underset { \substack { \mathbb { E } _ { z } \\ \sigma _ { 1 } } } { \mathbb { E } _ { z } } \left [ \underset { \substack { \mathbb { E } _ { z } \\ \sigma _ { 1 } } } { \mathbb { E } _ { z } } \left [ \underset { \substack { \mathbb { E } _ { z } \\ \sigma _ { 1 } } } { \mathbb { E } _ { z } } \left [ \underset { \substack { \mathbb { E } _ { z } \\ \sigma _ { 1 } } } { \mathbb { E } _ { z } } \left [ \text{$\Gamma_{n}$} } \right ] \text{$\chi_{\beta_{1}$} } \right ] \chi_{\beta_{2}$} \left ( \text{$\Gamma$} \circ \right ) \chi_{\beta_{2}$} \left ( \underset { \substack { \mathbb { E } _ { z } \\ \sigma _ { 1 } } \circ \overline { x } \right ) } \right ] \chi_{\beta_{2}$} \left ( \text{$\Gamma$} \circ \right ) \chi_{\beta_{2}$} \left ( \underset { \substack { \mathbb { E } _ { z } \\ \sigma _ { 1 } } \circ \overline { x } \right ) } \right ] \\ \text{Note that since } F \in \mathbb { F } _ { 2 } ^ { r \times r } \text{ is chosen uniformly at random,} \\ \mathbb { E } \, \chi _ { \alpha _ { 1 } } ( F \circ \pi ) \chi _ { \alpha _ { 2 } } ( F \circ \pi ) \chi _ { \beta _ { 1 } } ( F \circ \sigma ) \chi _ { \beta _ { 2 } } ( F \circ \sigma ) = \mathbb { E } ( - 1 ) ^ { ( \pi ( \alpha _ { 1 } + \alpha _ { 2 } ), F ) + ( \sigma ( \beta _ { 1 } + \beta _ { 2 } ), F ) }$$
Note that since F ∈ F r × r 2 is chosen uniformly at random,
$$\frac { \mathbb { E } } { F } \chi _ { \alpha _ { 1 } } ( F \circ \pi ) \chi _ { \alpha _ { 2 } } ( F \circ \pi ) \chi _ { \beta _ { 1 } } ( F \circ \sigma ) \chi _ { \beta _ { 2 } } ( F \circ \sigma ) = \frac { \mathbb { E } ( - 1 ) ^ { \langle \pi ( \alpha _ { 1 } + \alpha _ { 2 } ), F \rangle + \langle \sigma ( \beta _ { 1 } + \beta _ { 2 } ), F \rangle } } { }$$
is zero unless π α ( 1 + α 2 ) = σ β ( 1 + β 2 ) . Let ν α ( ) := ( -1) 〈 α,e m ⊗ e m 〉 . Now taking expectations over x, y, z, x , y , z ′ ′ ′ , and noting that 〈 α, x ⊗ 〉 y = 〈 αx,y 〉 , we obtain
$$\cdots \cdots \cdots \mp \cdots \cdots \cdots \\ \text{Term} _ { u, v, w } ( \alpha _ { 1 }, \alpha _ { 2 }, \beta _ { 1 }, \beta _ { 2 } ) = ( - 1 ) ^ { \nu ( \beta _ { 1 } + \beta _ { 2 } ) } \widehat { A } _ { v } ( \alpha _ { 1 } ) ^ { 2 } \widehat { A } _ { v } ( \alpha _ { 2 } ) ^ { 2 } \widehat { A } _ { w } ( \beta _ { 1 } ) ^ { 2 } \widehat { A } _ { w } ( \beta _ { 2 } ) ^ { 2 } \\ \Pr \left [ \alpha _ { 1 } \overline { x } = 0 \wedge \alpha _ { 2 } \overline { x } = \alpha _ { 2 } e _ { m } \right ] \cdot \\ \Pr \overline { x } ^ { \prime } \left [ \beta _ { 1 } \overline { x } ^ { \prime } = 0 \wedge \beta _ { 2 } \overline { x } ^ { \prime } = \beta _ { 2 } e _ { m } \right ]$$
when π α ( 1 + α 2 ) = σ β ( 1 + β 2 ) and 0 otherwise. Define:
$$\sum _ { \substack { \cdot \cdot \\ \cdot \cdot \cdot \cdot \\ \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \colon } }$$
$$\text{when $\pi(\alpha_{1}+\alpha_{2})=\sigma(\beta_{1}+\beta_{2})$ and $0$ otherwise. Define} \colon \\ \Theta _ { 0 } & = \mathbb { E } _ { u, w } \sum _ { \substack { \text{rank}(\alpha_{1}+\alpha_{2}),rank(\beta_{1}+\beta_{2})\leq k \\ \pi(\alpha_{1}+\alpha_{2})=\sigma(\beta_{1}+\beta_{2}) \\ \nu(\beta_{1}+\beta_{2})=0 } } \text{Term} _ { u, v, w } ( \alpha_{1}, \alpha_{2}, \beta_{1}, \beta_{2}) \\ \Theta _ { 1 } & = \mathbb { E } _ { u, w, w } \sum _ { \substack { \text{rank}(\alpha_{1}+\alpha_{2}),rank(\beta_{1}+\beta_{2})\leq k \\ \pi(\alpha_{1}+\alpha_{2})=\sigma(\beta_{1}+\beta_{2}) \\ \nu(\beta_{1}+\beta_{2})=1 } } \text{Term} _ { u, v, w } ( \alpha_{1}, \alpha_{2}, \beta_{1}, \beta_{2}) \\ \Theta _ { 2 } & = \mathbb { E } _ { u, w, w } \sum _ { \max \{ \text{rank}(\alpha_{1}+\alpha_{2}),rank(\beta_{1}+\beta_{2})\}>k \\ \pi(\alpha_{1}+\alpha_{2})=\sigma(\beta_{1}+\beta_{2}) } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{\ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text { } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text` \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{} \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{
\text{lower bound $\Theta_{0}$ by $s^{\8}, \, \,upper bound | \Theta_{1} | \text{ by $\delta$ and $| \Theta_{2}| \text{ by $1/2^{k/2+1}$ below. Along with $(3.1),}$$
We lower bound Θ 0 by s 8 , upper bound | Θ 1 | by δ and | Θ 2 | by 1 2 / k/ 2+1 below. Along with (3.1), this will prove Lemma 3.1.
## 3.2.1 Lower bound on Θ 0
Note that all terms in Θ 0 are positive. Now consider the term corresponding to α 1 = α 2 = β 1 = β 2 = 0 .
## 3.2.2 Upper bound on | Θ 1 |
$$\Phi _ { u, v, w } ^ { \cdot } \widehat { A } _ { v } ^ { 4 } ( 0 ) \widehat { A } _ { w } ^ { 4 } ( 0 ) = \mathbb { E } _ { u } \left ( \mathbb { E } _ { v } \widehat { A } _ { v } ^ { 4 } ( 0 ) \right ) ^ { 2 } \geq \left ( \mathbb { E } _ { u v } \widehat { A } _ { v } ( 0 ) \right ) ^ { 8 } \geq s ^ { 8 }. \\ \text{per bound on } | \Theta _ { 1 } |$$
We can upper bound | Θ 1 | by
$$\iota _ { P P } & \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \iota _ { \i i _ { \iota _ { \iota _ { \iota _ { \iota _ { \i i _ { \iota _ { \iota _ { \iota _ { \iota _ { \i i _ { \i i _ { \iota _ { \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i. i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \ i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \j i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \ j i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \sigma _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ { \i i _ \i i _ { \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \i i _ \j i } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } \ } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } _ } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } { } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } }. } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } ) } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } \\ } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } }, } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } ( } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } }$$
Consider the following strategy for labeling vertices u ∈ U and v ∈ V . For u ∈ U , pick a random neighbor v , choose ( α , α 1 2 ) with probability ̂ A 2 v ( α 1 ) ̂ A 2 v ( α 2 ) and set its label to π α ( 1 + α 2 ) . For w ∈ V , choose ( β , β 1 2 ) with probability ̂ A 2 w ( β 1 ) ̂ A 2 w ( β 2 ) and set its label to β 1 + β 2 . Since A w is folded, both β 1 and β 2 are symmetric and satisfies C v . Since these constraints are homogeneous, β 1 + β 2 is also symmetric and satisfies C v . Also π maps symmetric matrices to symmetric matrices. Note that (3.7) gives the probability that a random edge ( u, w ) of the label cover is satisfied by this labeling. Hence (3.7) and | Θ 1 | are upper bounded by δ .
## 3.2.3 Upper bound on | Θ 2 |
Note that if the rank( α ) > k , for any fixed b , Pr [ x αx = b ] ≤ 1 2 / k +1 . All terms in Θ 2 has max rank( { α 1 ) , rank( α 2 ) , rank( β 1 ) , rank( β 2 ) } > k/ . 2 From (3.2) we have that, for any fixed choice of u, v, w each term in Θ 2 has absolute value at most 1 2 / k/ 2+1 . Since A,B are { 0 1 , } valued functions, sum of their squared coefficients is upper bounded by 1 (i.e. Parseval's inequality). Thus | Θ 2 | ≤ 1 2 / k/ 2+1 .
Proof of Theorem 1.1. We already saw in Section 3.1 that an YES instance of label cover is mapped to a 2 -colorable hypergraph. Since k = (log N ) 1 8 / -2 /epsilon1 and δ = 2 -(log N ) 1 / 4 -2 /epsilon1 , s ≤ 2 -(log N ) 1 / 8 -3 /epsilon1 . Also the number of vertices in G ,
$$n \leq N 2 ^ { m ^ { 2 } } \leq N \cdot 2 ^ { ( \log N ) ^ { 1 0 / 4 + 2 \epsilon } }.$$
From Lemma 3.1 and above, a NO instance of label cover is mapped to a hypergraph G that has no independent set of relative size 2 -(log n ) 1 / 20 -4 /epsilon1 .
## 4 4 -colorable 4 -uniform hypergraphs
In this section, we modify the reduction in the previous section, so that the uniformity of the hypergraph produced is decreased to 4 at the cost of increasing the number of colors required in the YES case to 4 . This method was proposed by Guruswami et. al. [GHH 14]. The hypergraph + G = ( V E , ) constructed will have vertices
$$\mathcal { V } = V \times \left ( \mathbb { F } _ { 2 } ^ { m \times m } \times \mathbb { F } _ { 2 } ^ { m \times m } / \mathcal { H } _ { v } \times \mathcal { H } _ { v } \right ). \\ \text{be expressed as a collection of functions}$$
Any 4 -coloring of G can be expressed as a collection of functions
$$\Phi \times \mathcal { G } \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \sin \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \tan \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \cos \, \ sin \, \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \sin \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \ sin \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \ cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \osigma \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \lesstharpo \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \sec \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \loss \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \exp \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \prod \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \cos \end{}$$
Wecan uniquely extend such functions to get A v : F m m × 2 × F m m × 2 →{ 0 1 , } 2 which is constant over cosets of H ×H v v . This ensures that A satisfies the following: if α = ( α , α 1 2 ) ∈ F m m × 2 × F m m × 2 is such that ̂ A α ( ) is non-zero, then α , α 1 2 are both symmetric and satisfies C v . The set of edges E will be defined by the test mentioned below.
## 4 -Colorable 4 -Uniform Test
- 1. Sample v, w and { X , Y i i } 4 i =1 from the distribution T 2 8 , as described by the test in the previous section.
- 2. Accept if and only if the following 4 values are not all equal :
$$\ A _ { v } ( X _ { 1 }, X _ { 2 } ) \ A _ { v } ( X _ { 3 }, X _ { 4 } ) \ A _ { w } ( Y _ { 1 }, Y _ { 2 } ) \ A _ { w } ( Y _ { 3 }, Y _ { 4 } )$$
## 4.1 YES Case
Given a perfectly satisfying labeling y v ⊗ y v for v ∈ V and x u ⊗ x u for u ∈ U , we define the following 4 -coloring for G : for each v ∈ V ,
$$A _ { v } ( X _ { 1 }, X _ { 2 } ) \coloneqq \left ( \langle X _ { 1 }, \overline { y } _ { v } \otimes \overline { y } _ { v } \rangle, \langle X _ { 2 }, \overline { y } _ { v } \otimes \overline { y } _ { v } \rangle \right ).$$
Note that such a function is constant over cosets of H v . Using the arguments from Section 3.1, it is easy to see that A is a valid 4 -coloring of G .
## 4.2 NOCase
The analysis of the NO case is similar to Section 3.2.
## 5 Acknowledgements
The author would like to thank Prahladh Harsha, Subhash Khot, Rishi Saket and Srikanth Srinivasan for helpful discussions.
## References
[BGH + 12] BOAZ BARAK, PARIKSHIT GOPALAN, JOHAN H ˚ ASTAD, RAGHU MEKA, PRASAD RAGHAVENDRA, and DAVID STEURER. Making the long code shorter . In Proc. 53 th IEEE Symp. on Foundations of Comp. Science (FOCS) , pages 370-379. 2012. arXiv:1111.0405 doi:10.1109/FOCS.2012.83 , .
[BGS98]
MIHIR BELLARE, ODED GOLDREICH, and MADHU SUDAN. Free bits, PCPs, and nonapproximability-towards tight results . SIAM J. Computing, 27(3):804-915, June 1998. (Preliminary version in 36th FOCS , 1995). eccc:TR95-024 doi:10.1137/S0097539796302531 , .
[DG14]
IRIT DINUR and VENKATESAN GURUSWAMI. PCPs via low-degree long code and hardness for constrained hypergraph coloring . In Proc. 54 th IEEE Symp. on Foundations of Comp. Science (FOCS) , pages 340-349. 2014. eccc:TR13-122 doi:10.1109/FOCS.2013.44 , .
+
[GHH 14] VENKAT GURUSWAMI, PRAHLADH HARSHA, JOHAN H ˚ ASTAD, SRIKANTH SRINIVASAN, and GIRISH VARMA. Super-polylogarithmic hypergraph coloring hardness via low-degree long codes . In Proc. 46 th ACM Symp. on Theory of Computing (STOC) , pages 614-623. 2014. arXiv:1311.7407 , doi:10.1145/2591796.2591882 .
[GHS00]
V. GURUSWAMI, J. HASTAD, and M. SUDAN. Hardness of approximate hypergraph coloring . In Foundations of Computer Science, 2000. Proceedings. 41st Annual Symposium on , pages 149-158. 2000. doi:10.1109/SFCS.2000.892074 .
[KS14]
SUBHASH KHOT and RISHI SAKET. Harndess of coloring 2-colorable 12-uniform hypergraphs . In Proc. 55 th IEEE Symp. on Foundations of Comp. Science (FOCS) . 2014. (To appear). eccc:TR14-051 . | null | [
"Girish Varma"
] | 2014-08-01T18:49:47+00:00 | 2014-12-11T06:18:40+00:00 | [
"cs.CC"
] | Reducing uniformity in Khot-Saket hypergraph coloring hardness reductions | In a recent result, Khot and Saket [FOCS 2014] proved the quasi-NP-hardness
of coloring a 2-colorable 12-uniform hypergraph with $2^{(\log n)^{\Omega(1)}}$
colors. This result was proved using a novel outer PCP verifier which had a
strong soundness guarantee. In this note, we show that we can reduce the arity
of their result by modifying their 12-query inner verifier to an 8-query inner
verifier based on the hypergraph coloring hardness reductions of Guruswami et.
al. [STOC 2014]. More precisely, we prove quasi-NP-hardness of the following
problems on n-vertex hypergraphs.
- Coloring a 2-colorable 8-uniform hypergraph with $2^{(\log n)^{\Omega(1)}}$
colors.
- Coloring a 4-colorable 4-uniform hypergraph with $2^{(\log n)^{\Omega(1)}}$
colors. |
1408.0263v4 | ## IS THE BLAZHKO EFFECT THE BEATING OF A NEAR-RESONANT DOUBLE-MODE PULSATION?
Paul H. Bryant
BioCircuits Institute (formerly Institute for Nonlinear Science), University of California, San Diego, La Jolla, CA 92093, USA
[email protected]
## ABSTRACT
In this paper it is shown that the Blazhko effect may result from a near-resonant type of multi-mode pulsation, where two (or sometimes more) periodic oscillations with slightly different frequencies gradually slip in phase, producing a beat frequency type of modulation. Typically one of these oscillations is strongly non-sinusoidal. Two oscillations are sufficient for the standard Blazhko effect; additional oscillations are needed to explain multi-frequency modulation. Previous work on this hypothesis by Arthur N. Cox and others is extended in this paper by developing a simple (non-hydro) model that can accurately reproduce several important features found in Kepler data for RR Lyr, including the pulsation waveform, the upper and lower Blazhko envelope functions and the motion, disappearance and reappearance of the bump feature. The non-sinusoidal oscillation is probably generated by the fundamental mode and the other oscillations are probably generated by nonradial modes. This model provides an explanation for the strong asymmetry observed in the side peak spectra of most RR Lyrae stars. The motion and disappearance of the bump feature are shown to be an illusion, just an artifact of combining the oscillations. V445 Lyr is presented as an example with dual modulation. The mysterious double-maxima waveform observed for this star is explained, providing additional strong evidence that this beating-modes hypothesis is correct. Problems with other recent explanations of the Blazhko effect are discussed in some detail.
Subject headings: instabilities - stars: oscillations (including pulsations) - stars: variables: RR Lyrae
## 1. INTRODUCTION
Variable stars of the RR Lyrae type often exhibit a slow modulation of their pulsations known as the Blazhko effect. First reported in 1907 by Russian astronomer Sergey Nikolaevich Blazhko (Blazhko 1907), the effect remains mysterious, with none of the numerous proposed explanations achieving widespread acceptance. See the recent reviews by Kovacs (2009) and Kolenberg (2012). Recent attempts to explain the Blazhko effect include:
- · 'The ninth overtone model' which involves an unstable resonance interaction between between the fundamental mode and the ninth radial overtone, the oscillations of which lead to modulation of the fundamental (Szabo et al. 2010; Buchler & Kollath 2011; Kollath et al. 2011)
- · 'The Stothers model' in which an oscillatory magnetic field induces variations in the strength of turbulent convection which in turn modulates the fundamental (Stothers 2006, 2010)
- · 'The shock model' in which a shock wave periodically generates turbulence in the atmosphere which in turn modulates the fundamental (Gillet 2013).
All of these models have significant problems which are discussed in Section 4.
In this paper a simple explanation is put forth for the Blazhko effect which involves (at least) two periodic oscillations, PO1 and PO2, of slightly different frequency. Typically one of these oscillations is strongly non-sinusoidal. The slow phase slip between these oscillations results in a combined waveform that displays a beat frequency or apparent modulation at the Blazhko frequency. We refer to this as 'the beating-modes model'. It is the only model where there is no actual modulation of the amplitude of the primary mode, rather the apparent modulation is an illusion produced by the partial cancellation that results from adding to it the other mode (or modes) which have nearly the same frequency but differing phases. This hypothesis is supported by a simple (non-hydro) model of the dynamics that accurately reproduces Kepler data for RR Lyr, as shown in Figure 1. In addition, the beating-modes model provides an explanation for the mysterious 'double-maxima' waveform of V445 Lyr (See Section 3.1), providing additional strong evidence in its favor.
Referring to Figure 1, note the many features of the data that are reproduced by the model including the shape of the pulsation waveform, the phase offset between the upper and lower modulation envelope functions, the size of the bump feature, the behavior of the bump: hesitating near the bottom of the pulsation waveform, moving upwards for a while, fading out and disappearing for a while and then reappearing near the bottom to complete the cycle. The disappearance of the bump lasts for about a third of the Blazhko cycle starting near the Blazhko minimum. The bump behavior of the model is a little surprising since it is entirely brought about by the addition of PO2, a pure sinusoid. PO1 by itself displays a bump at a fixed location about one quarter up from the bottom.
PO1 and PO2 may be generated by two unstable modes of the star. It is generally believed that the pulsation frequency for Blazhko stars corresponds to the fundamental mode, so this would be the likely choice for PO1, while PO2 may be generated by a nonradial mode of nearly the same frequency. An alternate possibility is that PO2 is a stable nonradial mode which becomes unstable by acquiring additional energy from the fundamental. (This is only possible if a mechanism exists that can accomplish this without phase-locking.) Since nonradial modes are involved, the usual 1D radial hydrocode software will not be able to test this model. Multi-D codes are now being
developed which may be able to accomplish this at some future date (see e.g. Mundprecht et al. 2013; Geroux & Deupree 2014).
There have been several previous papers in support of the beating-modes model, but the current work goes much further in terms of finding a simple model that actually provides a remarkably good fit to the observed Blazhko data, including the mysterious motion and disappearance of the socalled bump feature. Guggenberger et al. (2008), building on earlier work by Breger & Kolenberg (2006), proposed a test of the phasing behavior in an effort to determine if beating was responsible for the Blazhko effect. They applied this to the two well-observed Blazhko RR Lyrae stars SS For and RR Lyr, with results consistent with the beating of two oscillations. Cox (1993), based on a combination of nonradial modal stability analysis combined with radial hydrocode analysis, proposed a near-resonant beating process between the fundamental and a nonradial mode, most likely the g 4 , l = 1 mode. (Note that the stability analysis was for a static star and so might be significantly impacted by the presence of the high amplitude fundamental.) Sixteen years later, Cox (2009), working with improved code, finds that nonradial modes near the fundamental are always just slightly stable, and proposes that the g 4 , l = 1 mode becomes unstable through nonlinear interaction with the fundamental. He still refers to this as a beating process not as a resonant mode interaction. Borkowski (1980) proposed a near-resonant double-mode pulsation involving the fundamental and either the second or third radial overtone, where the overtone frequency is higher than the second harmonic of the fundamental by the Blazhko frequency. Because the overtone is near the second harmonic instead of the fundamental itself, the Borkowski model is not validated by the results of this paper. Also since it only involves radial modes, one might have expected it to have been observed in 1D hydrocode models, but apparently it has not.
A variant of the beating-modes model is found in the work of Nowakowski & Dziembowski (2001). Their approach involves a resonance between the fundamental mode and a pair of nonradial modes with opposite orders ± m . Modes with nonzero m have oscillations as a function of the azimuthal angle. These can be thought of as traveling wave modes that move around the star either with or against the rotational motion of the star depending on the sign of m . As a result, these are split in frequency by the rotation. These modes can derive energy through nonlinear interaction with the fundamental if they satisfy the resonance condition that the sum of their frequencies must be exactly twice the fundamental frequency. As a result they generate a triplet spectrum with one peak on either side of the fundamental. According to Equation 77 of Nowakowski & Dziembowski (2001), the resulting Blazhko frequency is some fraction of the rotational frequency, most likely around 1/2. The pair are assumed initially stable and made unstable through the resonance. The resonant interaction between the modes is presumed stable and therefore the mode amplitudes and frequencies do not vary in time. Since all three peaks of the triplet correspond to actual modes, this is another type of beating-modes model, where the combined waveform appears modulated even though no actual modulation taking place. The main problem with this model is that there seems to be no good explanation for the amplitude disparity that is often observed for the opposite side peaks. However it seems reasonable to consider, due to the wide variation of behavior observed
Fig. 1.- Comparison of data with the model. (a) Data from the Kepler project for RR Lyr (KIC 7198959, second quarter, long cadence corrected flux data starting at TJD 55032.7625503). 72 pulsation cycles are displayed which constitutes approximately 1 Blazhko cycle. There are two visible gaps where good data is not available. (b) Model output generated by adding 2 periodic oscillations, PO1 and PO2. PO1 is a highly non-sinusoidal waveform, generated as discussed in the text and shown in Figure 2. PO2 is a simple sine wave of slightly higher frequency than PO1. (c) Corrected model as discussed in the text. Note that the model and corrected model match many features of the data including the phase offset of the upper and lower Blazhko envelope curves (with the lower one leading), the shape of the pulsation waveform, the apparent motion of the bump feature from the bottom upwards towards the middle of the waveform, its subsequent disappearance and then reappearance at the bottom. The corrected model also matches the dissimilar amplitudes of the upper and lower Blazhko envelopes.
among Blazhko stars, that this model could apply to some stars which have a relatively balanced triplet.
The spectra for Blazhko stars includes progressions of side-peaks to the left and right of the fundamental and each of its harmonics, all evenly spaced apart by the Blazhko frequency. These tend to diminish in amplitude when moving away to the left or right and often only the first sidepeak is visible above the noise on each side. Most side-peaks are nonlinear mixing product peaks and do not correspond to actual excited modes of the star. Usually these side-peaks are attributed to modulation of the fundamental mode. However, if at least one of the side-peaks immediately adjacent to the fundamental corresponds to an actual mode, this is sufficient through nonlinear processes (as discussed below) to generate the entire set of side-peaks. The other frequencies are integer linear combinations of these two frequencies:
$$\omega _ { n m } = n \omega _ { 1 } + m \omega _ { 2 },$$
where ω nm is a side-peak frequency, n and m are integers (positive or negative), ω 1 is the frequency of PO1 (the fundamental) and ω 2 is the frequency of PO2. So, for example, the side-peak that is opposite to ω 2 (on the other side of the fundamental) has n = 2 and m = -1. If one side-peak is notably larger than the others then it is the likely choice for PO2. In the case of RR Lyr, there is a significantly larger peak on the higher frequency side. A spectral analysis (using the Period04 software package) of the Kepler data used in Figure 1(a) gives the amplitude (in units of 10 ) of 7 the upper peak as 0.0487 and of the lower peak as 0.0097, giving an amplitude ratio of about 5.02 and corresponding power ratio of 25.2. The central peak amplitude is about 0.1777.
It is well known that amplitude and frequency modulation can generate a side-peak spectrum similar to the one observed (see e.g. Benko et al. 2011; Szeidl & Jurcsik 2009; Szeidl et al. 2012), however a quasiperiodic dynamical system will also generate such a spectrum. A quasiperiodic system is one whose oscillations contain two base frequencies that are not rationally related. These would correspond to the frequencies of PO1 and PO2 and the apparent modulation frequency is the difference between them. The complete spectrum of such a system consists of the set of frequencies given by Equation 1. It can be obtained by a two-dimensional Fourier series expansion as is shown in Appendix B. Nonlinear processes can change the amplitudes of the (side-peak) frequencies including those which are initially zero. We take as a starting point a spectrum that includes PO1 (the fundamental mode frequency with its harmonics) and PO2 (a sine wave). There are two nonlinear mechanisms for generating the complete side-peak array. One is the process that translates the pulsational motion of the star into a variation in the light output. There is no reason to assume this process is entirely linear. So, for example, if the light output depends slightly on the square of the sum of the radial velocities of the two oscillations, then there will be cross terms generated by the square which will introduce components at the triplet (or first order) side-peaks for the fundamental and all of its harmonics. Higher order nonlinear terms will generate higher order side-peaks. A second mechanism is generated by nonlinear coupling between the active modes. Although such coupling is (by assumption) insufficient to produce phase-locking between the modes, it can introduce side-peaks that were previously missing, change the amplitudes
of peaks that were already present and shift the base frequencies. In the first mechanism, the sidepeak frequencies are only present in the light curve, while in the second they are also physically present in the stellar vibration. Both mechanisms may be active at the same time.
## 2. MODEL
For purposes of matching the Kepler data, PO1 and PO2 are approximated as constant in amplitude and frequency throughout the Blazhko cycle, which is equivalent to saying that the nonlinear interactions between them are neglected as relatively small. PO1 is a highly non-sinusoidal waveform whose frequency ω 1 is the pulsation frequency. Based on success at fitting the data with the model, it appears that this oscillation is made up of two sub-components which we will call PO1a and PO1b both of which are sawtooth-like waveforms. PO1a is larger in amplitude, while PO1b is shifted in phase and generates the bump feature. The most likely possibility is that PO1 is generated by a single mode with complicated dynamics. It is generally thought that the bump is an echo, either a reflection from the stellar core or the atmosphere (Guggenberger & Kolenberg 2006) and thus PO1b is the echo of PO1a. An alternate, but less likely, possibility is that PO1b is actually a separate mode, probably nonradial, that is phase-locked to the fundamental. This would make it similar to Cepheids which are thought to exhibit a bump due to a resonance with another mode (Simon 1977; Buchler et al. 1990; Bono, Marconi & Stellingwerf 2000). But regardless of the reason for the bump, the important point is that the model generates a reasonably good approximation to the observed waveform, enabling a test of the beating-modes hypothesis. The model output for PO1 is shown in Figure 2. Details of how it is generated are given below. PO2, on the other hand, is approximated as a pure sinusoid as this seems to do a good job of reproducing the observed Blazhko behavior. The frequency of PO2 is ω 2 = ω 1 + ω B , where ω B is the Blazhko frequency. In Section 5 it is discussed how the current work relates to the 'hybrid mode model' developed previously by the author (Bryant 2014).
For PO1a and PO1b we will use the same one-zone model developed in Bryant (2014), although here we are only using the model to generate the sawtooth-like waveform needed for PO1. The model details are given in Appendix A. This one-zone stellar model is similar to one developed by Baker (1966) and used extensively by Stellingwerf and others (Stellingwerf 1972, 1986; Stellingwerf, et al. 1987; Munteanu et al. 2005) to study Cepheid and RR Lyrae variable stars. When excited to high amplitude, this model produces a sawtooth-like velocity function, a waveform that is commonly seen in the light curves of RR Lyrae and Cepheid variable stars. For simplicity, we will use identical model parameters to generate PO1a and PO1b, the only differences are that PO1b has its amplitude reduced by a scale factor and it is shifted in phase relative to PO1a. Both sub-components are made equally non-sinusoidal, which seems to produce a very good fit to the observational data.
Let v c be the combined velocity of PO1 and PO2. For PO1 we use v 1 ( ) from Equation A4 in t Appendix A and for PO2 we use a sine wave, with phase and scale factor chosen to fit the Kepler
Fig. 2.- Shows the first periodic oscillation PO1, offset by the RR Lyr base flux level of 1 25 . × 10 . 7 When this is added to PO2, a simple sine wave, it produces the model output shown in Figure 1(b). PO2 is slightly higher in frequency than PO1, the difference being the Blazhko frequency. Note that the bump feature is stationary in PO1. The apparent motion and disappearance of the bump in Figure 1 is simply an artifact produced by the addition of PO2.
data:
$$v _ { c } ( t ) = v _ { 1 } ( t ) + 0. 0 5 s i n ( \omega _ { 2 } t - 0. 6 )$$
The light flux output L t ( ), as plotted in Figure 1(b), is obtained from:
$$L ( t ) / 1 0 ^ { 7 } = 1. 2 5 + 1. 3 v _ { c } ( t ).$$
The coefficients used were selected to obtain good visual agreement with the Kepler data in Figure 1(a)
We observe that for the Kepler data, the upper and lower Blazhko envelopes have differing amplitudes with the lower one being considerably smaller than the upper. The model on the other hand has upper and lower amplitudes of the envelope functions that are equal to each other and to the amplitude of PO2. We take a phenomenological approach to this problem and assume that the observed stellar flux is approximately equal to some nonlinear function of the combined radial velocity. In particular we add a correction term proportional to v 2 c ( ) to the expression for the light t flux. The result is shown in Figure 1(c) and obtained from
$$L _ { c o r r } ( t ) / 1 0 ^ { 7 } = 1. 1 8 + 1. 3 v _ { c } ( t ) + 1. 5 6 v _ { c } ^ { 2 } ( t )$$
The effect of the correction term is to stretch the upper part of the graph increasing the amplitude of the upper envelope and compress the lower part decreasing the amplitude of the lower envelope. Note that there may be other ways to achieve the same effect, e.g. it could conceivably result from a nonlinear interaction between the modes that generate PO1 and PO2.
## 3. OTHER BLAZHKO CASES
## 3.1. Multiple Frequency Blazhko
Recent work (Skarka 2014; Benko et al. 2014) has shown that it is fairly common to find Blazhko cases that have two or more simultaneous modulation frequencies. This is accompanied with the appearance in the spectra of two or more sets of side-peaks with different spacings. This could be the result of having an additional excited mode or modes near the fundamental that is not phase-locked to one of the peaks in the original side-peak array. Nonlinear processes can then generate additional side peaks that are integer linear combinations of these with the other base frequencies.
A good example of a complex case is V445 Lyr (Guggenberger et al. 2012), which exhibits a dual frequency Blazhko modulation. The spectrum shows two strong peaks just above the fundamental and much weaker peaks below it. Thus it is likely that these strong peaks correspond to two active modes in addition to the fundamental. Each mode has an associated set of side-peaks. Modeling in this case would require the addition of all three modal oscillations, which could be labeled PO1, PO2 and PO3. With this simple change, the model will generate the observed dual
frequency Blazhko modulation. When the amplitude is large, the observed waveform has the typical sawtooth-like character (see e.g. Figure 1(b) in Guggenberger et al. 2012) although its rising section is not nearly as sharp as in the waveform for RR Lyr.
Occasionally the Blazhko minimum of V445 Lyr is very low, at which point it exhibits a 'double-maxima' waveform, see (see e.g. Figures 1(b) and 1(d) in Guggenberger et al. 2012). This can be easily explained as a cancellation of the base frequency, leaving the harmonic content behind. The first oscillation, PO1 is strongly non-sinusoidal and hence has a significant harmonic content. It can be thought of as the sum of two components: a sine wave at the base frequency and the harmonic content (2nd harmonic and above). The other two oscillations PO2 and PO3 have little harmonic content and so can be approximated as sine waves with frequencies that are very close to the base frequency of PO1. If the sum of the amplitudes of PO2 and PO3 is approximately equal to the amplitude of the base frequency component of PO1, then approximate cancellation is possible. It will occur when PO2 and PO3 are approximately in phase with each other and approximately 180 degrees out of phase with the base component of PO1. Since the relative phases of PO1, PO2 and PO3 are slowly changing, the near cancellation will only persist for a short time interval (a few pulsation cycles). The harmonic content of PO1 has nothing to cancel with and therefore remains, producing for that short interval a waveform that appears to be almost entirely harmonics. This is the observed double-maxima waveform. The doubling of the maximum suggests a strong second harmonic content in PO1.
Explanations such as the ninth overtone model that involve an actual modulation of the fundamental would appear to be unable explain this effect. Normally when a nonlinear oscillation is at a low amplitude the nonlinear effects become dramatically less important. In the limit of small amplitude, the oscillation is expected to approach a perfect sine wave. But here the opposite is apparently happening-the amplitude is very small, but the harmonics, and particularly the second harmonic, are stronger than the base frequency.
Assuming for the sake of argument that the ninth overtone is a viable model (in spite of the serious problems outlined in Section 4.1), one might try to explain the strong harmonic content as resulting from interactions between the fundamental and the ninth overtone modes. But these interactions are nonlinear and in the small amplitude limit all such interactions vanish and the vibrational modes become completely independent. In this regard we note that the double-maxima waveform appears only when the apparent amplitude of the fundamental mode is at it smallest, probably about 10 times smaller that at its maximum. We also note that the peak in the observed spectrum that (supposedly) corresponds to the ninth overtone mode is extremely small in amplitude, see f H +3 f 0 in Table 2 in Guggenberger et al. (2012). Thus, within the ninth overtone model, it does not seem reasonable that sufficient nonlinear distortion can be present to account for the double-maxima waveform.
## 3.2. Different Triplet Symmetry
A study of variable stars in the Large Magellanic Cloud (Alcock et al. 2003) found 74% of Blazhko stars had a larger side-peak on the higher frequency side. These cases are potentially compatible with the RR Lyr results, but what about the other 26%? Since nothing presented here depends on the active side-peak being on the high side, one possibility that the active mode is on the lower frequency side.
It is also possible that some stars may have additional side-peaks that are active modes. If another mode is phase-locked to one of the side-peaks (or to one of the harmonics of the fundamental) it may be difficult to determine with certainty that it is present. One indication of this could be that the side-peak in question is notably higher in amplitude than expected from the pattern of its neighbors.
In cases where the side peaks are nearly symmetric in amplitude, the model of Nowakowski & Dziembowski (2001) could be at work. This is a beating-modes model that is generated by a three mode resonance. See the discussion of this model in Section 1.
Cases of extremely low amplitude Blazhko modulation could potentially be generated by turbulence noise. This could be the case when there is a stable resonance between the fundamental and a nonradial mode at the same frequency. These modes would normally be locked at a particular relative phase. However, if the phase is offset from equilibrium it may oscillate back and forth as it decays back to the equilibrium point. Such oscillations are typically seen in amplitude equation simulations, see e.g. the introduction of Buchler & Goupil (1984). So there is phase modulation, but when adding the two modes together a beating process will occur generating the appearance of amplitude modulation. If driven constantly by noise, this might generate a symmetric triplet with fuzzy side-peaks. This is of course rather speculative as it is not known if such phase locked states exist nor how much excitation could be achieved by the turbulence noise.
## 3.3. Long Period Blazhko
There are also cases with very long periods, up to 10 years (see e.g. Figure 5 in Soszynski et al. 2011), which seem to be at odds with the idea that nonlinear coupling should cause phase-locking for modes that are close together. One explanation is that the nonlinear coupling between the modes could be too weak for phase locking to occur. The coupling coefficient depends on the spacial and temporal characteristics of the modes and the order of the resonance involved and so it can vary substantially depending on the specific modes involved. There can also be an element of chance involved: If the coefficient is expected to lie in a range from negative to positive values then for certain stellar parameters it may 'accidentally' have a value very near zero.
Another explanation is that these stars could have extremely low rotation rates. In this case, a dipolar nonradial mode could have its axis drift due to a combination of random forces from
turbulence plus the effects of the slow rotation. If this mode was phase-locked to the fundamental, then this could be described as a noisy version of the Nowakowski & Dziembowski (2001) model (see discussion in Section 1) and the observed Blazhko frequency would be related to (but not equal to) the rotational frequency. As noted before, this model is also a beating-modes model since the amplitudes of the modes are not actually being modulated.
This long period case is also problematic for Blazhko models of the 'unstable resonance' type such as the ninth overtone model (Section 4.1). One might expect the system in that case simply go to stable phase-locking of the two modes rather than oscillate with an exceedingly long period-probably far longer than characteristic time constants of the system, such as the damping time constants for the modes involved. The time scale is also excessively long for the shock model (Section 4.2). It might be compatible with the Stothers model which seems to be more viable for very long period modulation (Section 4.3).
## 4. PROBLEMS WITH OTHER BLAZHKO EXPLANATIONS
## 4.1. The Ninth Overtone Model
One explanation for the Blazhko effect (arguably the leading one) involves an unstable resonant interaction between the fundamental mode and the ninth radial overtone (Szabo et al. 2010; Buchler & Kollath 2011; Kollath et al. 2011). This is a very high order resonance with the frequency ratio of the overtone to the fundamental being 9 to 2. As a result, correspondingly high powers appear in the amplitude equations (Equation 2 in Buchler & Kollath 2011). This can cause extreme changes in coupling strength in response to moderate changes in amplitude.
Probably the most serious problem is that the spectral peak that they identify with the overtone is exceedingly small compared to the fundamental. Resonance based models like theirs depend on a nonlinear interaction between the modes to generate the Blazhko oscillation. This interaction must exhibit an instability that results in a substantial amount of energy being transferred back and forth at the Blazhko frequency between the fundamental and the other mode or modes. It is this energy shift that causes the observed amplitude changes in the fundamental during the Blazhko cycle. When the energy is shifted away from the fundamental it goes into the ninth overtone. One can see this quite clearly in the mode simulations of Buchler & Kollath (2011). In their Figure 1, note that the amplitude curve of mode B is roughly an inverted copy of mode A and is comparable in size. Also Buchler & Goupil (1984) state in their introduction '... one can have periodic or 'chaotic' energy transfer between the resonant modes.' But this appears to be inconsistent with the negligibly small size of the ninth overtone peak in the observational spectrum. To be specific, from Szabo et al. (2010) Table 1 and Figure 8, the amplitude of the peak at f 0 (the fundamental) is about 158 mmag and the peak at 9 / f 2 0 (the presumed ninth overtone) is about 1.65 mmag, about 100 times smaller. (The size discrepancy is even worse than it appears, since the kinetic energy in a mode is proportional to the square of the velocity, and the luminosity is known to vary roughly in
proportion to velocity, making the energy ratio on the order of 10000.) When there is a large size discrepancy between the two modes there is a corresponding imbalance in the interaction between them. This is clear from the amplitude equations of Buchler & Kollath (2011) (their Equation 2) which shows that when B is much smaller than A , the effect of B on the dynamics of A is negligible, while the effect of A on the dynamics of B is quite strong. To make an analogy, an ocean wave can have a big effect on the motion of a surfer, while the surfer has a negligible effect on the wave. Thus the dynamics of the fundamental will proceed as if the overtone did not exist and there can be no Blazhko effect.
A second problem, is that the ninth overtone is being identified as a half integer harmonic of f 0 , but in addition to the expected peak at 9 / f 2 0 the spectrum also shows peaks at 1 / f 2 0 , 3 2 / f 0 , 5 2 / f 0 , 7 2 / f 0 , etc. Of these, peak at 3 / f 2 0 is the largest by far, and is close to 3 times the amplitude of the one at 9 / f 2 0 . This suggests that 3 / f 2 0 is the peak associated with an excited mode, while the others, including the peak at 9 / f 2 0 , may be nonlinear mixing product peaks. The fact that the 9 2 / f 0 peak is slightly elevated compared to the 7 / f 2 0 peak may simply reflect the fact that 9 / f 2 0 is the third harmonic of 3 / f 2 0 .
A third problem is that the alternating peak heights in the Kepler time series data appear and disappear in an intermittent, seemingly random fashion, rather than varying in synchronization with the Blazhko cycle. If the half integer harmonics are directly related to the Blazhko effect then they must play an active role in the Blazhko dynamics. They cannot be exhibiting their own separate dynamics and still be held responsible for the Blazhko effect. This is in contrast to the Smolec et al. (2011) test of the Stothers model, where they state 'We note that in our models period doubling is strictly repetitive and always appears at the same phase of the Blazhko cycle.'
A fourth problem is that since the ninth overtone model is entirely radial, the Blazhko effect should be reproducible with the available 1D hydrocodes. Apparently this has not occurred.
## 4.2. The Shock Model
The shock model of Gillet (2013) presents just the framework of a possible theory with all of the needed details either missing or fatally flawed. The basic idea is that shock waves build up in the atmosphere while the fundamental is growing in amplitude, then these shocks generate turbulence affecting the excitation mechanism for the fundamental, then the fundamental decays in amplitude and finally the atmosphere becomes stable again and the process repeats. Gillet claims this process depends critically on the presence of the 'transient' first overtone, but this concept is not clearly defined. Does transient mean the mode is only present for part of the Blazhko cycle? If so what portion of the cycle and why does Gillet not look at the observational data to try and prove that the overtone has a transient presence? Gillet admits that the observed amplitude of the first overtone is about 100 times smaller than the fundamental. From this one might estimate that the energy could be on the order of 10000 times smaller since energy typically is proportional to
the square of the velocity while the light signal tends to be roughly proportional to the velocity. So how can something so weak and insignificant be expected to play a major role in the stellar dynamics? Gillet also admits that 'The physical origin of shock s3' has not been determined yet' (i.e. it is complete speculation that is has anything to do with the first overtone). Also there is the fact that the first overtone has a different frequency than the fundamental so if it is supposed to be generating a shock wave at a certain phase of the fundamental pulsation, there will be a problem since the overtone will have a different relative phase each successive pulsation. The results also depend on the ability of the shock wave to suddenly cause the complete 'desynchronization' of the photospheric layer; is there evidence that a shock wave can in fact do this and do it suddenly rather than gradually? One should bear in mind that these shock waves are not caused by an outside source, but simply appear as part of the dynamics of the pulsational motion. So it seems surprising that such a shock wave would suddenly disrupt the motion that is generating it. It would seem more likely that if turbulence was emerging this would limit the strength of the shock wave and an equilibrium would be reached. Aside from the turbulence associated with the desynchronization this would seem to be an entirely radial model and so there is also the question as to why the proposed mechanism is not seen in 1D hydrocode simulations. Fossati et al. (2014) discuss this model-they find that the changes in the level of turbulence required by the model appears higher than what should be expected from their measurements of variations in the microturbulent velocity for RR Lyr.
## 4.3. The Stothers Model
The Stothers model has received a fair amount of attention recently. It is discussed at length in Kovacs (2009) and the papers by Smolec et al. (2011) and Molnar et al. (2012) are entirely devoted to it. It depends on several assumptions that are very difficult to test. In particular it is assumed that there is a magnetic field whose amplitude varies at the Blazhko frequency and that the changes in magnetic field in turn affect the level of turbulence sufficiently to modulate the amplitude of the fundamental mode. In an effort to test the theory, Smolec et al. (2011) use a hydro model into which the assumed periodic changes in the turbulence are introduced in an ad hoc manner. They are able to recreate some of the desired Blazhko features including asymmetric side-peaks and the disappearance of the bump during the ascending Blazhko phase, but they think that the amount of turbulence changes required for this may unrealistic. The simulation also fails to reproduce the upward motion of the bump as seen in Figure 1 of the current paper. The behavior of the bump in the Stothers case is of course a function of the level of turbulence. Also there is no significant amplitude differential between the upper and lower Blazhko envelope functions. Molnar et al. (2012) using a different hydro model find the Stothers model could be workable but only for Blazhko periods longer than 100 days due to the slow response of the system. Fossati et al. (2014) discuss this model-they find that the changes in the level of turbulence required by the model appears higher than what should be expected from their measurements of variations in the microturbulent velocity for RR Lyr.
## 5. RELATION TO EARLIER WORK BY THE AUTHOR
In a previous paper (Bryant 2014) it was shown that an accurate fit to second quarter Kepler data for RR Lyr could be achieved by a hybrid mode consisting of two component modes of the same frequency. The first of these components is a highly non-sinusoidal sawtooth-like waveform, while the second is approximated as a sinusoid. Although locked in frequency, the two components are allowed to change independently as a function of the Blazhko phase. The first component varies little in amplitude or frequency throughout the Blazhko cycle, while the second changes quite strongly both in amplitude and in phase relative to the first component. By optimizing the fit to the Kepler data, the amplitude and phase behavior was obtained and displayed in Figure 4 of Bryant (2014).
The current work is related to that earlier work in the following way: It is found that the phase and amplitude dynamics of the second component of the earlier description can be approximately generated by adding together two constant amplitude sine waves, one with frequency ω 1 and a second of lower amplitude with frequency ω 1 + ω B . This was noted previously by Guggenberger et al. (2008) who found this type of phase and amplitude behavior in several RR Lyrae stars and attributed it to this type of beating process. The first component corresponds to PO1a in the current model. The second sine wave corresponds to PO2. The first sine wave corresponds to PO1b, although in the current model this is made non-sinusoidal. Making this non-sinusoidal helps to correct one of the problems with that earlier model, in that the bump feature generated by that model was not as pronounced as in the Kepler data.
## 6. CONCLUSION
It has been shown that a simple beating-modes model of the Blazhko effect can provide a remarkably good fit to Kepler data for RR Lyr. The model consists of two periodic oscillations of slightly different frequency that are simply added together. This is in contrast to the much more complex picture of resonant mode models that depend on nonlinear coupling between modes leading to an oscillatory exchange of energy between them. The current model reproduces for the first time the apparent motion, disappearance and reappearance of the bump feature, the phase difference between the upper and lower modulation envelope functions and their relationship to the motion of the bump. In addition, it has been shown that the amplitude disparity between the upper and lower envelope functions can be reproduced if the light curve is assumed to be a nonlinear function of the radial velocity. Previous work by Cox (1993, 2009) suggests that the modes involved could be the fundamental and the g 4 , l = 1 nonradial mode. The model is easily extended to cases exhibiting multiple modulation frequencies by simply associating additional large near-resonant peaks in the spectrum with active modes of the star, i.e. one additional mode is required for each additional modulation frequency. Application of this extended model was discussed for the case of V445 Lyr.
The mysterious double-maxima waveform observed for that case is explained providing very strong
evidence that the hypothesis of this paper is correct.
In addition, problems with other recent models of the Blazhko effect have been discussed in some detail, including the ninth overtone model, the shock model and the Stothers model.
The author thanks Katrien Kolenberg for very helpful discussion and comments about the model, the manuscript and the spectral analysis. The author thanks Elisabeth Guggenberger for helpful comments on the manuscript.
## A. ONE-ZONE MODEL AND PO1 DETAILS
The equations of motion are derived from a simplified analysis of pressure and gravity acting on the portion of a star above a certain radius R 0 . We may take R 0 to be the effective bottom of the stellar envelope if we are considering the fundamental mode (as we are in this paper) or the outermost radial node for other modes such as the first radial overtone. For large amplitude oscillations, the motion is ballistic or gravity dominant for most of the cycle, with pressure becoming dominant only near the minimum position of the cycle. This leads to a 'bouncing ball' type of oscillation whose derivative is the sawtooth-like waveform mentioned previously. Regardless of the accuracy of this simplified picture of the dynamics, the ability to reproduce the sawtooth-like waveform shows that the one-zone model seems to capture an essential aspect of the true dynamics.
The model takes the form of an ordinary differential equation for a harmonic oscillator, but with the usual linear forcing term replaced with a nonlinear function f ( x ) that, in a very simplified way, captures the essence of the gravity and pressure forces, i.e. ˙ x = v and ˙ = v f ( x ) where x and v are position and velocity. This model is conservative, excitation and damping of the oscillations are omitted, but could be added to the model if desired. Also currently assumed to be small and ignored are nonlinear interactions between modes that could lead to energy transfer or phase shift. In Bryant (2014) the following expression is derived for the forcing function:
$$f ( x ) = A ( R _ { 1 } + x ) ^ { 2 } / ( ( R _ { 1 } + x ) ^ { 3 } - 1 ) ^ { \gamma } - B / ( R _ { 1 } + x ) ^ { 2 },$$
where:
$$A = \omega ^ { 2 } V _ { 1 } ^ { \gamma } ( 3 \gamma R _ { 1 } ^ { 4 } V _ { 1 } ^ { - 1 } - 4 R _ { 1 } ) ^ { - 1 },$$
$$B = \omega ^ { 2 } R _ { 1 } ^ { 4 } ( 3 \gamma R _ { 1 } ^ { 4 } V _ { 1 } ^ { - 1 } - 4 R _ { 1 } ) ^ { - 1 },$$
γ = 1 + 1 /n p , V 1 = R 3 1 -1, R 1 is a characteristic radius for the stellar material that lies above r = R 0 and n p is the polytropic index. The problem has been rescaled so that R 0 = 1. To keep the values near unity, time is in units of 0.1 days. The values used in the model are ω = 1 5 rad/0.1 day, . n p = 1 5, . R 1 = 1 1, . ω 1 = 1 108328 and . ω B = 0 015393. The Blazhko period is 72 pulsation periods. . Initial condition for PO1a: x 1 a (0) = -0 0691, . v 1 a (0) = 0. Initial condition for PO1b is chosen so that it produces the identical oscillation except that it is leading in phase by 2.1058232 radians or
0.19 days: x 1 b (0) = 0 185258, . v 1 b (0) = 0 078770. . Note that ω is the frequency for small amplitude oscillations; the actual oscillation frequency drops as the amplitude increases and is equal to ω 1 for the specified initial conditions. The velocity for PO1, v 1 ( ), is the combined velocity of PO1a and t PO1b, with scale factors chosen to fit the Kepler data:
$$v _ { 1 } ( t ) = v _ { 1 a } ( t ) + 0. 1 5 v _ { 1 b } ( t ) & & ( A 4 )$$
This function is plotted in Figure 2.
## B. FREQUENCY SPECTRA FOR QUASIPERIODICITY
When a nonlinear dynamical system is quasiperiodic, it will have two base frequencies which are incommensurate, i.e. they do not have a rational ratio. In the phase space the dynamics will lie on a two-torus or doughnut surface. The dynamics will wind its way around this surface, taking an infinite amount of time to completely cover it. Motion on a 2-torus can be described by two angle-like variables, θ 1 and θ 2 , corresponding to the two ways in which one can go around the torus. We can define the angles so that they satisfy θ 1 = ( ω t 1 ) mod (2 π ) and θ 2 = ( ω t 2 ) mod (2 π ). The state of the system can be expressed as a function of these two angles:
$$X = X ( \theta _ { 1 }, \theta _ { 2 } ) = \text{state vector of the system}$$
The state is 2 π periodic in both angles (since a 2 π shift in either angle corresponds to one pass around the torus). Thus it can be expanded as a two dimensional Fourier series:
$$X = \sum _ { m = - \infty } ^ { \infty } \sum _ { n = - \infty } ^ { \infty } A _ { m n } \exp ( i m \theta _ { 1 } + i n \theta _ { 2 } )$$
where
$$A _ { m n } & = \frac { 1 } { ( 2 \pi ) ^ { 2 } } \int _ { 0 } ^ { 2 \pi } \mathrm d \theta _ { 1 } \int _ { 0 } ^ { 2 \pi } \mathrm d \theta _ { 2 } X ( \theta _ { 1 }, \theta _ { 2 } ) \exp ( - i m \theta _ { 1 } - i n \theta _ { 2 } )$$
In terms of time this becomes:
$$X = \sum _ { m = - \infty } ^ { \infty } \sum _ { n = - \infty } ^ { \infty } A _ { m n } \exp ( i \omega _ { n m } t )$$
where the frequencies in the spectrum are
$$\omega _ { m n } = m \omega _ { 1 } + n \omega _ { 2 }$$
The amplitudes A mn can also be obtained as an integral over time:
$$A _ { m n } = \lim _ { T \to \infty } \frac { 1 } { T } \int _ { 0 } ^ { T } \mathrm d t X ( t ) \exp ( - i \omega _ { m n } t )$$
The amplitude of these components must fall off as m and n become large.
Cases of three frequency quasiperiodicity (or higher) require a simple extension of the above analysis to a higher dimensional Fourier series.
## REFERENCES
Alcock, C., Alves, D. R., Becker A., et al. 2003, ApJ., 598, 597
Baker, N. 1966, in Stellar Evolution: Proc. of International Conf., November 13-15, 1963, ed. R. F. Stein & A. G. W. Cameron eds., (New York: Plenum), 333
Benko, J. M., Szabo, R. & Paparo, M. 2011, MNRAS 417, 974
Benko, J. M., Plachy, E., Szabo, R., et al. 2014, ApJ Supplement Series, 213, 31
Blazhko, S. N. 1907, Astron. Nachr., 175, 325
Bono, G., Marconi, M. & Stellingwerf, R. F. 2008, A&A 2000, 360, 245
Borkowski, K. J. 1980, Space Science Reviews 27, 511
Breger, M. & Kolenberg, K. 2006, A&A 460, 167
Bryant, P. H. 2014, ApJ 783, L15
Buchler, J. R. & Goupil, K. 1984, ApJ 279,394
Buchler, J. R. & Kollath, Z. 2011, ApJ 731, 24
Buchler, J. R., Moskalik, P. & Kovacs, G. 1990, ApJ, 1990, 351, 617
Cox, A. N. 1993, in Proc. IAU Coll. 139, New Perspectives on Stellar Pulsation and Pulsating Variable Stars, Eds. J. Nemec , J.M. Matthews, Cambridge Univ. Press, p. 409
Cox A. N. 2009, in AIP Conf. Ser. Vol. 1170, Stellar Pulsation: Challenges for Theory and Observation. Am. Inst. Phys., New York, p. 276
Fossati E., Kolenberg, K., Shulyak, D. V., et al. 2014, MNRAS Accepted, arXiv:1410.4380 [astroph.SR]
Geroux, C. M. & Deupree, R. G. 2014, ApJ, 783, 107
Gillet, D. 2012, 2013, A&A, 554, 46
Guggenberger, E. & Kolenberg, K. 2006, Comm. in Asteroseismology., 148, 21
Guggenberger, E., Kolenberg, K. & Breger M. 2008, Journal of Physics: Conference Series, 118, 012056
Guggenberger, E., Kolenberg, K., Nemec, J. M., et al. 2012, MNRAS 424, 649
Kolenberg, K. 2012, JAAVSO, 40, 481
Kolenberg, K., Szabo, R., Kurtz, D. W., et al. 2010, Astrophys. J. Lett., 713, L198
Kollath, Z., Molnar, L. & Szabo R. 2011, MNRAS 414, 1111 Kovacs, G. 2009, AIP Conference Proceedings 1170, 261 Molnar, L., Kollath, Z. & Szabo R. 2012a, MNRAS, 424, 31 Mundprecht, E., Muthsam, H. J. & Kupka, F. 2013, MNRAS, 435, 3191 Munteanu, A., Bono, G., Jose, J., et al. 2005, ApJ, 627, 454 Nowakowski, R. M. & Dziembowski, W. A. 2001, Acta Astronomica, 51,5 Simon, N. 1977, ApJ, 1977, 217, 160 Skarka, M. 2014, A&A 562, A90 Smolec, R., Moskalik, P., Kolenberg, K., et al. 2011, MNRAS, 414, 2950 Soszynski, I., Dziembowski, W. A., Udalski, A., et al. 2011, Acta Astron., 61, 1 Stellingwerf, R. F. 1972, A&A, 21, 91 Stellingwerf, R. F. 1986, ApJ 303, 119 Stellingwerf, R. F., Gautschi, A., & Dickens, R. J. 1987, ApJ, 313, L75 Stothers, R. B. 2006, MNRAS 365, 178. Stothers, R. B. 2010, PASP 122, 536. Szabo R., Kollath, Z., Molnar, L., et al. 2010, MNRAS, 409, 1244
Szeidl, B. & Jurcsik, J. 2009, Comm. in Asteroseismology, 160, 17
Szeidl, B., Jurcsik, J., Sodor, A., Hajdu, G. & Smitola, P. 2012, MNRAS 424, 3094
This preprint was prepared with the AAS LT E X macros v5.2. A | 10.1088/0004-637X/802/1/52 | [
"Paul H. Bryant"
] | 2014-08-01T18:53:58+00:00 | 2015-01-22T08:33:44+00:00 | [
"astro-ph.SR"
] | Is the Blazhko effect the beating of a near-resonant double-mode pulsation? | In this paper it is shown that the Blazhko effect may result from a
near-resonant type of multi-mode pulsation, where two (or sometimes more)
periodic oscillations with slightly different frequencies gradually slip in
phase, producing a beat frequency type of modulation. Typically one of these
oscillations is strongly non-sinusoidal. Two oscillations are sufficient for
the standard Blazhko effect; additional oscillations are needed to explain
multi-frequency modulation. Previous work on this hypothesis by Arthur N. Cox
and others is extended in this paper by developing a simple (non-hydro) model
that can accurately reproduce several important features found in Kepler data
for RR Lyr, including the pulsation waveform, the upper and lower Blazhko
envelope functions and the motion, disappearance and reappearance of the bump
feature. The non-sinusoidal oscillation is probably generated by the
fundamental mode and the other oscillations are probably generated by nonradial
modes. This model provides an explanation for the strong asymmetry observed in
the side peak spectra of most RR Lyrae stars. The motion and disappearance of
the bump feature are shown to be an illusion, just an artifact of combining the
oscillations. V445 Lyr is presented as an example with dual modulation. The
mysterious double-maxima waveform observed for this star is explained,
providing additional strong evidence that this beating-modes hypothesis is
correct. Problems with other recent explanations of the Blazhko effect are
discussed in some detail. |
1408.0264v1 | ## Liquid Argon Dielectric Breakdown Studies with the MicroBooNE Purification System
## R. Acciarri, B. Carls, C. James, B. Johnson, H. Jostlein, S. Lockwitz, B. Lundberg, ∗ J.L. Raaf, R. Rameika, B. Rebel, G.P. Zeller, and M. Zuckerbrot
Fermi National Accelerator Laboratory, P.O. Box 500, Batavia, IL, 60510, USA
ABSTRACT: The proliferation of liquid argon time projection chamber detectors makes the characterization of the dielectric properties of liquid argon a critical task. To improve understanding of these properties, a systematic study of the breakdown electric field in liquid argon was conducted using a dedicated cryostat connected to the MicroBooNE cryogenic system at Fermilab. An electrode sphere-plate geometry was implemented using spheres with diameters of 1.3 mm, 5.0 mm, and 76 mm. The MicroBooNE cryogenic system allowed measurements to be taken at a variety of electronegative contamination levels ranging from a few parts-per-million to tens of parts-pertrillion. The cathode-anode distance was varied from 0.1 mm to 2.5 cm. The results demonstrate a geometric dependence of the electric field strength at breakdown. This study is the first time that the dependence of the breakdown field on stressed cathode area has been shown for liquid argon.
KEYWORDS: Liquid Argon; Time Projection Chambers; Dielectric Strength, Electric Breakdown; Stressed Area.
∗ Corresponding author: [email protected] (S. Lockwitz)
## Contents
| 1. | Motivation | Motivation | 1 |
|------|-------------------------------------|--------------------------------------------------|-----|
| 2. | Experimental Setup | Experimental Setup | 2 |
| | 2.1 | HV breakdown test device | 2 |
| | 2.2 | Liquid argon filtration and recirculation system | 4 |
| | 2.3 | Experimental Procedure | 5 |
| 3. | Results | Results | 6 |
| 4. | Comparison to Previous Measurements | Comparison to Previous Measurements | 10 |
| 5. | Conclusion | Conclusion | 12 |
| 6. | Acknowledgements | Acknowledgements | 12 |
## 1. Motivation
Liquid argon time projection chambers (LArTPCs) are a popular and effective detection technique for neutrino interactions thanks to their capability to perform three-dimensional imaging of charged particle interactions with fine spatial resolution and accurate calorimetry. Neutrino experiments are driving the design of these detectors to contain increasingly larger volumes of argon. The increase in argon volumes implies both the application of higher voltages and the capability to maintain such voltages in the detector, avoiding discharge.
In this context, a precise characterization of the dielectric properties of liquid argon becomes extremely important. Several research activities have been devoted to this subject [1, 2, 3, 4]. Nonetheless, parameters related to dielectric breakdown through liquid argon are still not well known due to the varied values found among studies characterized by different liquid argon electronegative contamination levels, electrode sizes, electrode shapes, and materials.
To address the origin of these differences, a systematic study was conducted of the breakdown electric field in liquid argon as a function of electrode size, liquid argon electronegative contamination level, and cathode-anode distance. These studies were conducted at Fermilab's Liquid Argon Test Facility using a dedicated test stand linked to the MicroBooNE cryogenic system.
Figure 1. Schematic overlaid on a photo of the test stand with spherical-tipped high-voltage feedthrough and adjustable distance between sphere and grounded anode plate.
## 2. Experimental Setup
## 2.1 HV breakdown test device
The experimental setup shown in Fig. 1 consists of a breakdown test device housed in a cryostat that was connected to the MicroBooNE liquid argon filtration and circulation system. The breakdown measurements reported here were made using a sphere-plate geometry as shown in the figure. The plate was held at ground and the sphere, attached to the end of a high-voltage (HV) feedthrough, was brought to the desired voltage. The 74 cm diameter grounded plate was mechanically polished to a mirror finish, then electropolished. Two mirrored, angled wings were attached to the edges of the grounded plate, as seen in Fig. 2. The wings allow inspection of the sphere-plane gap through viewports on the top of the cryostat. Both the size of the sphere and its distance from the plate were varied in the tests.
A Glassman LX150N12 power supply capable of delivering up to -150 kV was used for these measurements [5]. It was equipped with a Glassman-supplied microcontroller which allowed control by a LABVIEW program [6]. The output of the power supply was connected to a resistive filter pot that served to isolate the stored energy in the setup. The output of the filter pot was then
H
G
F
E
D
C
B
A
Figure 2. Breakdown test device. The stainless steel ground plate was mechanically polished and then electropolished. Angled, mirrored wings were installed for viewing from above. The spherical-tipped HV feedthrough was brought within proximity of the plate for the tests. The white block seen in the photograph was used to calibrate the device and was removed during operation.
connected to a high-voltage feedthrough with swappable spherical probe tips.
(a) Fabrication begins with the aquisition of PE insulator material. Supplers of extruded polyethelene insulator cannot guarantee precise tolerances on diameters. One needs to purchase more than will be used, and down-select based on surface quality and uniformity of extrusion. The ID and OD of selected PE extrusions are measured at room temperature and at cryogenic temperature. These measurements dictate the selection of SS materials for the central conductor and for the ground tube. Dimensions provided in this drawing set represent as-builts, and can be used as guidelines Figure 3. Photograph (a) and drawing (b) of the production HV feedthrough. The spherical probe tip attached to the end of the inner conductor, on the right side of these figures.
(b) Assemble the Central Conductor, Sheets 2 & 3. Fabricate the Warm Fit fixture, Sheet 8.
(c) The assembled Central Conductor is forced at room temperature into the PE insulator, termed the "warm tight fit.
PARTS LIST
Further details in Notes on Sheet 4.
UNITED STATES DEPARTMENT OF ENERGY FERMI NATIONAL ACCELERATOR LABORATORY ORIGINATOR DRAWN CHECKED APPROVED SCALE SHEET NO. DRAWING NUMBER REV. MicroBooNE High Voltage Feed Through Assembly and Notes DESCRIPTION NO. REQ PART NO. 1 Assembly Central Conductor Sheet 2 & 3 1 1 1 1 2 3 4 Assembly Assembly 25% 1 of 8 3 PE Insulator Sheet 4 Ground Tube Sheet 5 & 6 Spring Tip Sheet 7 (d) The PE-Central Conductor assembly is submerged in LN2 to determine OD-cold and compare to OD-warm of the "tight fit" portion at the location of the central conductor slug. These values are used to select the ground tube material ID and wall thickness. Further details in Notes on Sheet 4. (f) Assemble the Ground Tube, Sheets 5 & 6. (g) Cold-fit the Ground Tube assembly onto the PE-Central Conductor assembly. Fit is performed by immersing the PE-Conductor assembly into LN2, upside-down (grooved portion of PE pointing up). The ground tube, flange-end first, is pushed over the PE-Conductor. A PE spacer ring at the bottom of the LN2 vessel is used to give the PE just enough extra length so that when the full assembly is warm, the PE expands with its upper edge flush with the top flange of the ground tube (e) Finish the PE, length and grooves, as on Sheet 4. (h) Other important final parameters : (i) there must be a minimum of 11-in of grooved PE between the ground tube base and the exposed central conductor (ii) the grooves must extend up into the ground tube for a minimum of 2 in. (iii) the PE cannot cover the screw holes for the spring tip The feedthrough was a prototype of the MicroBooNE design, shown in Fig. 3. The design was based on the ICARUS feedthrough [7], with a stainless steel center conductor surrounded by insulating ultra high molecular weight polyethylene (UHMW PE) that was inserted into an outer ground tube. The swappable spherical probe tips attached directly to the center conductor of the feedthrough. The tests reported in this paper evaluated 1.3 mm, 5.0 mm, and 76 mm diameter
ITEM
QTY
8
7
6
5
4
3
2
1
H
G
F
E
D
C
B
A
spherical probe tips, which are shown in Fig. 4.
Figure 4. Three spherical probe tips used in the testing. The two smaller tips had a chrome finish, while the larger was mirror-finished stainless steel.
The feedthrough was attached to a motorized linear translator on the cryostat. The translator was fit with a bellows that allowed the feedthrough to move vertically up and down. An OMEGA force transducer model LC703-150 [8] was connected between the grounded plate and its supports. Force registered by the transducer could then be used to determine contact between the probe tip attached to the feedthrough and the grounded plate. The vertical distance travelled was read out by an AccuRemote digital readout #35-812 with 0.01 mm resolution [9].
## 2.2 Liquid argon filtration and recirculation system
The cryostat was a vacuum-jacketed vessel held between 0-8 psig or 101-157 kPa for the measurements. The pressure was held at a constant value during each measurement by allowing the argon boil off to vent through a check valve. The pressure varied slightly from one measurement to another. The cryostat was plumbed into MicroBooNE's cryogenic system for access to pure argon. The MicroBooNE cryogenic system consists of two filter vessels to remove water and oxygen from the argon, a pump to circulate the argon, and a dewar refrigerated by liquid nitrogen. The first filter was filled with a 4A molecular sieve supplied by Sigma-Aldrich to remove water [10]. The second filter was filled with BASF CU-0226 S, a dispersed copper oxide impregnated on a high surface area alumina to remove oxygen [11]. The filtration system consistently delivered better than 100 partsper-trillion (ppt) oxygen equivalent contamination to the test vessel. The contamination level in the vessel was adjusted by adding higher contamination argon from a separate supply dewar. This
allowed the production of oxygen equivalent contaminations ranging from 1.5 parts-per-million (ppm) to less than 100 ppt for the testing.
The level of oxygen contamination in the argon was measured using two Servomex gas analyzers, a DF-310E and a DF-560E [12]. Combined, the two oxygen analyzers covered a range of 0.1 parts-per-billion (ppb) to 5000 ppm. The level of nitrogen contamination in the argon was measured using an LD8000 Trace Impurity Analyzer [13].
Additionally, contamination values ranging between 300 and 50 ppt oxygen equivalent were measured with a double-gridded ion chamber, henceforth referred to as a purity monitor, immersed in liquid argon. The purity monitor was based on the design in Reference [14]. A measure of electronegative impurities was determined from looking at the fraction of electrons generated at the cathode that arrived at the anode ( QA / QC ) after a drift time, t . The fraction can also be interpreted as an electron lifetime, t such that
$$Q _ { A } / Q _ { C } = e ^ { - t / \tau }.$$
A thorough description of the purity monitor and the data acquisition hardware and software can be found in Reference [15]. The purity monitor had a total drift distance from cathode to anode of 19 cm. It resided in a vessel downstream of the cryostat used for the breakdown tests. Before and after breakdown measurements were taken in the cryostat, argon was allowed to flow into the purity monitor vessel for a purity measurement.
The systematic uncertainties of the purity measurement had the effect of decreasing the fraction of QA / QC , consequently decreasing the observed lifetime [15]. Therefore, the reported electron lifetimes from the purity monitor should be interpreted as a lower bound on the true lifetime.
## 2.3 Experimental Procedure
Measurements over the range of distances between 0 and 2.5 cm were collected over a 12 hour period for a given probe and contamination level. First, the zero spacing point was found using the force transducer described in Section 2.1, and the current was monitored as voltage was applied. The probe was then moved to a distance of 0.1 mm away from the ground plate and the voltage ramped up at a rate of 50 V/s until breakdown occurred. Breakdown voltage is defined as the voltage where the current drawn by the power supply changes its value by more than +100% for a fraction of a second.
The cycle of ramping the voltage up until breakdown and recording the voltage at which the breakdown occurred was repeated multiple times at each distance. The distances considered in these tests are 0.0, 0.1, 0.3, 0.5, and 0.8 mm; 1 through 10 mm in millimeter steps; and 15, 20, and 25 mm. The zero-point of the probe with respect to the ground plate was measured each time the probe was moved. For gap spacings of 1 mm and above, the HV ramp rate was increased from 50 V/s to 250 V/s. No significant effect on the breakdown voltages was found by making this change for the larger distances. Increasing the rate for the spacings above 1 mm allowed the full set of gap spacing measurements to be completed for a given test configuration before a significant amount of argon boiled off.
With the geometry of the test setup, it was found that the liquid argon boil-off caused the liquid level to decrease to nearly the bottom edge of the ground tube at times during testing. In
this situation, for gap spacings greater than 1 cm, breakdown would sometimes occur along the feedthrough instead of the sphere-plane gap as seen in Fig. 5. The breakdown was captured using a camera mounted above the view ports. This effect was mitigated by increasing the liquid argon level in the cryostat. However, in order to avoid changing the contamination level of the liquid during a given test configuration, the liquid argon level was not increased until the end of datacollection for that configuration. In the case of the 5.0 mm probe, data for gap spacings larger than 1.5 cm were not considered because the liquid level was too low, and similarly in the case of the 76 mm probe, data for gap spacings larger than 1 cm were not considered.
Figure 5. A view looking down at the mirrored grounding plate inside the liquid argon-filled test cryostat comparing (a) the effect of a spark along the feedthrough, versus (b) from the electrode to the grounded plate. In the first case, spark light is only visible along the feedthrough and not in the mirrors. In the second case, light is present in the mirrors and not along the feedthrough.
## 3. Results
A full list of the collected data, including probe size and contamination levels, is shown in Table 1. Fig. 6 shows the average breakdown voltage versus distance for a given probe and electronegative contamination level. For each configuration, the average at each distance was computed from multiple breakdown voltages. The extrema of the uncertainties are the lowest and highest values for the given measurement. Systematic uncertainties are not shown here since they would be dwarfed by the spread of the data. A 0.05 mm uncertainty was estimated for the gap spacing measurement based on the resolution of the digital linear readout. A 500 V uncertainty was estimated on the breakdown voltage measurement based on the ramp rate and step size.
As can be seen from Fig. 6, a dependence of the breakdown voltage on liquid argon electronegative contamination level is most evident with the 1.3 mm probe. At gap spacings of about 1 cm with the 1.3 mm probe, the breakdown voltages for oxygen contaminations between 0.2 and 1.4 ppm are about a factor of 1.5 more than for oxygen contaminations between 0.29 and 1.8 ppb. No such dependence of the breakdown voltage on liquid argon electronegative contamination level is observed for the 76 mm probe in the same ranges of oxygen equivalent contamination. This difference in behavior can be seen in Fig. 6b, where the average breakdown voltage versus distance is plotted for only the highest and lowest electronegative contaminations levels achieved with the 1.3 mm and 76 mm probe.
Table 1. The configuration and contamination levels present during the breakdown tests. Errors on the contamination level are based on gas analyzer resolution and measurement time. Values marked with † are upper limits for oxygen equivalent contamination as derived from the purity monitor measurements. Lifetime values listed are derived solely from the purity monitor.
| Probe (mm) | O (ppb) | N (ppm) | Lifetime (ms) |
|--------------|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|-----------------|
| 1.3 | 2 1200 + 50 - 50 1200 + 50 - 50 1200 + 50 - 50 744 + 2 - 2 744 + 2 - 2 . 0 1 . 8 + 1 . 0 - 0 . 7 0 . 35 + 0 . 15 - 0 . 15 200 + 20 - 10 < 0 . 29 † < 0 . 29 † 1400 + 100 | 2 1 . 3 + 0 . 1 - 0 . 1 1 . 7 + 0 . 2 - 0 . 2 1 . 7 + 0 . 2 - 0 . 2 3 . 1 + 0 . 2 - 0 . 5 3 . 1 + 0 . 2 - 0 . 5 6 . 0 + 0 . 2 - 0 . 5 5 . 2 + 0 . 2 - 0 . 5 5 . 5 + 0 . 2 - 0 . 5 6 . 4 + 0 . 2 - 0 . 2 6 . 4 + 0 . 2 - 0 . 2 28 + 2 - 2 | > 1 . 0 > 1 . 0 |
| 5.0 | - 50 1300 + 50 - 50 > 900 775 + 50 | 2 . 2 + 0 . 1 - 0 . 1 22 + 5 - 5 22 + 5 | |
| 76 | - 50 1400 + 50 - 50 70 + 100 - 20 10 . 20 + 0 . 05 - 3 . 00 | - 5 1 . 4 + 0 . 1 - 0 . 1 3 . 2 + 0 . 2 - 0 . 5 + 0 . 2 | |
| | 0 . 60 + 0 . 10 - 0 . 10 < 0 . 23 † < 0 . 14 † < 0 . 13 † < 0 . 13 † < 0 . 09 † | 5 . 5 - 0 . 5 5 . 0 + 0 . 2 - 0 . 5 5 . 9 + 0 . 2 - 0 . 5 5 . 9 + 0 . 2 - 0 . 5 5 . 9 + 0 . 2 - 0 . 5 + 0 . 2 5 . 9 + 0 . 2 | > 2 . 1 > 3 . 5 |
| | | | > 1 . 3 |
| | | | > 2 . 3 |
| | | 5 . 9 - 0 . 5 | > 2 . 3 |
| | 1500 + 100 - 50 | - 0 . 5 24 + 2 - 2 | |
In an attempt to understand the spread of the data, many breakdown voltages were recorded for a fixed distance and contamination level with a given probe. In Fig. 7, 174 breakdown voltages are plotted for the 76 mm probe with a 7 mm gap spacing. The data in Fig. 7 are shown fit with a Weibull function [16], which demonstrates good agreement. The underlying mechanism that produces an asymmetric distribution of voltages is not understood. Previous measurements in other liquid insulators [17, 18, 19] observed similar distributions that were also well described by a Weibull function, suggesting that the process of electrical breakdown has a stochastic nature in liquid insulators.
For each configuration, the electric field was computed using OPERA-2D. The field for the 5.0 mm probe at a spacing of 0.1 mm is shown as an example in Fig. 8. The average breakdown voltage values from Fig. 6 were translated to maximum electric field values using the OPERA model and the results are shown as average maximum breakdown fields in Fig. 9.
Gerhold et al. suggest a minimum electric field value, below which, no breakdown will occur [18]. The lowest breakdown field observed out of all probes and distances examined was
Figure 6. (a) The measured average breakdown voltage recorded as a function of distance for the three probes. The 76 mm sphere data are shown in green, the 5.0 mm sphere data are in blue, and the 1.3 mm sphere data are in red. Different electronegative contamination levels are indicated by the solid, dashed, and dotted lines specified in the legend. (b) The highest and lowest electronegative contaminations levels data for the 1.3 mm and 76 mm spheres.
## Breakdown Voltage at 7 mm Spacing (76 mm Ball)
Figure 7. The distribution of breakdown values for the 76 mm ball for a fixed 7 mm gap spacing. A fit to the Weibull function is shown by the solid red curve.
38 kV/cm. The value of 38 kV/cm can then be interpreted as an upper bound on the minimum electric field for breakdown to occur. Beyond providing a possible lower threshold, the maximum electric field is likely not a parameter of interest by itself, given the differing values among the probes.
Gerhold et al. [18] additionally suggest a dependence between the maximum electric field at breakdown and a stressed cathode area in liquid helium data. The stressed area is defined as the area on the cathode surface with electric field exceeding some percentage, e , of the maximum electric field. In Figs. 10a and 10b, the average maximum electric field is plotted versus stressed area for e values 80% and 90% respectively. This representation unifies the breakdown behavior across the geometries tested in this study. A trend in the data is observed, which appears at any
Figure 8. Sample electric field from OPERA-2D. This is for the 5.0 mm diameter probe at a 0.01 cm gap spacing.
contamination level. An example is given in Fig. 11, where the plots of Fig. 10 are reported only for contaminations above 800 ppb oxygen equivalent. The cluster of points seen for the largest areas of the 1.3 mm and 5 mm probes is a result of the stressed area covering the entire surface of the probe's bottom hemisphere; any further increase in gap spacing does not significantly increase the stressed area.
The combined data for the three probes appears to be consistent with the function suggested in [18]:
$$E _ { m a x } = C ( A ) ^ { - 0. 2 6 }$$
where Emax is the maximum electric field at breakdown, A the stressed area, and C a constant term. The red line shown in Figs. 10 and 11 is a representation of Eq. 3.1.
Gerhold et al. [18] further suggest a dependence of breakdown maximum electric field versus stressed volume. Preliminary calculations show a similar trend as the area effect. This is expected since stressed volume and area are highly correlated in this geometrical setup, however a different geometry could be designed to better decouple the two parameters in a future test setup.
## Average Peak Breakdown Field vs. Distance
Figure 9. The average maximum breakdown field as a function of distance. The 76 mm sphere data are shown in green, the 5.0 mm sphere data are in blue, and the 1.3 mm sphere data are in red. The solid, dashed, and dotted lines are indications of the different electronegative contamination levels specified in the legend.
Figure 10. The average maximum breakdown field versus stressed area of the cathode. (a) The stressed area is defined as the area with an electric field greater than 80% of the maximum electric field. (b) The stressed area is defined as the area with an electric field greater than 90% of the maximum electric field.
## 4. Comparison to Previous Measurements
The data presented in this paper affirm the growing consensus in the literature that breakdown voltages in liquid argon present a strong dependence on the geometry of the experimental setup and a weaker dependence on argon electronegative contamination level. Fig. 12 shows the data from this study along with two others that span a range of parameter space in terms of electrode distances and contamination levels.
Early studies [1, 2] first indicated a possible influence of electronegative contaminants on breakdown voltages, but not for contamination levels or distances important for understanding
Figure 11. The average maximum breakdown field versus stressed area of the cathode for contaminations above 800 ppb oxygen. (a) The stressed area is defined as area with an electric field greater than 80% of the maximum electric field. (b) The stressed area is defined as area with an electric field greater than 90% of the maximum electric field.
Figure 12. A comparison to previous measurements [3]. The measurements described in this paper are shown in red, blue, and green and result from the highest and lowest contamination level measurement taken with each probe. The oxygen values with an asterisk are extrapolated from purity monitor measurements.
LArTPCs. The results from these early studies indicated that at oxygen levels above 0.2 ppm, the breakdown electric fields were above 1 MV/cm at sub-mm spacings for a sphere-sphere geometry. Those same studies showed that the breakdown electric field decreased with lower levels of oxygen. These older data have often been cited as justification for the claim that liquid argon could sustain
electric fields on the order of 1 MV/cm. Recent studies using the same sphere-plate geometry as reported in this paper [3] found that argon electronegative contamination level had a relatively small impact in comparison to geometry in influencing the breakdown field.
## 5. Conclusion
A study of the dielectric strength of liquid argon has been performed wherein both the argon electronegative contamination level and the electrode size and separation have been controlled. The results show a dependence of the electric field strength at breakdown on the geometry of the test. Breakdown occurs at lower field strengths as the separation between electrodes increases from sub-millimeter to centimeter scales.
Breakdown also occurs at lower field strength as the stressed surface area of the electrode increases; the maximum electric field at breakdown scales roughly as the stressed area to the -0 3 . power. This study is the first time that the dependence of the breakdown field on stressed cathode area has been shown for liquid argon.
There is some dependence between breakdown field and argon electronegative contamination level, but the effect from contamination level is small relative to that of the geometries used in this study. The 1.3 mm electrode experienced breakdown at lower electric fields when the argon had lower contamination level: breakdown electric field values for oxygen contaminations between 0.2 and 1.4 ppm were about a factor of 1.5 more than for oxygen contaminations between 0.29 and 1.8 ppb at gap spacings of about 1 cm. However, the 76 mm electrode showed no clear dependence between electric field at breakdown and argon electronegative contamination level.
Electrical breakdown in liquid argon is a stochastic process; the chain of events which lead to the breakdown is not precisely repeatable. A given geometric setup experienced breakdown over a range of voltages; marks left by sparks on the surface of the larger electrodes show that the location of the breakdown can shift when all else is held static. This observation is consistent with the hypothesis mentioned in [3] that space charge effects in the stressed volume near the cathode contribute to breakdown probability.
The observed geometric effects are of critical importance in the design of LArTPC detectors for neutrino experiments, where drift distances must be on the scale of meters. Ideally, one would use a well-known breakdown probability distribution to design LArTPCs. A suitable operating point could then be chosen to minimize damage or dead time while accounting for signal-to-noise requirements. It is difficult in practice to extrapolate these measurements to TPCs given their unique and complicated geometries. However, it is possible that stressed area may need to be minimized in TPC design to prevent dielectric breakdown.
## 6. Acknowledgements
The authors wish to thank the staff at Fermilab for their technical assistance, specifically Jim Walton. Additionally, a great deal of thanks is due to the authors of [3] from the Laboratory for High Energy Physics at University Bern for graciously providing their data.
Fermilab is operated by Fermi Research Alliance, LLC under Contract No. De-AC02-07CH11359 with the United States Department of Energy.
## References
- [1] D. Swan and T. Lewis, 'Influence of electrode surface conditions on the electrical strength of liquified gases,' J. Electrochem. Soc. 107 (1960) 180.
- [2] D. Swan and T. Lewis, 'The influence of cathode and anode surfaces on the electric strength of liquid argon,' Proc. Phys. Soc. 78 (1961) 448.
- [3] A. Blatter et al. , 'Experimental study of electric breakdowns in liquid argon at centimeter scale,' JINST 9 (2014) P04006, arXiv:1401.6693 [physics.ins-det] .
- [4] F. Bay et al. , 'Evidence of electrical breakdown induced by bubbles in liquid argon,' arXiv:1401.2777 [physics.ins-det] .
- [5] Glassman High Voltage Inc., PO Box 317, 124 West Main Street, High Bridge, NJ 08829-0317, U.S.A.
- [6] National Instruments Corp., 11500 N Mopac Expwy, Austin, TX 78759-3504.
- [7] S. Amerio et al. , 'Design, construction and tests of the ICARUS T600 detector,' Nucl. Instrum. Methods 527 (2004) 329.
- [8] OMEGA Engineering Inc., One Omega Drive, P.O. Box 4047, Stamford, Connecticut 06907-0047.
- [9] AccuRemote, 1301 Calle Avanzado, San Clemente, CA 92673, USA.
- [10] Sigma-Aldrich, P.O. Box 14508, St. Louis, MO 63178 USA.
- [11] BASF Corp., 100 Park Avenue, Florham Park, NJ 07932 USA.
- [12] Spectris plc, Heritage House, Church Road, Egham, TW20 9QD, United Kingdom.
- [13] LDetek Inc., 271 Saint-Alphonse Sud, Thetford Mines, Quebec, Canada G6G 3V7.
- [14] G. Carugno et al. , 'Electron lifetime detector for liquid argon,' Nucl. Instrum. Meth. A 292 (1990) 580.
- [15] M. Adamowski et al. , 'The Liquid Argon Purity Demonstrator,' JINST 9 (2014) P07005, arXiv:1403.7236 [physics.ins-det] .
- [16] W. F. Gauster, 'Über oberflächeneffekte beim elektrischen durchbruch von flüssigkeiten,' Ingenieur Archiv 10 (1956) 160.
- [17] J. Gerhold, M. Hubmann, and E. Telser, 'Breakdown probability and size effect in liquid helium,' IEEE Transactions on Dielectrics and Electrical Insulation 5 no. 3, (1998) 321.
- [18] J. Gerhold, M. Hubmann, and E. Telser, 'Gap size effect on liquid helium breakdown,' Cryogenics 34 no. 7, (1994) 579.
- [19] A. L. Kupershtokh et al. , 'Stochastic regularities of electrical breakdown initiation in transformer oil,' Industry Applications Conference. Conference Record of the 2001 IEEE 4 (2001) 2729. | 10.1088/1748-0221/9/11/P11001 | [
"R. Acciarri",
"B. Carls",
"C. James",
"B. Johnson",
"H. Jostlein",
"S. Lockwitz",
"B. Lundberg",
"J. L. Raaf",
"R. Rameika",
"B. Rebel",
"G. P. Zeller",
"M. Zuckerbrot"
] | 2014-08-01T18:54:15+00:00 | 2014-08-01T18:54:15+00:00 | [
"physics.ins-det"
] | Liquid Argon Dielectric Breakdown Studies with the MicroBooNE Purification System | The proliferation of liquid argon time projection chamber detectors makes the
characterization of the dielectric properties of liquid argon a critical task.
To improve understanding of these properties, a systematic study of the
breakdown electric field in liquid argon was conducted using a dedicated
cryostat connected to the MicroBooNE cryogenic system at Fermilab. An electrode
sphere-plate geometry was implemented using spheres with diameters of 1.3 mm,
5.0 mm, and 76 mm. The MicroBooNE cryogenic system allowed measurements to be
taken at a variety of electronegative contamination levels ranging from a few
parts-per-million to tens of parts-per- trillion. The cathode-anode distance
was varied from 0.1 mm to 2.5 cm. The results demonstrate a geometric
dependence of the electric field strength at breakdown. This study is the first
time that the dependence of the breakdown field on stressed cathode area has
been shown for liquid argon. |
1408.0265v1 | ## THE STABILIZED SET OF p 'S IN KRIVINE'S THEOREM CAN BE DISCONNECTED
KEVIN BEANLAND, DANIEL FREEMAN, AND PAVLOS MOTAKIS
In memory of Edward Odell
Abstract. For any closed subset F of [1 , ∞ ] which is either finite or consists of the elements of an increasing sequence and its limit, a reflexive Banach space X with a 1-unconditional basis is constructed so that in each block subspace Y of X /lscript , p is finitely block represented in Y if and only if p ∈ F . In particular, this solves the question as to whether the stabilized Krivine set for a Banach space had to be connected. We also prove that for every infinite dimensional subspace Y of X there is a dense subset G of F such that the spreading models admitted by Y are exactly the /lscript p for p ∈ G .
## 1. Introduction
In the past, many of the driving questions in the study of Banach spaces concerned the existence of 'nice' subspaces of general infinite dimensional Banach spaces. Finding counterexamples to these questions involved developing new ideas for constructing Banach spaces. B. Tsirelson's construction of a reflexive infinite dimensional Banach space which does not contain /lscript p for any 1 < p < ∞ [T] and W.T. Gowers and B. Maurey's construction of an infinite dimensional Banach space which does not contain an unconditional basic sequence [GM] are two important examples. On the other hand, after Tsirelson's construction, J-L. Krivine proved that every basic sequence contains /lscript p for some 1 /lessorequalslant p /lessorequalslant ∞ finitely block represented [K] (where the case p = ∞ refers to c 0 ), and it is not difficult to show that every normalized weakly null sequence in a Banach space has a subsequence with a 1-suppression unconditional spreading model. Thus, though we cannot always find these properties in infinite dimensional subspaces, they are still always present in certain finite block or asymptotic structure.
2010 Mathematics Subject Classification: Primary 46B03, 46B06, 46B07, 46B25, 46B45 Key words: Spreading models, Finite block representability, Krivine's Theorem The first named author acknowledges support from the Lenfest Summer Grant at Wash-
ington and Lee University.
Research of the second author was supported by the National Science Foundation.
The authors would like to acknowledge the support of program APIΣTEIA-1082.
In his paper on Krivine's Theorem, Rosenthal proved that given any Banach space, the set of p 's such that /lscript p is finitely block represented in the Banach space can be stabilized on a subspace [R] (for a simplified proof of the stability result see also [M, page 133]). That is, given any infinite dimensional Banach space X , there exists an infinite dimensional subspace Y ⊆ X with a basis and a nonempty closed subset I ⊆ [1 , ∞ ] such that for every block subspace Z of Y , /lscript p is finitely block represented in Z if and only if p ∈ I . Rosenthal concluded his paper by asking if this stabilized Krivine set I had to be a singleton. E. Odell and Th. Schlumprecht answered this question by constructing a Banach space X with an unconditional basis which had the property that every unconditional basic sequence is finitely block represented in every block sequence in X [OS1]. Thus, the stabilized Krivine set for this space is the interval [1 , ∞ ]. Later, Odell and Schlumprecht constructed a Banach space with a conditional basis which had the property that every monotone basic sequence is finitely block represented in every block sequence in X [OS2]. At this point, the known possible stabilized Krivine sets for a Banach space are singletons and the entire interval [1 , ∞ ]. P. Habala and N. Tomczak-Jaegermann proved that if 1 /lessorequalslant p < q /lessorequalslant ∞ and X is an infinite dimensional Banach space such that /lscript p and /lscript q are finitely block represented in every block subspace of X then X has a quotient Z so that every r ∈ [ p, q ] is finitely block represented in Z [HT]. They then asked if the stabilized Krivine set for a Banach space is always connected [HT], which was later included as problem 12 in Odell's presentation of 15 open problems in Banach spaces at the Fields institute in 2002 [O]. We solve the stabilized Krivine set problem with the following theorem.
Theorem. Let F ⊆ [1 , ∞ ] be either a finite set or a set consisting of an increasing sequence and its limit. Then there exists a reflexive Banach space X with an unconditional basis such that for every infinite dimensional block subspace Y of X :
- (i) For all 1 /lessorequalslant p /lessorequalslant ∞ , the space /lscript p is finitely block represented in Y if and only if p ∈ F .
- (iii) If F is an increasing sequence with limit p ω then every spreading model admitted by Y is isomorphic to /lscript p for some p ∈ F and for every p ∈ F \ { p ω } /lscript p is admitted as a spreading model by Y .
- (ii) If F is finite then the spreading models admitted by Y are exactly the spaces /lscript p for p ∈ F .
This theorem is somewhat surprising in that the corresponding question for finite representability instead of finite block representability is very different. Indeed, if /lscript p is finitely representable in a Banach space X for some 1 /lessorequalslant p < 2 then /lscript r is finitely representable in X for all r ∈ [ p, 2]. However, for 2 < p < ∞ the Banach space /lscript r is finitely representable in /lscript p if and only if r = 2 or r = p . Thus the position in [1 , ∞ ] of the set F of p 's that are finitely represented in a space X determines whether F is an interval.
Our results show that in the case of block finitely represented, the position of F in the interval [1 , ∞ ] does not matter.
Theorem 1 also solves several open questions on spreading models raised by G. Androulakis, E. Odell, Th. Schlumprecht, and Tomczak-Jaegermann [AOST]. They asked in particular the following three questions: Does there exist a Banach space so that every subspace has exactly n many different spreading models? Does there exist a Banach space so that every subspace has exactly countably infinitely many different spreading models? If a Banach space admits /lscript 1 and /lscript 2 spreading models in every subspace must it admit uncountably many spreading models? In [AM2], S.A. Arygros and the third author have constructed a space so that every subspace admits every unconditional basis as a spreading model. In [ABM], S.A. Argyros with the first and third named authors created a Banach space such that every infinite dimensional subspace admits exactly two spreading models up to isomorphism, namely /lscript 1 and c 0 . Theorem 1 includes the case that F is an increasing sequence and its limit, and so it is natural to question if the case of a decreasing sequence is possible. However, B. Sari proved that if a Banach space admits a countable collection of spreading models which form a strictly increasing sequence in terms of domination, then the Banach space admits uncountably many spreading models [S]. Thus, Theorem 1 (iii) would be impossible in the case that F is a decreasing sequence and its limit as the spaces { /lscript p } p ∈ F would include an increasing sequence in terms of domination.
Given a Banach space X with a basis, one may consider the set of all p 's, such that /lscript p is admitted as a spreading model by X . Although this set may fail to coincide with the Krivine set of the space, or may even be empty [OS1], it is always contained in the Krivine set. Therefore, for a given subset F of [1 , ∞ ] when constructing a space X , one way to ensure that F is contained in any stabilized Krivine set of X , is to have /lscript p admitted as a spreading model by every subspace of the space for every p ∈ F . For any single 1 /lessorequalslant p < ∞ , Tsirelson's method allows one to build a reflexive space not containing /lscript p that is asymptotic /lscript p . Every spreading model admitted by this space is isomorphic to /lscript p and the Krivine set of every infinite dimensional subspace is the singleton { p } . In this paper, for any finite set of p 's, we build a space with exactly these /lscript p 's hereditarily as spreading models and exactly these p 's hereditary as Krivine p 's. Moreover, for any increasing sequence of p 's we get almost the same result. In this case, the only caveat is that, although the basis of the space X admits only the limit p as a spreading model, we did not prove that the limit p is admitted in every subspace.
In the case of two distinct p 's, our construction is rooted in the convexified Tsirelson's spaces in the sense of T. Figiel and W.B. Johnson's description [FJ] and the work of Odell and Schlumprecht [OS1], [OS2]. The methods we follow are based on the ones from [ABM]. In particular, for the simplest case of F = 1 { , ∞} , our construction reduces to a small modification of the space X 1 0 1 , , which is the simplest case of the construction defined in that paper,
and in which it is shown that X 1 0 1 , admits only c 0 and /lscript 1 spreading models in every subspace. In recent literature, the spaces in [OS1], [OS2], [ABM], [AM1] and [AM2] are referred to as Tsirelson spaces with constraints or multi-layer Tsirelson spaces. For the sake of understanding our construction in the simplest case F = { 1 , ∞} , the norm satisfies the following implicit equation for x ∈ c 00 :
where the supremum is over successive intervals ( E i ) n i =1 and ( m i ) n i =1 with
$$\lambda \lambda \\ \| x \| & = \| x \| _ { 0 } \ve$$
(min E i ) n i =1 ∈ S 1 , min E > i (max E i -1 ) 2 and m > i max E i -1 , and for each m ∈ N , where the supremum is over successive intervals ( F i ) m i =1 . These m -norms and the way they are combined above are the previously mentioned contraints. Since the constraints are based on averages, local and asymptotic c 0 structure occurs in every subsapce. Furthermore, in contrast to Tsirelson space, which has homogeneous asymptotic /lscript 1 structure, the above construction hereditarily provides both /lscript 1 and c 0 local and asymptotic structure.
$$\| x \| _ { m } & = \frac { 1 } { 4 m } \sup \$$
In the case that F = { p 1 < · · · < p n } ⊂ [1 , ∞ ] we present a space X , admitting hereditarily /lscript p 1 , . . . , /lscript p n asymptotic structure and nothing more. For this purpose we define a new norm, which has n -many layers, each one corresponding to an /lscript p k structure, for k = 1 , . . . , n . The base layer corresponds to the /lscript p n norm, while for k = 1 , . . . , n -1 the k th layer corresponds to norm /lscript p k and it is defined using the previous layers. To avoid the domination of some of these layers over the rest, to each of these layers, except for the basic one, some constraints have to be applied. The constraints are based on p ′ n -averages, where p ′ n is the conjugate exponent of p n .
When F consists of an increasing sequence ( p k ) k and its limit p ω , countably many layers of the norm are used. In this case, the norm ‖ · ‖ ∗ of the space is defined through the following formulas. We state here the implicit equations for the norms for the sake of giving insight into our construction, but we will actually use a different definition in Section 3 in terms of norming functionals. For 0 < θ /lessorequalslant 1 4 and / x ∈ c 00 ( N ) we define:
where both suprema are are taken over all d ∈ N and successive intervals ( E q ) d q =1 of the natural numbers. These ‖ · ‖ 0 ,m norms are the constraints applied to the norm of the space. If for some k > 0 the norms ‖ · ‖ i,m have
$$\begin{array} { l l$$
been defined for every 0 /lessorequalslant i < k and m ∈ N , for x ∈ c 00 ( N ) and m ∈ N we define:
where the supremum is taken over all d ∈ N , 0 /lessorequalslant i q < k for q = 1 , . . . , d and ( E q ) d q =1 , ( m q ) d q =1 which satisfy certain growth conditions depending on m . The norm of the space satisfies the following implicit equation:
$$\| x \| _ { k, m } = \theta \sup \left ( \sum _$$
$$\left \| x \right \| _ { * } & = \max \left \{ \left$$
Using the above description of the norm, it is easy to see that any block sequence in our space satisfies a lower /lscript p ω estimate with constant θ . Likewise, in section 4 we prove that any block sequence satisfies an upper /lscript p 1 estimate with constant 2. In the case F is finite with F = { p 1 < · · · < p n } then the norm of the space satisfies the same formula, where p ω is replaced with p n .
The paper is organized as follows. In section 2 we give a few preliminary definitions. Section 3 contains the definition of the spaces. In section 4 we set notation that we will use in our subsequent evaluations and prove upper and lower estimates on normalized block sequences. In sections 5 and 6 we prove the spaces have the desired spreading model structure. Finally, in section 7 we show that in every block subspace the only Krivine p 's are those admitted as spreading models.
The majority of the research included in this paper was conducted while the first two authors where visiting the National Technical University of Athens. We sincerely thank Spiros Argyros for his hospitality and enlightening conversations.
## 2. Preliminaries
We begin with some preliminary definitions. Two basic sequences ( x i ) and ( y i ) are C -equivalent for some C /greaterorequalslant 1 if √ C -1 ‖ ∑ a x i i ‖ /lessorequalslant ‖ ∑ a y i i ‖ /lessorequalslant √ C ‖ ∑ a x i i ‖ for all scalar sequences ( a i ). A basic sequence ( e i ) ∞ i =1 is finitely block represented in a basic sequence ( x i ) ∞ i =1 if for all N ∈ N and ε > 0 there exists a finite block sequence ( y i ) N i =1 of ( x i ) ∞ i =1 which is (1 + ε )-equivalent to ( e i ) N i =1 . For 1 /lessorequalslant p /lessorequalslant ∞ , we say that /lscript p is finitely block represented in ( x i ) ∞ i =1 if the unit vector basis of /lscript p is finitely block represented in ( x i ) ∞ i =1 (where we use the case p = ∞ to mean c 0 ).
We say that a basic sequence ( e i ) ∞ i =1 is a spreading model of a basic sequence ( x i ) ∞ i =1 if for all finite sequences of scalars ( a i ) n i =1 we have that
$$\imath = \imath \sum _ { i = 1 } ^ { n } a _ { i } e _ { i } & \| = \lim _ { t _ { 1 } \to \infty } \cdots \lim _ { t _ { n } \to \infty } \| \sum _ { i = 1 } ^ { n } a _ { i } x _ { t _ { i } } \|. \\ \intertext { a Banach space $X$ adds $(e_{i})\nolimits_{i-1}$ as a spreading moc}$$
We say that a Banach space X admits ( e i ) ∞ i =1 as a spreading model if ( e i ) ∞ i =1 is equivalent to a spreading model of some basic sequence in X . We say that X admits /lscript p as a spreading model if X admits a spreading model equivalent
to the unit vector basis for /lscript p . In the literature, the basic sequence ( e i ) ∞ i =1 as well as the Banach space formed by its closed span are both often referred to as spreading models.
## 3. The definition of the space X .
In this section we give the definition of the norming set of the space X . We apply a variation of the method of saturation under constraints, introduced by Odell and Schlumprecht in [OS1], [OS2]. The way this method is applied is similar to the one in [ABM] and it allows /lscript p structure to appear hereditarily in the space, for a predetermined set of p 's, which is either finite or consists of an increasing sequence and its limit.
## Notation. Let G be a subset of c 00 ( N ).
- (i) A finite sequence ( f q ) d q =1 of elements of G will be called admissible if f 1 < · · · < f d and d /lessorequalslant minsupp f 1 .
- (ii) Assume that ( f q ) q is a sequence of functionals in G , and each functional f q has been assigned a positive integer s f ( q ), called the size of f q . Then ( f q ) q will be called very fast growing if (max supp f q -1 ) 2 < minsupp f q and s f ( q ) > maxsupp f q -1 for all q > 1.
Let ξ 0 ∈ [2 , ω ] and let F = { p k : 1 /lessorequalslant k < ξ 0 }∪{ p ξ 0 } ⊂ [1 , ∞ ] with p k ↑ p ξ 0 in the case that ξ 0 = ω and p 1 < p 2 < · · · < p ξ 0 -1 < p ξ 0 otherwise.
We now define the norming set of the space X . We do so inductively by defining an increasing sequence of subsets of c 00 ( N ). To some of the functionals that we construct we shall assign an order, a size or both. Fix a positive real number 0 < θ /lessorequalslant 1 4. / Let W 0 = {± e ∗ j } j ∈ N . To the functionals in W 0 we don't assign an order or size. Assume that for some m ∈ N ∪ { 0 } the set W m has been defined, below we describe how the set W m +1 is defined.
## Functionals of order- , or of order0 ξ 0 . Define
$$W _ { m + 1 } ^ { 0 } & = \left \{ \theta \sum _ { q = 1 } ^ { d } c _ { q } f _ { q } \colon \, f _ { 1 } < \dots < f _ { d } \in W _ { m }, \, \sum _ { q = 1 } ^ { d } | c _ { q } | ^ { p _ { \xi _ { 0 } } ^ { \prime } } \leqslant 1 \right \}.$$
If a functional f of order-0 has the form f = θ ∑ d q =1 (1 /n ) 1 /p ′ ξ 0 f q with d /lessorequalslant n , then the size of f is defined to be s f ( ) = n . If a functional f of order-0 is not of this form then we do not assign any size to it.
A functional f = θ ∑ d q =1 c f q q as above will be called of order-0. In some cases, for convenience these functionals shall also be referred to as functionals of orderξ 0 .
If m +1 = 1 then we define W 1 = W 0 ∪ W 0 1 , otherwise m +1 /greaterorequalslant 2 and we shall include more functionals in W m +1 , as described below.
Functionals of orderk , with 1 /lessorequalslant k < ξ 0 . Define
$$W _ { m + 1 } ^ { k } = \begin{cases} \theta \sum _ { q = 1 } ^ { d } c _ { q } f _ { q } \colon \sum _ { q = 1 } ^ { d } | c _ { q } | ^ { p _ { k } ^ { \prime } } \leqslant 1, & ( f _ { q } ) _ { q = 1 } ^ { d } \text{ is an admissible and very} \end{cases}.$$
$$\text{of which has order strictly smaller than $k$} \}.$$
$$\overset { r } { \overset { k } { \text{$m+1} } } = \left \{ \theta \sum _ { q = 1 } ^ { d } c _ { q } f _ { q } \colon \sum _ { q = 1 } ^ { d } | c _ { q } | ^ { p _ { k } ^ { \prime } } \leqslant 1, \ ( f _ { q } ) _ { q = 1 } ^ { d } \text{ is an admissible and very} \\ \text{fast growing sequence of functions in $W_{m}$, each one} \\ \text{ of which has order strictly smaller than $k$} \right \}.$$
A functional f = θ ∑ d q =1 c f q q as above will be called of orderk with size s f ( ) = min { s f ( q ) : q = 1 , . . . , d } .
If k = 1 and p 1 = 1, we replace the condition ∑ d q =1 | c q | p ′ 1 /lessorequalslant 1 with the condition max {| c q | : q = 1 , . . . , d } /lessorequalslant 1. Note that if ξ 0 is finite and ξ 0 = k 0 +1, then very fast growing sequences of functionals of orderk 0 are not used. Observe also that some functionals may be of more than one order or have multiple sizes, however, this shall not cause any problems.
If m +1 /greaterorequalslant 2, let W m +1 = ( ∪ 0 /lessorequalslant k<ξ 0 W k m +1 ) ∪ W m and W = ∪ ∞ m =0 W . m The space X is the completion of c 00 ( N ) under the norm induced by W , i.e. for x ∈ c 00 ( N ) the norm of x is equal to sup {| f ( x ) | : f ∈ W } .
Remark 3.1. The following are clear from the definition of the norming set.
- (i) For every f 1 < · · · < f d in W and real numbers ( c q ) d q =1 with d q =1 | c q 0 | p ′ ξ 0 /lessorequalslant 1, the functional f = θ d q =1 c f q q is also in W .
Remark 3.2. For every f ∈ W and subset E of the natural numbers, we have that f | E , the restriction of f onto E , is also in W . In particular, if f is of orderk , then f | E is also of orderk and s f ( | E ) /greaterorequalslant s f ( ). One can also check that the norming set W is closed under changing signs, i.e. if f ∈ W and g is such that | f | = | g | , then g is also in W . Therefore, the unit vector basis of c 00 ( N ) forms a 1-unconditional basis for X .
- ∑ ∑ (ii) For every 1 /lessorequalslant k < ξ 0 and every admissible and very fast growing f 1 < · · · < f d in W , each one of which has order strictly smaller than k and real numbers ( c q ) d q =1 with ∑ d q =1 | c q 0 | p ′ k /lessorequalslant 1, the functional f = θ ∑ d q =1 c f q q is also in W .
Recall that functionals of order-0 are also called functionals of orderξ 0 .
Remark 3.3. For every m > 0, 1 /lessorequalslant ζ /lessorequalslant ξ 0 and f ∈ W m , which is of orderζ , there exist f 1 < · · · < f d in W m -1 and real numbers c 1 , . . . , c d with ∑ d q =1 | c q | p ′ ζ /lessorequalslant 1 such that f = θ ∑ d q =1 c f q q . If moreover ζ = k < ξ 0 , then ( f q ) d q =1 is an admissible and very fast growing sequence of functionals, each one of which has order strictly smaller than k .
Before proceeding to the study of the properties of the space X , let us briefly explain the ingredients of the norming set W , without getting into
too many details. If 1 < ξ 0 /lessorequalslant ω and we have determined a set F = { p 1 < · · · < p ξ 0 } ⊂ [1 , ∞ ], then every element f of the norming set falls into one of the following three categories:
- (i) The functional f is an element of the basis, i.e. f ∈ {± e i } i .
- (iii) The functional f is of orderk with 1 /lessorequalslant k < ξ 0 , i.e. f = θ ∑ d q =1 c f q q where the sequence f 1 < · · · < f d are successive elements of the norming set satisfying certain constraints, while the coefficients ( c q ) q are in the unit ball of /lscript p ′ k .
- (ii) The functional f is of order-0, i.e. f = θ ∑ d q =1 c f q q where f 1 < · · · < f d can be any successive elements of the norming set, combined with coefficients ( c q ) q in the unit ball of /lscript p ′ ξ 0 .
The functionals of order-0 provide /lscript p ξ 0 structure to the space and, since the /lscript p ξ 0 is the smallest of the /lscript p norms for p ∈ F , their construction is not subject to any constraints. On the other hand, for 1 /lessorequalslant k < ξ 0 , the functionals of order k provide /lscript p k structure. One has to define these functionals carefully, in order not to demolish the desired /lscript p ζ structure, for k < ζ /lessorequalslant ξ 0 . This is the role of the constraints, which become more restrictive as k becomes smaller.
One can verify that the norm induced by the norming set W is alternatively described by the implicit formula given in the introduction.
## 4. Basic norm evaluations on block sequences of X .
In this section we prove a simple, but useful, lemma and we also prove that block sequences in X have an upper /lscript p 1 estimate and a lower /lscript p ξ 0 estimate. We start with some notation, which in conjunction with the next lemma, will be used frequently throughout the paper. Here, the range of a vector is the smallest closed interval containing the support.
Notation. Let x 1 < · · · < x m be a finite block sequence in X and f = θ ∑ d q =1 c f q q be a functional of orderζ , 1 /lessorequalslant ζ /lessorequalslant ξ 0 . Define the following :
- A 2 = { 1 , . . . , d } \ A , 1
- A 1 = { q ∈ { 1 , . . . , d } : ran f q ∩ ran x j = ∅ for at most one 1 /lessorequalslant j /lessorequalslant m , }
/negationslash
- B = { j ∈ { 1 , . . . , m } : there exists q ∈ A 1 with ran f q ∩ ran x j = ∅ }
/negationslash
- C j = ∥ ∥ ∥ ( c q ) q ∈ A j 1 ∥ ∥ ∥ /lscript p ζ ′ for j ∈ B,
- A j 1 = { q ∈ A 1 : ran f q ∩ ran x j = ∅ } for j ∈ B,
/negationslash
- g j = θ ∑ q ∈ A j 1 ( c /C q j ) f q for j ∈ B and
- E q = { j ∈ { 1 , . . . , m } : ran f q ∩ ran x j = ∅ } , for q ∈ A . 2
/negationslash
The following lemma follows immediately from our choice of notation. As in Remark 3.3, here we also use the fact that order-0 functionals can be referred to as orderξ 0 functionals.
Lemma 4.1. Let x 1 < . . . < x m be a finite block sequence in X and f = θ ∑ d q =1 c f q q be a functional of orderζ for some 1 /lessorequalslant ζ /lessorequalslant ξ 0 . The functionals ( g j ) j ∈ B are orderζ functionals in W , we have that ( ∑ j ∈ B C p ′ ζ j ) 1 /p ′ ζ /lessorequalslant 1, and the following holds:
$$& ( 1 ) \quad \left | f \left ( \sum _ { j = 1 } ^ { m } x _ { j } \right ) \right | \leqslant \left ( \sum _ { j \in B } C _ { j } | g _ { j } ( x _ { j } ) | \right ) + \theta \left ( \sum _ { q \in A _ { 2 } } | c _ { q } | \cdot \left | f _ { q } \left ( \sum _ { j \in E _ { q } } x _ { j } \right ) \right | \right ). \\ & \quad \text{Moreover, if } A _ { 2 } \, = \, \{ q _ { 1 } < \cdots < q _ { r } \} \text{ then } \max E _ { q _ { i } } \leqslant \min E _ { q _ { i + 1 } } \text{ for } i = 1 \text{ } r - 2 \text{ Thus for each}$$
∣ ∣ ∣ ∣ Moreover, if A 2 = { q 1 < · · · < q r } then max E q i /lessorequalslant min E q i +1 for i = 1 , . . . , r -1 and max E q i < min E q i +2 for i = 1 , . . . , r -2. Thus, for each 1 /lessorequalslant j /lessorequalslant m there exists at most two sets E q such that x j ∈ E q .
Note that applying H¨lder's inequality to (1), gives the following inequalo ity, which in most cases will be more convenient for us than (1).
$$\left | f \left ( \sum _ { j = 1 } ^ { m } x _ { j } \right ) \right | \leqslant \left \| \left ( g _ { j } ( x _ { j } ) \right ) _ { j \in B } \right \| _ { \ell _ { p _ { \zeta } } } + \theta \left \| \left ( f _ { q } ( \sum _ { j \in E _ { q } } x _ { j } ) \right ) _ { q \in A _ { 2 } } \right \| _ { \ell _ { p _ { \zeta } } }. \\ \text{Proposition 4.2. Let } x _ { 1 } < \cdots < x _ { m } \text{ be a normalized finite block sequence} \\ \text{in } X \text{ and } ( \lambda _ { \lambda } ) ^ { m } \text{. be scalars. The following holds:}$$
∣ ∣ Proposition 4.2. Let x 1 < · · · < x m be a normalized finite block sequence in X and ( λ j ) m j =1 be scalars. The following holds:
$$\theta \| ( \lambda _ { j } ) _ { j } \| _ { \ell _ { p _ { \xi _ { 0 } } } } \leqslant \left \| \sum _ { j = 1 } ^ { m } \lambda _ { j } x _ { j } \right \| \leqslant 2 \| ( \lambda _ { j } ) _ { j } \| _ { \ell _ { p _ { 1 } } }. \\ \text{ first prove the lower inequality. Note that this is tri} \\ n _ { \epsilon } = \infty \,. \, \text{ thus we assume that $n_{\epsilon}<\infty$. For each $i\in\}$}$$
$$\left \| \sum _ { j = 1 } ^ { m } \lambda _ { j } x _ { j } \right \| \geqslant \theta \sum _ { j = 1 } ^ { m } | \lambda _ { j } | ^ { p _ { \xi _ { 0 } } / p _ { \xi _ { 0 } } ^ { \prime } } f _ { j } \left ( \sum _ { i = 1 } ^ { m } \lambda _ { i } x _ { i } \right ) = \theta \left ( \sum _ { i = 1 } ^ { m } | \lambda _ { i } | ^ { p _ { \xi _ { 0 } } ^ { \prime } } \right ) = \theta. \\ \text{The upper inequality clearly follows from the following claim that we will} \\ \text{prove} \, \text{by induction on } n \in \mathbb { N }. \, \text{For all } n \in \mathbb { N } \cup \{ 0 \} \text{ and } f \in W _ { \lambda } \text{ (see Remark}$$
Proof. We first prove the lower inequality. Note that this is trivial in the case that p ξ 0 = ∞ , thus we assume that p ξ 0 < ∞ . For each j ∈ { 1 , . . . , m } find f j so that f j ( x j ) = 1 and supp f j = supp x j . Without loss of generality, we may assume that ( ∑ m i =1 | λ i | p ξ 0 ) 1 /p ξ 0 = 1 and λ i ≥ 0 for all 1 /lessorequalslant i /lessorequalslant m . Thus, θ ∑ m j =1 | λ j | p ξ 0 /p ′ ξ 0 f j ∈ W . Therefore
∥ The upper inequality clearly follows from the following claim that we will prove by induction on n ∈ N . For all n ∈ N ∪{ } 0 and f ∈ W n (see Remark 3.3) we have
$$f \left ( \sum _ { j = 1 } ^ { m } \lambda _ { j } x _ { j } \right ) \leqslant 2 \left ( \sum _ { j = 1 } ^ { m } | \lambda _ { j } | ^ { p _ { 1 } } \right ) ^ { 1 / p _ { 1 } }.$$
The case of f ∈ W 0 = {± e ∗ j } is trivial. Assume that the above holds for some n /greaterorequalslant 0. Let f ∈ W n +1 . Then f = θ ∑ d q =1 c f q q is of orderζ and f 1 < · · · < f d are in in W n and d q =1 | c q | p ′ ζ /lessorequalslant 1.
∑ By (2) after Lemma 4.1 then applying the inductive hypothesis, we obtain the following:
$$& \text{the following:} \\ & \left | f ( \sum _ { j = 1 } ^ { m } \lambda _ { j } x _ { j } ) \right | \, & \leqslant \, \left ( \sum _ { j \in B } | \lambda _ { j } | ^ { p } \right ) ^ { 1 / p _ { \zeta } } + \theta \left ( \sum _ { q \in A _ { 2 } } \left | f _ { q } \left ( \sum _ { j \in E _ { q } } \lambda _ { j } x _ { j } \right ) \right | ^ { p _ { \zeta } } \right ) ^ { 1 / p _ { \zeta } } \\ & \leqslant \, \left ( \sum _ { j \in B } | \lambda _ { j } | ^ { p _ { \zeta } } \right ) ^ { 1 / p _ { \zeta } } + 2 \theta \left ( \sum _ { q \in A _ { 2 } } \left ( \sum _ { j \in E _ { q } } | \lambda _ { j } | ^ { p _ { 1 } } \right ) ^ { p _ { \zeta } / p _ { 1 } } \right ) ^ { 1 / p _ { \zeta } }. \\ & \text{By the last part of Lemma 4.1, for each $j$ there exists at most two distinct} \\ & q \in A _ { 2 } \text{ such that $j \in E_{q}$. This fact together with $p_{1} \leqslant p_{\zeta}$ imply that}$$
$$\mathfrak { x } & \prec \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \colon \\ & \left ( 5 \right ) & \left ( \sum _ { q \in A _ { 2 } } \left ( \sum _ { j \in E _ { q } } | \lambda _ { j } | ^ { p _ { 1 } } \right ) ^ { p _ { \zeta } / p _ { 1 } } \right ) ^ { 1 / p _ { \zeta } } \, < 2 ^ { 1 / p _ { 1 } } \left ( \sum _ { j = 1 } ^ { m } | \lambda _ { j } | ^ { p _ { 1 } } \right ) ^ { 1 / p _ { 1 } } \,. \\ & \text{Combining relations } ( 4 ) \text{ and } ( 5 ) \text{ together with } 0 < \theta \leqslant 1 / 4, \, \text{ we obtain the}$$
By the last part of Lemma 4.1, for each j there exists at most two distinct q ∈ A 2 such that j ∈ E q . This fact together with p 1 /lessorequalslant p ζ imply that
Combining relations (4) and (5) together with 0 < θ /lessorequalslant 1 4, we obtain the / desired bound in (3).
/square
## 5. Spreading models of X .
In this section we define the α -indices in a very similar manner as they have been defined in [ABM], [AM1] and [AM2]. Although previously the α -indices were used to describe the action of certain averages of functionals on a block sequence, in our case this is not exactly the same. Here, the indices are used to study the action of functionals of a certain order on a block sequence. However, the principle is the same and we retain this notation. As is the case in these papers, the indices determine the spreading models admitted by a block sequence in the space X . As a consequence we prove that every spreading model admitted by a weakly null sequence in X must equivalent to the unit vector basis of /lscript p ζ for some ζ ∈ [1 , ξ 0 ].
Definition 5.1. Let ( x j ) j be a block sequence in X and let 1 /lessorequalslant k < ξ 0 . Assume that for every very fast growing sequence ( f q ) q of functionals in W , each one of which has order strictly smaller than k , and every subsequence ( x j i ) i of ( x j ) j we have that lim i | f i ( x j i ) | = 0. Then we say that the α <k -index of ( x j ) j is zero and write α <k { ( x j ) j } = 0. Otherwise we write α <k { ( x j ) j } > 0.
Remark 5.2. Let ( x j ) j be a block sequence in X and 1 /lessorequalslant m < k < ξ 0 . If α <k { ( x j ) j } = 0 then also α <m { ( x j ) j } = 0.
The following characterization has appeared in similar forms in [ABM], [AM1] and [AM2]. We omit the proof as it is simple and straightforward.
Proposition 5.3. Let 1 /lessorequalslant k < ξ 0 and ( x j ) j be a block sequence in X . The following assertions are equivalent:
- (i) α <k { ( x j ) j } = 0.
- (ii) For every ε > 0 there exist j 0 , i 0 ∈ N such that for every f ∈ W of order strictly smaller than k with s f ( ) /greaterorequalslant i 0 and every j /greaterorequalslant j 0 we have that | f ( x j ) | < ε .
Lemma 5.4. Let 1 /lessorequalslant k < ξ 0 and ( x j ) j be a bounded block sequence in X such that α <k { ( x j ) j } > 0. Then ( x j ) j has a subsequence with a spreading model that dominates the unit vector basis for /lscript p k . That is, there exists ε > 0 and a subsequence ( x j i ) i of ( x j ) j such that for every natural numbers m /lessorequalslant i 1 < · · · < i m and every real numbers λ , . . . , λ 1 m the following holds:
Proof. By the definition of the α <k index, there exists ε ′ > 0, a subsequence of ( x j ) j , again denoted by ( x j ) j and a very fast growing sequence ( f j ) j of functionals of order strictly smaller than k , such that | f j ( x j ) | > ε ′ for all j ∈ N . We may also assume that ran f j ⊂ ran x j for all j ∈ N . Set ε = θε ′ and note that for every m /lessorequalslant j 1 < · · · < j m and every real numbers ( c t ) m t =1 with ∑ m t =1 | c t | p ′ k /lessorequalslant 1 the functional f = θ ∑ m t =1 c f t j t is of orderk . Let m /lessorequalslant j 1 < · · · < j m be natural numbers and λ , . . . , λ 1 m be real numbers. We have the following estimate:
$$\begin{array} { c } \ e q \cup \nu _ { j } \ e q \colon \ e q \colon \ e q \colon \ e q \colon \\ \left \| \sum _ { t = 1 } ^ { m } \lambda _ { t } x _ { j _ { i _ { t } } } \right \| \geqslant \varepsilon \left \| ( \lambda _ { t } ) _ { t } \right \| _ { \ell _ { p _ { k } } }. \\ \\ \intertext { \text{ition of the $\alpha_{<k}$ index, there exists $\varepsilon^{\prime}_{\cdot}$} } \quad \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ e q \colon \ a$$
$$m \leqslant j _ { 1 } < \dots < j _ { m } & \text{ be natural numbers and } \lambda _ { 1 }, \dots, \lambda _ { m } \text{ be real numbers. } \text{ We have the following estimate: } \\ \left \| \sum _ { t = 1 } ^ { m } \lambda _ { t } x _ { j _ { t } } \right \| & \geqslant \sup \left \{ \theta \sum _ { t = 1 } ^ { m } c _ { t } f _ { j _ { t } } \left ( \sum _ { t = 1 } ^ { m } \lambda _ { t } x _ { j _ { t } } \right ) \colon \sum _ { t = 1 } ^ { m } | c _ { t } | ^ { p ^ { \prime } } \leqslant 1 \right \} \\ & = \sup \left \{ \theta \sum _ { t = 1 } ^ { m } | c _ { t } \lambda _ { t } | \cdot | f _ { j _ { t } } ( x _ { j _ { t } } ) | \colon \sum _ { t = 1 } ^ { m } | c _ { t } | ^ { p ^ { \prime } } \leqslant 1 \right \} \\ & \geqslant \quad \theta \varepsilon ^ { \prime } \sup \left \{ \sum _ { t = 1 } ^ { m } | c _ { t } \lambda _ { t } | \colon \sum _ { t = 1 } ^ { m } | c _ { t } | ^ { p ^ { \prime } } \leqslant 1 \right \} \\ & = \quad \varepsilon \left ( \sum _ { t = 1 } ^ { m } | \lambda _ { t } | ^ { p _ { k } } \right ) ^ { 1 / p _ { k } }. \\ \text{ Lemma $5.5. Let $(x_{i})\i$ be a normalized block sequence and $2\leqslant \zeta\leqslant \varepsilon$ such}$$
/square
Lemma 5.5. Let ( x j ) j be a normalized block sequence and 2 /lessorequalslant ζ /lessorequalslant ξ 0 such that α <k { ( x j ) j } = 0 for every 1 /lessorequalslant k < ζ . Then ( x j ) j has a subsequence with a spreading model that is 2-dominated by the unit vector basis for /lscript p ζ . In particular, there exists a subsequence ( x j i ) i of ( x j ) j such that for every
natural numbers m /lessorequalslant i 1 < · · · < i m and every real numbers λ , . . . , λ 1 m the following holds:
$$\left \| \sum _ { t = 1 } ^ { m } \lambda _ { t } x _ { j _ { i _ { t } } } \right \| \leqslant 3 \left \| ( \lambda _ { t } ) _ { t } \right \| _ { \ell _ { p _ { \zeta } } }. \\ \text{sider the case in which $\zeta$ is finite, i.}$$
∥ ∥ Proof. We first consider the case in which ζ is finite, i.e. ζ = k ′ + 1 with 1 /lessorequalslant k ′ < ξ 0 .
Using Proposition 5.3 we pass to a subsequence, again denoted by ( x j ) j such that for any j /greaterorequalslant j 0 /greaterorequalslant 2 and any f ∈ W of order strictly smaller than k ′ with s f ( ) /greaterorequalslant minsupp x j 0 we have that
$$& \left | f ( x _ { j } ) \right | < ( j _ { 0 } \max \text{supp} \, x _ { j _ { 0 } - 1 } ) ^ { - 1 } \,. \\ & \mathfrak { w r } \cdot \mathfrak { m } \, \mathfrak { w } \cdot \mathfrak { w } \cdot \mathfrak { w } \cdot \mathfrak { w r } \cdot \mathfrak { w r } \cdot \mathfrak { w r } \cdot \mathfrak { w r } \cdot \mathfrak { w r } \cdot \mathfrak { w r } \cdot \mathfrak { w r } \cdot \mathfrak { w r } \cdot \mathfrak { w r } \cdot \mathfrak { w r } \cdot \mathfrak { w r } \cdot \mathfrak { w r } \cdot \mathfrak { w r } \cdot \mathfrak { w r } \cdot \mathfrak { w r } \cdot \mathfrak { w r } \cdot \mathfrak { w r } \cdot \mathfrak { w r } \cdot \mathfrak { w r }$$
We will show by induction on n , where W = ∪ n W n (see Remark 3.3) that for every m /lessorequalslant j 1 < · · · < j m , every real numbers λ , . . . , λ 1 m and every f ∈ W n the following holds:
$$\left | f \left ( \sum _ { t = 1 } ^ { m } \lambda _ { t } x _ { j _ { t } } \right ) \right | & \leqslant 2 \left \| \left ( \lambda _ { t } \right ) _ { t = 1 } ^ { m } \right \| _ { \ell _ { p _ { \zeta } } }. \\ \text{For $f\in W_{0}$ the result holds. Let $f=\theta\sum_{q=1}^{d}c_{q}f_{q}$ be a functi}.$$
Case 1: The functional f is of orderη with ζ /lessorequalslant η /lessorequalslant ξ 0 . By Inequality (2) after Lemma 4.1, we have that
∣ ∣ For f ∈ W 0 the result holds. Let f = θ ∑ d q =1 c f q q be a functional in W n . We distinguish two cases, concerning the order of f .
$$& \left | f \left ( \sum _ { t = 1 } ^ { m } \lambda _ { t } x _ { j t } \right ) \right | \leqslant \left ( \sum _ { t \in B } | \lambda _ { t } | ^ { p _ { \eta } } \right ) ^ { 1 / p _ { \eta } } + \theta \left ( \sum _ { q \in A _ { 2 } } \left | f _ { q } \left ( \sum _ { t \in E _ { q } } \lambda _ { t } x _ { j t } \right ) \right | ^ { p _ { \eta } } \right ) ^ { 1 / p _ { \eta } } \\ & \quad \leqslant \left ( \sum _ { t \in B } | \lambda _ { t } | ^ { p _ { \eta } } \right ) ^ { 1 / p _ { \eta } } + \theta 2 \left ( \sum _ { q \in A _ { 2 } } \left ( \sum _ { t \in E _ { q } } | \lambda _ { t } | ^ { p _ { k } } \right ) ^ { p _ { \eta } / p _ { k } } \right ) ^ { 1 / p _ { \eta } } \text{by } ( 7 ). \\ & \quad \text{The fact that $p_{c} \leqslant p_{\eta}$ and for each $1\leqslant t\leqslant m$ there exists at most two} \\ & \text{distinct $q \in A_{2}$ such that $t \in E_{q}$ implies that}$$
$$\begin{array} { c } \dots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \\ ( 9 ) & \left ( \sum _ { q \in A _ { 2 } } \left ( \sum _ { t \in E _ { q } } | \lambda _ { t } | ^ { p _ { c } } \right ) ^ { p _ { \eta } / p _ { c } } \right ) ^ { 1 / p _ { \eta } } \leqslant 2 ^ { 1 / p _ { c } } \left ( \sum _ { t = 1 } ^ { m } | \lambda _ { t } | ^ { p _ { c } } \right ) ^ { 1 / p _ { c } }. \\ \text{Combining relations } ( 8 ) \text{ and } ( 9 ) \text{ with } 0 < \theta \leqslant 1 / 4, \, we \, \text{ get that } | f ( \sum _ { t = 1 } ^ { m } \lambda _ { t } x _ { j _ { t } } ) | \\ \vdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \\ \end{array}$$
The fact that p ζ /lessorequalslant p η and for each 1 /lessorequalslant t /lessorequalslant m there exists at most two distinct q ∈ A 2 such that t ∈ E q implies that
Combining relations (8) and (9) with 0 < θ /lessorequalslant 1 4, we get that / | f ( ∑ m t =1 λ x t j t ) | is bounded by the desired value. Note that for convenience we have implicitly assumed that p η < ∞ , but the case that p η = ∞ would only require trivial modification.
Case 2: The functional f is of orderk ′′ with 1 /lessorequalslant k ′′ /lessorequalslant k ′ .
Set
$$t _ { 0 } \ & = \ \min \{ t \, \colon \, \text{ran} \, f \cap \text{ran} \, x _ { j _ { t } } \neq \varnothing \} \text{ and} \\ q _ { 0 } \ & = \ \min \{ q \, \colon \, \max \, \text{supp} \, f _ { q } \geq \min \, \text{supp} \, x _ { j _ { t _ { 0 } + 1 } } \}.$$
/negationslash
$$\colon \min \{ q \, \colon \max \, \text{supp} \, f _ { q } \geqslant \min \, \text{supp} \, x _ { j _ { t _ { 0 } + 1 } } \}.$$
We shall prove the following:
$$\begin{array} { c } \cdots \, \dots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \circ \\ ( 1 0 ) & \theta \left | \sum _ { q > q _ { 0 } } c _ { q } f _ { q } \left ( \sum _ { t = 1 } ^ { m } \lambda _ { t } x _ { j _ { t } } \right ) \right | < \theta \max \{ | \lambda _ { t } | \colon t > t _ { 0 } \}. \\ \intertext { Since } ( f _ { q } ) _ { q = 1 } ^ { d } \text{ is admissible, we have that } d \leqslant \max \text{supp} \, x _ { j _ { t _ { 0 } } }. \text{ Also, } ( f _ { q } ) _ { t } ^ { \prime } \\ \text{is very fact overflow and hence for } a \succ a \text{ was how that } \end{array}$$
∣ ∣ Since ( f q ) d q =1 is admissible, we have that d /lessorequalslant maxsupp x j t 0 . Also, ( f q ) d q =1 is very fast growing and hence for q > q 0 we have that
$$s ( f _ { q } ) \geqslant \max \text{supp} \, f _ { q _ { 0 } } \geqslant \min \text{supp} \, x _ { j _ { t _ { 0 } + 1 } }.$$
Moreover the functionals f q are of order strictly smaller than k ′ , therefore for q > q 0 and t > t 0 , (6) yields that | f q ( x j t ) | < / j 1 ( t 0 +1 maxsupp x j t 0 ) and since d /lessorequalslant maxsupp x j t 0 , by keeping t fixed, we obtain ∑ q>q 0 | c f q q ( x j t ) | < 1 /j t 0 +1 . Similarly, summing over the t which are strictly greater than t 0 , since m /lessorequalslant j t 0 +1 we obtain:
$$\sin c \, m \leqslant j _ { t _ { 0 } + 1 } \, \text{ we obtain} \colon \\ \left | \sum _ { q > q _ { 0 } } c _ { q } f _ { q } \left ( \sum _ { t = 1 } ^ { m } \lambda _ { t } x _ { j _ { t } } \right ) \right | \ & = \ \left | \sum _ { q > q _ { 0 } } c _ { q } f _ { q } \left ( \sum _ { t > t _ { 0 } } ^ { m } \lambda _ { t } x _ { j _ { t } } \right ) \right | \\ & \leqslant \ \sum _ { t > t _ { 0 } } | \lambda _ { t } | \sum _ { q > q _ { 0 } } | c _ { q } f _ { q } \left ( x _ { j _ { t } } \right ) | \\ & < \ \max _ { t > t _ { 0 } } | \lambda _ { t } | ( m / j _ { t _ { 0 } + 1 } ) \leqslant \max _ { t > t _ { 0 } } | \lambda _ { t } |. \\ \text{ Thus, (10) holds. We now observe the following} \colon$$
Thus, (10) holds. We now observe the following:
$$\Phi _ { 1 } \left ( \sum _ { q < q _ { 0 } } c _ { q } f _ { q } \left ( \sum _ { t = 1 } ^ { m } \lambda _ { t } x _ { j _ { t } } \right ) \right | = \theta \left | \sum _ { q < q _ { 0 } } c _ { q } f _ { q } \left ( \lambda _ { t _ { 0 } } x _ { j _ { t _ { 0 } } } \right ) \right | & \leqslant | \lambda _ { t _ { 0 } } |. \\ \text{Moreover, the inductive assumption yields that} \\ | \quad \int _ { q } ^ { m } \quad \| \dots \dots \dots$$
∣ ∣ Combining (10), (11) and (12) with 0 < θ /lessorequalslant 1 4 we get that / | f ( ∑ m t =1 λ x t j t ) | is bounded by the desired value.
$$\left | f _ { q _ { 0 } } \left ( \sum _ { t = 1 } ^ { m } \lambda _ { t } x _ { j _ { t } } \right ) \right | \leqslant 2 \left \| \left ( \lambda _ { t } \right ) _ { t = 1 } ^ { m } \right \| _ { \ell _ { p _ { \zeta } } }. \\ \text{Combining (10), (11) and (12) with 0 < \theta \leqslant 1 / 4 \text{ we get that } | f ( \sum \\ \text{is bounded by the desired value.}$$
The proof for the case in which ζ is finite is complete. Assume now that ζ = ξ 0 = ω and pass to a subsequence of ( x j ) j generating as a spreading model some sequence ( z j ) j . The previous case implies that ( z j ) j is 2-dominated by the unit vector basis of /lscript p k for all k < ξ 0 and hence, by taking a limit, it is also 2-dominated by the unit vector basis of /lscript p ξ 0 which yields the desired result. /square
The next result explains that the α <k -indices of a given block sequence determine the spreading models admitted by it.
Proposition 5.6. Let ( x j ) j be a normalized block sequence in X . Then ( x j ) j admits an /lscript p ζ spreading model, for some ζ ∈ [1 , ξ 0 ]. The following describes more precisely the spreading models admitted by ( x j ) j .
- a. Let 2 /lessorequalslant k < ξ 0 , then the following assertions are equivalent:
- (ii) There exists a subsequence of ( x j ) j that generates an /lscript p k spreading model, while no subsequence of ( x j ) j generates an /lscript p k ′ spreading model for 1 /lessorequalslant k ′ < k .
- (i) α <k { ( x j ) j } > 0 and α <k ′ { ( x j ) j } = 0 for 1 /lessorequalslant k ′ < k .
- b. The following are equivalent
- (i) α < 1 { ( x j ) j } > 0.
- (ii) There exists a subsequence of ( x j ) j that generates an /lscript p 1 spreading model.
- c. The following are also equivalent:
- (i) For every 1 /lessorequalslant k < ξ 0 we have that α <k { ( x j ) j } = 0.
- (ii) Every subsequence of ( x j ) j has a further subsequence generating an /lscript p ξ 0 spreading model.
Note that in the case ξ 0 is finite and ξ 0 = k 0 +1, then c.(i) is equivalent to α <k 0 { ( x j ) j } = 0.
Proof. We shall only prove a. as the others are proved similarly, using Proposition 4.2 and Lemmas 5.4, 5.5. Assume that the first assertion of a. holds. Note that on every subsequence of ( x j ) j the α <k -1 -index is zero, and hence, applying Lemma 5.5, it has a further subsequence which admits a spreading model dominated by the unit vector basis of /lscript p k . This in particular implies that no subsequence of ( x i ) i generates an /lscript p k ′ spreading model for 1 /lessorequalslant k ′ < k . Moreover, applying Lemma 5.4 we pass to a subsequence ( x j i ) , i of ( x j ) j , generating some spreading model dominating the usual vector basis of /lscript p k . Since α <k -1 { ( x j ) j } = 0, Lemma 5.5 implies that this spreading model has to be /lscript p k .
We assume now that the second assertion of a. holds. We first note that α <k { ( x j ) j } > 0. If this were not the case, then on every subsequence of ( x j ) j the α <k -index would be zero and hence, by Lemma 5.5, every spreading model admitted by it is dominated by the unit vector basis of /lscript p k +1 . This means that no subsequence of ( x j ) j can generate an /lscript p k spreading model, which is absurd. Therefore the natural number k 0 = min { k ∈ [1 , ξ 0 ) : α <k { ( x j ) j } > 0 } is well defined and k 0 /lessorequalslant k . We shall prove that k 0 = k and this will complete the proof.
Assume that k 0 < k and apply Lemma 5.4 to pass to a subsequence ( x j i ) i of ( x j ) j generating some spreading model which dominates the usual basis of /lscript p k 0 . If k 0 = 1 then by Proposition 4.2 we conclude that ( x j i ) i generates an /lscript p 1 spreading model, where 1 = k 0 < k , which is absurd. If 1 < k 0 , then α <k 0 -1 { ( x j i ) i } = 0 by and Lemma 5.5 we conclude that ( x j i ) i generates an /lscript p k 0 spreading model, which is absurd for the same reasons.
/square
Remark 5.7. It is not hard to check that for every 1 /lessorequalslant k < ξ 0 , the α <k -index of the basis ( e i ) i is zero and hence it only admits /lscript p ξ 0 as a spreading model.
## 6. Spreading models of infinite dimensional subspaces of X .
In the previous section we showed that every spreading model admitted by X must be /lscript p for some p ∈ F . In this section we show that, starting with a block sequence generating some spreading model, one may pass to a block sequence of it generating an other spreading model. We conclude that, in the case in which F is finite, the spreading models admitted by every infinite dimensional subspace of X are exactly the /lscript p for p ∈ F . In the case which F consists of an increasing sequence and its limit p ξ 0 , the spreading models admitted by every infinite dimensional subspace of X may either be the /lscript p for p ∈ F , or the /lscript p for p ∈ F \ { p ξ 0 } . We start with two lemmas that describe the kind of block vectors one has to consider when switching from one spreading model to an other.
Lemma 6.1. Let k be in [1 , ξ 0 ), x 1 < · · · < x K be a finite normalized block sequence in X that is 3-dominated by the unit vector basis of /lscript K p k , and set x = K -1 /p k ∑ K j =1 x j . If f is a functional of order-0 in W with s f ( ) = m then the following holds:
$$| f ( x ) | < \frac { K ^ { 1 / p _ { \xi _ { 0 } } } } { K ^ { 1 / p _ { k } } } + 2 \frac { m ^ { 1 / p _ { \xi _ { 0 } } } } { m ^ { 1 / p _ { k } } },$$
where in the case p ξ 0 = ∞ we set 1 /p ξ 0 = 0.
Proof. Let f = θ ∑ d q =1 (1 /m ) 1 /p ′ ξ 0 f q be a functional of order-0 with s f ( ) = m (recall that d /lessorequalslant m ). For convenience, we assume that p ξ 0 < ∞ , and the proof for the case ξ 0 = ∞ requires only trivial modifications. By Lemma 4.1, following the notation used there, applying H¨lder's inequality for the o pair ( p ξ 0 , p ξ ′ 0 ), we obtain the following:
$$\text{pair $(p_{\varepsilon_{0}},p_{\varepsilon_{0}})$, we obtain the following:} \\ | f ( x ) | \ \leqslant \ \ & K ^ { - 1 / p _ { k } } \left ( \left ( \sum _ { j \in B } | g _ { t } ( x _ { j } ) | ^ { p _ { \varepsilon _ { 0 } } } \right ) ^ { 1 / p _ { \varepsilon _ { 0 } } } \\ \quad \ + \theta \left ( \sum _ { q \in A _ { 2 } } ( 1 / m ) ^ { 1 / p _ { \varepsilon _ { 0 } } } \left | f _ { q } \left ( \sum _ { j \in E _ { q } } x _ { j } \right ) \right | \right ) \right ) \\ ( 1 4 ) \ \leqslant \ \ & K ^ { - 1 / p _ { k } } \left ( K ^ { 1 / p _ { \varepsilon _ { 0 } } } + 3 \theta ( 1 / m ) ^ { 1 / p _ { \varepsilon _ { 0 } } } \left ( \sum _ { q \in A _ { 2 } } ( \# E _ { q } ) ^ { 1 / p _ { k } } \right ) \right ).$$
Recall that # A 2 /lessorequalslant d /lessorequalslant m and the last part of Lemma 4.1 gives that ∑ q ∈ A 2 # E < q 2 K . Combing these two facts gives us
$$\sum _ { q \in A _ { 2 } } \# E _ { q } ^ { \prime } < 2 K \,. \, \text{ Combine two facts gives us} \\ ( 1 / m ) ^ { 1 / p _ { \xi _ { 0 } } ^ { \prime } } \left ( \sum _ { q \in A _ { 2 } } ( \# E _ { q } ) ^ { 1 / p _ { k } } \right ) & \ \leq \ \ ( 1 / m ) ^ { 1 / p _ { \xi _ { 0 } } ^ { \prime } } m ^ { 1 / p _ { k } } \left ( \sum _ { q \in A _ { 2 } } \# E _ { q } \right ) ^ { 1 / p _ { k } } \\ & < \ \frac { m ^ { 1 / p _ { k } ^ { \prime } } } { m ^ { 1 / p _ { \xi _ { 0 } } } } 2 ^ { 1 / p _ { k } } K ^ { 1 / p _ { k } } \\ ( 1 5 ) & = \ \ 2 ^ { 1 / p _ { k } } \frac { m ^ { 1 / p _ { \xi _ { 0 } } } } { m ^ { 1 / p _ { k } } } K ^ { 1 / p _ { k } }. \\ \text{By combining relations (14) and (15) we achieve the desired upper bound.}$$
By combining relations (14) and (15) we achieve the desired upper bound. /square
Lemma 6.2. Let ( x j ) j be a normalized block sequence in X and 2 /lessorequalslant k +1 < ξ 0 with α <k { ( x j ) j } = 0. Then there exists a subsequence ( x j i ) i of ( x j ) j such that for every K /lessorequalslant j i 1 < · · · < j i K and every f ∈ W of order at most k with s f ( ) = m , we have that if x = K -1 /p k +1 ∑ K t =1 x j i t then
$$| f ( x ) | < \frac { 3 + K ^ { 1 / p _ { \xi _ { 0 } } } } { K ^ { 1 / p _ { k + 1 } } } + 2 \frac { m ^ { 1 / p _ { \xi _ { 0 } } } } { m ^ { 1 / p _ { k + 1 } } },$$
where in the case p ξ 0 = ∞ we set 1 /p ξ 0 = 0.
Proof. By Lemma 5.5 we may assume for every K /lessorequalslant j 1 < · · · < j K that ( x j i ) K i =1 is 3-dominated by the unit vector basis of /lscript K p k +1 . Using Proposition 5.3 we pass to a subsequence, again denoted by ( x j ) j such that for any j /greaterorequalslant j 0 /greaterorequalslant 2, for any f ∈ W of order strictly smaller than k with s f ( ) /greaterorequalslant minsupp x j 0 we have that
$$\left ( 1 7 \right ) & & \left | f ( x _ { j } ) \right | < \left ( j _ { 0 } \max \text{supp} \, x _ { j _ { 0 } - 1 } \right ) ^ { - 1 }. \\ & & \quad \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cd.$$
Let K /lessorequalslant j 1 < · · · < j K , x = K -1 /p k +1 ∑ K i =1 x j i and f = θ ∑ d q =1 c f q q be a functional of order at most k . Using a finite induction on 0 /lessorequalslant k ′ /lessorequalslant k , we shall prove that for every f = θ ∑ d q =1 c f q q of order at most k ′ there is i ∈ N such that the following holds:
$$& ( 1 8 ) & \quad | f ( x ) | < \left ( \frac { 1 - \theta ^ { i } } { 1 - \theta } \right ) \frac { 2 } { K ^ { 1 / p _ { k + 1 } } } + \frac { K ^ { 1 / p _ { 0 } } } { K ^ { 1 / p _ { k + 1 } } } + 2 \frac { m ^ { 1 / p _ { \xi _ { 0 } } } } { m ^ { 1 / p _ { k + 1 } } }. \\ & \intertext { T h o o h o i n o m i t i o n w i t h o o } & \times a < 1 / A \, \text{allowline} \, \text{inline to the odd odd odd} }$$
The above in conjunction with 0 < θ /lessorequalslant 1 4 clearly implies the desired result. / If a functional f is of order-0. Then, by Lemma 6.1 for i = 1 we have that (18) holds. Assume that f = θ ∑ d q =1 c f q q is of orderk ′ with 0 < k ′ /lessorequalslant k and that (18) holds for every functional with order strictly smaller than k ′ . Set
/negationslash
$$t _ { 0 } \ & = \ \min \{ t \, \colon \, \text{ran} \, f \cap \text{ran} \, x _ { j _ { t } } \neq \varnothing \} \text{ and} \\ q _ { 0 } \ & = \ \min \{ q \, \colon \, \max \, \text{supp} \, f _ { q } \geq \min \, \text{supp} \, x _ { j _ { t _ { 0 } + 1 } } \}.$$
$$\colon \min \{ q \, \colon \max \, \text{supp} \, f _ { q } \geqslant \min \, \text{supp} \, x _ { j _ { t _ { 0 } + 1 } } \}.$$
The same argument used to obtain (10) and (11) in the proof of Lemma 5.5 gives us the following:
/negationslash
$$\circ \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \colon \\ ( & \int _ { q \neq q _ { 0 } } c _ { q } f _ { q } ( x ) \Big | < 2 / K ^ { 1 / p _ { k + 1 } }. \\ & \text{By the inductive assumption there exists $i\in\mathbb{N}$ such that:} \\ & / / / / / \int _ { q \neq q _ { 0 } } \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \ddots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots.$$
$$& ( 2 0 ) \quad \theta | f _ { q _ { 0 } } ( x ) | < \theta \left ( \left ( \frac { 1 - \theta ^ { i } } { 1 - \theta } \right ) \frac { 2 } { K ^ { 1 / p _ { k + 1 } } } + \frac { K ^ { 1 / p _ { 0 } } } { K ^ { 1 / p _ { k + 1 } } } + 2 \frac { s ( f _ { q _ { 0 } } ) ^ { 1 / p _ { \xi _ { 0 } } } } { s ( f _ { q _ { 0 } } ) ^ { 1 / p _ { k + 1 } } } \right ).$$
∣ ∣ By the inductive assumption there exists i ∈ N such that:
By the definition of size for functionals which are not of order-0 we have that s f ( q 0 ) /greaterorequalslant s f ( ) and hence combining (19) and (20) we conclude that:
$$\cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdot \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \colon \\ | f ( x ) | < \left ( \frac { 1 - \theta ^ { i + 1 } } { 1 - \theta } \right ) \frac { 2 } { K ^ { 1 / p _ { k + 1 } } } + \theta \left ( \frac { K ^ { 1 / p _ { 0 } } } { K ^ { 1 / p _ { k + 1 } } } + 2 \frac { s ( f ) ^ { 1 / p _ { \xi _ { 0 } } } } { s ( f ) ^ { 1 / p _ { k + 1 } } } + 2 \frac { s ( f ) ^ { 1 / p _ { \xi _ { 0 } } } } { s ( f ) ^ { 1 / p _ { k + 1 } } } \right ).$$
The next proposition allows us to pass from a block sequence admitting an /lscript p ξ 0 spreading model to a further block admitting /lscript p 1 spreading model and from block sequence admitting an /lscript p k spreading model to a further block admitting an /lscript p k +1 spreading model. In the case that ξ 0 < ω , we use this to show that the spreading models in every subspace are exactly /lscript p for p ∈ { p , p 1 2 , · · · , p ξ 0 -1 , p ξ 0 } . In the case that ξ 0 = ω we require an additional argument to show that we have /lscript p k spreading model for every k < ω since we are not able to show that every block subspace admits an /lscript p ξ 0 spreading model.
Proposition 6.3. Let ( x j ) j be a normalized block sequence in X .
- (i) If ( x j ) j generates an /lscript p ξ 0 spreading model, then there exists a further normalized block sequence ( y j ) j of ( x j ) j that generates an /lscript p 1 spreading model.
- (ii) If 1 /lessorequalslant k < ξ 0 and ( x j ) j generates an /lscript p k spreading model, then there exists a further normalized block sequence ( y j ) j of ( x j ) j that generates an /lscript p k +1 spreading model.
Proof. Let ( x j ) j be a normalized block sequence in X , generating an /lscript p ζ spreading model, for some 1 /lessorequalslant ζ /lessorequalslant ξ 0 . Note that by Proposition 4.2 and Lemma 5.5 we may assume that for every K /lessorequalslant j 1 < · · · < j K we have that ‖ x j 1 + · · · + x j K ‖ /lessorequalslant 3 · K 1 /p ζ . We distinguish three cases concerning ζ , namely ζ = ξ , ζ 0 = 1, and 2 /lessorequalslant ζ < ξ 0 . We shall only consider the first two cases, as the last one is proved in an identical manner as the case ζ = 1 and uses Lemma 6.2 instead of Lemma 6.1.
Case 1: ζ = ξ 0 . For every j ∈ N choose f j ∈ W with f j ( x j ) = 1 and ran f j ⊂ ran x j . Choose an increasing sequence of finite subsets of the natural numbers ( E j ) j with # E j /lessorequalslant min E j and lim j # E j = ∞ . For j ∈ N define
y ′ j = (# E j ) -1 /p ξ 0 ∑ i ∈ E j x i , y j = ‖ y ′ j ‖ -1 y ′ j and g j = θ ∑ i ∈ E j (# E j ) -1 /p ′ ξ 0 f i . Then we have the following:
- (b) The functional g j is of order-0 with s g ( j ) = # E j for all j ∈ N .
- (a) The sequence ( y j ) j is a normalized block sequence of ( x j ) j and for every j ∈ N we have that g j ( y j ) /greaterorequalslant θ/ 3.
Note that lim j s g ( j ) = ∞ and therefore, passing to a subsequence, we may assume that ( g j ) j is a very fast growing sequence of functionals of order-0. We conclude that α < 1 { ( x j ) } > 0 and by Proposition 5.6 we have that ( y j ) j is the desired sequence.
Case 2: ζ = 1. By Proposition 5.6 we have that α < 1 { ( x j ) j } > 0 and hence, by passing to a subsequence, there exists ε > 0 and a very fast growing sequence ( f j ) j of order-0 functionals such that ran f j ⊂ ran x j and f j ( x j ) > ε for all j ∈ N . Choose an increasing sequence of finite subsets of the natural numbers ( E j ) j with min E j /lessorequalslant # E j and lim j # E j = ∞ . For j ∈ N define y ′ j = (# E j ) -1 /p 1 ∑ i ∈ E j x i , y j = ‖ y ′ j ‖ -1 y ′ j and g j = θ ∑ i ∈ E j (# E j ) -1 /p ′ 1 f j (if p 1 = 1 take g j = θ i ∈ E j f j instead). Then we have the following:
- (b') The functional g j is of order-1 with s g ( j ) /greaterorequalslant max { s f ( i ) : i ∈ E j } for all j ∈ N .
- ∑ (a') The sequence ( y j ) j is a normalized block sequence of ( x j ) j and for every j ∈ N we have that g j ( y j ) /greaterorequalslant εθ/ 3.
Once more, lim j s g ( j ) = ∞ and as before we conclude that α < 2 { ( x j ) } > 0. By Proposition 5.6 it remains to observe that α < 1 { ( y j ) } j = 0, which is an easy consequence of the definition of the y j 's and Lemma 6.1. /square
Remark 6.4. The proof of Proposition 6.3 implies that the space X does not admit an /lscript S 2 p ζ spreading model for any 1 /lessorequalslant ζ /lessorequalslant ξ 0 . For the definition of an /lscript S k p spreading model see [ABM, Definition 1.1].
Corollary 6.5. The space X is reflexive.
Proof. Proposition 6.3 implies that neither c 0 nor /lscript 1 embed into X . By James' well known theorem for spaces with an unconditional basis we conclude that X is reflexive. /square
Remark 6.6. If ( z j ) j is a spreading model generated by a non-norm convergent (not necesssarily Schauder basic) sequence in X , then [AKT, Remark 5, page 581] the reflexivity of the space and Proposition 5.6 imply that, although the sequence ( z j ) j need not be a Schauder basis for Z = 〈{ z j : j ∈ N }〉 , the space Z must be isomorphic to /lscript p , for some p ∈ F .
Lemma 6.7. Let 1 /lessorequalslant k < ξ , K 0 ∈ N and ( x j ) j be a sequence in X generating an /lscript p k spreading model. Then for every j 0 ∈ N there exists a normalized vector x ∈ span( x j ) j /greaterorequalslant j 0 and a functional f of order-0 with s f ( ) = K such that
$$f ( x ) \geqslant \frac { \theta } { 3 } K ^ { 1 / p _ { \xi _ { 0 } } - 1 / p _ { k } },$$
where in the case p ξ 0 = ∞ we set 1 /p ξ 0 = 0.
Proof. We may clearly assume that ( x j ) j is normalized. Proposition 5.6 in conjunction with 5.5 imply that we may choose j 0 /lessorequalslant j 1 < · · · < j K such that if y = K -1 /p k ∑ K i =1 x j i then ‖ y ‖ /lessorequalslant 3. Choose f , . . . , f 1 K with ran f i ⊂ ran x j i and f i ( x j i ) = 1 for i = 1 , . . . , K and define f = θ ∑ K i =1 (1 /K ) -1 /p ′ ξ 0 f i and x = y/ y ‖ ‖ . /square
Theorem 6.8. Let F = { p ζ : 1 /lessorequalslant ζ /lessorequalslant ξ 0 } and let Y be an infinite dimensional subspace of X . Then there exists a dense subset G of F such that the spreading models admitted by Y are exactly the /lscript p , for p ∈ G . In particular, /lscript p is finitely block represented in every block subspace of X for every p ∈ F .
Proof. Let Y be a block subspace of X . We observe the following:
- (i) Every spreading model admitted by Y is equivalent to the unit vector basis of /lscript p ζ for some 1 /lessorequalslant ζ /lessorequalslant ξ 0 . In particular, there exists 1 /lessorequalslant ζ 0 /lessorequalslant ξ 0 such that Y admits an /lscript p ζ 0 spreading model.
- (ii) If 1 /lessorequalslant k < ξ 0 and Y admits an /lscript p k spreading model and, then Y also admits an /lscript p k +1 spreading model.
- (iii) If Y admits an /lscript p ξ 0 spreading model then Y also admits an /lscript p 1 spreading model.
- (iv) There exists 1 /lessorequalslant k 0 < ξ 0 such that Y admits an /lscript p k 0 spreading model.
The statement (i) follows from Proposition 5.6. Statements (ii) and (iii) follow from Proposition 6.3, while (iv) follows from the first and the third. We now distinguish two cases regarding whether ξ 0 is finite or not.
Case 1: If ξ 0 is finite, statement (i), statement (ii), and a finite inductive argument yield that Y admits an /lscript p ξ 0 spreading model. By (iii) we have that Y admits an /lscript p 1 spreading model. Once more, by a finite induction we obtain that G = F is the desired set.
Case 2: If ξ 0 = ω we shall prove that for every 1 /lessorequalslant k < ξ 0 , Y admits an /lscript p k spreading model. This in particular implies that G = F or G = F \ { p ξ 0 } is the desired set. By (ii) it is sufficient to show that Y admits an /lscript p 1 spreading model. By Proposition 5.6 it is enough to find a normalized block sequence ( x j ) j in Y and a very fast growing sequence of functionals ( f j ) j of order-0 with f j ( x j ) > θ/ 4, i.e. α < 1 { ( x j ) j } > 0.
Choose a normalized vector x 1 in Y and a functional f ∈ W with f ( x 1 ) = 1 and set f 1 = θf . Then f 1 is of order-0 with s f ( ) = 1 and f 1 ( x 1 ) > θ/ 4. Assume that we have chosen normalized vectors x 1 < · · · < x j and a very fast growing sequence of functionals f , . . . , f 1 j of order-0 with f i ( x i ) > θ/ 4 for i = 1 , . . . , j . By (iv), there exists 1 /lessorequalslant k 0 < ξ 0 such that Y admits an /lscript p k 0 spreading model. Fix K > maxsupp f j and choose k 0 /lessorequalslant k < ξ 0 such that K 1 /p ξ 0 -1 /p k > / 3 4 (recall that lim k p k = p ξ 0 ). By (ii) we may choose a sequence ( y i ) i in Y generating an /lscript p k spreading model. Choose i 0 ∈ N with
minsupp y i 0 /greaterorequalslant (maxsupp x j ) 2 and apply Lemma 6.7 to find the desired pair x j +1 , f j +1 . /square
Remark 6.9. In the case that F is finite, then clearly the spreading models admitted by every block subspace of X are exactly the /lscript p , for p ∈ F . In the case that F consists of an increasing sequence and its limit p ξ 0 , then it is easily checked that exactly one of the following holds:
- (i) The spreading models admitted by every block subspace of X are exactly the /lscript p , for p ∈ F .
We were unable to determine which one of the above holds, in either case however on some subspace Y of X , the set of spreading models admitted by every further subspace of Y is stabilized.
- (ii) There exists a block subspace Y of X , such that the spreading models admitted by every further block subspace of Y are exactly the /lscript p , for p ∈ F \ { p ξ 0 } .
## 7. The set of Krivine p 's of the space X .
In this section we prove that for any p / F ∈ = { p ζ : 1 /lessorequalslant ζ /lessorequalslant ξ 0 } , /lscript p is not finitely block represented in the space X . We conclude that the set of p 's that are finitely block represented in every block subspace of X is exactly the set F , which is not connected.
We begin with the following Lemma, whose proof we omit as it follows from the same argument as the proof of Lemma 6.1.
Lemma 7.1. Let p ∈ [ p , p 1 ξ 0 ] \ F . Suppose ε > 0 and ( x j ) N j =1 is a finite block sequence in X which is (1 + ε )-equivalent to the unit vector basis of /lscript N p . If 1 /lessorequalslant ζ /lessorequalslant ξ 0 is such that p < p ζ and f is a functional of orderζ , then we have the following estimate:
$$\text{$we have the following estimate:} \\ ( 2 2 ) & \left | f \left ( \frac { 1 } { N ^ { 1 / p } } \sum _ { j = 1 } ^ { N } x _ { j } \right ) \right | < ( 1 + \varepsilon ) \left ( \frac { N ^ { 1 / p _ { \zeta } } } { N ^ { 1 / p } } + 2 \theta \right ). \\ \text{The next lemma follows directly from the above lemma.} \\ \text{Lemma} \, 7. 2. \, Suppose that ( x _ { j } ) _ { i = 1 } ^ { N } \text{ is a finite block sequence in $X$} \, \text{th}.$$
Lemma 7.2. Suppose that ( x j ) N j =1 is a finite block sequence in X that is (1+ )-equivalent to the unit vector basis of ε /lscript N p , 1 /lessorequalslant k < ξ 0 satisfies p < p k +1 and N satifies
$$N ^ { 1 / p _ { k + 1 } - 1 / p } + 2 \theta < ( 1 + \varepsilon ) ^ { - 2 }.$$
If f ∈ W satisfies f ( N -1 /p ∑ N j =1 x j ) /greaterorequalslant 1 (1 + / ε ) then f has non-zero order less than or equal to k .
We are now ready to prove the second main theorem.
Theorem 7.3. For all p ∈ [1 , ∞ \ ] F there exists K ∈ N and ε > 0 such that no block sequence ( x j ) K j =1 in X is (1 + ε )-equivalent to the unit vector basis of /lscript K p .
Proof. Let p ∈ [1 , ∞ \ ] F . If p / ∈ [ p , p 1 ξ 0 ], then the result clearly follows from Proposition 4.2. Otherwise, we have that p ∈ [ p , p 1 ξ 0 ] \ F . Find k ∈ N so that p k < p < p k +1 . Find N,M ∈ N and ε > 0 as follows:
$$N ^ { 1 / p } \ > \ 2 + \theta ( N - 2 ) ^ {$$
Choose N ∈ N such that
$$\begin{matrix} \cdot & \cdot & \cdot & \cdot &$$
Now that N is fixed, we choose ε > 0 such that:
$$\frac { 1 } { 1 + \varepsilon } N ^ { 1 / p } \$$
$$\begin{array} {$$
$$\frac { N ^ { 1 / p _ { k + 1 } } } { N ^ { 1$$
We set
$$\Theta & = \frac { 1 } { 1 + \varepsilon } N ^ { 1 / p } - ( 1 + \varepsilon ) ( N - 1 ) ^ { 1 / p }.$$
Notice (26) implies that Θ > 0. Finally, let M ∈ N so that
$$M ^ { 1 / p _ { k } } \Theta > ( 1 + \varepsilon ) M ^ { 1 / p }.$$
Let K = ( N -1) M + 1 and consider the following normalized block sequence which, towards a contradiction, we assume is (1 + ε )-equivalent to the unit vector basis of /lscript K p and that M < minsupp x 1 .
$$x _ { 1 } & < x _ { 2 } ^ { 1 } < x _ { 3 } ^ { 1 } < \dots < x _ { N } ^ { 1 } < x _ { 2 } ^ { 2 } < x _ { 3 } ^ { 2 } < \dots < x _ { N } ^ { M - 1 } < x _ { 2 } ^ { M } < \dots < x _ { N } ^ { M }. \\ \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \hat { \ \hat { \hat { \ \hat { \hat { \hat { \ \hat { \ \hat { \ \hat { \ \hat { \ \hat { \ \hat { \ \hat { \ \hat { \ \hat { \ \hat { \ \hat { \ \hat { \ \hat { \ \hat { \ \hat { \ \hat { \ \ \hat { \ \hat { \ \hat { \ \hat { \ \hat { \ \hat { \ \hat { \ \ \hat { \ \hat { \ \hat { \ \hat { \ \hat { \ \hat { \ \ \hat { \ \hat { \ \hat { \ \ \hat { \ \ \hat { \ \ \ \ \ \ \ \ } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } { } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } } }.. \quad \, \, \,$$
(i.e. x m j < x m i for i < j and x m N < x m +1 2 ). Let us mark the following, which is obviously true.
- (a) For each m with 1 /lessorequalslant m /lessorequalslant M the block sequence x 1 < x m 2 < · · · < x m N is (1 + ε )-equivalent to the unit vector basis of /lscript N p .
Fix m with 1 /lessorequalslant m /lessorequalslant M . For notational reasons we set x m 1 = x 1 . Find g m ∈ W with g m ( ∑ N i =1 x m i ) /greaterorequalslant N 1 /p (1+ ) ε . By Lemma 7.2 and (27) we conclude that g m has non-zero order less than or equal to k . Let
$$g _ { m } = \theta \sum _ { q = 1 } ^ { d _ { m } } c _ { m, q } f _ { m, q }. \\ \tt n o \sin \, \text{accordin} \, \text{to} \, \text{Rem}$$
be the functionals decomposition according to Remark 3.3, i.e.:
- (b) the coefficients ( c m,q ) d m q =1 are in the unit ball of /lscript p ′ k for m = 1 , . . . , M and
- (c) the sequence ( f m,q ) d m q =1 is an admissible and very fast growing sequence of functionals, each one of which has order strictly smaller than k .
Define q m = min { q : minsupp x m 2 /lessorequalslant maxsupp f m,q }
We will prove the following three claims:
- (i) g m ( x 1 ) > Θ, g m ( x m N ) > Θ and d m /lessorequalslant maxsupp x 1 .
- (iii) (min supp x m 2 ) 2 < minsupp f m,q m +1 and minsupp x m 2 < s f ( m,q m +1 ).
- (ii) The number q m exists, max supp f m,q m < minsupp x m N and q m < d m .
Item (i): Using that g m ( ∑ N i =1 x m i ) /greaterorequalslant N 1 /p (1+ ) ε , (a) and (28) we have
$$\geqslant \frac { 1 } { 1 + \varepsilon } N ^ { 1 / p } - ( 1 + \varepsilon ) ( N - 1 ) ^ { 1 / 2 }$$
$$g _ { m } ( x _ { 1 } ) & = g _ { m } \left ( \sum _ { j = 1 } ^ { N } x _ { j } ^ { m } \right ) - g _ { m } \left ( \sum _ { j = 2 } ^ { N } x _ { j } ^ { m } \right ) \\ & \geqslant \frac { 1 } { 1 + \varepsilon } N ^ { 1 / p } - ( 1 + \varepsilon ) ( N - 1 ) ^ { 1 / p } = \Theta > 0.$$
The same argument works to show that g m ( x m N ) > Θ. If d m > maxsupp x 1 then max supp x 1 < minsupp g m which implies that g m ( x 1 ) = 0. This contradiction tells us that d m /lessorequalslant maxsupp x 1 .
∑ If max supp f m,q m /greaterorequalslant minsupp x m N then we have:
Item (ii): If q m did not exist then min supp x m 2 > maxsupp f m,q for all q and so g m ( ∑ N j =1 x m j ) = g m ( x 1 ) /lessorequalslant 1. On the other hand we clearly have g m ( N j =1 x m j ) > 2 and so q m exists.
$$\Phi _ { m } \left ( \sum _ { j = 1 } ^ { N } x _ { j } ^ { m } \right ) & = g _ { m } ( x _ { 1 } ) + \theta c _ { m, q _ { m } } f _ { m, q _ { m } } \left ( \sum _ { j = 2 } ^ { N - 1 } x _ { j } ^ { m } \right ) + g _ { m } ( x _ { N } ^ { m } ) \\ & \leqslant 2 + ( 1 + \varepsilon ) \theta ( N - 2 ) ^ { 1 / p } < \frac { N ^ { 1 / p } } { ( 1 + \varepsilon ) }. \\ \Phi _ { m } \left ( \sum _ { j = 1 } ^ { N } x _ { j } ^ { m } \right ) & = \Phi _ { m } ( x _ { m } ^ { m } ) \right ) \Phi _ { m } \left ( \sum _ { j = 2 } ^ { N - 1 } x _ { j } ^ { m } \right ) \Phi _ { m } \left ( \sum _ { j = 2 } ^ { N - 1 } x _ { j } ^ { m } \right ) \Phi _ { m } \left ( \sum _ { j = 2 } ^ { N - 1 } x _ { j } ^ { m } \right ) \Phi _ { m } \left ( \sum _ { j = 2 } ^ { N - 1 } x _ { j } ^ { m } \right ) \Phi _ { m } \left ( \sum _ { j = 2 } ^ { N - 1 } x _ { j } ^ { m } \right ) \Phi _ { m } \left ( \sum _ { j = 2 } ^ { N - 1 } x _ { j } ^ { m } \right ) \Phi _ { m } \left ( \sum _ { j = 3 } ^ { N - 1 } x _ { j } ^ { m } \right ) \Phi _ { m } \left ( \sum _ { j = 3 } ^ { N - 1 } x _ { j } ^ { m } \right ) \Phi _ { m } \left ( \sum _ { j = 3 } ^ { N - 1 } x _ { j } ^ { m } \right ) \Phi _ { m } \left ( \sum _ { j = 3 } ^ { N - 1 } x _ { j } ^ { m } \right ) \Phi _ { m } \left ( \sum _ { j = 3 } ^ { N - 1 } x _ { j } ^ { m } \right ) \Phi _ { m } \left ( \sum _ { j = 3 } ^ { N - 1 } x _ { j } ^ { m } \right )$$
The last inequality uses (25). This contradicts that fact that g m ( ∑ N j =1 x m j ) /greaterorequalslant N 1 /p (1+ ) ε .
Item (iii): By definition of q m and the fact that ( f m,q ) d m q =1 is very fast growing
Using item (i) we have g m ( x m N ) > Θ. This fact combined with the fact that maxsupp f m,q m < minsupp x m N gives us q m < d m .
$$( \min \text{supp} \, x _ { 2 } ^ { m } ) ^ { 2 } \leqslant ( \max \text{supp} \, f _ { m, q _ { m } } ) ^ { 2 } < \min \text{supp} \, f _ { m, q _ { m } + 1 } \text{ and}$$
$$\min \text{supp} \, x _ { 2 } ^ { m } < s ( f _ { m, q _ { m } + 1 } ).$$
This proves (iii).
Note that q m +1 /lessorequalslant d m by item (ii). Define
$$f _ { m } \coloneqq g _ { m } \Big | _ { \text{ran} x _ { N } ^ { m } } = \theta \sum _ { q = q _ { m } + 1 } ^ { d _ { m } } c _ { m, q } f _ { m, q } \Big | _ { \text{ran} x _ { N } ^ { m } }.$$
We claim that
$$& \frac { 1 } { M ^ { 1 / p _ { k } ^ { \prime } } } \sum _ { m = 1 } ^ { M } f _ { m } \in W. \\ & \text{We first assume that } ( 3 2 ) \text{ holds and finish the proof of our}$$
We first assume that (32) holds and finish the proof of our theorem as follows:
$$& M ^ { 1 / p } ( 1 + \varepsilon ) \geqslant \frac { 1 } { M ^ { 1 / p _ { k } ^ { \prime } } } \sum _ { m = 1 } ^ { M } f _ { m } \left ( \sum _ { m = 1 } ^ { M } x _ { N } ^ { m } \right ) = \frac { 1 } { M ^ { 1 / p _ { k } ^ { \prime } } } \sum _ { m = 1 } ^ { M } f _ { m } ( x _ { N } ^ { m } ) \geqslant M ^ { 1 / p _ { k } } \Theta. \\ & \text{This contradicts (29).}$$
This contradicts (29).
All that remains to prove is (32). Note that
$$\frac { 1 } { M ^ { 1 / p _ { k } ^ { \prime } } } \sum _ { m = 1 } ^ { M } f _ { m } = \theta \sum _ { m = 1 } ^ { M } \sum _ { q = q _ { m } + 1 } ^ { d _ { m } } \left ( c _ { m, q } / M ^ { 1 / p _ { k } ^ { \prime } } \right ) f _ { m, q } | _ { \text{ran} } x _ { N } ^ { m }. \\ \text{Using } ( b ) \text{ we obtain that } \sum _ { m = 1 } ^ { M } \sum _ { q = q _ { m } + 1 } ^ { d _ { m } } ( c _ { m, q } / M ^ { 1 / p _ { k } ^ { \prime } } ) ^ { p _ { k } ^ { \prime } } \leqslant 1 \text{ and therefore}$$
Using (b) we obtain that ∑ M m =1 ∑ d m q = q m +1 ( c m,q /M 1 /p ′ k ) p ′ k /lessorequalslant 1 and therefore it suffices to show that (( f m,q ) d m q = q m +1 1 ) /lessorequalslant m M /lessorequalslant is an admissible and very fast growing sequence of functionals, each one of which has order strictly smaller than k , which will imply that f is a funtional in W of orderk . First we check admissibility:
$$\sum _ { m = 1 } ^ { M } d _ { m } & \leqslant M \max \text{supp} x _ { 1 } \\ & \leqslant \min \text{supp} x _ { 2 } ^ { 1 } \cdot \min \text{supp} x _ { 2 } ^ { 1 } \\ & \leqslant \min \text{supp} f _ { 1, q _ { 1 } + 1 }. \\ \tau _ { 1, q _ { 1 } + 1 }. \tau _ { 2, q _ { 2 } + 1 }. \tau _ { 3, q _ { 2 } + 1 }. \tau _ { 4, q _ { 3 } + 1 }.$$
The first inequality follows from item (i), the second from that fact that M < maxsupp x 1 < minsupp x 1 2 and the third comes from item (iii) (for m = 1).
Note that by (b) the functionals under consideration have order strictly smaller than k and for each m with 1 /lessorequalslant m /lessorequalslant M the collection ( f m,q ) d m q = q m +1 is very fast growing. At last, it suffices to show for each m ∈ N with 2 /lessorequalslant m /lessorequalslant M that
$$( \max \text{supp} \, f _ { m - 1, d _ { m - 1 } } ) ^ { 2 } < \min \text{supp} \, f _ { m, q _ { m } + 1 } \text{ and }$$
$$\max \text{supp} \, f _ { m - 1, d _ { m - 1 } } < s ( f _ { m, q _ { m } + 1 } ).$$
$$& \vdots _ { m - 1, d _ { m - 1 } } ) ^ { 2 } < \min \text{supp} \, f _ { m, q _ { m } + 1 } \text{ and} \\ & \text{supp} \, f _ { m - 1, d _ { m - 1 } } < s ( f _ { m, q _ { m } + 1 } ). \\ & \quad. \quad \dots \quad.$$
This, however, follows from item (iii) since
$$\max \text{supp} \, f _ { m - 1, d _ { m - 1 } } < \min \text{supp} \, x _ { 2 } ^ { m }.$$
This proves the claim and finishes the proof of the theorem. /square
We are interested in three problems related to the present work.
Problem 1. Let 1 < p 1 < p 2 < ∞ . Is the space X p ,p 1 2 constructed in this paper super-reflexive?
Problem 2. Let 1 /lessorequalslant p 1 < p 2 /lessorequalslant ∞ . Does there exist a space so that in every block subspace the Krivine set is [ p , p 1 2 ]? More generally, which types of closed sets can be hereditary Krivine sets?
Problem 3. Let F ⊂ [2 , ∞ ) be finite. Does there exist a Banach space X such that for every infinite dimensional subspace Y of X /lscript , p is finitely represented in Y if and only if p ∈ { 2 } ∪ F ? In particular, does our construction satisfy this?
## References
- [ABM] S.A. Argyros, K. Beanland, and P. Motakis, The strictly singular operators in Tsirelson like reflexive spaces , to appear in Illinois J. Math.
- [AKT] S. A. Argyros, V. Kanellopoulos and K. Tyros, Higher Order Spreading Models , Fund. Math. 221 (2013), 23-68.
- [AM1] S. A. Argyros and P. Motakis, A reflexive hereditarily indecomposable space with the hereditary invariant subspace property , Proc. London Math. Soc. 108 , no. 6 (2014), 1381-1416.
- [AM2] S. A. Argyros and P. Motakis, A hereditarily indecomposable Banach space with rich spreading model structure , to appear in Israel J. Math.
- [AOST] G. Androulakis, E. Odell, Th. Schlumprecht, and N. Tomczak Jaegermann, On the structure of the spreading models of a Banach space , Canad. J. Math. 57 (2005), no. 4, 673-707.
- [FJ] T. Figiel and W. B. Johnson, A uniformly convex Banach space which contains no /lscript p , Compositio Math. 29 (1974), 179-190.
- [GM] W.T. Gowers and B. Maurey, The unconditional basic sequence problem , Journal A.M.S. 6 (1993), 851-874.
- [HT] P. Habala and N. Tomczak-Jaegermann, Finite representability of /lscript p in quotients of Banach spaces , Positivity 5 (2001), 75-94.
- [K] J. L. Krivine, Sous espaces de dimension finie des espaces de Banach r´ticul´s e e , Ann. of Math. 104 (1976), 1-29.
- [M] B. Maurey, A remark about distortion , Operator Theory: Advances and Applications 77 (1995), 131-142.
- [O] E. Odell, Panel discussion: Future directions , Workshop on Geometry of Banach spaces and infinite dimensional Ramsey theory, Fields Institute, The University of Toronto, November 13 2002.
- [OS1] E. Odell and Th. Schlumprecht, On the richness of the set of p 's in Krivine's theorem , Operator Theory, Advances and Applications 77 (1995), 177-198.
- [OS2] E. Odell and Th. Schlumprecht, A Banach space block finitely universal for monotone bases , Trans. Amer. Math. Soc. 352 , no. 4 (2000), 1859-1888.
- [R] H.P. Rosenthal, On a theorem of Krivine concerning block finite representability of /lscript p in general Banach spaces , J. Func. Anal. 28 (1978), 197-225.
- [S] B. Sari, On Banach spaces with few spreading models , Proceedings Amer. Math. Soc. 134 , no. 5 (2005), 1339-1345.
- [T] B. S. Tsirelson, Not every Banach space contains /lscript p or c 0 , Functional Anal. Appl. 8 (1974), 138-141.
Department of Mathematics, Washington and Lee University, Lexington, VA 24450
E-mail address : [email protected]
Department of Mathematics and Computer Science, Saint Louis University, St Louis, MO 63103
E-mail address : [email protected]
National Technical University of Athens, Faculty of Applied Sciences, Department of Mathematics, Zografou Campus, 157 80, Athens, Greece
E-mail address
: [email protected] | 10.1016/j.aim.2015.05.005 | [
"Kevin Beanland",
"Daniel Freeman",
"Pavlos Motakis"
] | 2014-08-01T18:54:39+00:00 | 2014-08-01T18:54:39+00:00 | [
"math.FA",
"46B03, 46B06, 46B07, 46B25, 46B45"
] | The stabilized set of $p$'s in Krivine's theorem can be disconnected | For any closed subset $F$ of $[1,\infty]$ which is either finite or consists
of the elements of an increasing sequence and its limit, a reflexive Banach
space $X$ with a 1-unconditional basis is constructed so that in each block
subspace $Y$ of $X$, $\ell_p$ is finitely block represented in $Y$ if and only
if $p \in F$. In particular, this solves the question as to whether the
stabilized Krivine set for a Banach space had to be connected. We also prove
that for every infinite dimensional subspace $Y$ of $X$ there is a dense subset
$G$ of $F$ such that the spreading models admitted by $Y$ are exactly the
$\ell_p$ for $p\in G$. |
1408.0266v1 | ## Hadroproduction of t ¯ bb t ¯ final states at LHC: predictions at NLO accuracy matched with Parton Shower
## M. V. Garzelli, A. Kardos and Z. Tr´cs´nyi o a
MTA-DE Particle Physics Research Group and Institute of Physics, University of Debrecen, H-4010 Debrecen P.O.Box 105, Hungary
## Abstract
Wepresent predictions for the hadroproduction of t ¯ b ¯ final states t b at the LHC with collision energies √ s = 8TeV and 14TeV at NLO accuracy matched with parton shower, as obtained with PowHel + PYTHIA . We quantify the effects of parton shower and hadronization. We find these are in general moderate except the effect of the decay of heavy particles, which can modify significantly some distributions, like that of the invariant mass of the two leading b-jets. We also show kinematic distributions obtained with cuts inspired to those recently employed by the CMS collaboration. For these predictions, we present the theoretical uncertainty bands, related to both scale and PDF variations. We find that these uncertainties are only moderately affected by the change of the collision energy.
## 1 Introduction
The t ¯ b ¯ t b hadroproduction process was one of the first four-parton hadroproduction processes studied a few years ago in order to test and show the capabilities of new unitarity-inspired methods for evaluating one-loop amplitudes, as well as the potential of new developments in the more traditional tensorreduction approach. Pioneering works in this direction were performed by Bredenstein et al. [1, 2], as well as by the HELAC-NLO collaboration [3]. In both cases the role of next-to-leading-order (NLO) QCD corrections was shown, by considering rather loose systems of cuts. The agreement between predictions obtained by different methods was taken as a prove of the correctness of the new conceptual developments, that were subsequently applied successfully to other multiparticle hadroproduction processes, reaching a level of unforeseen complexity [4, 5, 6, 7].
First predictions concerning t ¯ b ¯ t b computations with NLO QCD accuracy matched with PS were presented by our group in Ref. [12, 13], making use of PowHel . More recently, t ¯ b ¯ t b predictions obtained by using the MC@NLO matching algorithm [14], as encoded in SHERPA [15], and massive b-quarks, with a fixed pole mass m b = 4 75 GeV, have been reported in Ref. [16], using . 1-loop amplitudes computed by OpenLoops [17]. Our computation differs from that presented in Ref. [16] in the following aspects: we use (i) massless b-quarks in the generation of NLO matrix elements, (ii) a different matching algorithm (POWHEG [18, 19]), and (iii) a different Shower Monte Carlo program ( PYTHIA ).
A further application was represented by the study of t ¯ b ¯ t b production at the TeVatron [8]. Nowadays, a renewed interest for this process has arisen, driven not only by further theoretical developments, including the possibility of matching complex NLO QCD calculations to parton shower (PS) approaches, but also by the requests of the experimental collaborations working at LHC, that need input to reduce the uncertainties on the estimate of the t ¯ b ¯ t b contribution to the total background in their searches for the Higgs boson in the t ¯ t H channel, with the H boson decaying in a b ¯ pair [9, 10, 11]. b The theoretical uncertainty quoted in recent experimental analyses on the contribution of this irreducible non-resonant background component amounts to a few tens percent ( ∼ 50 % in Ref. [11]). We expect that theoretical estimates with the best possible theoretical accuracy within present reach of the most advanced event generators can help reduce this uncertainty. Thus, one of the aims of this paper is to present predictions for differential cross sections at NLO QCD accuracy matched to PS and hadronization, resulting in predictions at the hadron level, using cuts similar to those employed by the experimental collaborations.
As pointed out in Ref. [16], the finite mass of the b-quarks allows for computing the contribution of t ¯ jj events with the jets originating from two t gluons that both separately split into a collinear b ¯ b pair. Each pair itself becomes part of one b -jet (as the b-quarks are sufficiently collinear to be unresolved by the jet algorithm), so the final state contains a t ¯ -quark pair t in association with two b -jets. Strictly speaking, such contributions are higher order in perturbation theory and are not included in our computation with NLO+PS accuracy, where collinear bb ¯ pairs can only be produced by gluon splitting in the PS. On the other hand, in Ref. [16], one of those g → b ¯ b splittings is included in the real radiation matrix elements (at LO accuracy for these contributions), while the other is generated by the parton shower, and such double g → b ¯ b splittings were found to give a significant contribution to the t ¯ +2 t b -jet final states when the t-quarks are kept stable.
This paper represents an extension of our previous studies, relying on a more advanced PowHel implementation, on the basis of recent POWHEG-BOX developments, and includes a detailed study of the theoretical uncertainties affecting our results in the most general case, including top decay and the full SMC chain up to hadronization and hadron decay.
## 2 Method
Predictions presented in this paper were obtained using events generated by PowHel [20], and stored in Les Houches event (LHE) files [21]. Our computational framework, PowHel , is an event generator that uses the HELAC-NLO codes [22] ( HELAC-1LOOP [23] and HELAC-Dipole [24]) for the computation of the amplitudes required as input by the POWHEG-BOX [25]. We use the latter for matching NLO QCD computations with PS, according to the POWHEG method, generating events stored in the Les Houches format (LHE). Details of event generation for t ¯ b ¯ t b final states were presented in Ref. [12], where thorough checks of the correctness of the event files were also discussed. In particular, we treated the b-quarks massless, and used five massless flavours, but neglected the small contribution to the cross section from the b-quarks in the initial state. Feynman graphs containing g → b ¯ b splittings are included in our computations. Such splittings are singular when the massless b-quarks become collinear. To avoid this divergence, we employ a generation cut on the invariant mass of the b ¯ b pair, which is chosen such that it is harmless when physical cuts are also employed [12]. We set the mass of the t-quark to m t = 173 2 GeV for the events at 8 TeV and to . m t = 172 6 GeV for those . at 14 TeV. The latter choice was made in the first calculation at 14 TeV [3], against which we checked our predictions at NLO accuracy in Ref. [12].
Subsequent radiation emissions on top of the LHEs are generated by shower Monte Carlo (SMC) programs. SMCs can be further used to generate hadronization and hadron decay, as well as the decay of the heavy elementary particles in the narrow width approximation (NWA). For making predictions in this paper, we used the PYTHIA SMC code [26], in the Fortran version 6.4.28. However, other SMCs can be used to further shower the events generated by PowHel .
Following Ref. [12], we fix the default value of the renormalization and factorization scales equal to µ R = µ F = µ 0 = H / T 2, where H T is the sum of the transverse masses of partons in the final state, using the underlying Born kinematics. Then we consider their simultaneous variation by a factor of two around this value, leading to scale uncertainty bands corresponding to scales in the range [ µ / , 0 2 2 µ 0 ]. We also consider µ R and µ F variations separately in this range and we verify that the uncertainties associated to off diagonal variations ( µ , µ R F / 2), ( µ / , µ R 2 F ), ( µ , R 2 µ F ), (2 µ , µ R F ) lie within the bands of diagonal ones. Our scale uncertainty bands represent the envelope coming from these variations, leaving out the antipodal choices ( µ / , R 2 2 µ F ) and (2 µ , µ F R / 2) according to the prescription of Ref. [27].
We used the CT10NLO , MSTW2008NLO and NNPDF sets of PDFs with 2-loop running α S and 5 active flavours, as available in the LHAPDF interface [28], drawing a PDF uncertainty band as the envelope of the results corresponding to the three central values of these PDF sets. For practical reasons, due to the intensive CPU requirements of our calculations, we neglected uncertainties coming from considering the statistical variations around each central PDF distribution. We expect that differences between different PDF sets are more pronounced than differences within each PDF set, as already found for the hadroproduction of t ¯ t H final states in the HXSWG reports [13, 29].
As discussed in Ref. [12], suppression factors are used in order to suppress the generation of the events in those regions of the phase space that are expected to be less relevant from the experimental point of view, considering the typical cuts applied in the experimental analyses. These factors however, lead to events characterized by a wide weight distribution, extending without any explicit limit both in the negative and in the positive weight region. This may lead to spikes in the kinematic distributions. We expect these spikes would disappear with significantly more statistics, but the present capabilities of computer resources limit the total number of events we can generate to several millions. We thus implemented an automated spikeelimination procedure applied after SMC, relying on the fact that the same LHE can lead to different SMC emissions, i.e. can populate different bins of the final differential distributions at the hadron level, depending on the random number sequence in the SMC generator.
## 3 Phenomenology
Using PowHel one can make predictions at four different stages in the evolution of the final state: (i) at the parton level using NLO accuracy, (ii) from the pre-showered POWHEG simulation (referred to Les Houches events, or LHEs), formally at the NLO accuracy, (iii) after decay of the heavy particles, (iv) at the hadron level after full SMC. A possible alternative to the last stage is (iv') after PS switching off hadronization and hadron decay, but in this case the heavy quarks are automatically kept stable by PYTHIA. Utilizing these options, we study the effect of these various stages of event evolution on several differential distributions, before making predictions at the hadron level.
## 3.1 SMC effects
To understand the effect of the PS and hadronization, we first performed a phenomenological analysis where the cuts are applied on the LHEs, whereas distributions are plotted after different stages of the evolution. The cuts applied on jets formed from the LHEs, reconstructed using the k ⊥ algorithm with R = 0 4, . as implemented in F astJet 3.0.6 [30] are the following: we require at least one b - and one ¯ -jet with (a) b p T > 20 GeV, (b) | η | < 2.5, and (c) m bb ¯ > 100 GeV. We do not apply cuts on b -jets emerging from the decay of top quarks, implemented at LO accuracy in the NWA in the SMC, which neglects spin correlations.
In Figs. 1-6 we present the comparisons of distributions at different stages of event evolution. Each plot contains five predictions, corresponding to the list (i-iv') presented in the introduction of this section.
First we study the effect of the PS by comparing differential distributions from the LHEs and after PS, keeping heavy particles stable. The corresponding predictions are marked as 'LHE' and 'PS'. In the middle section of each plot we also show the ratio PS/LHE of these predictions together with the ratio of the prediction at NLO accuracy to that from the LHEs, already studied in Ref. [12], where a fairly uniform increase of the distributions was found from the LHEs. Here we find that the effect of the PS is to 'bring back' the predictions close to the NLO ones with two exceptions: (i) the rapidity distribution of the hardest b -jet, which remains close to the prediction from LHEs even after shower evolution, and (ii) the distributions of transverse momenta for small p ⊥ (below about 50 GeV). Thus, the effect of the PS is in general small on top of the predictions at NLO level when the t-quarks are kept stable, or looking from an experimental point of view, when those are reconstructed from their decay products.
⊥
Figure 1: Distribution of transverse momentum of the hardest b -jet at the LHC at √ s = 14TeV using PowHel . Distributions from LHEs are denoted LHE, while those at NLO accuracy by NLO. The middle panel shows the predictions at NLO accuracy, as well as after PS, normalized by the predictions from LHEs. The lower panel shows the predictions after decay normalized by the predictions after full SMC. The errorbars in both plots represent the combined statistical accuracy of the numerical integrations.
Secondly we study the effect of the full SMC by comparing differential distributions from the LHEs, after decay and after SMC, corresponding to the full simulation including PS, hadronization and hadron decay. The corresponding predictions are labelled as 'decay' and 'SMC' and the ratio of these is also shown in the lower panel of each plot. Here we find significant changes in almost all distributions. The decay of the heavy quarks modifies the shapes, which are further modified by the SMC in the case of distributions of transverse momenta, or invariant mass of the bb ¯ -jet pair. In case of rapidity distributions the effect of the SMC is negligible.
We select isolated leptons with p ⊥ ,glyph[lscript] > 25 GeV, | η glyph[lscript] | < 2 5 . and mini-
Figure 2: Same as Fig. 1, as for the rapidity distribution of the hardest b -jet.
⊥
Figure 3: Same as Figs. 1 and 2, as for the second hardest b -jet.
mal separation from all jets in the pseudorapidity-azimuthal angle plane ∆ R = 0 4. . Looking at kinematic distributions of those leptons, such as
⊥
Figure 4: Same as Fig. 1, as for the distribution of transverse momentum of the bb ¯ -jet pair.
the transverse momentum and pseudorapidity of the hardest isolated lepton plotted in Fig. 7, it is clear that these are not affected by SMC effects, as expected. Such leptons appear in the decay of the heavy quarks, and neither the PS nor the hadronization change those. The same is true for the W transverse mass, reconstructed for W 's decaying leptonically, not shown here.
On the other hand, the distributions most affected by the SMC are the total number of ( b + ¯ )-jets, the total number of jets ( b b and non- ), the b p ⊥ of the hardest non-b jet. We show the last of these in Fig 8. In this case the prediction at NLO accuracy diverges for small transverse momentum, which is screened by the Sudakov-factor, in case of parton shower matched prediction. The decay of the heavy quarks hardens this distribution significantly through jets emerging from the hadronic decay of the W bosons, which is subsequently softened by the PS and hadronization.
Figure 5: Same as Fig. 1, as for the rapidity distribution of the hardest bb ¯ -jet pair.
Figure 6: Same as Fig. 1, as for the distribution of (a) ∆ R bb ¯ separation and (b) invariant mass of the hardest bb ¯ -jet pair.
⊥
Figure 7: PowHel + PYTHIA predictions after decay and after SMC for (a) the transverse momentum and (b) the pseudorapidity of the hardest lepton. The lower panel shows the ratio of the predictions after SMC to those after decay.
⊥
-
Figure 8: Same as Fig. 1, as for the distribution of the transverse momentum of the hardest nonb jet.
## 3.2 Analyses with cuts at the hadron level
In order to make predictions at the hadron level including scale and PDF uncertainties, we decided to use selection cuts at the hadron level, inspired by the CMS note Ref. [31], to identify events in the dileptonic decay mode. These are rather involved, which amply shows the flexibility of our method:
- 1. We require the events to have at least one pair of isolated opposite sign leptons with p ⊥ ,glyph[lscript] > 20 GeV and | η glyph[lscript] | < 2 4. . Dilepton candidate events with invariant mass m glyph[lscript]glyph[lscript] ′ < 12 GeV are removed to suppress multijet final states, at essentially no penalty for the collected signal. Alepton is considered isolated if the sum of the transverse momenta of the charged hadrons, neutral hadrons and photons in a cone of ∆ R < 0 3 around it . divided by its transverse momentum, I rel , does not exceed I max rel = 0 15. .
- 2. To remove the large background from events with Z -boson and jets, with Z decaying into leptons, we require that the invariant mass of the lepton pair defined above ( e + e -or µ + µ -) falls outside a ± 15 GeV window centered at m Z .
- 3. Signal events typically contain large missing transverse energy due to the presence of neutrinos from decays of the W -bosons. We require a minimum missing transverse energy, / p ⊥ > 30 GeV for the e + e -or µ + µ -dilepton final states, but not for the mixed flavour dilepton final state.
- 4. We cluster jets from hadrons, photons and non-isolated leptons using the antik ⊥ algorithm [32] with R = 0 5. . The jets are all required to have | η j | < 2 5 and . p ⊥ ,j > 40 or 20 GeV. We performed the analysis with both cuts, but present distributions here for the former only.
- 5. We require at least four well-separated b -jets after SMC, ∆ R j , j ( i k ) > 0 5 for . i, k ∈ { 1, 2, 3, 4 } , as well as ∆ R j , glyph[lscript] ( i k ) > 0 5 for . i ∈ { 1, 2, 3, 4 } , and k ∈ { 1, 2 } for the selected opposite-sign leptons. We assume 100 % b-tagging efficiency using MCTRUTH .
With these set of cuts the cross section for various choices of the default scale and for different PDF sets is presented in Tables 1 and 2 for a collider energy of 14 TeV and in Table 3 for a collider energy of 8 TeV. These cross-section values were obtained at the hadron level by showering events generated by PowHel with a p T -ordered version of the PYTHIA PS, by using the Perugia 2011 tune [33]. Using the default version of PYTHIA , where the PS is virtuality-ordered in absence of tunes, gives rise to values a few percent lower.
Table 1: Total cross sections in fb at 14 TeV in the different dilepton channels with cuts listed in the text. The minimum transverse momentum of the jets is set to 20 GeV. The quoted uncertainties are of statistical origin.
| µ 0 | PDF | e + e - | µ + µ - | e ± µ ∓ | Total |
|--------|-------|--------------|--------------|--------------|--------------|
| H ⊥ /4 | CT10 | 10.93 ± 0.36 | 11.10 ± 0.25 | 32.23 ± 1.01 | 54.26 ± 1.10 |
| H ⊥ /2 | CT10 | 7.77 ± 0.33 | 7.77 ± 0.27 | 22.40 ± 0.58 | 37.94 ± 0.72 |
| H ⊥ | CT10 | 5.47 ± 0.21 | 5.76 ± 0.31 | 15.87 ± 0.44 | 27.10 ± 0.58 |
| H ⊥ /2 | NNPDF | 8.57 ± 0.40 | 8.79 ± 0.42 | 24.51 ± 0.86 | 41.87 ± 1.04 |
| H ⊥ /2 | MSTW | 9.04 ± 0.45 | 8.65 ± 0.24 | 24.48 ± 0.86 | 42.17 ± 1.00 |
In Table 4 we exhibit the number of expected events for 19.6 fb -1 at 8 TeV. These marginally agree with the number of events expected in the CMS data sample, obtained on the basis of a LO+PS computation by MadGraph [34] + PYTHIA (first line in Table 2 of Ref. [31]). However, this comparison can only be indicative for two reasons. On the one hand our predictions assume 100 % b-tagging efficiency, while the experimental analysis has smaller. On the other hand, we require at least four b -jets, treating all b -jets on the same footing, while the experimental analysis requires only two b -jets, because they do not include the b -jets coming from t-quark decays in this count. Furthermore, in order to identify b -jets we kept the lowest lying B-hadrons stable in PYTHIA for simplicity, tagging as b -jets those including at least a B-hadron, whereas the experiment reconstructs b -jets from their decay products, using displaced vertex information.
Recently an independent study of t ¯ b ¯ t b and t ¯ jj t hadroproduction, also inspired by the same CMS note, appeared in Ref. [35]. This work differs from our one because it is an NLO study, including a simplified sets of cuts applied on parton level events (also including top decays, but neglecting parton shower, hadronization and hadron decay effects).
In Figs. 9-12 we present several kinematic distributions obtained with these selection cuts: the transverse momenta and rapidities of the two hardest b -jets, the transverse momentum, rapidity and invariant mass distribution of the hardest b -jet pair and the invariant mass distribution of the hardest b -jet and hardest charged lepton. In each figure we show two predictions, one for 8 TeV c.m. energy (smaller cross sections) and one for 14 TeV (larger cross sections). The shapes of the rapidity distributions are very similar at the two energies, the result of the larger collision energy just amounting to about a nine-fold increase of the cross section. On the other hand, in case of the invariant mass and transverse momentum distributions, the spectra at 14
| µ 0 | PDF | e + e - | µ + µ - | e ± µ ∓ | Total |
|--------|-------|-------------|-------------|-------------|--------------|
| H ⊥ /4 | CT10 | 2.71 ± 0.04 | 2.72 ± 0.04 | 7.81 ± 0.06 | 13.24 ± 0.09 |
| H ⊥ /2 | CT10 | 1.97 ± 0.03 | 1.97 ± 0.03 | 5.62 ± 0.06 | 9.56 ± 0.08 |
| H ⊥ | CT10 | 1.43 ± 0.03 | 1.42 ± 0.02 | 4.19 ± 0.10 | 7.04 ± 0.11 |
| H ⊥ /2 | NNPDF | 2.15 ± 0.04 | 2.14 ± 0.04 | 6.17 ± 0.14 | 10.46 ± 0.15 |
| H ⊥ /2 | MSTW | 2.19 ± 0.04 | 2.18 ± 0.04 | 6.41 ± 0.14 | 10.78 ± 0.15 |
Table 2: Total cross sections in fb at 14 TeV in the different dilepton channels with cuts listed in the text. The minimum transverse momentum of the jets is set to 40 GeV.
Table 3: Total cross sections in fb at 8 TeV in the different dilepton channels with cuts listed in the text. The minimum transverse momentum of the jets is set to 40 GeV.
| µ 0 | PDF | e + e - | µ + µ - | e ± µ ∓ | Total |
|--------|-------|---------------|---------------|---------------|---------------|
| H ⊥ /4 | CT10 | 0.490 ± 0.007 | 0.485 ± 0.006 | 1.421 ± 0.010 | 2.396 ± 0.013 |
| H ⊥ /2 | CT10 | 0.346 ± 0.006 | 0.343 ± 0.004 | 1.002 ± 0.006 | 1.691 ± 0.010 |
| H ⊥ | CT10 | 0.246 ± 0.004 | 0.242 ± 0.003 | 0.714 ± 0.008 | 1.204 ± 0.010 |
| H ⊥ /2 | NNPDF | 0.361 ± 0.006 | 0.360 ± 0.004 | 1.060 ± 0.013 | 1.781 ± 0.015 |
| H ⊥ /2 | MSTW | 0.374 ± 0.005 | 0.376 ± 0.004 | 1.098 ± 0.017 | 1.848 ± 0.018 |
Table 4: Number of expected t ¯ b ¯ t b events at 8 TeV in the different dilepton channels with cuts listed in the text. The minimum transverse momentum of the jets is set to 40 GeV. The uncertainty corresponds to the variation of the renormalization and factorization scales around their default value by factors of half and two.
| e + e - | µ + µ - | e ± µ ∓ | Total |
|------------------|------------------|-----------------------|------------------------|
| 6 . 8 +2 . 8 - 2 | 6 . 7 +2 . 8 - 2 | 19 . 6 +8 . 2 - 5 . 4 | 33 . 1 +13 . 8 - 9 . 4 |
⊥
Figure 9: Distribution of (a) transverse momentum (b) rapidity of the hardest b -jet at the LHC at √ s = 14TeV using PowHel + PYTHIA after full SMC, under the cuts (1 - 5) listed in the text. The band represent the envelope of the scale uncertainties. The lower panels show the ratio of predictions obtained with MSTW2008NLO and NNPDF sets to that obtained with CT10NLO .
⊥
Figure 10: Same as Fig. 9, as for the second hardest b -jet.
TeV also become considerably harder with respect to those at 8 TeV.
In the same figures we exhibit two bands: one corresponding to the envelope of standard variations of the renormalization and factorization scales, between half and twice the default scale µ 0 , while the other to the envelope of dependence on the PDF set ( CT10NLO , MSTW2008NLO and NNPDF ). In order to see these better, in the lower panels we show all predictions normalized to our default choice: predictions with µ 0 = H / ⊥ 2 and CT10NLO PDF set. The middle panels corresponds to the cross sections at 14 TeV, while the lower
⊥
Figure 11: Same as Fig. 9, as for the hardest bb ¯ -jet pair.
Figure 12: Same as Fig. 9, as for the invariant mass of (a) the hardest bb ¯ -jet pair, (b) the hardest b -jet and positron.
ones to those at 8 TeV.
We see that the scale dependence for these distributions follows the scale dependence found at the NLO accuracy [12]. It is similar in size (about +38 % -26 % at 14 TeV and +41 % -29 % at 8 TeV, for a jet p ⊥ cut at 40 GeV) and also uniform in shape, supporting our choice for the default scale.
The PDF bands are much narrower. In general, we find that up to statistical fluctuations, the smallest cross sections are obtained with CT10NLO , while MSTW2008NLO and NNPDF give similar, at most ∼ 10 % larger values than our default, with MSTW predictions slightly larger than NNPDF ones. These results show that this complicated final state with rather exclusive experimental se-
Figure 13: Same as Fig. 9, as for the transverse momentum of (a) the hardest lepton, (b) the t ¯ b ¯ t b system.
lection cuts can be modelled reliably in perturbation theory at NLO accuracy matched with SMC.
We present an example for lepton distributions in Fig. 13.a where the spectrum of the transverse momentum of the hardest lepton is depicted. This lepton emerges in the decay of one of the t-quarks, i.e. it is not present in the NLO matrix elements. Nevertheless, all what we pointed out for the b -jet distributions concerning the scale and PDF uncertainties are valid for this (and other leptonic distributions, not shown here), too.
Finally, in Fig. 13.b we present the distribution of the transverse momentum of the t ¯ b ¯ t b system, emphasizing the low p ⊥ -region, where we see the effect of Sudakov damping. This distribution is divergent for p ⊥ = 0 at fixed order in perturbation theory, while it is finite in the matched prediction. We find decreasing scale dependence with decreasing p ⊥ , although with naturally worsening statistics.
## 4 Conclusions
We have studied the t ¯ b ¯ t b process including NLO QCD corrections computed with five massless flavours matched with parton shower, using the PowHel event generator interfaced with the PYTHIA SMC. This computation, challenging due to the presence of two massless b's in the final state, producing singular underlying Born configurations, opens the road to a unified treatment of t ¯ b ¯ t b and t ¯ jj. t The latter process can also produce t ¯ +2 t b -jet final states when both jets are tagged as b -jets due to g → b ¯ b splittings. Such
contributions are not included in our computation.
We presented predictions at different stages of event evolution in order to show the effect of the SMC in modifying NLO distributions. We found that the effect of the parton shower over the predictions at NLO accuracy are usually very small except for rapidity distributions and small values of transverse momenta. The effect of the hadronization is to soften the transverse momentum or invariant mass spectra and can be up to 30 %. However, the largest effect is due to the decay of the heavy quarks. Kinematic distributions of the leptonic decay products of the heavy quarks are not affected by the hadronization.
We paid special attention to the identification of the sources of uncertainties and in the estimate of the uncertainty bands at the hadron level. We found that the scale dependences change only very moderately with c.m. energy, and are very similar to those of the predictions at the NLO accuracy reported in Ref. [12], altough the latter were obtained with looser systems of cuts. These scale dependences are fairly uniform for all distributions shown when we employ our default scale - the half of the sum of transverse masses in the final state -, which supports our choice. The PDF uncertainties are much smaller, with CT10NLO yielding the smallest cross section in general, while MSTW2008NLO and NNPDF giving very similar results that are up to about 10 % higher. These results show that our events can model reliably the t ¯ b ¯ t b final states.
Final states consisting of a t ¯ and a b ¯ t b quark pair constitute important backgrounds for Higgs boson production in association with a t ¯ -pair. t Our events can be used to optimize the selection of the signal events to find the best signal/background ratio in the various decay channels. For this purpose sets of LHEs can be downloaded from our web-page for both the t ¯ t H signal ( http://grid.kfki.hu/twiki/bin/view/DbTheory/TthProd ) and for the background ( http://grid.kfki.hu/twiki/bin/view/DbTheory/TtbbProd ), or new ones can be requested for different choices of scale, PDF, and parameter values.
## Acknowledgements
We are grateful to Malgorzata Worek, Giuseppe Bevilacqua and Tae Jeong Kim for useful discussions on several aspects of their t ¯ bb t ¯ studies. This research was supported by the Hungarian Scientific Research Fund grant K-101482, the SNF-SCOPES-JRP-2014 grant 'Preparation for and exploitation of the CMS data taking at the next LHC run', the European Union and the European Social Fund through LHCPhenoNet network PITN-GA-2010264564, and the Supercomputer, the national virtual lab TAMOP-4.2.2.C11/1/KONV-2012-0010 project.
## References
| [1] | A. Bredenstein, A. Denner, S. Dittmaier, and S. Pozzorini, NLO QCD corrections to pp → t anti-t b anti-b + X at the LHC , Phys.Rev.Lett. 103 (2009) 012002, [ arXiv:0905.0110 ]. |
|-------|---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [2] | A. Bredenstein, A. Denner, S. Dittmaier, and S. Pozzorini, NLO QCD Corrections to Top Anti-Top Bottom Anti-Bottom Production at the LHC: 2. full hadronic results , JHEP 1003 (2010) 021, [ arXiv:1001.4006 ]. |
| [3] | G. Bevilacqua, M. Czakon, C. Papadopoulos, R. Pittau, and M. Worek, Assault on the NLO Wishlist: pp → t anti-t b anti-b , JHEP 0909 (2009) 109, [ arXiv:0907.4723 ]. |
| [4] | G. Bevilacqua, M. Czakon, C.G. Papadopoulos, and M. Worek, Dominant QCD Backgrounds in Higgs Boson Analyses at the LHC: A Study of pp → t anti-t + 2 jets at Next-To-Leading Order , Phys.Rev.Lett. 104 (2010) 162002, [ arXiv:1002.4009 ]. |
| [5] | G. Cullen, H. van Deurzen, N. Greiner, G. Luisoni, P. Mastrolia, et al., Next-to-Leading-Order QCD Corrections to Higgs Boson Production Plus Three Jets in Gluon Fusion , Phys.Rev.Lett. 111 (2013) 131801, [ arXiv:1307.4737 ]. |
| [6] | J. Alwall, R. Frederix, S. Frixione, V. Hirschi, F. Maltoni, et al., The automated computation of tree-level and next-to-leading order differential cross sections, and their matching to parton shower simulations , JHEP 1407 (2014) 079, arXiv:1405.0301 . |
| [7] | Z. Bern, L.J. Dixon, F. Febres Cordero, S. Hoeche, H. Ita et al., Universality of W+Multijet Production , arXiv:1407.6564 . |
| [8] | M. Worek, On the next-to-leading order QCD K -factor for ttbb ¯ ¯ production at the TeVatron , JHEP 1202 (2012) 043, [ arXiv:1112.4325 ]. |
| [9] | ATLAS Collaboration, Search for the standard model higgs boson produced in association with top quarks in proton-proton collisions at √ s = 7 tev using the atlas detector , Tech. Rep. ATLAS-CONF-2012-135, CERN, Geneva, Sep, 2012. |
| [10] | CMS Collaboration, S. Chatrchyan et al., Search for the standard model Higgs boson produced in association with a top-quark pair in pp collisions at the LHC , JHEP 1305 (2013) 145, [ arXiv:1303.0763 ]. |
| [11] | CMS Collaboration, Search for higgs boson production in association with a top-quark pair and decaying to bottom quarks or tau leptons , Tech. Rep. CMS-PAS-HIG-13-019, CERN, Geneva, 2013. |
|--------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [12] | A. Kardos and Z. Tr´cs´nyi, o a Hadroproduction of t anti-t pair with a b anti-b pair using PowHel , J.Phys. G41 (2014) 075005, [ arXiv:1303.6291 ]. |
| [13] | LHC Higgs Cross Section Working Group Collaboration, S. Heinemeyer et al., Handbook of LHC Higgs Cross Sections: 3. Higgs Properties , arXiv:1307.1347 . |
| [14] | S. Frixione and B. R. Webber, Matching NLO QCD computations and parton shower simulations , JHEP 0206 (2002) 029, [ hep-ph/0204244 ]. |
| [15] | T. Gleisberg, S. Hoeche, F. Krauss, M. Schonherr, S. Schumann, et al., Event generation with SHERPA 1.1 , JHEP 0902 (2009) 007, [ arXiv:0811.4622 ]. |
| [16] | F. Cascioli, P. Maierhoefer, N. Moretti, S. Pozzorini, and F. Siegert, NLO matching for ttbb production with massive b-quarks , arXiv:1309.5912 . |
| [17] | F. Cascioli, P. Maierhofer, and S. Pozzorini Scattering Amplitudes with Open Loops , Phys.Rev.Lett. 108 (2012) 111601, [ arXiv:1111.5206 ]. |
| [18] | P. Nason, A New method for combining NLO QCD with shower Monte Carlo algorithms , JHEP 0411 (2004) 040, [ hep-ph/0409146 ]. |
| [19] | S. Frixione, P. Nason, and C. Oleari, Matching NLO QCD computations with Parton Shower simulations: the POWHEG method , JHEP 11 (2007) 070, [ arXiv:0709.2092 ]. |
| [20] | M. Garzelli, A. Kardos, and Z. Tr´cs´nyi, o a NLO Event Samples for the LHC , PoS EPS-HEP2011 (2011) 282, [ arXiv:1111.1446 ]. |
| [21] | J. Alwall et al., A standard format for Les Houches event files , Comput. Phys. Commun. 176 (2007) 300-304, [ hep-ph/0609017 ]. |
| [22] | G. Bevilacqua, M. Czakon, M. Garzelli, A. van Hameren, A. Kardos, et al., HELAC-NLO , Comput.Phys.Commun. 184 (2013) 986-997, [ arXiv:1110.1499 ]. |
| [23] | G. Bevilacqua, M. Czakon, M. Garzelli, A. van Hameren, Y. Malamos, et al., NLO QCD calculations with HELAC-NLO , Nucl.Phys.Proc.Suppl. 205-206 (2010) 211-217, [ arXiv:1007.4918 ]. |
|--------|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [24] | M. Czakon, C. Papadopoulos, and M. Worek, Polarizing the Dipoles , JHEP 0908 (2009) 085, [ arXiv:0905.0883 ]. |
| [25] | S. Alioli, P. Nason, C. Oleari, and E. Re, A general framework for implementing NLO calculations in shower Monte Carlo programs: the POWHEG BOX , JHEP 06 (2010) 043, [ arXiv:1002.2581 ]. |
| [26] | T. Sjostrand, S. Mrenna, and P. Z. Skands, PYTHIA 6.4 Physics and Manual , JHEP 0605 (2006) 026, [ hep-ph/0603175 ]. |
| [27] | M. Cacciari, S. Frixione, N. Houdeau, M. L. Mangano, P. Nason, et al., Theoretical predictions for charm and bottom production at the LHC , JHEP 1210 (2012) 137, [ arXiv:1205.6344 ]. |
| [28] | M. Whalley, D. Bourilkov, and R. Group, The Les Houches accord PDFs (LHAPDF) and LHAGLUE , hep-ph/0508110 . |
| [29] | LHC Higgs Cross Section Working Group Collaboration, S. Dittmaier et al., Handbook of LHC Higgs Cross Sections: 1. Inclusive Observables , arXiv:1101.0593 . |
| [30] | M. Cacciari, G. P. Salam, and G. Soyez, FastJet user manual , Eur.Phys.J. C72 (2012) 1896, [ arXiv:1111.6097 ]. |
| [31] | CMS Collaboration, Measurement of the cross section ratio σ ( ttbb ¯ ¯ )/ σ ( ttjj ¯ ) in pp collisions at √ s = 8 tev , Tech. Rep. CMS-PAS-TOP-13-010, CERN, Geneva, 2013. |
| [32] | M. Cacciari, G. P. Salam, and G. Soyez, The Anti-k(t) jet clustering algorithm , JHEP 0804 (2008) 063, [ arXiv:0802.1189 ]. |
| [33] | P. Z. Skands, Tuning Monte Carlo Generators: The Perugia Tunes , Phys.Rev. D82 (2010) 074018, [ arXiv:1005.3457 ]. |
| [34] | J. Alwall, M. Herquet, F. Maltoni, O. Mattelaer and T. Stelzer MarGraph 5: Going Beyond , JHEP 1106 (2011) 128, [ arXiv:1106.0522 ]. |
| [35] | G. Bevilacqua and M. Worek, On the ratio of ttbb and ttjj cross sections at the CERN Large Hadron Collider , JHEP 1407 (2014) 135, [ arXiv:1403.2046 ]. | | 10.1007/JHEP03(2015)083 | [
"M. V. Garzelli",
"A. Kardos",
"Z. Trocsanyi"
] | 2014-08-01T18:59:16+00:00 | 2014-08-01T18:59:16+00:00 | [
"hep-ph",
"hep-ex"
] | Hadroproduction of $t\,\bar{t}\,b\,\bar{b}$ final states at LHC: predictions at NLO accuracy matched with Parton Shower | We present predictions for the hadroproduction of $t\bar{t}b\bar{b}$ final
states at the LHC with collision energies $\sqrt{s}$ = 8TeV and 14TeV at NLO
accuracy matched with parton shower, as obtained with PowHel+ PYTHIA. We
quantify the effects of parton shower and hadronization. We find these are in
general moderate except the effect of the decay of heavy particles, which can
modify significantly some distributions, like that of the invariant mass of the
two leading b-jets. We also show kinematic distributions obtained with cuts
inspired to those recently employed by the CMS collaboration. For these
predictions, we present the theoretical uncertainty bands, related to both
scale and PDF variations. We find that these uncertainties are only moderately
affected by the change of the collision energy. |
1408.0267v3 | ## Quantum Corrections in Galileons from Matter Loops
Lavinia Heisenberg, a,b
a Perimeter Institute for Theoretical Physics,
31 Caroline St. N, Waterloo, Ontario, Canada, N2L 2Y5
b D´ epartment de Physique Th´orique and Center for Astroparticle Physics, e
Universit´ de Gen`ve, 24 Quai E. Ansermet, CH-1211 Gen` eve, Switzerland e e
E-mails: [email protected]
Abstract. Galileon interactions represent a class of effective field theories that have received much attention since their inception. They can be treated in their own right as scalar field theories with a specific global shift and Galilean symmetry or as a descendant of a more fundamental theory like massive gravity. It is well known that the Galileon theories are stable under quantum corrections thanks to the non-renormalization theorem which is not due to the symmetry. We consider different covariant couplings of this Galileon scalar field with the matter field: the conformal coupling, the disformal coupling and the longitudinal coupling. We compute the one-loop quantum corrections to the Galileon interactions from the coupling to the external matter fields. In all the considered cases of covariant couplings we show that the terms generated by one-loop matter corrections not only renormalize the Galileon interactions but also give rise to higher order derivative ghost interactions. However, the renormalized version of the Galileon interactions as well as the new interactions come at a scale suppressed by the original classical coupling scale and hence are harmless within the regime of validity of the effective field theory.
## Contents
| 1 Introduction | 1 Introduction | 1 Introduction | 1 |
|------------------|----------------------------|----------------------------|-----|
| 2 Setup | 2 Setup | 2 Setup | 4 |
| 3 Galileon Loops | 3 Galileon Loops | 3 Galileon Loops | 7 |
| 4 Matter Loops | 4 Matter Loops | 4 Matter Loops | 8 |
| 4.1 | | Conformal coupling | 9 |
| 4.2 | | Disformal coupling | 14 |
| 4.3 | | Longitudinal coupling | 16 |
| 4.4 | Mixed | couplings | 17 |
| 5 | Stability of the couplings | Stability of the couplings | 17 |
| 6 | Summary and discussion | Summary and discussion | 19 |
## 1 Introduction
Cosmology has progressively developed from a philosophical to an empirical scientific discipline. Given the high precision achieved by the cosmological observations, cosmology is now a suitable arena to test fundamental physics. It is witnessing promising attempts to unite the physics of the large scale structures in the Universe with the physics of the small scale. Latest observations from Planck [1] and BICEP2 [2] with the exquisite precision have driven the models of early universe into a corner. Inflation, which describes a phase of accelerated expansion in the very early Universe might explain the initial conditions of the Universe.
Observations of the Cosmic Microwave Background, supernovae Ia, Baryon Acoustic Oscillations and lensing have led to the cosmological standard model which also requires an accelerated expansion of the late Universe, driven by dark energy [3-6]. The physical origin of the accelerated expansion is still a mystery. In the Standard Model of particles the detection of the missing fundamental particle, the Higgs boson, was a revolutionary event. In a similar way, the missing particles in the Standard Model of cosmology, like the graviton, and the resolution to the puzzle of accelerated expansion and its origin would be as revolutionary. There are several promising explanatory attempts to provide an explanation for the accelerated expansion of the late Universe.
One promising approach corresponds to explaining the acceleration of the Universe by modifying the geometrical part of Einstein's equations. Particularly, weakening gravity on cosmological scales might not only be responsible for a late-time speed-up of the Hubble expansion, but could also tackle the cosmological constant problem, which reflects the large discrepancy between observations and theoretical predictions. Such promising scenarios arise in infrared modifications of general relativity like massive gravity or in higher-dimensional frameworks [7-10]. All these models of infrared modifications of general relativity are united by the common feature of invoking new degrees of freedom. These degrees of freedom in the considered space-time are particles characterized by their masses and spins (or equivalently helicities). These particles are the excited quanta of the underlying fields. Their Lagrangian are constructed based on the requirement of yielding second order equations of motion, hence with a bounded Hamiltonian from below. With these requirements another successful and interesting class of infrared modifications was introduced: the Galileons [11]. This Galileon model relies strongly on the symmetry of the new scalar degree of freedom, namely the invariance under internal Galilean and shift transformations φ → φ + b µ x µ + , and on the c ghost absence. Interestingly Galileon interactions naturally arise in theories of massive gravity which has also been constructed to be ghost-free [8, 10, 12]. The Galileon theory has also been generalized to the non-flat background case. Direct covariantization might give rise to ghost-like terms in the equation of motion, which has motivated the introduction of non-minimal coupling between φ and the curvature [13-15]. Even if this covariantization is ghost-free, the Galileon symmetry is lost in curved backgrounds. However, there has been also successful generalizations to the maximally symmetric backgrounds with generalized Galileon symmetry [16, 17]. Interestingly enough, there has been a recent construction of a covariant Galileon theory without the requirement of non-minimal interactions and still avoiding ghost instabilities [18-21]. Another interesting point worth to mention is the fact that the covariant Galileon interactions also arise from the covariantization of the decoupling limit of massive gravity [22, 23]. Similarly, there has been attempts to find the vector Galileons besides the Maxwell kinetic term with second order equations of motion [24-26].
The Galileon theory exhibits a broad and interesting phenomenology. However one potentially worrying phenomenon is the fact that fluctuations of the Galileon field can propagate superluminally in the regime of interest [11, 27, 28], i.e. faster than light. Nevertheless, it has been shown in [29] that closed time like curves can never arise since the Galileon inevitably becomes infinitely strongly coupled and breaks down the effective field theory. Additionally, a dual description to the Galileon interactions has been discovered by [30-32] in which the original Galileon theory gets mapped to another Galileon theory by a non trivial field redefinition ˜ x µ = x µ + ∂ φ x µ ( ). For a given very specific Galileon coefficients the Galileon interactions are dual to a free massless scalar field. Thus, this mapping between a free luminal theory and the superluminal Galileon theory suggests that the naive existence of superluminal propagation
can still give rise to causal theory with analytic and unitary S-matrix [30]. These properties together with the presence of a Vainshtein mechanism could be tied to theories which allow for an alternative to UV completion such as classicalization.
Galileon interactions can be considered as an effective field theory constructed by the above mentioned restrictions of symmetry and ghost absence. In order for the theory to be viable, the Vainshtein mechanism is needed, which on the other hand relies on the presence of interactions at an energy scale Λ 3 glyph[lessmuch] M Pl . From a traditional effective field theory point of view these interactions are irrelevant operators which renders the theory non-renormalizable, but to contrary to the traditional case, within the Galileon theory these irrelevant operators need to be large in the regime of interest, in the so called strong coupled regime ∂ φ 2 ∼ Λ . 3 3 Therefore, one might have concerns that the effective field theory could go out of control in this strong coupling regime where the irrelevant operators need to be large. Nevertheless, the Galileon theories are not typical effective field theories in the sense that it is organized in the small parameter expansion of the whole operator but rather it has to be reorganized in a way that the derivative now plays the role of the small parameter rather than the whole operator itself [11, 33-35]. There exist a regime of interest for which φ ∼ Λ , 3 ∂φ ∼ Λ 2 3 and ∂ φ 2 ∼ Λ 3 3 even though any further derivative is suppressed ∂ φ 3 glyph[lessmuch] Λ , meaning that the effective field 4 3 expansion is reorganized such that the Galileon interactions are the relevant operators with equations of motion with only two derivatives, while all other interactions with equations of motion with more than two derivatives are treated as negligibale corrections. For a similar discussions see also [35-37].
Galileon interactions are protected against quantum corrections via the non-renormalization theorem. First of all, the Galileon and shift symmetry will prevent to generate local operators by Galileon loop corrections which would explicitly break these symmetries, like potential interactions. But this is not enough for the non-renormalization theorem. Quantum corrections might still generate local operators which are invariant under shift and Galileon transformations, either renormalizing Galileon interactions themselves and giving rise to large quantum corrections of the strong coupling scale Λ 3 or generating operators of higher derivative interactions. It is the non-renormalization theorem, which ensures that the Galileon interactions themselves are not renormalized at all and that the higher derivative operators are irrelevant corrections in the regime of validity of the effective field theory. The specific form of the counter terms arising in the 1-loop effective action coming from Galileon loops is such that they all come with at least one extra derivative as compared to the original interactions. Therefore there is no counter term which takes the Galileon form, and the Galileon interactions are hence not renormalized. This would mean that the Galileon coupling constants may be technically natural tuned to any value and remain radiatively stable.
In this work we address the question of one loop quantum corrections coming from matter loops. We will consider the Galileon scalar field as a scalar field in its own right, without restricting it to the massive gravity case. The Galileon scalar field can couple to matter as a conformal mode, φT at the linear level but also as a longitudinal mode, ∂ ∂ φT µ ν µν (even though this coupling would vanish for a conserved source), where T µν is the stress-energy tensor of the external matter field and T is the trace of it. At the non-linear level one could also consider more generic conformal couplings like φ T 2 or f ( φ T ) even though the symmetry would be broken at the level of the equations of motion. Naturally, one could also consider derivative couplings of the form ∂ φ∂ φT µ ν µν known as disformal coupling (which also arises in massive gravity), or more generally f ( φ ∂ φ∂ φT ) µ ν µν . We will be mainly concentrating on the cases of conformal, disformal and longitudinal couplings.
## 2 Setup
In this section we will set up our framework and notations. For convenience, we will work in Euclidean space such that the Galileon interactions live on top of a flat Euclidean metric δ ab . Furthermore, we will use units for which ¯ h = 1 and we will use the ( -, + + +) signature , , convention. For simplicity, we will assume a massive scalar field for our matter field which covariantly couples to the Galileon field.
Our starting point is the action for the Galileon φ and a massive scalar field χ ,
$$S & = \int \mathrm d ^ { 4 } x \left ( \mathcal { L } _ { G a l } + \mathcal { L } _ { \text{matter} } \right ) \\ & = \int \mathrm d ^ { 4 } x \mathcal { L } _ { G a l } + \int \mathrm d ^ { 4 } x \left ( \frac { 1 } { 2 } \left ( \partial \chi \right ) ^ { 2 } + \frac { 1 } { 2 } M ^ { 2 } \chi ^ { 2 } \right ) \,.$$
where L Gal is the Lagrangian for the Galileon interactions
$$\mathcal { L } _ { G a l } = \sum _ { n = 0 } ^ { 4 } c _ { n } \phi \ \mathcal { U } _ { n } ( \Phi ( x ) )$$
with Φ µν = ∂ ∂ φ µ ν and the characteristic symmetric polynomial invariants
$$\mathcal { U } [ \Phi ] = \mathcal { E } ^ { \mu _ { 1 } \dots \mu _ { 4 } } \mathcal { E } ^ { \nu _ { 1 } \dots \nu _ { 4 } } \prod _ { j = 1 } ^ { n } \Phi _ { \mu _ { j } \nu _ { j } } \prod _ { k = n + 1 } ^ { 4 } \eta _ { \mu _ { k } \nu _ { k } }$$
In order to have the standard kinetic term for the Galileon scalar field, we can choose c 2 = -1 / 12 and in order to make the scalar field to have the dimension of mass we can scale the parameters respectively c 3 = ˜ c / 3 Λ 3 3 . . . etc and reabsorb the dimensionless parameters ˜ . c n The Galileon field φ can couple to matter χ with an arbitrary coupling. In this work we will consider the important representatives of commonly used coupling classes. First of all, a
natural way of coupling the Galileon field to the matter is through a conformal coupling of the form φT where T = T µ µ is the trace of the associated stress energy tensor of the matter field χ . As we mentioned, we assume that the fields are propagating on flat Euclidean space-times and therefore indices are raised and lowered by the Euclidean metric T = δ µν T µν . The stress energy tensor for our matter field is simply given by
$$\Phi _ { \mu \nu } = \frac { - 2 } { \sqrt { - g } } \frac { \delta \mathcal { L } _ { \text{matter} } \sqrt { - g } } { \delta g ^ { \mu \nu } } | _ { g = \delta } = - \partial _ { \mu } \chi \partial _ { \nu } \chi + \frac { 1 } { 2 } \delta _ { \mu \nu } \left ( ( \partial \chi ) ^ { 2 } + M ^ { 2 } \chi ^ { 2 } \right )$$
with its trace being T = (( ∂χ ) 2 +2 M χ 2 2 ), such that the conformal coupling reads
$$\frac { \phi ( ( \partial \chi ) ^ { 2 } + 2 M ^ { 2 } \chi ^ { 2 } ) } { M _ { C } }$$
We suppressed the interaction by a so far arbitrary scale M C , which quotes when this interaction becomes important. In the context of massive gravity, the coupling of the Galileon in massive gravity (the helicity-0 degree of freedom of the massive graviton) with the matter field comes from the coupling h µν T µν /M Pl , which would mean that the φT coupling is Planck mass suppressed M C = M Pl . However, if we consider the Galileon field as a scalar field in its own right, then this scale can be arbitrarily different from the Planck mass. This conformal coupling can be extended to the non-linear level φ T 2 at the prize of loosing the Galileon symmetry. From the next order stress energy tensor one can also construct this type of non-linear conformal couplings φ T 2 µ α µ α where the tensor T µναβ arises from the variation of T µν with respect to the metric,
$$\ x x \, \text{we use menu}, \\ T _ { \mu \nu \alpha \beta } & = \frac { - 2 } { \sqrt { - g } } \frac { \delta \sqrt { - g } T _ { \mu \nu } } { \delta g ^ { \alpha \beta } } | _ { g = \delta } = - \partial _ { \mu } \chi \partial _ { \nu } \chi \delta _ { \alpha \beta } - \partial _ { \alpha } \chi \partial _ { \beta } \chi \delta _ { \mu \nu } \\ & \quad + \frac { 1 } { 2 } \left ( \delta _ { \mu \alpha } \delta _ { \nu \beta } + \delta _ { \nu \alpha } \delta _ { \mu \beta } + \delta _ { \mu \nu } \delta _ { \alpha \beta } \right ) \left ( \left ( \partial \chi \right ) ^ { 2 } + M ^ { 2 } \chi ^ { 2 } \right ), \\. &. &. &. &. &. &. &. &.$$
such that the non-linear conformal coupling would become
$$\frac { \phi ^ { 2 } ( 4 ( \partial \chi ) ^ { 2 } + 1 2 M ^ { 2 } \chi ^ { 2 } ) } { M _ { N C } ^ { 2 } }$$
suppressed by the scale M 2 NC . Again, in the context of massive gravity, this coupling would arise from h µν h αβ T µναβ /M 2 Pl but here we shall keep the scale arbitrary. Another important class of possible couplings is the derivative coupling. At the linear level we can also have longitudinal coupling of the form ∂ ∂ φT µ ν µν which would not contribute in the case of conserved sources, and at the non-linear level we would respectively have the disformal coupling of the form ∂ φ∂ φT µ ν µν . This derivative coupling naturally arises in the massive gravity and usually comes hand in hand with interesting features. [8, 10]. In massive gravity, this interaction
would come in at a scale ∂ φ∂ φT µ ν µν / M ( Pl Λ ), but here we shall consider this coupling at an 3 3 arbitrary scale 1 /M 4 D .
$$\frac { - \partial ^ { \mu } \phi \partial ^ { \nu } \phi \partial _ { \mu } \chi \partial _ { \nu } \chi + \frac { 1 } { 2 } ( \partial \phi ) ^ { 2 } \left ( ( \partial \chi ) ^ { 2 } + M ^ { 2 } \chi ^ { 2 } \right ) } { M _ { D } ^ { 4 } }$$
In the context of Horndeski interactions and screening mechanisms this disformal coupling has also received much attention [38-42].
For the one loop calculations we will need the Feynman propagators for the Galileon and the scalar field. The Galileon field is a massless scalar field and therefore its Feynman propagator is simply given by
$$G _ { \phi } = \langle \phi ( x _ { 1 } ) \phi ( x _ { 2 } ) \rangle = \int \frac { \mathrm d ^ { 4 } p } { ( 2 \pi ) ^ { 4 } } \frac { e ^ { i p _ { \mu } \left ( x _ { 1 } ^ { \mu } - x _ { 2 } ^ { \mu } \right ) } } { p ^ { 2 } } \,.$$
On the other hand the matter field we consider here is a simple massive scalar field with the mass M
$$G _ { \chi } = \langle \chi ( x _ { 1 } ) \chi ( x _ { 2 } ) \rangle = \int \frac { \mathrm d ^ { 4 } k } { ( 2 \pi ) ^ { 4 } } \frac { e ^ { i k _ { \mu } \left ( x _ { 1 } ^ { \mu } - x _ { 2 } ^ { \mu } \right ) } } { k ^ { 2 } + M ^ { 2 } } \,.$$
Before starting the computation, let us first remind us of the the following useful general property due symmetry when computing integrals over loops:
$$\frac { 1 } { M ^ { 4 } } \int \frac { \mathrm d ^ { 4 } k } { ( 2 \pi ) ^ { 4 } } \frac { k ^ { 2 n } k _ { a _ { 1 } } k _ { b _ { 1 } } \cdots k _ { a _ { m } } k _ { b _ { m } } } { ( k ^ { 2 } + M ^ { 2 } ) ^ { n + m } } = \frac { 1 } { 2 ^ { m } ( m + 1 )! } \delta _ { a _ { 1 } b _ { 1 } \cdots a _ { m } b _ { m } } J _ { n + m } \,.$$
where J n stands for the notation
$$J _ { n } = \frac { 1 } { M ^ { 4 } } \int \frac { \mathrm d ^ { 4 } k } { ( 2 \pi ) ^ { 4 } } \frac { k ^ { 2 n } } { ( k ^ { 2 } + M ^ { 2 } ) ^ { n } } \,.$$
The for our purpose specifically useful cases are mostly
$$\frac { 1 } { M ^ { 4 } } \int \frac { \mathrm d ^ { 4 } k } { ( 2 \pi ) ^ { 4 } } \frac { k ^ { 2 ( n - 1 ) } k _ { \mu } k _ { \nu } } { ( k ^ { 2 } + M ^ { 2 } ) ^ { n } } = \frac { 1 } { 4 } \delta _ { \mu \nu } J _ { n }$$
$$\frac { 1 } { M ^ { 4 } } \int \frac { \partial ^ { j } k } { ( 2 \pi ) ^ { 4 } } \frac { k ^ { 2 ( n - 2 ) } k _ { \mu } k _ { \nu } k _ { \alpha } k _ { \beta } } { ( k ^ { 2 } + M ^ { 2 } ) ^ { n } } = \frac { 1 } { 2 4 } \delta _ { \mu \nu \alpha \beta } J _ { n }$$
$$\frac { 1 } { M ^ { 4 } } \int \frac { \partial ^ { \cdot } k } { ( 2 \pi ) ^ { 4 } } \frac { k ^ { 2 ( n - 3 ) } k _ { \mu } k _ { \nu } k _ { \alpha } k _ { \beta } k _ { \delta } k _ { \gamma } } { ( k ^ { 2 } + M ^ { 2 } ) ^ { n } } = \frac { 1 } { 1 9 2 } \delta _ { \mu \nu \alpha \beta \delta \gamma } J _ { n } \,,$$
$$\delta _ { \mu \nu \alpha \beta } = ( \delta _ { \mu \nu } \delta _ { \alpha \beta } + \delta _ { \mu \alpha } \delta _ { \nu \beta } + \delta _ { \mu \beta } \delta _ { \nu \alpha } )$$
$$\delta _ { \mu \nu \alpha \beta \delta \gamma } = ( \delta _ { \mu \nu } \delta _ { \alpha \beta \delta \gamma } + \delta _ { \mu \alpha } \delta _ { \nu \beta \delta \gamma } + \delta _ { \mu \beta } \delta _ { \alpha \nu \delta \gamma } + \delta _ { \mu \delta } \delta _ { \alpha \beta \nu \gamma } + \delta _ { \mu \gamma } \delta _ { \alpha \beta \delta \nu } ) \.$$
with
In this work we will be interested in the running of the interactions such that we will use dimensional regularization (or equivalently focuse on the log contribution from cutoff regularization). In this case we have the relation
$$J _ { n } = \frac { n ( n + 1 ) } { 2 } J _ { 1 } \,.$$
Another very useful formula that we will be using throughout the paper is the following identity
$$\frac { 1 } { A _ { 1 } ^ { \alpha _ { 1 } } A _ { 2 } ^ { \alpha _ { 2 } } \cdots A _ { n } ^ { \alpha _ { n } } } = \int _ { 0 } ^ { 1 } d x _ { 1 } \cdots d x _ { n } \delta \left ( \sum x _ { i } - 1 \right ) \frac { \prod x _ { i } ^ { \alpha _ { i } - 1 } } { ( \sum x _ { i } A _ { i } ) ^ { \sum \alpha _ { i } } } \frac { \Gamma ( \alpha _ { 1 } + \cdots \alpha _ { n } ) } { \Gamma ( \alpha _ { 1 } ) \cdots \Gamma ( \alpha _ { n } ) } \quad \ ( 2. 2 1 )$$
with the Feynman parameters x i .
## 3 Galileon Loops
The quantum corrections coming from the Galileon self interactions have already been extensively studied in the literature and it was successfully shown that they are protected under quantum corrections [11, 33, 34, 43-45]. First of all, the symmetry of the theory, namely the shift and Galileon symmetry, prevents to generate local operators by loop corrections which breaks explicitly this symmetry. Nevertheless, this does not forbid to generate local operators which are invariant under shift and Galileon transformations, either renormalizing Galileon interactions themselves or generating operators of higher derivative interactions. It is rather the non-renormalization theorem, which ensures that the Galileon interactions themselves are not renormalized and that the higher derivative operators are irrelevant corrections in the regime of validity of the effective field theory ∂ n φ glyph[lessmuch] Λ n +1 3 for n ≥ 3. One can show the nonrenormalization theorem in a straightforward way by realizing that each vertex in an arbitrary Feynman diagram gives rise to interactions with at least one more derivative.
Without loss of generality, let us for a moment concentrate on the cubic Galileon interaction φ E µαρσ E νβ ρσ Φ µν Φ αβ / Λ . 3 3 We can quickly compute the 2-point function contributions coming from this interaction. We will let the field with the two derivatives run in the loop.
Figure 1 . 1-loop contribution to the 2-point correlation function from the cubic Galileon interaction. Solid lines denote the Galileon scalar field.
$$\mathcal { M } ^ { ( 2 p t, 3 v t ) } _ { \Lambda ^ { 3 } _ { 3 } } & = \frac { ( - 1 ) ^ { 2 } } { 2! \Lambda ^ { 6 } _ { 3 } } \, ( 2 \cdot 2 ) \int \frac { \mathrm d ^ { 4 } k } { ( 2 \pi ) ^ { 4 } } \frac { ( 4 k ^ { \alpha } k ^ { \beta } q _ { 1 \mu } q _ { 1 \nu } ( k _ { \alpha } k _ { \beta } q _ { 1 } ^ { \mu } q _ { 1 } ^ { \nu } + k ^ { \mu } q _ { 1 \beta } ( k ^ { \nu } q _ { 1 \alpha } - 2 k _ { \alpha } q _ { 1 } ^ { \nu } ) ) } { ( k ^ { 2 } + q _ { 1 } ^ { 2 } ) } \\ & = \frac { 2 } { \Lambda ^ { 6 } _ { 3 } } \left ( \frac { p _ { 1 } ^ { 8 } } { 8 } ( 3 J _ { 1 } ) \right ) \,, \\ \Re \quad. \quad. \quad. \quad. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cd:. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots. \cdots........................................................................................................................................................................................................ \cdots...................................................................................................................................................................................................$$
The counter term generated by this diagram contains more derivatives per field ( ∂ φ 4 ) 2 / Λ than 6 3 the classical interaction ✷ φ ∂φ ( ) 2 / Λ . 3 3 We can generalize this result to an arbitrary Feynman diagram with this interaction at a given vertex. We can contract an external Galileon field φ with momentum p µ with the Galileon field coming without derivatives in this vertex from the cubic Galileon while letting the other two φ -particles run in the loop with momenta k µ and ( p + k ) µ . The contribution of this vertex to the scattering amplitude is [34]
$$\mathcal { A } \in \int \frac { \mathrm d ^ { 4 } k } { ( 2 \pi ) ^ { 4 } } G _ { k } \, G _ { k + p } \ \mathcal { E } ^ { \mu \alpha \rho \sigma } \mathcal { E } ^ { \nu \beta } _ { \ \rho \sigma } \ k _ { \mu } \, k _ { \nu } \, ( p + k ) _ { \alpha } \, ( p + k ) _ { \beta } \cdots,$$
where G k is the Feynman massless propagator for the Galileon field 2.11. Now, it is a trivial observation that all the terms which are linear in the external momentum E µαρσ E νβ ρσ k k k µ ν α p β as well as all the contributions which are independent of it E µαρσ E νβ ρσ k k k k α β µ ν will cancel owing to the antisymmetric nature of the vertex (carried by the indices in the Levi-Civita symbol). Therefore, the only non-vanishing term will come in with at least two powers of the external Galileon field with momentum p α p β . This is the essence of the non-renormalization theorem of the Galileon interactions. In a similar way, if we contract the external leg with the derivative free field in vertices of quartic and quintic Galileon, the same argument straightforwardly leads to the same conclusion regarding the minimal number of derivatives on external fields.
$$\mathcal { A } \in \int \frac { d ^ { 4 } k } { ( 2 \pi ) ^ { 4 } } G _ { k _ { 1 } } G _ { k _ { 2 } } \cdots \, \mathcal { E } ^ { \mu _ { 1 } \cdots \mu _ { 4 } } \mathcal { E } ^ { \nu _ { 1 } \cdots \nu _ { 4 } } \prod _ { i = 1 } ^ { n - 1 } k _ { \mu _ { i } } k _ { \nu _ { i } } \left ( p + k + \cdots \right ) _ { \mu _ { n } } ( p + k + \cdots ) _ { \mu _ { n } }, \prod _ { j = n + 1 } ^ { 4 } \eta _ { \mu _ { j } \nu _ { j } }$$
There is no counterterm which takes the Galileon form, and the Galileon interactions are hence not renormalized. This guaranties the stability of the Galileon interactions under quantum corrections if one only considers Galileon loops. However, this is not the case when one starts considering couplings to other matter fields. In the following section we will study in great detail the quantum corrections coming from matter loops.
## 4 Matter Loops
In this section we would like to have a look to the quantum corrections coming from matter one-loops. We will compute explicitly the first order loop corrections coming from the tadpole
and two point functions ...etc. and try to generalize the results to the case of n-point functions. In General Relativity the helicity-2 degree of freedom can couple to the matter fields as h µν T µν /M Pl and h µν h αβ T µναβ /M 2 Pl ... etc. Now, if we have an additional propagating scalar degree of freedom as the Galileon field, it can also couple to the matter fields unless we finetune the coupling to be zero. The natural way of coupling this additional scalar field to the matter field is via a conformal coupling φT µ µ /M C and φ T 2 µ α µ α /M 2 NC . . . etc. If this scalar degree of freedom couples to ordinary matter, then it can mediate a fifth force with a long range of interaction which has never been detected in Solar System gravity tests or laboratory experiments. On that account, it is crucial to find ways to hide this extra degree of freedom on small scales. One could fine-tune its coupling to matter which is less satisfactory. Fortunately, there exist alternatives to fine-tunings thanks to the screening mechanisms that allow to hide the scalar field on small scales while being unleashed on large scales to produce cosmological effects. Typical examples of screening mechanisms are Vainshtein, chameleon or symmetron [46-48]. One can basically use the mass term, the coupling to matter or the kinetic term of the scalar field to screen its effect. For the Galileon scalar field it is the Vainshtein mechanism which is at work. Besides these two couplings we will also consider couplings of the form 1 M 2 glyph[star] φ T 2 µ µ suppressed with a different scale M glyph[star] . As we already mentioned Galileon interactions naturally arise in the decoupling limit of massive gravity with a very specific way of coupling to matter. Therefore motivated by massive gravity, we will also consider disformal couplings of the form 1 M 4 D ∂ φ∂ φT µ ν µν between the Galileon field and the matter fields. Finally we will also consider longitudinal couplings of the form 1 M 3 L ∂ ∂ φT µ ν µν .
## 4.1 Conformal coupling
Our main interest in this work is the quantum corrections coming from the matter loops. We will consider one loop corrections where only matter field runs in the loop. In this subsection we will first study the quantum corrections coming from conformal couplings. To start let us have a close look at the 1-loop contributions to the tadpole and 2-point correlation function. The corresponding Feynman diagrams are represented in Fig. 2. We designate by M (1pt) the 1-loop contribution to the tadpole, by M (2pt 3vt) , the 1-loop correction to the 2-point correlation function arising from the cubic vertex φT/M C and by M (2pt 4vt) , µναβ the 1-loop correction to the 2-point correlation function arising from the quartic vertex φ T 2 µ α µ α /M 2 NC and φ T/M 2 2 glyph[star] .
## Zeroth order in external momenta:
To zeroth order in the external momenta we can easily compute the 1-loop contribution to the tadpole from the conformal coupling φT/M C (as represented on the first line of Fig. 2) yielding
$$\mathcal { M } _ { p = 0, M _ { C } } ^ { ( 1 \text{pt} ) } = \frac { ( - 1 ) } { M _ { C } } \int \frac { \mathrm d ^ { 4 } k } { ( 2 \pi ) ^ { 4 } } \frac { ( k ^ { 2 } + 2 M ^ { 2 } ) } { k ^ { 2 } + M ^ { 2 } } = \frac { 1 } { 2 } \frac { M ^ { 4 } } { M _ { C } } J _ { 1 } \,.$$
Figure 2 . 1-loop contributions to the tadpole and 2-point correlation function from the matter coupling. Solid lines denote the Galileon scalar field whereas dashed lines the matter field.
Now turning to the 2-point correlation function, the 1-loop contribution arising for the cubic vertex φT M C to zeroth order in the external momentum is given by
$$\mathcal { M } _ { p = 0, M _ { C } } ^ { ( 2 \text{pt}, 3 \text{vt} ) } & = \frac { ( - 1 ) ^ { 2 } } { 2! M _ { C } ^ { 2 } } \, ( 2 \cdot 2 ) \int \frac { \mathrm d ^ { 4 } k } { ( 2 \pi ) ^ { 4 } } \frac { ( k ^ { 4 } + 4 k ^ { 2 } M ^ { 2 } + 4 M ^ { 4 } ) } { ( k ^ { 2 } + M ^ { 2 } ) ^ { 2 } } \\ & = \frac { 2 } { M _ { C } ^ { 2 } } \left ( - \frac { M ^ { 4 } } { 2 } ( 3 J _ { 1 } ) \right ) = - 3 \frac { M ^ { 4 } } { M _ { C } ^ { 2 } } J _ { 1 } \,, \\. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad..$$
where the terms in bracket in the first line are combinatory factors. Similarly, focusing on the contribution from the quartic vertex, φ T 2 µ α µ α M 2 NC , we obtain
$$\mathcal { M } _ { p = 0, M _ { \text{NC} } } ^ { \text{-}, 4 \text{vt} ) } & = \frac { \text{-} \right ) \text{(2)} } { M _ { \text{NC} } ^ { 2 } } \int \frac { \text{d} ^ { 4 } k } { ( 2 \pi ) ^ { 4 } } \frac { ( 4 k ^ { 2 } + 1 2 M ^ { 2 } ) } { k ^ { 2 } + M ^ { 2 } } \\ & = - \frac { 2 } { M _ { \text{NC} } ^ { 2 } } \left ( - 4 M ^ { 4 } J _ { 1 } \right ). \\ \hat { \ }$$
Finally, from the non-linear conformal coupling φ T 2 M 2 glyph[star] we have a second contribution to the two point function from the quartic vertex
$$\Phi _ { p = 0, M ^ { * } } ^ { ( 2 p t, 4 v t ) } & = \frac { ( - 1 ) \, ( 2 ) } { M ^ { * } } \int \frac { d ^ { 4 } k } {$$
The total 1-loop contribution to the tadpole and 2-point function is thus given by
$$\mathcal { M } ^ { ( 1 \text{pt} ) } = \frac { 1 } { 2 } \frac { M ^ { 4 } } { M _ { \text{C} } } J _ { 1 } \quad & \quad \quad \quad$$
$$\mathcal { M } ^ { ( 2 p t ) } = \mathcal { M } ^ { ( 2 p t, 3 v t ) } + \mathcal { M } ^ { ( 2 p t, 4 v t ) } = M ^ { 4 } \left ( \$$
This corresponds to the following counter-terms at the level of the action
$$\L _ { C T } = - \left ( \mathcal { M } ^ { ( 1 \text{pt} ) } ( \phi ) + \mathcal { M } ^ { ( 2 \text{pt} ) } ( \phi ^ { 2 } ) + \cd$$
$$= - M ^ { 4 } J _ { 1 } \left ( \frac { 1 } { 2 M _ { C } } ( \phi ) + \left ( \frac { - 3 } { M _ { C } ^ { 2 } } + \frac { 8 } { M _$$
(4.9)
At this point it is not necessary to compute the 3-point or higher n-point functions explicitly. From the tadpole and the 2-point function it is already very suggestive that the coupling φT gives rise to appearance of φ φ , 2 , φ 3 , φ 4 terms..etc, potential interactions for φ to zeroth order in the external momenta.(The coupling φT breaks spontaneously the scale invariance). It is a trivial observation that the higher n-point functions will give rise to counter terms of higher potential terms
$$\L _ { C T } = M ^ { 4 } J _ { 1 } \sum _ { n } \left ( \frac { 1 } { M _ { C } ^ { n } } + \frac { 1 } { M _ { N C } ^ { n } } + \frac { 1 } { M _ { * } ^ { n } } \right ) \phi ^ { n }$$
These quantum contributions are all suppressed by their own scales M C , M NC and M glyph[star] and hence the scale at which they become important is when M ≈ M n/ 4 C , M ≈ M n/ 4 NC or M ≈ M n/ 4 glyph[star] respectively. If the scalar field φ is not an arbitrary scalar, but descends from a full-fledged tensor field h µν as in massive gravity, then all these counter terms coming from one-loop matter quantum corrections are Planck mass suppressed
$$\L _ { C T } ^ { M G } = M ^ { 4 } J _ { 1 } \sum _ { n } \frac { 1 } { M _ { P 1 } ^ { n } } \phi ^ { n }$$
and so for a Galileon coming from massive gravity one can ignore these quantum corrections as long as the mass of the external matter field is smaller than Planck mass.
## Non-vanishing external momenta:
Now we would like also to know what kind of operators are generated when we do not put the external momentum to zero. In this case we will not only obtain quantum corrections in form of potential interactions but also derivative interactions. These generated quantum corrections will renormalize the Galileon interactions but also give rise to higher derivative operators with a ghost. From the conformal coupling φT M C there will not be any contribution with non-zero external momenta to the tadpole since the momentum running in the loop is unaware of the external momentum. From the tadpole there will be only a contribution in form of a potential linear in the field which we already computed above. Therefore, the first contribution comes from the 2-point function.
glyph[negationslash]
$$\mathcal { M } _ { p \neq 0, M _ { C } } ^ { ( 2 p t, 3 v t ) } = \frac { ( - 1 ) ^ { 2 } } { 2! M _ { C } ^ { 2 } } \, ( 2 \cdot 2 ) \int \frac { d ^ { 4 } k } { ( 2 \pi ) ^ { 4 } } \frac { ( 4 M ^ { 4 } + k _ { \mu } k ^ { \mu } ( - 4 M ^ { 2 } + k ^ { \nu } ( k _ { \nu } - 2 p _ { \nu } ) ) + k ^ { \mu } p _ { \mu } ( 4 M ^ { 2 } + k ^ { \nu } p _ { \nu } ) ) } { ( k ^ { 2 } + M ^ { 2 } ) \left ( ( p - k ) ^ { 2 } + M ^ { 2 } \right ) }$$
(4.12)
In order to perform this integration, we will use the specific relation of 2.21 with two factors in the denominator
$$\frac { 1 } { A _ { 1 } ^ { \alpha } A _ { 2 } ^ { \beta } } = \int _ { 0 } ^ { 1 } d _ { x } \frac { x ^ { \alpha - 1 } ( 1 - x ) ^ { \beta - 1 } } { ( x A _ { 1 } + ( 1 - x ) A _ { 2 } ) ^ { \alpha + \beta } } \frac { \Gamma ( \alpha + \beta ) } { \Gamma ( \alpha ) \Gamma ( \beta ) }.$$
For the above integration we have α = 1, β = 1, A 1 = k 2 + M 2 and A 2 = ( p -k ) 2 + M 2 . The key point is now to perform a change of variable such that the mixing of the two momentas in ( p -k ) 2 disappears. This is indeed achieved by defining k µ = l µ -( x -1) p µ . In this way, the propagators in the denominator becomes simply ( l 2 +∆ ) 2 2 where ∆ = M 2 + p x 2 (1 -x ) and we can easily perform the integration. After using the relations 2.15 and performing the x -integration we obtain the following contribution glyph[negationslash]
$$\mathcal { M } _ { p \neq 0, M _ { C } } ^ { ( 2 p t, 3 v t ) } = \frac { 2 } { M _ { C } ^ { 2 } } \frac { ( 3 J _ { 1 } ) } { 8 } \left ( - 4 M ^ { 4 } - 2 M ^ { 2 } p ^ { 2 } + p ^ { 4 } \right ) \right )$$
The counter terms arising from the two point function would be thus of the form
$$\L _ { C T } \supset - \frac { 3 J _ { 1 } } { 4 M _ { C } ^ { 2 } } \left ( - 4 M ^ { 4 } ( \phi ) ^ { 2 } + 2 M ^ { 2 } ( \partial \phi ) ^ { 2 } + ( \Box \phi ) ^ { 2 } \right )$$
The first contribution in form of a potential corresponds to the contribution we computed above to zeroth order in external momentum. Now with a non-vanishing external momentum we obtain derivative interactions. The quantum corrections coming from this 2-point function renormalize the kinetic term of the Galileon field but also generates dangerous higher order derivative interactions ( ✷ φ ) 2 . This operator will become important and hence need to be considered at a scale ∂ 2 ∼ M C .
Similarly as the tadpole, the two point function with the quartic vertex (on the last line of Fig. 2 ) coming from the interactions φ T 2 µ α µ α M 2 NC and φ T 2 M 2 glyph[star] will not have any contribution with non-zero external momentum since again the internal momentum in the loop is unaware of the external momentum. At the two point function level the above contribution will be thus the only one. In order to capture better the general behavior of the n-point function contribution, we shall also perform at this stage the contribution of the 3-point function with non-zero external momentum.
In Fig. 3 we see the three different contributions to the 3-point function. The first diagram constitutes three vertices with the conformal coupling φT/M C giving rise to a contribution of the following form glyph[negationslash]
$$\mathcal { M } _ { p \neq 0 } ^ { ( 3 \text{pt}, 3 \text{vt} ) } = \frac { ( - 1 ) ^ { 3 } } { 3! M _ { \text{C} } ^ { 3 } } \left ( 3 \cdot 2 \right ) \left ( 4 \cdot 2 \right ) \int \frac { \text{d} ^ { 4 } k } { ( 2 \pi ) ^ { 4 } } \frac { \mathcal { J } } { \left ( k ^ { 2 } + M ^ { 2 } \right ) \left ( q _ { 1 } ^ { 2 } + M ^ { 2 } \right ) \left ( q _ { 2 } ^ { 2 } + M ^ { 2 } \right ) }$$
Figure 3 . 1-loop contributions to the 3-point correlation function from the matter coupling.
with the momentum conversation relations q 1 = ( p 1 -k ) and q 2 = ( k + p 2 ) and the shortcut notation
$$\mathcal { J } = 8 M ^ { 6 } + 4 M ^ { 4 } q _ { 1 } ^ { \mu } q _ { 2 \mu } + k ^ { \mu } ( 2 M ^ { 2 } q _ { 2 \mu } ( 2 M ^ { 2 } + q _ { 1 } ^ { \nu } q _ { 2 \nu } ) + q _ { 1 \mu } ( 4 M ^ { 4 } + q _ { 2 \nu } ( 2 M ^ { 2 } ( k ^ { \nu } + q _ { 1 } ^ { \nu } ) + k ^ { \nu } q _ { 1 } ^ { \alpha } q _ { 2 \alpha } ) ) ) \\ _ { - } \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \colon$$
In order to perform this integration involving a denominator with three factors we again use equation 2.21
$$\Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \, \Psi \,
\quad \frac { 1 } { A _ { 1 } A _ { 2 } A _ { 3 } } = \int _ { 0 } ^ { 1 } d x _ { 1 } d x _ { 2 } \frac { ( 3 - 1 )! } { ( x _ { 1 } A _ { 1 } + x _ { 2 } A _ { 2 } + ( 1 - x _ { 1 } - x _ { 2 } ) A _ { 3 } ) ^ { 3 } }$$
Similarly as we did for the 2-point function, we use the trick of completing the square by shifting the integration variable k to k µ = l µ -( p 2 µ -p 2 µ x 1 -p 1 µ x 2 -p 2 µ x 2 ) in order to absorb the terms which are linear in p 1 and p 2 . The denominator then again simply become ( l 2 +∆ ) 2 3 with this time ∆ = M 2 -p 2 2 ( x 1 -1) x 1 +( p 1 + p 2 )( p 1 + p 2 -2 p x 2 1 ) x 2 -( p 1 + p 2 ) 2 x 2 2 . We can now perform the integration over the momentum l µ and the Feynman parameters x 1 and x 2 . This results in glyph[negationslash]
$$\mathcal { M } _ { p \neq 0 } ^ { ( 3 p t, 3 v t ) } & = \frac { - ( 6 J _ { 1 } ) } { 6 M _ { C } ^ { 3 } } \left ( 1 9 2 M ^ { 4 } + 3 p _ { 1 } ^ { 4 } + 4 p _ { 1 } ^ { 2 } p _ { 2 } ^ { 2 } + 1 3 2 p _ { 2 } ^ { 4 } + 1 8 M ^ { 2 } ( p _ { 1 } ^ { 2 } - 1 1 p _ { 2 } ^ { 2 } ) \\ & \quad + 2 p _ { 1 } \cdot p _ { 2 } ( 3 ( - 4 1 M ^ { 2 } + p _ { 1 } ^ { 2 } + 4 4 p _ { 2 } ^ { 2 } ) + 6 7 p _ { 1 } \cdot p _ { 2 } ) \right )$$
In an analog way, to the second diagram in fig. 3 the interactions φ T 2 µ α µ α M 2 NC and φT/M C will contribute at the respective vertex yielding glyph[negationslash]
$$\mathcal { M } _ { p \neq 0 } ^ { ( 3 p t, 4 v t - 3 v t ) } & = \frac { ( - 1 ) ^ { 2 } } { 2! M _ { \mathbb { N } C } ^ { 2 } M _ { C } } \left ( 3 \right ) ( 2 ) \int \frac { d ^ { 4 } k } { ( 2 \pi ) ^ { 4 } } \frac { 4 \left ( 6 M ^ { 4 } + k ^ { \mu } ( p _ { 1 \mu } + p _ { 2 \mu } - k _ { \mu } ) ( 5 M ^ { 2 } + k ^ { \nu } ( p _ { 1 \mu } + p _ { 2 \mu } - k _ { \nu } ) ) \right ) } { \left ( k ^ { 2 } + M ^ { 2 } \right ) \left ( ( p _ { 1 } + p _ { 2 } - k ) ^ { 2 } + M ^ { 2 } \right ) } \\ & = \frac { 3 J _ { 1 } } { 2 M _ { \mathbb { N } C } ^ { 2 } M _ { C } } \left ( 2 2 8 M ^ { 4 } + 1 4 7 p _ { 1 } ^ { 4 } - 5 8 p _ { 1 } ^ { 2 } p _ { 2 } ^ { 2 } + 3 p _ { 2 } ^ { 4 } + 4 8 M ^ { 2 } ( - 1 0 p _ { 1 } ^ { 2 } + p _ { 2 } ^ { 2 } ) \\ & \quad + 4 p _ { 1 } \cdot p _ { 2 } ( 3 ( 8 M ^ { 2 } - 7 p _ { 1 } ^ { 2 } + p _ { 2 } ^ { 2 } ) + 7 p _ { 1 } \cdot p _ { 2 } ) \right )$$
From the 2- and 3- point function one clearly sees that the one loop contributions are always proportional to M 4 , M ∂ 2 2 or ∂ 4 . The contributions proportional to M 4 we can ignore here at this stage since they correspond to the contributions from 4.22 in the zero external momentum limit. On the other hand the contributions with the scaling M ∂ 2 2 will be harmless in the sense that they will give rise to few derivatives per field. The dangerous terms are hence the terms
with higher derivatives per field. Thus, the coupling to matter will give rise to counter terms with the following dangerous higher order derivatives acting on the Galileon field.
$$\L _ { C T } & = - \left ( \mathcal { M } ^ { ( 2 p t ) } ( \phi ^ { 2 } ) + \mathcal { M } ^ { ( 3 p t ) } ( \phi ^ { 3 } ) \cdots \right ) \\ & \quad \supset M ^ { 2 } \frac { ( \partial \phi ) ^ { 2 } } { M _ { C } ^ { 2 } } + \frac { ( \partial ^ { 2 } \phi ) ^ { 2 } } { M _ { C } ^ { 2 } } + \phi \frac { ( \partial ^ { 2 } \phi ) ^ { 2 } } { M _ { \text{NC} } ^ { 3 } } + \frac { \Box \phi ( \partial \phi ) ^ { 2 } } { M _ { \text{NC} } ^ { 2 } M _ { C } } + \phi \frac { ( \partial ^ { 2 } \phi ) ^ { 2 } } { M _ { \text{NC} } ^ { 2 } M _ { C } } + \phi \frac { ( \partial _ { \mu } \partial _ { \nu } \phi ) ^ { 2 } } { M _ { \text{C} } ^ { 3 } } \cdots ( 4. 2 1 ) \\ \text{----} \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \ \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \ } \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \ \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \,. \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \$$
The quantum corrections coming from the 2-point function renormalizes the kinetic term with a scale M /M 2 2 C , whereas the 3-point functions renormalizes the cubic Galileon interaction with a scale 1 /M 3 C and 1 /M 2 NC M C respectively. Therefore, the scale of the interactions need to be chosen above the scale Λ 3 below which the Galileon interactions are important. For scales M C ≈ Λ , the nice structure of the Galileon interaction would be detuned at an unacceptable 3 scale. For the restricted Galileon field coming from massive gravity, the scale at which this ghost would become accessible is close to the Planck scale. The higher n-point functions would give rise to counter terms of the form
$$\L _ { C T } = J _ { 1 } \sum _ { n } \left ( \frac { \partial ^ { 2 } \phi } { M _ { \text{Pl} } } \right ) ^ { n }$$
Near sources these quantum corrections are even more suppressed due to the Vainshtein mechanism ( ✷ φ ) n -2 / M ( n Pl Z ). In massive gravity this is completely harmless since all these quantum corrections will start playing a role close to the Planck mass M Pl .
## 4.2 Disformal coupling
Wewould like now to draw our attention to the derivative coupling of the form ∂ φ∂ φT µ ν µν /M 4 D . This coupling has to be considered in the context of massive gravity. There it arises naturally after shifting the metric perturbations by h µν → φη µν + ∂ φ∂ φ µ ν in order to diagonalize some of the interactions between the helicity-2 and helicity-0 degrees of freedom. In massive gravity this derivative disformal coupling brings along important consequences. It plays a very crucial role for the existence of degravitating solutions [10] which is absent in the usual Galileon theories. Furthermore, it plays an important role in the lensing measurements since photons can now couple to the scalar degree of freedom [49].
We would like now to study the quantum corrections coming from this derivative coupling. To zeroth order in the external momentum this coupling will give zero contributions and hence the potential terms from 4.22 will not be generated. Also since already two φ fields appear in the coupling, there will not be any contribution to the tadpole either, or to any odd n-point function. Thus, there will be only contributions to the even number n-point functions. The first contribution will be to the 2-point function tadpole depicted in the last line of fig. 2.
$$\mathcal { M } _ { D } ^ { ( 2 p t, 4 v t ) } = \frac { ( - 1 ) ^ { 1 } ( 2 ) p _ { \mu } p _ { \nu } } { M _ { D } ^ { 4 } } \int \frac { d ^ { 4 } k } { ( 2 \pi ) ^ { 4 } } \frac { ( - k ^ { \mu } k ^ { \nu } + \frac { 1 } { 2 } \eta ^ { \mu \nu } ( k ^ { 2 } + M ^ { 2 } ) ) } { k ^ { 2 } + M ^ { 2 } } = \frac { M ^ { 4 } } { 4 M _ { D } ^ { 4 } } p ^ { 2 } \. \quad ( 4. 2 3 )$$
which would again simply renormalize the kinetic term of the Galileon scalar field φ . As next, let us have a look at the next leading contribution coming from the 4-point function.
Figure 4 . One-loop contributions to the 4-point function coming from the derivative coupling.
$$\mathcal { M } _ { D } ^ { ( 4 p t, 4 v t ) } = \frac { ( - 1 ) ^ { 2 } ( 4 \cdot 2 \cdot 2 ) p ^ { 1 \mu } p ^ { 2 \nu } p ^ { 3 \alpha } p ^ { 4 \beta } } { 2! M _ { \text{D} } ^ { 8 } } \int \frac { \text{d} ^ { 4 } k } { ( 2 \pi ) ^ { 4 } } \frac { 1 } { 4 } \frac { \mathcal { H } } { ( k ^ { 2 } + M ^ { 2 } ) ( M ^ { 2 } + ( p _ { 1 } + p _ { 2 } - k ) ^ { 2 } ) } \, ( 4. 2 4 )$$
where the shortcut H stands for the following tensor
$$\mathcal { H } & = - 2 k _ { \mu } q _ { 1 \nu } ( - 2 k _ { \alpha } q _ { 1 \beta } + g _ { \alpha \beta } ( M ^ { 2 } + k \cdot q _ { 1 } ) ) + g _ { \mu \nu } ( - 2 k _ { \alpha } q _ { 1 \beta } ( M ^ { 2 } + k \cdot q _ { 1 } ) \\ & \quad + g _ { \alpha \beta } ( M ^ { 4 } + k \cdot q _ { 1 } ( 2 M ^ { 2 } + k \cdot q _ { 1 } ) ) )$$
The energy conservation forces q 1 = p 1 + p 2 -k and p 4 = p 1 + p 2 -p 3 . After performing the integral we obtain the long expression
$$1$$
$$\begin{array} { l } \text{integral we obtain the long expression} \\ \mathcal { M } _ { p } ^ { ( 4 p t ) } = \frac { 1 } { 6 0 M _ { D } ^ { \delta } } \left ( - 2 p _ { 1 } ^ { 2 } p _ { 2 } ^ { 2 } ( 1 0 M ^ { 2 } + 3 ( p _ { 1 } ^ { 2 } + p _ { 2 } ^ { 2 } ) ) p _ { 3 } \cdot p _ { 4 } - 2 p _ { 2 } ^ { 2 } ( 1 0 M ^ { 2 } + p _ { 1 } ^ { 2 } - 3 p _ { 2 } ^ { 2 } ) ( p _ { 2 } \cdot p _ { 3 } ) p _ { 2 } \cdot p _ { 4 } \\ + p _ { 1 } ^ { \alpha } ( p _ { 2 } ^ { \beta } ( ( 3 0 M ^ { 4 } - 5 p _ { 1 } ^ { 4 } - 2 p _ { 1 } ^ { 2 } p _ { 2 } ^ { 2 } - 5 p _ { 2 } ^ { 4 } - 3 0 M ^ { 2 } ( p _ { 1 } ^ { 2 } + p _ { 2 } ^ { 2 } ) ) p _ { 3 \alpha } p _ { 4 \beta } \\ + p _ { 3 \beta } ( ( 3 0 M ^ { 4 } + 5 p _ { 1 } ^ { 4 } + 1 8 p _ { 1 } ^ { 2 } p _ { 2 } ^ { 2 } + 5 p _ { 2 } ^ { 4 } + 3 0 M ^ { 2 } ( p _ { 1 } ^ { 2 } + p _ { 2 } ^ { 2 } ) ) p _ { 4 \alpha } - 4 ( 1 0 M ^ { 2 } + p _ { 1 } ^ { 2 } ) p _ { 2 \alpha } p _ { 2 } \cdot p _ { 4 } ) \right ) \\ + 2 ( ( 1 0 5 M ^ { 4 } - 6 p _ { 1 } ^ { 2 } p _ { 2 } ^ { 2 } + 2 0 M ^ { 2 } ( p _ { 1 } ^ { 2 } + p _ { 2 } ^ { 2 } ) ) p _ { 2 \alpha } p _ { 3 } \cdot p _ { 4 } + p _ { 1 } ^ { \beta } ( - p _ { 2 } ^ { 2 } ( 1 0 M ^ { 2 } - 3 p _ { 1 } ^ { 2 } + p _ { 2 } ^ { 2 } ) p _ { 3 \alpha } p _ { 4 \beta } \\ + p _ { 2 \alpha } ( ( - 2 ( 1 0 M ^ { 2 } + p _ { 2 } ^ { 2 } ) p _ { 1 } ^ { 2 } p _ { 3 \beta } + ( 5 0 M ^ { 2 } + 3 ( p _ { 1 } ^ { 2 } + p _ { 2 } ^ { 2 } ) ) p _ { 2 \beta } p _ { 3 } ^ { 3 } + 8 p _ { 1 } ^ { 1 } p _ { 2 \beta } p _ { 2 } \cdot p _ { 3 } ) p _ { 4 \gamma } \\ - 2 p _ { 1 } ^ { 2 } p _ { 2 \beta } ( 2 ( p _ { 1 } ^ { \delta } + 3 p _ { 2 } ^ { \delta } ) p _ { 3 \gamma } - 3 p _ { 2 \gamma } p _ { 3 } ^ { \delta } ) p _ { 4 \delta } + p _ { 2 } ^ { \gamma } ( - ( 4 0 M ^ { 2 } + 9 ( p _ { 1 } ^ { 2 } + p _ { 2 } ^ { 2 } ) ) p _ { 3 \beta } p _ { 4 \gamma } \\ + p _ { 3 \gamma } ( ( 2 0 M ^ { 2 } + 1 1 ( p _ { 1 } ^ { 2 } + p _ { 2 } ^ { 2 } ) ) p _ { 4 \beta } - 4 p _ { 2 \beta } p _ { 2 } \cdot p _ { 4 } ) ) ) ) ) ) \end{array} \right ) \\ \text{(4.26)} \\ \text{Even if the above contribution is a very long expression we can still extract useful information.}$$
Even if the above contribution is a very long expression we can still extract useful information. The derivative disformal coupling gives counter terms of the following form
$$\L _ { C T } = - \left ( \mathcal { M } _ { D } ^ { ( 2 p t ) } ( \phi ^ { 2 } ) + \mathcal { M } _ { D } ^ { ( 4 p t ) } ( \phi ^ { 4 } ) + \cdots \right ) \supseteq \frac { M ^ { 4 } } { M _ { D } ^ { 4 } } ( \partial \phi ) ^ { 2 } + \frac { M ^ { 2 } } { M _ { D } ^ { 8 } } ( \Box \phi ) ^ { 2 } ( \partial \phi ) ^ { 2 } \frac { 1 } { M _ { D } ^ { 8 } } \Box \phi \partial ^ { 4 } \phi ( \partial \phi ) ^ { 2 } + \cdots$$
We see immediately, that the kinetic term and the quartic Galileon interactions get renormalized. Besides that, the quantum corrections generate counter terms with too many derivatives acting per field which do not belong to the Galileon class of interactions and contain a ghost degree of freedom. The kinetic term receives a quantum correction scaling as M /M 4 4 D which becomes important for M ∼ M D . The quartic Galileon on the other hand gets renormalized by an operator which scales as M /M 2 8 D . The higher derivative operator remains negligible as long as ∂ /M 4 8 D << Λ . 4 3 Again in the case of massive gravity these contributions would be harmless since they would be Planck mass suppressed instead of the M D suppressed.
## 4.3 Longitudinal coupling
Even if longitudinal couplings of the form 1 M 3 L ∂ ∂ φT µ ν µν would vanish for conserved sources and do not originate from theories like massive gravity, it is worth studying quickly the counter terms arising from this type of couplings. The scalar field comes already with two derivatives in this longitudinal coupling and we expect that the counter terms will result in operators with higher derivatives per field. The easiest contribution to compute is the tadpole contribution
$$\mathcal { M } ^ { ( 1 \text{pt} ) } & = \frac { ( - 1 ) } { M _ { L } ^ { 3 } } p _ { \mu } p _ { \nu } \int \frac { \mathrm d ^ { 4 } k } { ( 2 \pi ) ^ { 4 } } \frac { \left ( - k ^ { \mu } k ^ { \nu } + \frac { 1 } { 2 } \eta ^ { \mu \nu } ( k ^ { 2 } + M ^ { 2 } ) \right ) } { k ^ { 2 } + M ^ { 2 } } \\ & = \frac { M ^ { 4 } } { 8 M _ { L } ^ { 3 } } p ^ { 2 } \\ & \quad. \quad \, _ { \cdot }$$
Thus, the tadpole already generates a term of the form ( ✷ φ M /M ) 4 3 L . The contribution to the 2-point function can also be performed easily glyph[negationslash]
$$\mathcal { M } _ { p \neq 0 } ^ { ( 2 p t, 3 v t ) } = \frac { ( - 1 ) ^ { 2 } } { 2! M _ { L } ^ { 6 } } \left ( 2 \cdot 2 \right ) p ^ { \mu } p ^ { \nu } p ^ { 2 \alpha } p ^ { 2 \beta } \int \frac { \mathrm d ^ { 4 } k } { ( 2 \pi ) ^ { 4 } } \frac { 1 } { 4 } \frac { \mathcal { H } } { ( k ^ { 2 } + M ^ { 2 } ) \left ( q _ { 1 } ^ { 2 } + M ^ { 2 } \right ) }$$
where H is given by 4.25 and where q 1 = ( k -p ) and p 2 = p because of momentum conversation. After performing the integration this gives glyph[negationslash]
$$\mathcal { M } _ { p \neq 0 } ^ { ( 2 \text{pt}, 3 \text{vt} ) } = \frac { 2 } { M _ { L } ^ { 6 } } ( 3 J _ { 1 } ) \frac { 9 M ^ { 4 } } { 1 6 } p ^ { 4 }$$
The longitudinal coupling gives rise to quantum corrections with the following counter terms glyph[negationslash]
$$\L _ { C T } = - \left ( \mathcal { M } _ { p \neq 0 } ^ { ( 1 \text{pt}, 3 \text{vt} ) } ( \phi ) + \mathcal { M } _ { p \neq 0 } ^ { ( 2 \text{pt}, 3 \text{vt} ) } ( \phi ^ { 2 } ) + \cdots \right ) \supset \frac { M ^ { 4 } } { M _ { L } ^ { 3 } } \Box \phi + \frac { M ^ { 4 } } { M _ { L } ^ { 6 } } ( \Box \phi ) ^ { 2 } + \cdots \quad \ \ ( 4. 3 1 )$$
glyph[negationslash]
As you can see, this longitudinal coupling naturally generates higher derivatives acting per scalar field becoming important at a scale close to M ∼ M L .
## 4.4 Mixed couplings
We can also quickly study an example of Feynman diagrams in which both the conformal coupling and the derivative couplings contribute at the same time. These diagrams mix the scales of the two couplings, therefore the counter terms of the n-point function will be suppressed by powers of ( M M 4 D C ) n and ( M M 3 L C ) n respectively. For our purpose, it will be enough to study the first contributions of these mixed Feynman diagrams. Let us first start with the 3-point function contribution coming from the second diagram depicted in fig. 3 where at the one vertex the coupling φT/M C and the other vertex the coupling ∂ φ∂ φT µ ν µν /M 4 D have to be considered.
$$\mathcal { M } ^ { ( 3 p t ) } = \frac { ( - 1 ) ^ { 2 } ( 3 \cdot 2 ) ( 2 \cdot 2 ) p _ { 1 } ^ { \mu } p _ { 2 } ^ { \nu } } { 2! M _ { \text{L} } ^ { 4 } M _ { \text{C} } } \int \frac { d ^ { 4 } k } { ( 2 \pi ) ^ { 4 } } \frac { 1 } { 2 } \frac { ( - 2 k _ { \mu } q _ { 1 \nu } ( 2 M ^ { 2 } + k \cdot q _ { 1 } ) + g _ { \mu \nu } ( 2 M ^ { 4 } + k \cdot q _ { 1 } ( 3 M ^ { 2 } + k \cdot q _ { 1 } ) ) ) } { ( k ^ { 2 } + M ^ { 2 } ) ( M ^ { 2 } + ( p _ { 1 } + p _ { 2 } - k ) ^ { 2 } ) } \\ \dots \quad \cdot \quad \ e \quad \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \end{cases}$$
We can very easily perform this integral. This diagram gives the following contribution
$$\mathcal { M } ^ { ( 3 p t ) } & = \frac { 1 2 } { M _ { N L } ^ { 4 } M _ { C } } \frac { 1 } { 2 4 } \left ( - p _ { 1 } ^ { 2 } p _ { 2 } ^ { 2 } ( 6 M ^ { 2 } + p _ { 1 } ^ { 2 } + p _ { 2 } ^ { 2 } ) + p _ { 1 } \cdot p _ { 2 } ( 4 5 M ^ { 4 } - 2 p _ { 1 } ^ { 2 } p _ { 2 } ^ { 2 } + 6 M ^ { 2 } ( p _ { 1 } ^ { 2 } + p _ { 2 } ^ { 2 } ) \\ & \quad + p _ { 1 } \cdot p _ { 2 } ( 1 8 M ^ { 2 } + p _ { 1 } ^ { 2 } + p _ { 2 } ^ { 2 } + 2 p _ { 1 } \cdot p _ { 2 } ) ) \right )$$
Similarly, the contribution to the 2-point function with the one vertex being the coupling ∂ ∂ φT µ ν µν /M 3 L and the other vertex being the the coupling φT/M C gives
$$\mathcal { M } ^ { ( 2 p t ) } = \frac { ( - 1 ) ^ { 2 } ( 2 ) ( 2 \cdot 2 ) } { 2! M _ { L } ^ { 3 } M _ { C } } \frac { M ^ { 2 } } { 8 } p ^ { 2 } ( 1 5 M ^ { 2 } + 2 p ^ { 2 } )$$
Summarizing, the Feynman diagrams with the vertices of mixed couplings to the matter field generate in a very similar way contributions as we were obtaining from the purely conformal or disformal coupling diagrams
$$L _ { C T } & = - \left ( \mathcal { M } ^ { ( 2 p t ) } ( \phi ^ { 2 } ) + \mathcal { M } ^ { ( 3 p t ) } ( \phi ^ { 3 } ) \cdots \right ) \supset \frac { M ^ { 4 } ( \partial \phi ) ^ { 2 } } { M _ { L } ^ { 3 } M _ { C } } + \frac { M ^ { 2 } ( \Box \phi ) ^ { 2 } } { M _ { L } ^ { 3 } M _ { C } } + \frac { M ^ { 2 } ( \partial \phi ) ^ { 2 } \Box \phi } { M _ { D } ^ { 4 } M _ { C } } + \frac { ( \partial \phi ) ^ { 2 } ( \Box \phi ) ^ { 2 } } { M _ { D } ^ { 4 } M _ { C } } + \cdots \\ \psi \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \,.$$
It is a trivial observation that the contributions to the 3-point function take the form of Galileon interactions as well as new derivatives interactions not embedded in the Galileon interactions as we were obtaining above with the difference that this time the kinetic term scales as M / M M 4 ( 3 L C ), the cubic Galileon as M / M M 2 ( 4 D C ) and the higher order derivative operators as 1 /M M 4 D C .
## 5 Stability of the couplings
Another very interesting question is the stability of the classical couplings. If the couplings themselves receive large quantum corrections, it will consequently affect the results we presented above. In order to illustrate the impact of the quantum corrections of the couplings
themselves, let us focus on the conformal coupling φT M C . The question we want to quickly address is whether or not this coupling receives large quantum corrections. In order to study this question we will consider two specific diagrams with the coupling φT M C at the vertices and also the third Galileon interactions. The for us interesting diagrams are depicted in Fig. 5, where in the first diagram the conformal coupling φT M C acts on the vertices, whereas in the second diagram the third Galileon interaction acts on the second vertex.
Figure 5 . One-loop contributions to the 3-point function renormalizing the coupling itself.
The first diagram contributes to 〈 φχχ 〉 through the following integration
$$\mathcal { M } _ { 1 } ^ { \phi \chi \chi } = \frac { ( - 1 ) ^ { 3 } } { 3! M _ { C } ^ { 3 } } ( 3 ) ( 4 \cdot 2 \cdot 2 ) \int \frac { d ^ { 4 } k } { ( 2 \pi ) ^ { 4 } } \frac { \mathcal { W } } { ( k ^ { 2 } + M ^ { 2 } ) ( q _ { 1 } ^ { 2 } + M ^ { 2 } ) ( q _ { 2 } ^ { 2 } ) }$$
where W stands for
$$\mathcal { W } = 8 M ^ { 6 } + 4 M ^ { 4 } p _ { 3 } ^ { \mu } q _ { 1 \mu } + k ^ { \mu } ( 2 M ^ { 2 } q _ { 1 \mu } ( 2 M ^ { 2 } + p _ { 3 } ^ { \nu } q _ { 1 \nu } + p _ { 2 \mu } ( 4 M ^ { 4 } + q _ { 1 \nu } ( 2 M ^ { 2 } ( k ^ { \nu } + p _ { 3 } ^ { \nu } ) + k ^ { \nu } p _ { 3 } ^ { \alpha } q _ { 1 \alpha } ) ) ) \\ - \cdot \cdot \cdot \cdot \cdot \cdot$$
The conservation of energy dictates q 1 µ = p 1 µ -k µ , q 2 µ = p 2 µ + k µ and p 3 µ = p 1 µ + p 2 µ . Note that now in the propagator with the momentum q 2 the mass does not appear since it corresponds to the φ field running in the loop. Therefore, when we use the trick 4.18 in order to write the denominator as ( l 2 +∆ ) , an important difference will be also in the ∆ 2 3
$$\Delta ^ { 2 } = M ^ { 2 } ( x + x _ { 2 } ) - x _ { 2 } ^ { 2 } ( p _ { 1 } + p _ { 2 } ) ^ { 2 } + ( p _ { 1 } + p _ { 2 } ) ( p _ { 1 } + p _ { 2 } - 2 p _ { 2 } x ) x _ { 2 } - p _ { 2 } ^ { 2 } x ( - 1 + x ) \quad \ ( 5. 3 )$$
We can now perform the integration in l and the Feynman parameters easily and find
$$\mathcal { M } _ { 1 } ^ { \phi \chi \chi } & = \frac { ( - 8 ) } { M _ { C } ^ { 3 } } \frac { ( 3 J _ { 1 } ) } { 8 } \left ( - 1 6 M ^ { 4 } + 2 p _ { 2 } ^ { 4 } - M ^ { 2 } ( 4 p _ { 1 } ^ { 2 } + 5 p _ { 2 } ^ { 2 } ) + p _ { 1 } \cdot p _ { 2 } ( - 1 3 M ^ { 2 } - p _ { 1 } ^ { 2 } + 4 p _ { 2 } ^ { 2 } + p _ { 1 } \cdot p _ { 2 } ) \right ) \\ \cdot \quad \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot$$
In a similar way, we can compute the second diagram in Fig. 5. At the two vertices the conformal coupling φT/M C acts, whereas on the third vertex the cubic Galileon interaction φ E µναβ E ρσ αβ Φ µρ Φ νσ / Λ . The contribution of this diagram is given by 3 3
$$\mathcal { M } _ { 2 } ^ { \phi \chi \chi } = \frac { ( - 1 ) ^ { 3 } } { 3! M _ { \mathbb { C } } ^ { 2 } \Lambda _ { 3 } ^ { 3 } } ( 5 ) ( 2 ) ( 4 \cdot 2 \cdot 2 ) \int \frac { \mathrm d ^ { 4 } k } { ( 2 \pi ) ^ { 4 } } \frac { \mathcal { W } _ { 2 } } { ( k ^ { 2 } ) ( q _ { 1 } ^ { 2 } ) ( q _ { 2 } ^ { 2 } + M ^ { 2 } ) }$$
with this time
$$\mathcal { W } _ { 2 } & = 2 k ^ { \mu } ( k _ { \mu } ( 4 M ^ { 4 } q _ { 1 \nu } q _ { 1 } ^ { \nu } + q _ { 2 \nu } ( 2 M ^ { 2 } ( p _ { 2 } ^ { \nu } + p _ { 3 } ^ { \nu } ) q _ { 1 \alpha } q _ { 1 } ^ { \alpha } + p _ { 2 } ^ { \nu } p _ { 3 } ^ { \alpha } q _ { 1 \beta } q _ { 1 } ^ { \beta } q _ { 2 \alpha } ) ) \\ & \quad - k ^ { \nu } q _ { 1 \mu } q _ { 1 \nu } ( 4 M ^ { 4 } + q _ { 2 \alpha } ( 2 M ^ { 2 } ( p _ { 2 } ^ { \alpha } + p _ { 3 } ^ { \alpha } ) + p _ { 2 } ^ { \alpha } p _ { 3 } ^ { \beta } q _ { 2 \beta } ) ) )$$
Two φ -fields are running in the loops, therefore we have two massless propagators with momenta k µ and q 1 µ . We again complete the square by shifting the integration variable k to k µ = l µ -( p 2 µ -p 2 µ x 1 -p 1 µ x 2 -p 2 µ x 2 ) and so absorb the linear terms in p 1 and p 2 . The denominator again takes the desired form ( l 2 +∆ ) 2 3 with this time ∆ = -p 2 2 ( -1 + x 1 ) x 1 + ( p 1 + p 2 )( p 1 + p 2 -2 p x 2 1 ) x 2 -( p 1 + p 2 ) 2 x 2 2 -M 2 ( -1+ x 1 + x 2 ). After performing the integration over the momentum l µ and the Feynman parameters x 1 and x 2 we end up with
$$\mathcal { M } _ { 2 } ^ { \phi \chi \chi } & = \frac { ( - 8 0 ) } { 3 M _ { C } ^ { 2 } \Lambda _ { 3 } ^ { 3 } } \frac { ( 3 J _ { 1 } ) } { 1 6 } \left ( p _ { 1 } ^ { 2 } ( 4 8 M ^ { 4 } + 6 p _ { 1 } ^ { 2 } p _ { 2 } ^ { 2 } + 1 9 p _ { 2 } ^ { 4 } + 1 2 M ^ { 2 } ( p _ { 1 } ^ { 2 } + 4 p _ { 2 } ^ { 2 } ) ) \\ & \quad + p _ { 1 } \cdot p _ { 2 } ( p _ { 1 } ^ { 2 } ( 4 8 M ^ { 2 } + 3 p _ { 1 } ^ { 2 } + 3 4 p _ { 2 } ^ { 2 } ) + 4 p _ { 1 } \cdot p _ { 2 } ( 3 p _ { 1 } ^ { 2 } - p _ { 2 } ^ { 2 } - p _ { 1 } \cdot p _ { 2 } ) ) \right )$$
From these two diagrams we immediately observe that the conformal coupling itself receives quantum correction in terms of the following counter terms
$$\L _ { C T } = & - \left ( \mathcal { M } _ { 1 } ^ { \phi \chi \chi } ( \phi \chi ^ { 2 } ) + \mathcal { M } _ { 2 } ^ { \phi \chi \chi } ( \phi \chi ^ { 2 } ) + \cdots \right ) \\ & \supset \frac { M ^ { 4 } } { M _ { C } ^ { 3 } } \phi \chi ^ { 2 } + \frac { M ^ { 2 } } { M _ { C } ^ { 3 } } \phi \partial _ { \mu } \chi \partial ^ { \mu } \chi + \frac { 1 } { M _ { C } ^ { 3 } } \phi \partial _ { \mu } \partial _ { \nu } \chi \partial ^ { \mu } \chi + \frac { M ^ { 4 } } { M _ { C } ^ { 2 } A _ { 3 } ^ { 3 } } \phi \partial _ { \mu } \chi \partial ^ { \mu } \chi + \frac { M ^ { 2 } } { M _ { C } ^ { 2 } A _ { 3 } ^ { 3 } } ( \Box \phi ) \partial _ { \mu } \chi \partial ^ { \mu } \chi + \cdots$$
As you can see, the conformal coupling φT/M C receives quantum corrections that scale as M /M 4 3 C for the mass term and M /M 2 3 C for the kinetic term respectively. For masses of the scalar field χ close to M ∼ M C these quantum corrections to the conformal coupling become very important. The quantum corrections give also rise to couplings with more derivatives of the form φ∂ ∂ χ∂ µ ν µ ∂ χ ν .
## 6 Summary and discussion
Historically, Galileon interactions were discovered by the attempt of generalizing the interactions of the decoupling limit of DGP model [11]. They were constructed by insisting on the symmetry of the helicity-0 mode φ of the DGP model, namely the invariance under internal Galilean and shift transformations. From the perspectives of the higher dimensional induced gravity braneworld models these symmetries can be regarded as residuals of the 5-dimensional Poincar´ e invariance. The invariance under these transformations together with the postulate of ghost-absence restrict the construction of the effective φ Lagrangian. One can construct only five derivative interactions which fulfill these conditions as descendants of the Lovelock
invariants in the bulk of generalized braneworld models [15-17]. The Galileon interactions share a very important property: they do not get renormalized by the quantum corrections arising from the Galileon interactions themselves. It is a common misconception in the literature that the non-renormalization theorem is due to the symmetry of the theory. The symmetry guarantees only that there will not be any quantum corrections in form of potential interactions but the symmetry does not prohibit the generation of the Galileon interactions by the quantum corrections. Maybe it would be more fair to say that the fact that the symmetry is not realized exactly, meaning that the symmetry is fulfilled only up to total derivatives, plays a crucial role for the non-renormalization theorem.
Nevertheless, the non-renormalization theorem does not survive when one includes couplings to matter. One can couple the Galileon scalar field in three different ways to the matter fields: through linear or non-linear conformal coupling, through disformal coupling and finally through a longitudinal coupling. We have considered these three possible ways of coupling the Galileon to the matter field and computed the one-loop quantum corrections to study the counter terms arising from these couplings. Starting with the conformal coupling φT , we show how the one-loop quantum contributions give rise to potential interactions proportional to the mass of the matter field and suppressed by the scale of the coupling to zeroth order in the external momenta M φ /M 4 n n C . Including the contributions with non-vanishing external momentum we observe how the dangerous operators with higher operators per field are generated ( ✷ φ ) 2 /M 2 C . Another important class of couplings is the disformal coupling ∂ φ∂ φT µ ν µν which arises naturally in the context of massive gravity. Similarly as in the case of conformal couplings, the quantum corrections from this disformal coupling gives rise to higher derivative operators. The quantum corrections coming from a longitudinal coupling ∂ ∂ φT µ ν µν share the same destiny. Furthermore, we studied the interesting question whether or not the covariant couplings themselves are stable under quantum corrections. As an example we performed the one-loop quantum corrections to the conformal coupling and found that they get renormalized significantly for scalar masses close to M ∼ M 3 4 / C and M ∼ M 3 2 / C respectively. Additionally, the appearance of new couplings with more derivatives φ∂ ∂ χ∂ µ ν µ ∂ χ ν is unavoidable.
Summarizing, in this paper we have shown that quantum corrections coming from the coupling to the matter fields
- 1. renormalize the Galileon interactions: the non-renoramalization theorem protecting the Galileon interactions does not persist once one considers couplings to matter field. The standard covariant couplings give rise to quantum corrections with non-vanishing external momentum which renormalize the Galileon interactions.
- 2. generate new higher derivative interactions: in the same way as the Galileon interactions the coupling to matter fields generates ghost-like interactions with higher
derivative operators.
However, within the regime of the effective field theory these renormalized Galileons as well as the new higher derivative interactions come in suppressed by the coupling scale. They are harmless as long as
- 1. the mass of the external scalar field is smaller than the original classical coupling scale M glyph[lessmuch] M C
- 2. the derivatives applied on the Galileon field is smaller than the original classical coupling scale ∂ glyph[lessmuch] M C
In the context of massive gravity the original coupling scale is given by the Planck mass and therefore as long as the mass of the scalar field and the derivatives are smaller than the Planck mass, the quantum corrections are totally insignificant.
## Acknowledgments
We would like to thank Claudia de Rham for useful discussions. This work is supported by the Swiss National Science Foundation.
## References
- [1] Planck Collaboration Collaboration, P. Ade et al. , Planck 2013 results. XXII. Constraints on inflation , arXiv:1303.5082 .
- [2] BICEP2 Collaboration Collaboration, P. Ade et al. , BICEP2 I: Detection Of B-mode Polarization at Degree Angular Scales , arXiv:1403.3985 .
- [3] WMAP Collaboration, G. Hinshaw et al. , Nine-Year Wilkinson Microwave Anisotropy Probe (WMAP) Observations: Cosmological Parameter Results , Astrophys.J.Suppl. 208 (2013) 19, [ arXiv:1212.5226 ], [ doi:10.1088/0067-0049/208/2/19 ].
- [4] SDSS Collaboration Collaboration, W. J. Percival et al. , Baryon Acoustic Oscillations in the Sloan Digital Sky Survey Data Release 7 Galaxy Sample , Mon.Not.Roy.Astron.Soc. 401 (2010) 2148-2168, [ arXiv:0907.1660 ], [ doi:10.1111/j.1365-2966.2009.15812.x ].
- [5] O. Ilbert, P. Capak, M. Salvato, H. Aussel, H. J. McCracken, D. B. Sanders, N. Scoville, J. Kartaltepe, S. Arnouts, E. L. Floc'h, B. Mobasher, Y. Taniguchi, F. Lamareille,
- A. Leauthaud, S. Sasaki, D. Thompson, M. Zamojski, G. Zamorani, S. Bardelli,
- M. Bolzonella, A. Bongiorno, M. Brusa, K. I. Caputi, C. M. Carollo, T. Contini, R. Cook,
- G. Coppa, O. Cucciati, S. de la Torre, L. de Ravel, P. Franzetti, B. Garilli, G. Hasinger,
- A. Iovino, P. Kampczyk, J.-P. Kneib, C. Knobel, K. Kovac, J. F. L. Borgne, V. L. Brun, O. L. F` evre, S. Lilly, D. Looper, C. Maier, V. Mainieri, Y. Mellier, M. Mignoli, T. Murayama,
- R. Pell`, Y. Peng, E. P´rez-Montero, A. Renzini, E. Ricciardelli, D. Schiminovich, o e M. Scodeggio, Y. Shioya, J. Silverman, J. Surace, M. Tanaka, L. Tasca, L. Tresse, D. Vergani, and E. Zucca, Cosmos photometric redshifts with 30-bands for 2-deg2 , The Astrophysical Journal 690 (2009), no. 2 1236.
- [6] Planck Collaboration Collaboration, P. Ade et al. , Planck 2013 results. XVI. Cosmological parameters , arXiv:1303.5076 .
- [7] G. Dvali, G. Gabadadze, and M. Porrati, 4-D gravity on a brane in 5-D Minkowski space , Phys.Lett. B485 (2000) 208-214, [ arXiv:hep-th/0005016 ], [ doi:10.1016/S0370-2693(00)00669-9 ].
- [8] C. de Rham and G. Gabadadze, Generalization of the Fierz-Pauli Action , Phys.Rev. D82 (2010) 044020, [ arXiv:1007.0443 ], [ doi:10.1103/PhysRevD.82.044020 ].
- [9] C. de Rham, G. Gabadadze, and A. J. Tolley, Resummation of Massive Gravity , Phys.Rev.Lett. 106 (2011) 231101, [ arXiv:1011.1232 ], [ doi:10.1103/PhysRevLett.106.231101 ].
- [10] C. de Rham, G. Gabadadze, L. Heisenberg, and D. Pirtskhalava, Cosmic Acceleration and the Helicity-0 Graviton , Phys.Rev. D83 (2011) 103516, [ arXiv:1010.1780 ], [ doi:10.1103/PhysRevD.83.103516 ].
- [11] A. Nicolis, R. Rattazzi, and E. Trincherini, The galileon as a local modification of gravity , Phys.Rev. D79 (2009) 064036, [ arXiv:0811.2197 ], [ doi:10.1103/PhysRevD.79.064036 ].
- [12] C. de Rham, Massive Gravity , arXiv:1401.4173 .
- [13] C. Deffayet, S. Deser, and G. Esposito-Farese, Generalized Galileons: All scalar models whose curved background extensions maintain second-order field equations and stress-tensors , Phys.Rev. D80 (2009) 064015, [ arXiv:0906.1967 ], [ doi:10.1103/PhysRevD.80.064015 ].
- [14] C. Deffayet, G. Esposito-Farese, and A. Vikman, Covariant Galileon , Phys.Rev. D79 (2009) 084003, [ arXiv:0901.1314 ], [ doi:10.1103/PhysRevD.79.084003 ].
- [15] C. de Rham and A. J. Tolley, DBI and the Galileon reunited , JCAP 1005 (2010) 015, [ arXiv:1003.5917 ], [ doi:10.1088/1475-7516/2010/05/015 ].
- [16] C. Burrage, C. de Rham, and L. Heisenberg, de Sitter Galileon , JCAP 1105 (2011) 025, [ arXiv:1104.0155 ], [ doi:10.1088/1475-7516/2011/05/025 ].
- [17] G. Goon, K. Hinterbichler, and M. Trodden, Symmetries for Galileons and DBI scalars on curved space , JCAP 1107 (2011) 017, [ arXiv:1103.5745 ], [ doi:10.1088/1475-7516/2011/07/017 ].
- [18] J. Gleyzes, D. Langlois, F. Piazza, and F. Vernizzi, Healthy theories beyond Horndeski , arXiv:1404.6495 .
- [19] J. Gleyzes, D. Langlois, F. Piazza, and F. Vernizzi, Exploring gravitational theories beyond Horndeski , arXiv:1408.1952 .
- [20] X. Gao, Unifying framework for scalar-tensor theories of gravity , arXiv:1406.0822 .
- [21] M. Fasiello and S. Renaux-Petel, Non-Gaussian inflationary shapes in G 3 theories beyond Horndeski , arXiv:1407.7280 .
- [22] C. de Rham and L. Heisenberg, Cosmology of the Galileon from Massive Gravity , Phys.Rev. D84 (2011) 043503, [ arXiv:1106.3312 ], [ doi:10.1103/PhysRevD.84.043503 ].
- [23] L. Heisenberg, R. Kimura, and K. Yamamoto, Cosmology of the proxy theory to massive gravity , arXiv:1403.2049 .
- [24] C. Deffayet, A. E. Gumrukcuoglu, S. Mukohyama, and Y. Wang, A no-go theorem for generalized vector Galileons on flat spacetime , arXiv:1312.6690 .
- [25] L. Heisenberg, Generalization of the Proca Action , JCAP 1405 (2014) 015, [ arXiv:1402.7026 ], [ doi:10.1088/1475-7516/2014/05/015 ].
- [26] G. Tasinato, Cosmic Acceleration from Abelian Symmetry Breaking , arXiv:1402.6450 .
- [27] G. L. Goon, K. Hinterbichler, and M. Trodden, Stability and superluminality of spherical DBI galileon solutions , Phys.Rev. D83 (2011) 085015, [ arXiv:1008.4580 ], [ doi:10.1103/PhysRevD.83.085015 ].
- [28] P. de Fromont, C. de Rham, L. Heisenberg, and A. Matas, Superluminality in the Bi- and Multi- Galileon , JHEP 1307 (2013) 067, [ arXiv:1303.0274 ], [ doi:10.1007/JHEP07(2013)067 ].
- [29] C. Burrage, C. de Rham, L. Heisenberg, and A. J. Tolley, Chronology Protection in Galileon Models and Massive Gravity , JCAP 1207 (2012) 004, [ arXiv:1111.5549 ], [ doi:10.1088/1475-7516/2012/07/004 ].
- [30] C. de Rham, M. Fasiello, and A. J. Tolley, Galileon Duality , arXiv:1308.2702 .
- [31] C. de Rham, L. Keltner, and A. J. Tolley, Generalized Galileon Duality , arXiv:1403.3690 .
- [32] K. Kampf and J. Novotny, Unification of Galileon Dualities , arXiv:1403.6813 .
- [33] A. Nicolis and R. Rattazzi, Classical and quantum consistency of the DGP model , JHEP 0406 (2004) 059, [ arXiv:hep-th/0404159 ], [ doi:10.1088/1126-6708/2004/06/059 ].
- [34] C. de Rham, G. Gabadadze, L. Heisenberg, and D. Pirtskhalava, Nonrenormalization and naturalness in a class of scalar-tensor theories , Phys.Rev. D87 (2013), no. 8 085017, [ arXiv:1212.4128 ], [ doi:10.1103/PhysRevD.87.085017 ].
- [35] C. de Rham and R. H. Ribeiro, Riding on irrelevant operators , arXiv:1405.5213 .
- [36] C. de Rham, L. Heisenberg, and R. H. Ribeiro, On couplings to matter in massive (bi-)gravity , arXiv:1408.1678 .
- [37] C. de Rham, L. Heisenberg, and R. H. Ribeiro, Ghosts and Matter Couplings in Massive (bi-and multi-)Gravity , arXiv:1409.3834 .
- [38] T. S. Koivisto, D. F. Mota, and M. Zumalacarregui, Screening Modifications of Gravity through Disformally Coupled Fields , Phys.Rev.Lett. 109 (2012) 241102, [ arXiv:1205.3167 ], [ doi:10.1103/PhysRevLett.109.241102 ].
- [39] M. Zumalac´ arregui and J. Garc´ ıa-Bellido, Transforming gravity: from derivative couplings to matter to second-order scalar-tensor theories beyond the Horndeski Lagrangian , Phys.Rev. D89 (2014) 064046, [ arXiv:1308.4685 ], [ doi:10.1103/PhysRevD.89.064046 ].
- [40] D. Bettoni, Disformal invariance of second order scalar-tensor theories , arXiv:1405.3938 .
- [41] P. Brax and C. Burrage, Constraining Disformally Coupled Scalar Fields , arXiv:1407.1861 .
- [42] C. Burrage and J. Khoury, D-BIonic Screening of Scalar Fields , Phys.Rev. D90 (2014) 024001, [ arXiv:1403.6120 ], [ doi:10.1103/PhysRevD.90.024001 ].
- [43] K. Hinterbichler, M. Trodden, and D. Wesley, Multi-field galileons and higher co-dimension branes , Phys.Rev. D82 (2010) 124018, [ arXiv:1008.1305 ], [ doi:10.1103/PhysRevD.82.124018 ].
- [44] N. Brouzakis, A. Codello, N. Tetradis, and O. Zanusso, Quantum corrections in Galileon theories , Phys.Rev. D89 (2014) 125017, [ arXiv:1310.0187 ], [ doi:10.1103/PhysRevD.89.125017 ].
- [45] T. de Paula Netto and I. L. Shapiro, One-loop divergences in the Galileon model , Phys.Lett. B716 (2012) 454-460, [ arXiv:1207.0534 ], [ doi:10.1016/j.physletb.2012.08.056 ].
- [46] A. Vainshtein, To the problem of nonvanishing gravitation mass , Phys.Lett. B39 (1972) 393-394.
- [47] J. Khoury and A. Weltman, Chameleon cosmology , Phys.Rev. D69 (2004) 044026, [ arXiv:astro-ph/0309411 ], [ doi:10.1103/PhysRevD.69.044026 ].
- [48] K. Hinterbichler and J. Khoury, Symmetron Fields: Screening Long-Range Forces Through Local Symmetry Restoration , Phys.Rev.Lett. 104 (2010) 231301, [ arXiv:1001.4525 ], [ doi:10.1103/PhysRevLett.104.231301 ].
- [49] M. Wyman, Galilean-Invariant Scalar Fields Can Strengthen Gravitational Lensing , Physical Review Letters 106 (May, 2011) 201102, [ arXiv:1101.1295 ], [ doi:10.1103/PhysRevLett.106.201102 ]. | 10.1103/PhysRevD.90.064005 | [
"Lavinia Heisenberg"
] | 2014-07-31T18:10:30+00:00 | 2014-10-09T11:05:24+00:00 | [
"hep-th",
"gr-qc"
] | Quantum Corrections in Galileons from Matter Loops | Galileon interactions represent a class of effective field theories that have
received much attention since their inception. They can be treated in their own
right as scalar field theories with a specific global shift and Galilean
symmetry or as a descendant of a more fundamental theory like massive gravity.
It is well known that the Galileon theories are stable under quantum
corrections thanks to the non-renormalization theorem which is not due to the
symmetry. We consider different covariant couplings of this Galileon scalar
field with the matter field: the conformal coupling, the disformal coupling and
the longitudinal coupling. We compute the one-loop quantum corrections to the
Galileon interactions from the coupling to the external matter fields. In all
the considered cases of covariant couplings we show that the terms generated by
one-loop matter corrections not only renormalize the Galileon interactions but
also give rise to higher order derivative ghost interactions. However, the
renormalized version of the Galileon interactions as well as the new
interactions come at a scale suppressed by the original classical coupling
scale and hence are harmless within the regime of validity of the effective
field theory. |
1408.0268v2 | ## Horava-Lifshitz Black Hole Hydrodynamics
Christopher Eling 1 and Yaron Oz 2
- 1 Max Planck Institute for Gravitational Physics,
Albert Einstein Institute, Potsdam 14476, Germany and
- 2 Raymond and Beverly Sackler School of Physics and Astronomy,
Tel Aviv University, Tel Aviv 69978, Israel
(Dated: November 6, 2014)
## Abstract
We consider the holographic hydrodynamics of black holes in generally covariant gravity theories with a preferred time foliation. Gravitational perturbations in these theories have spin two and spin zero helicity modes with generically different speeds. The black hole solutions possess a spacelike causal boundary called the universal horizon. We relate the flux of the spin zero perturbation across the universal horizon to the new dissipative transport in Lifshitz field theory hydrodynamics found in arXiv:1304.7481. We construct in detail the hydrodynamics of one such black hole solution, and calculate the ratio of the shear viscosity to the entropy density.
## I. INTRODUCTION
A generic, generally covariant model of local Lorentz violating gravity is Einstein-aether theory [1]. In this theory the symmetry is broken by the aether covector v A , which is a dynamical field that is constrained to be unit timelike. As a consequence, the theory has in general spin-2, spin-1, and spin-0 gravitational wave polarizations traveling at different speeds. A particular choice for the aether field is to be hypersurface orthogonal, thus determining a preferred time foliation of space-time. In this case the Einstein-aether theory can be reduced [2, 3] to the Horava-Lifshitz theory [4].
Spherically symmetric asymptotically flat black hole solutions were first constructed in [5, 6]. The absence of local Lorentz symmetry has profound effects on black hole thermodynamics. Causality is no longer determined by the light cone, and the presence of multiple horizons seems to conflict with both the Zeroth and Second Laws [7-9]. Still, one can argue (see, e.g. [6, 10]) that there is a natural notion of causality defined by the preferred time foliation itself. At spatial infinity the time translation Killing vector and the aether are naturally aligned. However deep in the bulk, surfaces of constant preferred time bend and these two vectors eventually become orthogonal on a spacelike hypersurface. This is equivalent to the statement that the aether time τ →∞ on this hypersurface, called 'universal horizon'. Beyond the universal horizon even instantaneously propagating modes are causally disconnected from spatial infinity. Studies of the universal horizon suggest an associated temperature, and a first law of thermodynamics [11-14].
In this paper we will be interested in studying the Einstein-aether/Horava-Lifshitz theory with a negative cosmological constant, in the context of holography and the equivalence to a non-gravitational field theory. Holography is expected to relate these gravitational systems to field theories with broken Lorentz invariance in one lower space dimension. When the bulk aether field is hypersurface orthogonal, it induces a foliation at the boundary. The corresponding boundary field theories are known as Lifshitz field theories. In such field theories Lorentz invariance is broken.
Lifshitz field theories exhibit an anisotropic scaling of space and time (Lifshitz scaling) x i → λx , i i = 1 , ..., d, t → λ t z . z is called a dynamical exponent and is a measure of the anisotropy. In relativistic conformal field theories (CFTs) z = 1. The dynamical exponent differs from one in general Lifshitz theories. Examples of Lifshitz dual theories in 2 + 1
dimensional space-time are given by quantum critical points (QCPs). Such theories describe phase transitions at zero temperature driven by quantum fluctuations [15].
At zero temperature the correlation length ξ at the QCP diverges. Raising the temperature, one finds a quantum critical regime, where the system properties are constrained by the anisotropic scaling at the QCP. Denote by L a characteristic length scale of the system and by T the temperature. Hydrodynamics provides a good description in the quantum critical regime at the range of scales ξ /greatermuch L /greatermuch l T ∼ 1 T 1 z . The hydrodynamics expansion parameter is the dimensionless ratio l T L .
Since boost invariance is broken in Lifshitz field theories, the stress-energy tensor is no longer symmetric. The asymmetric term is associated with the foliation 1-form. With rotation invariance, the hydrodynamics of Lifshitz field theories exhibits one new dissipative transport coefficient at the first dissipative order found in [16, 17].
The gauge/gravity duality relates field theories at finite temperature to black holes in one higher space dimension. We will be interested in the out of equilibrium dynamics of such black holes. We will work in the hydrodynamic regime, which is described by black hole deformations, order by order in a derivative expansion. The field theory Navier-Stokes equations are the gravity constraint equations [18], which are also the horizon evolution equations [19].
Gravity is non-dissipative, however the horizon being a one way membrane introduces an effective dissipation: what goes in cannot go out. We propose that this boundary condition should be imposed at the universal horizon in these theories. Gravitational backgrounds with preferred foliation allow the propagation of spin-2 and spin-0 helicity gravitons. In thermal field theory language these two modes correspond to two possible channels of dissipation. The dissipation associated with the spin-2 helicity mode is seen in the standard relativistic hydrodynamics as the viscosity terms in the symmetric stress-energy tensor. The dissipation associated with the spin-0 helicity should be related to a new transport coefficient. We will argue that this is the new dissipation in Lifshitz field theory hydrodynamics appearing in the asymmetric part of the stress-energy tensor, and discovered in [16, 17].
In the special case where z = 1, an analytic asymptotically Lifshitz plane symmetric black hole solution to Horava-Lifshitz gravity is available [20, 21]. We will show that this new transport coefficient is zero is this case, and we will calculate the shear viscosity ratio η/s .
The paper is organized as follows. In section two we will discuss Einstein-aether and Horava-Lifshitz theories, the action and the classical field equations, and asymptotically Lifshitz solutions when a negative cosmological constant is added. In section three we will discuss Holographic Lifshitz Hydrodynamics. We will briefly review Lifshitz field theory hydrodynamics and construct the boundary stress-energy tensor. We will explain how its asymmetric part arises and exhibit the new dissipative transport coefficient associated with it. We will present the constraint equations and in particular the null focusing equation that is the gravitational counterpart of the entropy current divergence in field theory hydrodynamics. We will show where the new channel for dissipation comes from and connect it to the flux of the spin-0 helicity mode through the universal horizon. In section four we will analyze in detail the hydrodynamics at the first derivative order of a particular z = 1 analytic solution of Horava-Lifshitz black brane hydrodynamics. We will show that the new dissipative transport associated with the lack of Lorentz invariance vanishes in this case. We will calculate the ratio of the shear viscosity to the entropy density and show that it deviates from the general relativity result. Section five is devoted to a discussion of open problems.
## II. EINSTEIN-AETHER AND HORAVA-LIFSHITZ
## A. Einstein-aether action and field equations
In the following we will work in four-dimensional space-time dimensions (the generalization to other dimensions is straightforward). The action for Einstein-aether theory is given by
$$S _ { a e } = \frac { 1 } { 1 6 \pi G _ { a e } } \int d ^ { 4 } x \sqrt { - g } L _ { a e }$$
where L ae = R + L vec ,
$$- L _ { v e c } = K ^ { A B } _ { \, C D } \nabla _ { A } v ^ { C } \nabla _ { B } v ^ { D } - \lambda ( v ^ { 2 } + 1 )$$
with 'kinetic' tensor defined as
$$K ^ { A B } _ { \, C D } = c _ { 1 } g ^ { A B } g _ { C D } + c _ { 2 } \delta _ { \, C } ^ { A } \delta _ { \, D } ^ { B } + c _ { 3 } \delta _ { \, D } ^ { A } \delta _ { \, C } ^ { B } - c _ { 4 } v ^ { A } v ^ { B } g _ { C D } \.$$
This is the most general effective action for a timelike unit vector field at 2nd order in derivatives.
Varying this action with respect to the metric, vector field, and the Lagrange multiplier λ , one finds the following field equations
$$G _ { A B } = T _ { A B } ^ { a e }, \ \ E _ { A } = 0, \ \ v ^ { 2 } = - 1 \.$$
The aether stress tensor is given by
$$T _ { A B } ^ { a c } = \lambda v _ { A } v _ { B } + c _ { 4 } a _ { A } ^ { ( v ) } a _ { B } ^ { ( v ) } - \frac { 1 } { 2 } g _ { A B } Y ^ { C } _ { D } \nabla _ { C } v ^ { D } + \nabla _ { C } X ^ { C } _ { \ A B } + c _ { 1 } [ ( \nabla _ { A } v _ { C } ) ( \nabla _ { B } v ^ { C } ) - ( \nabla ^ { C } v _ { A } ) ( \nabla _ { C } v _ { B } ) ]$$
$$& ) ( \nabla _ { C } v _ { B } ) ] \, \\ & \quad ( 5 )$$
where
$$Y ^ { A } _ { \ B } = & K ^ { A C } _ { \ B D } \nabla _ { C } v ^ { D } \, \\ \varkappa \quad & \varkappa \quad & \varkappa \quad & \dots$$
$$X ^ { C } _ { \ A B } = & Y ^ { C } _ { \left ( A } v _ { B \right ) } - v _ { \left ( A } Y _ { B \right ) } ^ { C } + v ^ { C } Y _ { \left ( A B \right ) }$$
and a ( v ) A = v B ∇ B v A is the aether acceleration (which we distinguish from the fluid acceleration defined in the next section). The aether field equation is
$$E _ { A } = \nabla _ { B } Y ^ { B } _ { A } + \lambda v _ { A } + c _ { 4 } ( \nabla _ { A } v ^ { B } ) a _ { B } ^ { ( v ) } \.$$
## B. Hypersurface orthogonality and Horava-Lifshitz gravity
Consider now the case where the aether field is hypersurface orthogonal. This means that the twist vanishes
$$v _ { [ A } \nabla _ { B } v _ { C ] } = 0 \.$$
Since the squared twist also vanishes
$$\omega ^ { 2 } = ( \nabla _ { A } v _ { B } ) ( \nabla ^ { A } v ^ { B } ) - ( \nabla _ { A } v _ { B } ) ( \nabla ^ { B } v ^ { A } ) + a ^ { 2 }$$
adding a twist squared term to the action doesn't affect the solutions. We can therefore eliminate either c 1 , c 3 or c 4 in the action. Here we will choose to eliminate c 1 , i.e we take c 1 = 0 from now on.
Hypersurface orthogonality implies the co-vector is the gradient of a scalar
$$v _ { A } = \frac { - \partial _ { A } \phi } { \sqrt { g ^ { C D } \partial _ { C } \phi \partial _ { D } \phi } } \,.$$
One can show that hypersurface orthogonal solutions of Einstein-aether theory are also solutions to Horava-Lifshitz gravity [2]. The connection can be made explicit by choosing coordinates such that φ = τ , where τ is the preferred foliation of time. In this gauge the Einstein-aether action reduces to the generic 3+1 form of the Horava-Lifshitz action (e.g. [20])
$$S _ { H L } = \frac { 1 } { 1 6 \pi G _ { H } } \int d \tau d ^ { 3 } x \sqrt { \gamma } \left ( K _ { a b } K ^ { a b } - ( 1 - \lambda ) K ^ { 2 } + ( 1 + \beta ) R ^ { ( 3 ) } + \tilde { \alpha } \frac { \nabla _ { a } N \nabla ^ { a } N } { N ^ { 2 } } \right ) \. \quad ( 1 2 )$$
Here K ab is the extrinsic curvature of the preferred time slices, γ ab the spatial metric on the slices, R (3) the intrinsic Ricci scalar and N is the lapse function, i.e. v A = -Nδ τ A . The mapping between the constants is given by
$$1 + \lambda = \frac { 1 + c _ { 2 } } { 1 - c _ { 3 } }, \ \tilde { \alpha } = \frac { c _ { 4 } } { 1 - c _ { 3 } }, \ \frac { G _ { H } } { G _ { a e } } = 1 + \beta = \frac { 1 } { 1 - c _ { 3 } } \. \quad \quad \quad ( 1 3 ) \\. \ \sim. \ \dots \. \ \dots \. \ \dots \. \ \dots \. \$$
In generic Einstein-aether theory there are five propagating degrees of freedom with spin-2, spin-1, and spin-0 helicities [22]. In Horava-Lifshitz the spin-1 mode is non-propagating. The squared speeds of the remaining modes are given (in four dimensions) by [20]
$$s _ { 2 } ^ { 2 } = \frac { 1 } { 1 - c _ { 3 } }, \ s _ { 0 } ^ { 2 } = \frac { ( c _ { 2 } + c _ { 3 } ) ( 3 - c _ { 4 } ) } { c _ { 4 } ( 1 - c _ { 3 } ) ( 3 - 4 c _ { 2 } + c _ { 3 } ) } \.$$
## C. Asymptotically Lifshitz solutions
In [21] asymptotically Lifshitz and AdS solutions were studied in detail. For additional studies of these solutions, see [23-25]. In this case one adds a negative cosmological constant Λ to the action above
$$L _ { t o t a l } = \frac { 1 } { 1 6 \pi G _ { a e } } \left ( \Lambda + L _ { a e } \right ) \,.$$
We consider a metric and aether ansatz of the form
$$d s ^ { 2 } & = F ( \rho ) d t ^ { 2 } - 2 G ( \rho ) d t d \rho + \rho ^ { 2 } d x _ { i } d x ^ { i }, \\ & \quad \mathcal { C } / \wedge 2 \quad \mathcal { C } / \wedge \mathcal { C } / \wedge 2$$
$$v _ { A } d x ^ { A } = \frac { G ( \rho ) ^ { 2 } - F ( \rho ) K ( \rho ) ^ { 2 } } { 2 K ( \rho ) G ( \rho ) } d t + K ( \rho ) d \rho \.$$
Foreshadowing the holographic setup, we take x A = ( x , ρ µ ) with field theory coordinates x µ = ( t, x i ). The Lifshitz scaling reads ρ → λ -1 ρ, x i → λx , t i → λ t z .
Near infinity, the solution is required to approach
$$d s ^ { 2 } \sim - \rho ^ { 2 z } d t ^ { 2 } + 2 \rho ^ { z - 1 } d t d \rho + \rho ^ { 2 } d x _ { i } d x ^ { i } \\ 1$$
$$K ( \rho ) \sim \frac { 1 } { \rho }$$
It turns out that consistency of the field equations with this ansatz requires
$$c _ { 4 } = \frac { z - 1 } { z }, \Lambda = - \frac { ( 1 + z ) ( 2 + z ) } { 2 } \.$$
/negationslash
Generically, solutions with z = 1 can only be found numerically. In the case where z = 1 and c 4 = 0 (asymptotically AdS) an analytic solution was found in [21]. We will consider this case later in the paper. We expect that long wavelength, long time perturbations of these Lifshitz black brane solutions to be described by the hydrodynamics of Lifshitz field theories, which we describe in the following section.
## III. HOLOGRAPHIC LIFSHITZ HYDRODYNAMICS
## A. Lifshitz field theory hydrodynamics
Since boost invariance is explicitly broken in Lifshitz field theories, the conserved stressenergy tensor is not necessarily symmetric. In order to see its asymmetric part, we have to construct it not as a response of the action S to a change in a background metric h µν , but rather as a response to a change in the vielbein e µ a (by a we denote tangent space indices)
$$T _ { \mu } ^ { a } = - \frac { 1 } { e } \frac { \delta S } { \delta e _ { a } ^ { \mu } } \.$$
The vielbein encodes both the metric data h µν = e a µ e η b ν ab , and the foliation data v µ = e a µ v a , where v a = (1 0 , .., 0).
Using (21) one has
$$T _ { \mu \nu } = \Theta _ { \mu \nu } + J _ { \mu } v _ { \nu }$$
where
$$\Theta _ { \mu \nu } = \frac { 2 } { \sqrt { - h } } \frac { \delta S } { \delta h ^ { \mu \nu } }, \quad J _ { \mu } = \frac { 1 } { \sqrt { - h } } \frac { \delta S } { \delta v ^ { \mu } } \. \\ 2 ) \, \text{that the asymmetric part of the stress-energy tensor arises from } J. v. \text{ and}$$
We see from (22) that the asymmetric part of the stress-energy tensor arises from J v µ ν and is directly connected to the foliation data.
Consider next the hydrodynamics of Lifshitz field theories. The hydrodynamic stressenergy tensor in the uncharged case is expressed in terms of the temperature T , the velocity vector u µ normalized as u u µ µ = -1, and their derivatives via the constitutive relations. The hydrodynamics equations are the conservation law of the stress-energy tensor ∂ T µ µν = 0. As above, since boost invariance is explicitly broken, the stress-energy tensor can have an asymmetric part. Assuming rotation invariance, the asymmetric term shows up at the first viscous order [16, 17].
The energy-momentum tensor in the Landau frame T µν u ν = -εu µ takes the form
$$T ^ { \mu \nu } = \varepsilon u ^ { \mu } u ^ { \nu } + p P ^ { \mu \nu } + \pi _ { S } ^ { ( \mu \nu ) } + \pi _ { A } ^ { [ \mu \nu ] } + ( u ^ { \mu } \pi _ { A } ^ { [ \nu \sigma ] } + u ^ { \nu } \pi _ { A } ^ { [ \mu \sigma ] } ) u _ { \sigma },$$
with π ( µν ) S u ν = 0. At first order in derivatives π ( µν ) S includes the shear and bulk viscosities. The antisymmetric part of the stress-energy tensor reads at first order
$$\pi _ { A } ^ { [ \mu \nu ] } = - \alpha u ^ { [ \mu } a ^ { \nu ] }$$
where a µ = u ∂ u ν ν µ is the fluid acceleration, and α is a dissipative transport coefficient. It contributes to the divergence of the entropy current s µ = su µ
$$\partial _ { \mu } s ^ { \mu } = \frac { 2 \eta } { T } \sigma _ { \mu \nu } \sigma ^ { \mu \nu } + \frac { \zeta } { T } ( \partial _ { \mu } u ^ { \mu } ) ^ { 2 } + \frac { \alpha } { T } a _ { \mu } a ^ { \mu } \.$$
Here σ µν = P P ∂ λ µ σ ν ( λ u σ ) -1 3 P µν ∂ u λ λ is the fluid shear tensor, with P µν = h µν + u u µ ν the projection tensor orthogonal to u µ . The local form of the second law of thermodynamics ∂ s µ µ ≥ 0 requires that α ≥ 0, in addition to the usual positivity conditions on the shear and bulk viscosities η and ζ , respectively. We will argue that the a a µ µ entropy production term corresponds to the flux of spin-0 helicity graviton through the universal horizon.
## B. The boundary stress-energy tensor
In the following we derive the boundary stress-energy tensor from the gravity side. Suppose that we have the on-shell classical action S cl , which is a function of boundary data h µν and v µ . This classical action is invariant under diffeomorphisms in the boundary generated by ξ µ . One finds
$$\delta _ { \xi } S _ { c l } = \int \left ( \frac { \delta S _ { c l } } { \delta h _ { \mu \nu } } \mathcal { L } _ { \xi } h _ { \mu \nu } + \frac { \delta S _ { c l } } { \delta v _ { \mu } } \mathcal { L } _ { \xi } v _ { \mu } \right ) = 0 \.$$
Next, one identifies the canonical momenta Θ µν and J µ as in (23). Expanding out the Lie derivatives, one gets
$$D _ { \mu } ( \Theta ^ { \mu \nu } + J ^ { \mu } v ^ { \nu } ) = - J ^ { \mu } D ^ { \nu } v _ { \mu }$$
where D µ is the intrinsic covariant derivative on the slice. In general the momentum constraints do not need to be the divergence of a symmetric tensor. We identify the object in parentheses as the stress-energy tensor (22).
To compute the total stress-energy tensor using the gravitational variables, we focus on the boundary terms obtained by varying the Einstein-aether action (1) with respect to the metric and the aether fields. The variation of the usual Einstein-Hilbert action part yields the GR Brown-York stress tensor
$$\Theta _ { \mu \nu } ^ { B Y } = \frac { 1 } { 8 \pi G _ { a e } } \left ( h _ { \mu \nu } K - K _ { \mu \nu } \right )$$
where K µν is the extrinsic curvature tensor. Now consider the variation of the vector part of the action ∫ - √ -gd xK 4 ABCD ∇ A C v ∇ B v D . We find the boundary term
$$S _ { b d y } ^ { v e c } & = \frac { 1 } { 1 6 \pi G _ { a e } } \int d ^ { 3 } x \sqrt { h } \ n _ { C } \left ( 2 Y ^ { ( C A ) } v ^ { B } - Y ^ { A B } v ^ { C } \right ) \delta g _ { A B } \, \\ \cdot \. \. \. \. \. \. \. \. \. \. \. \. \. \. \. \. \.$$
where n A is the unit norm to the surface (here of constant bulk coordinate ρ = ρ 0 ) and h AB = g AB -n n A B . For the contribution to the boundary stress-energy tensor, we find in our coordinates x A = ( x , ρ µ )
$$\Theta _ { \mu \nu } ^ { v e c } & = \frac { 1 } { 8 \pi G _ { a e } } \left ( - Y _ { ( \mu \nu ) } v ^ { C } n _ { C } + 2 n ^ { C } Y _ { C ( \mu } v _ { \nu ) } + 2 n ^ { C } Y _ { ( \mu | C | } v _ { \nu ) } \right ) \. \\ \vdots & \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \.$$
From the vector action we also find the boundary term associated with the variation of the aether co-vector
$$S _ { b d y, v e c } ^ { v e c } = - \frac { 1 } { 8 \pi G _ { a e } } \int d ^ { 3 } x \sqrt { - h } \ n _ { C } Y ^ { C D } \delta v _ { D } \.$$
This yields the current
$$J _ { \mu } = - \frac { 1 } { 8 \pi G _ { a e } } n ^ { A } Y _ { A \mu } \.$$
Combining these results one gets the total boundary stress-energy tensor. Note, that evaluation on the asymptotic boundary at infinity will require the addition of counterterms in general to remove divergent terms.
## C. Constraint equations
Consider the constraint equations projected on generic surfaces of constant ρ = ρ 0 . In GR these take the form
$$C _ { \mu } = G _ { \mu B } n ^ { B } = 0 \.$$
In the fluid/gravity correspondence framework, one considers solutions constructed order by order in derivatives with respect to x µ . Assuming that all the field equations are imposed at ( n -1) order, the n th order Bianchi identity
$$\nabla ^ { A } _ { ( 0 ) } G ^ { ( n ) } _ { A \mu } = 0$$
can be written as a partial differential equation for the constraints as a function of radial direction C ( n ) µ ( ρ, x µ ). The solution of the differential equation is [26].
$$C _ { \mu } ^ { ( n ) } ( \rho, x ^ { \mu } ) = \frac { F _ { \mu } ( x ) } { A ( \rho ) }$$
where F µ ( x ) and A ρ ( ) are some functions. This off-shell analysis implies the constraint equations C µ = 0 have the same form on any constant ρ = ρ 0 slice. At the AdS boundary, the momentum constraints are equivalent to the fluid equations ∂ µ T µν = 0 [18], i.e. F ν ( x ) ∼ ∂ µ T µν . Thus, the factorization of the field theory and radial dependence means the constraint equations projected onto any radial surface yield the same hydrodynamics equations, with identical transport coefficients.
In Einstein-aether theory the generalized Bianchi identity takes the form
$$\nabla ^ { A } ( G _ { A B } - T _ { A B } ^ { a e } + v _ { A } E _ { B } ) + E _ { A } \nabla _ { B } v ^ { A } = 0$$
which implies the constraint equations are [27]
$$C _ { \mu } = ( G _ { A \mu } - T _ { A \mu } ^ { a e } + v _ { A } E _ { \mu } ) n ^ { A } = 0 \.$$
In this case the identity is no longer a simple conservation law, and the first term is a divergence of a non-symmetric tensor due to the v A E B piece. Nevertheless, repeating the same analysis as for GR outlined above, shows that the constraints factorize and one can study hydrodynamics by working on any radial slice.
Instead of considering the Einstein-aether constraint equations on the universal horizon, one can work at the Killing horizon. In this case the entropy balance law for the fluid
u ∂ T µ ν µν = 0, can be expressed in terms of horizon variables using the null Raychaudhuri equation. The hydrodynamic entropy balance law is equivalent to
$$( G _ { A B } - T _ { A B } ^ { a e } + v _ { A } E _ { B } ) \ell ^ { A } \ell ^ { B } = 0$$
where /lscript A the null normal to the horizon. For the first term we use g AB /lscript A B /lscript = 0 and the identity
$$G _ { A B } \ell ^ { A } \ell ^ { B } = R _ { A B } \ell ^ { A } \ell ^ { B } = \kappa \theta - \sigma _ { A B } \sigma ^ { A B } - \frac { 1 } { 2 } \theta ^ { 2 }$$
where κ the surface gravity at the Killing horizon, and θ and σ AB are the horizon expansion and shear tensor respectively.
Now it remains to evaluate the aether contributions using the field equations above. We concentrate on the contraction ( v A /lscript A ) E /lscript B B . Using the aether field equation (8) and the form of the stress tensor, we see that the ( v A /lscript A ) /lscript C ∇ B Y B C and λ v ( A /lscript A ) 2 terms cancel out. The remaining pieces are
$$( - T _ { A B } ^ { a e } + v _ { A } E _ { B } ) \ell ^ { A } \ell ^ { B } = c _ { 4 } ( v _ { A } \ell ^ { A } ) ( \ell ^ { C } \nabla _ { C } v ^ { B } ) a _ { B } - c _ { 4 } ( a _ { A } \ell ^ { A } ) ^ { 2 } - \\ \ell ^ { A } \ell ^ { B } ( Y ^ { C } _ { A } \nabla _ { C } v _ { B } - ( \nabla _ { C } v _ { A } ) Y _ { B } ^ { C } - v _ { A } \nabla _ { C } Y _ { B } ^ { C } + ( \nabla _ { C } v ^ { C } ) Y _ { A B } + v ^ { C } \nabla _ { C } Y _ { A B } ) \.$$
## IV. BLACK BRANE HYDRODYNAMICS WITH z = 1
In this section we will analyze the hydrodynamics of the black brane solution found in [21] when z = 1, that is c 4 = 0. The case z = 1 is special since the Lifshitz scaling symmetry of the boundary field theory is the same as that of relativistic CFTs, and implies the tracefree condition on the stress-energy tensor
$$T _ { \mu } ^ { \mu } = 0 \.$$
Boost invariance, however, is still expected to be broken in the boundary field theory. The gravitational solution is asymptotically AdS, but has a preferred time foliation and a universal horizon in the bulk interior.
In [28], it has been shown that in Horava-Lifshitz gravity the linearized spin-0 scalar perturbations around stationary background solutions generically do not propagate when c 4 = 0 (denoted by ˜ α = 0 in (12)). Moreover, as we discussed in the previous section,
the divergence of the fluid entropy current is equivalent to (39), which measures the flux of matter-energy across the horizon. The spin-0 flux is proportional to the energy density in the spin-0 waves times their speed s 0 . The spin-0 energy density scales like c 4 [29], while s 0 (14) goes like c -1 2 / 4 . Thus the spin-0 flux scales like √ c 4 and must vanish when z = 1. In the field theory language we expect this to translate into the statement that α = 0 in this case.
In the following we will study the first order hydrodynamics of this solution and show that this is indeed the case. We will calculate the ratio of the shear viscosity to entropy density and show that it deviates from that of Einstein gravity. Note, that while the first order hydrodynamics of the z = 1 solution is a CFT hydrodynamics, this does not necessarily imply that a non-relativistic behavior cannot be seen in the boundary field theory beyond the hydrodynamic regime.
## A. Equilibrium solution
We return to the metric and aether ansatz in (17). The c 4 = 0 solution is
$$F ( \rho ) & = - \, \rho ^ { 2 } + \frac { 2 \rho _ { h } ^ { 3 } } { \rho } + \frac { c _ { 3 } \rho _ { h } ^ { 6 } } { ( 1 - c _ { 3 } ) \rho ^ { 4 } } \\ \mathcal { C } ( \wedge - 1 ) & \quad \, \mathcal { H } ( \wedge )$$
$$G ( \rho ) = & 1$$
$$K ( \rho ) = & \frac { \rho ^ { 2 } } { ( \frac { 1 } { \sqrt { 1 - c _ { 3 } } } - 1 ) \rho _ { h } ^ { 3 } + \rho ^ { 3 } }. \\ \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots\slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slslots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slclots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots \slots$$
The parameter ρ h is the value of the universal horizon. This is defined as the value where the dot product of the timelike Killing vector and the aether vanishes χ v A A = 0. The above solution was obtained by demanding regularity at this point- that both ( χ v A A ) 2 and its first derivative vanish there. The solution does not depend on c 2 since the covariant divergence of the aether ∇ A v A = 0. This condition means all terms proportional to c 2 in the field equations vanish.
It has been argued [12, 21] that there is a Hawking temperature associated with this surface
$$T = \frac { 3 \rho _ { h } } { 2 \pi \sqrt { 1 - c _ { 3 } } }. \\ r m o d m a n i e a l a t i o n & \text{,\quad\ } o n e \, \text{can evaluate the boundary trace}$$
To determine the other thermodynamical variables, one can evaluate the boundary stress tensor found in Section III B on this solution in the limit as ρ → ∞ . The metric at the
AdS boundary is conformal to the flat metric, i.e. h µν = ρ η 2 0 µν , so one must normalize the expression by the overall conformal weight factor (in four dimensions)
$$T _ { \mu \nu } ^ { b d y } = \lim _ { \rho _ { 0 } \to \infty } \rho _ { 0 } T _ { \mu \nu } ^ { t o t }.$$
T bdy µν also contains divergent terms in the limit that must be subtracted off by the addition of appropriate counterterms to T tot µν . Computing the stress tensor using (29), (31), and (33) we find that the only counterterm needed to produce a finite answer is just the GR one,
$$T _ { \mu \nu } ^ { c o u n t e r } = - \frac { 1 } { 4 \pi G _ { a e } } h _ { \mu \nu },$$
independent of c 3 and proportional to the boundary metric.
The final result is a conformal perfect fluid stress tensor
$$T _ { \mu \nu } ^ { b d y } = p \left ( \eta _ { \mu \nu } + 3 u _ { \mu } u _ { \nu } \right )$$
with energy density
$$\epsilon = 2 p = \frac { \rho _ { h } ^ { 3 } } { 4 \pi G _ { a e } }.$$
This agrees with the value found by the on-shell Hamiltonian analysis in [21]. The value of thermal entropy density can be derived for example from the thermodynamic identity /epsilon1 + p = sT , giving
$$s = \frac { \rho _ { h } ^ { 2 } \sqrt { 1 - c _ { 3 } } } { 4 }.$$
Note, that in the weak field regime the effective Newton constant G ae = G N when z = 1, c 4 = 0. We will set this constant to be unity.
To probe the nature of the dual system, one can study perturbations of this solution in the fluid-gravity setting. One considers the equilibrium metric and aether in a generally boosted frame
$$d s ^ { 2 } = & F ( \rho ) u _ { \mu } u _ { \nu } d x ^ { \mu } d x ^ { \nu } - 2 u _ { \mu } d x ^ { \mu } d \rho + \rho ^ { 2 } P _ { \mu \nu } d x ^ { \mu } d x ^ { \nu } \\ & \quad \ r v \cdots \ r v \cdots 2$$
$$v _ { A } d x ^ { A } = & K ( \rho ) d \rho + \frac { 1 - F ( \rho ) K ( \rho ) ^ { 2 } } { 2 K ( \rho ) } u _ { \mu } d x ^ { \mu }$$
where u µ is the usual boost (fluid) velocity. Note that in GR, the boost we have implemented is a general coordinate transformation, which is a symmetry of the theory. However, in the
Horava-Lifshitz case the symmetry group is reduced to that of coordinate transformations that preserve the preferred time foliation. Since a boost naturally changes the notion of simultaneity, the solution in the boosted frame is not physically equivalent to the one in the rest frame.
The procedure is to then allow the velocity and Hawking temperature to be functions of the field theory coordinates: u µ ( x µ ) and ρ h ( x µ ). Since the metric and aether are no longer solutions, one must solve the field equations order by order in derivatives of x µ = ( t, x i ) subject to asymptotically AdS and regularity in the bulk interior.
## B. The first order corrections
We make the ansatz that the solution to the metric and aether field at first order in derivatives has the following form
$$g _ { A B } d x ^ { A } d x ^ { B } & = F ( \rho ) u _ { \mu } u _ { \nu } d x ^ { \mu } d x ^ { \nu } - 2 u _ { \mu } d x ^ { \mu } d \rho + \rho ^ { 2 } P _ { \mu \nu } d x ^ { \mu } d x ^ { \nu } \\ - 2 J ( \rho ) u _ { \mu } a _ { \nu } d x ^ { \mu } d x ^ { \nu } + L ( \rho ) u _ { \mu } u _ { \nu } ( \partial _ { \lambda } u ^ { \lambda } ) d x ^ { \mu } d x ^ { \nu } + H ( \rho ) \sigma _ { \mu \nu } d x ^ { \mu } d x ^ { \nu } \, \\ \dots \. & \. &.$$
while the aether is
$$v _ { A } d x ^ { A } = & K ( \rho ) d \rho + \frac { 1 - F ( \rho ) K ( \rho ) ^ { 2 } } { 2 K ( \rho ) } u _ { \mu } d x ^ { \mu } + M ( \rho ) a _ { \mu } d x ^ { \mu } - ( 1 / 2 ) K ( \rho ) L ( \rho ) u _ { \mu } ( \partial _ { \lambda } u ^ { \lambda } ) d x ^ { \mu$$
(55)
This assumes the standard fluid-gravity gauge choice that at all orders g ρρ = 0 and g ρµ = -u µ . Note that this form of the first order correction is consistent with the unit vector condition on the aether field v A v A = -1. This is the most general ansatz for the metric and aether one can write down using first order hydrodynamical variables (derivatives of temperature have been traded for derivatives of u µ using the zeroth order equations).
Using xAct [30] one can compute the full set of Einstein-aether field equations to first order in derivatives in order to solve for the unknown functions J ρ ( ), L ρ ( ), M ρ ( ), and H ρ ( ). We will first concentrate the solutions to J ρ ( ), L ρ ( ), and M ρ ( ). The solution for L ρ ( ) can be found from the F ρρ = 0, E ρ = 0, u F µ ρµ = 0 where F AB = G AB -3 g AB + T ae AB . These are a complicated set of ordinary differential equations. Solving in for example Maple, one finds the only solution is L ρ ( ) = ρ .
The field equations components P E ν µ ν = 0 and P F ν µ ρν = 0, are a very complicated coupled system of differential equations for M and J . In this case finding a simple solution is more
difficult. If we impose boundary conditions for the problem, e.g. J ρ ( ) ∼ ρ + const. +1 /ρ + · · · the solution is just J ρ ( ) = ρ and M ρ ( ) = K ρ ρ ( ) . As a result, the metric and aether turn out to be conformally covariant, following [31]. The combination
$$A _ { \mu } ^ { ( 1 ) } = a _ { \mu } - \frac { 1 } { 2 } u _ { \mu } ( \partial _ { \lambda } u ^ { \lambda } )$$
transforms like a connection under conformal transformations of the boundary metric. Explicitly, g µν → e 2 φ x ( ) g µν implies A µ → A µ + ∂ φ µ . Under the corresponding transformation of the radial coordinate ρ → e -φ x ( ) ρ , the combination
$$d \rho + \rho A _ { \mu } d x ^ { \mu }$$
transforms covariantly.
The presence of conformal symmetry in the solution is another hint that in this special case the transport coefficient α = 0. Since α is tied to the antisymmetric part of the hydrodynamic stress tensor, we consider the antisymmetric part of the boundary stress. In general this has the form
$$T _ { [ \mu \nu ] } ^ { b d y } = \frac { 1 } { 8 \pi G _ { a e } } \left ( c _ { 3 } n _ { A } \nabla _ { [ \mu } v ^ { A } v _ { \nu ] } - c _ { 4 } ( n _ { A } v ^ { A } ) a _ { [ \mu } ^ { ( v ) } v _ { \nu ] } \right )$$
Notice, that the second term proportional to c 4 is highly reminiscent of the a [ µ u ν ] term in the fluid stress, but a clear matching would require a specific solution. In our case, when c 4 = 0, we found using the first order solution that the antisymmetric part makes no contribution at the boundary. The first order corrections to the stress tensor will only involve shear terms and depend on the function H ρ ( ).
The result for the shear viscosity depends on the solution for the function H ρ ( ). To find H ρ ( ) we consider the following field equation P P F σ µ λ ν σλ = 0. One again we have a very complicated ordinary differential equation which we will not display here. In the limit as c 3 → 0, this equation reduces to
$$- \frac { 1 } { 2 } \rho ^ { - 3 } \left ( - 2 \rho ^ { 2 } \rho _ { h } ^ { 3 } H ^ { \prime \prime } - 8 \rho _ { h } ^ { 3 } H + 4 \rho ^ { 4 } + 6 H ^ { \prime } \rho \rho _ { h } ^ { 3 } + \rho ^ { 5 } H ^ { \prime \prime } - 2 H \rho ^ { 3 } \right ) = 0. \quad \quad \quad$$
The solution is
$$H ( \rho ) \, = \, - \frac { 1 } { 6 \rho _ { h } } \left ( - 1 8 A \rho _ { h } \ln ( \rho ) + 6 A \rho _ { h } \ln ( - 2 \rho _ { h } ^ { 3 } + \rho ^ { 3 } ) - 6 B \rho _ { h } + 2 ^ { 5 / 3 } \ln ( \rho - 2 ^ { 1 / 3 } \rho _ { h } ) \\ - 2 ^ { 2 / 3 } \ln ( \rho ^ { 2 } + 2 ^ { 1 / 3 } \rho _ { h } \rho + 2 ^ { 2 / 3 } \rho _ { h } ^ { 2 } ) + 2 ^ { 5 / 3 } \sqrt { 3 } \arctan ( ( 1 / 3 ) \sqrt { 3 } ( 2 ^ { 2 / 3 } \rho + \rho _ { h } ) / \rho _ { h } ) \right ) ( 6 0 )$$
Fixing AdS boundary conditions requires B = √ 3 · 2 2 3 / π/ ρ 6 h . The asymptotic solution near the boundary at x = 1 /ρ = 0 is then
$$H ( x ) = 2 / x + 2 \rho _ { h } ^ { 3 } A x + \rho _ { h } ^ { 3 } x ^ { 2 } + 2 \rho _ { h } ^ { 6 } A x ^ { 4 } + \cdots$$
Similarly, requiring that there is no divergence (curvature singularity) at the horizon fixes A = -2 2 3 / / 3 ρ h . This matches the solution found in the literature, e.g. [32].
For finite c 3 our strategy is to make the ansatz
$$H ( x ) = H _ { 0 } / x + H _ { 1 } + H _ { 2 } x + H _ { 3 } x ^ { 2 } + \cdots$$
and solve for the coefficients H i . The result is that
Up to order x 7 the solution is characterized by one free parameter H 2 . As in the pure GR case, we expect that one can tune the solution to a particular value of H 2 to avoid a curvature singularity at the (Killing) horizon.
$$\text{d solve for the coefficients } H _ { i }. \text{$The result is that} \\ H ( x ) = 2 / x + H _ { 2 } x + \frac { 1 } { 2 } \frac { \rho _ { h } ^ { 3 } ( 2 \sqrt { 1 - c _ { 3 } } - c _ { 3 } ) } { \sqrt { 1 - c _ { 3 } } } x ^ { 2 } + \rho _ { h } ^ { 3 } H _ { 2 } x ^ { 4 } + \cdots \\ \text{ to order } x ^ { 7 } \text{ the solution is characterized by one free parameter } H _ { i }. \text{ $As in the pure}$$
For small (positive) values of c 3 one can work with the differential equation for H to first (linear) order in c 3 corrections. The resulting equation can be solved analytically
$$\dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \\ H ( \rho ) \, = \, & \, H _ { G R } + \frac { 1 } { 1 2 \rho _ { h } } \left ( 2 c _ { 3 } 2 ^ { 2 / 3 } \rho \ln ( \rho - 2 ^ { 1 / 3 } \rho _ { h } ) - c _ { 3 } 2 ^ { 2 / 3 } \ln ( \rho ^ { 2 } + 2 ^ { 1 / 3 } \rho \rho _ { h } + 2 ^ { 2 / 3 } \rho _ { h } ^ { 2 } ) \\ & + 2 c _ { 3 } \sqrt { 3 } \cdot 2 ^ { 2 / 3 } \arctan ( ( 1 / 3 ) \sqrt { 3 } ( 2 ^ { 2 / 3 } \rho + \rho _ { h } ) / \rho _ { h } ) + 1 2 c _ { 3 } \rho _ { h } \right ) \,, \\ \text{where } H _ { \sim n } \text{ is given by } ( 6 ) \, \text{ as a result, the value of the coefficient } H _ { \, } \text{ is}$$
where H GR is given by (60). As a result, the value of the coefficient H 2 is
$$H _ { 2 } = & \frac { 2 ^ { 5 / 3 } \rho _ { h } ^ { 2 } } { 3 } ( c _ { 3 } / 2 - 1 ).$$
## C. Raychaudhuri equation and the entropy law
With a first order solution in hand, we now study the form of the Raychaudhuri equation for the Killing horizon. Since c 4 = 0,
$$Y _ { A B } = c _ { 2 } g _ { A B } \nabla _ { C } v ^ { C } + c _ { 3 } \nabla _ { B } v _ { A }.$$
We find that
$$\kappa \theta - \ell ^ { A } \nabla _ { A } \theta - \sigma _ { A B } \sigma ^ { A B } - \frac { 1 } { 2 } \theta ^ { 2 } \\ + c _ { 2 } ( \ell ^ { B } v _ { B } ) \ell ^ { A } \nabla _ { A } ( \nabla _ { C } v ^ { C } ) - c _ { 3 } ( \ell ^ { A } \nabla _ { A } v ^ { C } ) ( \ell ^ { B } \nabla _ { C } v _ { B } ) + c _ { 3 } ( \ell ^ { A } \nabla ^ { C } v _ { A } ) ( \ell ^ { B } \nabla _ { C } v _ { B } ) \\ - c _ { 3 } ( \nabla _ { C } v ^ { C } ) \ell ^ { A } \ell ^ { B } \nabla _ { A } v _ { B } + c _ { 3 } ( \ell ^ { C } v _ { C } ) \ell ^ { B } \nabla _ { A } \nabla ^ { A } v _ { B } - c _ { 3 } \ell ^ { A } \ell ^ { B } v ^ { C } \nabla _ { C } \nabla _ { B } v _ { A } = 0 \.$$
Note that the location of the Killing horizon is
$$\rho _ { K H } & = \left ( \frac { 1 - c _ { 3 } + \sqrt { 1 - c _ { 3 } } } { 1 - c _ { 3 } } \right ) ^ { 1 / 3 } \rho _ { h } \, \\ \text{and} \, \colon =$$
and the surface gravity is
$$\kappa _ { K H } = \frac { 3 \rho _ { h } ( 1 + \sqrt { 1 - c _ { 3 } } ) ( 1 - c _ { 3 } ) ^ { 2 / 3 } } { ( 1 - c _ { 3 } + \sqrt { 1 - c _ { 3 } } ) ^ { 5 / 3 } }. \\ \tau \tau \tau \text{transform to a odd} \text{or to the horizon and} \text{is fixed at}$$
It is also useful to transform to a coordinate system where the horizon radius is fixed at zero: ¯ = ρ ρ -ρ KH ( x ). Transforming to these coordinates, one finds
$$g _ { A B } d x ^ { A } d x ^ { B } = F ( \bar { \rho } ) u _ { \mu } u _ { \nu } d x ^ { \mu } d x ^ { \nu } + 2 u _ { \mu } d x ^ { \mu } d \bar { \rho } + ( \bar { \rho } + \rho _ { K H } ) ^ { 2 } P _ { \mu \nu } d x ^ { \mu } d x ^ { \nu } - \\ 2 \bar { \rho } u _ { \mu } A _ { \nu } d x ^ { \mu } d x ^ { \nu } + H ( \bar { \rho } ) \sigma _ { \mu \nu } d x ^ { \mu } d x ^ { \nu }$$
and
$$v _ { A } d x ^ { A } = & K ( \bar { \rho } ) d \bar { \rho } + \frac { 1 - F ( \bar { \rho } ) K ( \bar { \rho } ) ^ { 2 } } { 2 K ( \bar { \rho } ) } u _ { \mu } d x ^ { \mu } + \bar { \rho } K ( \bar { \rho } ) A _ { \mu } d x ^ { \mu } \.$$
In this coordinates, the normal to the Killing horizon is u µ to all orders.
Now we can use this ansatz to compute the Raychaudhuri equation to second order in derivatives and then evaluate at the Killing horizon, ¯ = 0. ρ The first step is consider the equation at first order in derivatives of u µ and ρ h . Combining terms from both the pure gravity and aether parts of the equation, we arrive at the following result
$$\frac { 3 } { \rho _ { h } ( 1 + \frac { 1 } { \sqrt { 1 - c _ { 3 } } } ) ^ { 2 / 3 } } \partial _ { \mu } ( \rho _ { h } ^ { 2 } u ^ { \mu } ) = 0. \\ \dots \quad. \quad. \quad. \quad. \quad.$$
Up to an overall factor, this equation matches that of the entropy conservation law
$$\partial _ { \mu } ( s u ^ { \mu } ) = 0.$$
with s given by (51).
The next step is to evaluate the Raychaudhuri equation at second order in derivatives. Here we must use the other projection of the constraint equations in the horizon limit
$$( G _ { A \mu } - T _ { A \mu } ^ { a e } + v _ { A } E _ { \mu } ) \ell ^ { A } = 0.$$
This is equivalent to the fluid equation P α ν ∂ T µ µν = 0,
$$P _ { \mu } ^ { \nu } \partial _ { \nu } \ln \rho _ { h } + a _ { \mu } = 0$$
and can be imposed in the second order expressions. The result can be put into the following form
$$\text{form} \\ & \frac { 3 } { \rho _ { h } ( 1 + \frac { 1 } { \sqrt { 1 - c _ { 3 } } } ) ^ { 2 / 3 } } \partial _ { \mu } ( \rho _ { h } ^ { 2 } u$$
where the functions appearing are to be evaluated at the Killing horizon. Consistent with results at the boundary, all a a µ µ terms cancel out of the final expression and only shear squared remains.
## D. The ratio of shear viscosity to entropy density
We now can determine the shear viscosity by matching (76) to the general hydrodynamic entropy balance law (26)
$$\partial _ { \mu } ( s u ^ { \mu } ) = \frac { 2 \eta } { T } \sigma _ { \mu \nu } \sigma ^ { \mu \nu }.$$
with s given in (51) and T in (46). We read off that
$$& \nu \dots \nu \dots \nu \dots \nu \dots \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \$$
For c 3 > 1 the formula is ill-defined, but in this regime the theory is known to suffer from negative energies, unstable linearized wave modes, etc. See the review in [1]. Working to linear order in c 3 and using the results above for H , we find
$$\eta = \frac { 2 ^ { 2 / 3 } } { 1 6 \pi } \rho _ { h } ^ { 2 } \left ( 1 - c _ { 3 } / 2 + O ( c _ { 3 } ^ { 2 } ) \right ) \,. \quad \quad \quad$$
As a check, we also evaluated the shear viscosity from the boundary stress tensor. In this case, one inserts the first order metric solution into (47) with counterterm (48) and reads off the shear viscosity as the coefficient of the shear term in the hydrodynamic stress tensor. The result depends on the value of H 2 in the asymptotic expansion (65) for H ρ ( )
$$\eta = - \frac { 3 } { 3 2 \pi } H _ { 2 },$$
matching with (79), as expected.
Dividing by the entropy density s (51), we find the shear viscosity to entropy density ratio
$$\frac { \eta } { s } = \frac { 2 ^ { 2 / 3 } } { 4 \pi } \left ( 1 + O ( c _ { 3 } ^ { 2 } ) \right ). \quad & ( 8 1 ) \\ \intertext { on for $U(\lambda\to\text{when and in $e$ and function to solution}$}$$
Expanding out the solution for H ρ ( ) to higher orders in c 3 and repeating the calculation indicates that the ratio is independent of c 3 up to fourth order. Therefore we conjecture that in general
$$\eta = \frac { 2 ^ { 2 / 3 } \sqrt { 1 - c _ { 3 } } } { 1 6 \pi } \rho _ { h } ^ { 2 }.$$
Curiously, in the limit as c 3 → 0 the shear viscosity to entropy density ratio differs from the GR value of 1 / 4 π by a factor of 2 2 3 / . The discrepancy arises because as c 3 → 0 the universal horizon entropy density s goes to ρ / 2 h 4, while the Killing horizon entropy density expected in pure GR gives s KH = A KH / 4 = 2 2 3 / s . This suggests that simply taking c 3 → 0 limit in Horava-Lifshitz gravity does not reduce to GR. The causal boundary in GR is associated with the Killing horizon, while that of Horava-Lifshitz gravity with the universal horzion. Thus, one should in the limit also shift variables to s KH .
## V. DISCUSSION
In the following we outline several open questions and directions. The new hydrodynamic transport associated with the breaking of boost invariance vanishes in the particular z = 1 solution that we studied. It would be valuable to construct gravitational solutions with z = 1, where one expects it to be generically non-vanishing.
/negationslash
Superfluid Lifshitz hydrodynamics has been analyzed in [34]. It would be interesting to construct dual gravitational solutions. These are black branes in Horava-Lifshitz gravity with hair corresponding to the condensate that breaks a U (1) global symmetry.
A simple formula for the ratio of the bulk viscosity to shear viscosity in holographic Lifshitz hydrodynamics has been derived in [35]. It is a generalization of [36] and is based on the horizon focusing equation. It is not clear that similar formula can be worked out for the new transport coefficient associated with broken boost invariance, but it is worth exploring this further.
If in addition to breaking boost invariance, one also allows a breakdown of rotational symmetry then there are new expected transports in field theory hydrodynamics. These
should presumably correspond to the spin-1 helicity mode in the gravitational description. It would be interesting to work out this relation.
Finally, while the z = 1 black brane hydrodynamics is conformal, it would be of interest to know whether there are Lorentz violating aspects that it exhibits beyond the hydrodynamic limit, both in the bulk and in the boundary field theory.
## Acknowledgements
We would like to thank Igal Arav and Adiel Meyer for valuable discussions. This work is supported in part by the I-CORE program of Planning and Budgeting Committee (grant number 1937/12), and by the US-Israel Binational Science Foundation.
- [1] T. Jacobson and D. Mattingly, Phys. Rev. D 64 , 024028 (2001) [gr-qc/0007031]; for a review, see T. Jacobson, PoS QG -PH , 020 (2007) [arXiv:0801.1547 [gr-qc]].
- [2] T. Jacobson, Phys. Rev. D 81 , 101502 (2010) [Erratum-ibid. D 82 , 129901 (2010)] [arXiv:1001.4823 [hep-th]].
- [3] D. Blas, O. Pujolas and S. Sibiryakov, JHEP 1104 , 018 (2011) [arXiv:1007.3503 [hep-th]].
- [4] P. Horava, Phys. Rev. D 79 , 084008 (2009) [arXiv:0901.3775 [hep-th]].
- [5] C. Eling and T. Jacobson, Class. Quant. Grav. 23 , 5643 (2006) [Erratum-ibid. 27 , 049802 (2010)] [gr-qc/0604088].
- [6] E. Barausse, T. Jacobson and T. P. Sotiriou, Phys. Rev. D 83 , 124043 (2011) [arXiv:1104.2889 [gr-qc]].
- [7] S. L. Dubovsky and S. M. Sibiryakov, Phys. Lett. B 638 , 509 (2006) [hep-th/0603158].
- [8] C. Eling, B. Z. Foster, T. Jacobson and A. C. Wall, Phys. Rev. D 75 , 101502 (2007) [hep-th/0702124 [HEP-TH]].
- [9] T. Jacobson and A. C. Wall, Found. Phys. 40 , 1076 (2010) [arXiv:0804.2720 [hep-th]].
- [10] D. Blas and S. Sibiryakov, Phys. Rev. D 84 , 124043 (2011) [arXiv:1110.2195 [hep-th]].
- [11] P. Berglund, J. Bhattacharyya and D. Mattingly, Phys. Rev. D 85 , 124019 (2012) [arXiv:1202.4497 [hep-th]].
- [12] P. Berglund, J. Bhattacharyya and D. Mattingly, Phys. Rev. Lett. 110 , no. 7, 071301 (2013) [arXiv:1210.4940 [hep-th]].
- [13] A. Mohd, arXiv:1309.0907 [gr-qc].
- [14] B. Cropp, S. Liberati, A. Mohd and M. Visser, Phys. Rev. D 89 , 064061 (2014) [arXiv:1312.0405 [gr-qc]].
- [15] S. Sahdev, 'Quantum Phase Transitions', Cambridge University Press (2011).
- [16] C. Hoyos, B. S. Kim and Y. Oz, JHEP 1311 , 145 (2013) [arXiv:1304.7481 [hep-th]].
- [17] C. Hoyos, B. S. Kim and Y. Oz, JHEP 1403 , 029 (2014) [arXiv:1309.6794 [hep-th], arXiv:1309.6794].
- [18] S. Bhattacharyya, V. E Hubeny, S. Minwalla and M. Rangamani, JHEP 0802 , 045 (2008) [arXiv:0712.2456 [hep-th]].
- [19] C. Eling and Y. Oz, JHEP 1002 , 069 (2010) [arXiv:0906.4999 [hep-th]].
- [20] T. Griffin, P. Horava and C. M. Melby-Thompson, Phys. Rev. Lett. 110 , no. 8, 081602 (2013) [arXiv:1211.4872 [hep-th]].
- [21] S. Janiszewski, arXiv:1401.1463 [hep-th].
- [22] T. Jacobson and D. Mattingly, Phys. Rev. D 70 , 024003 (2004) [gr-qc/0402005].
- [23] S. Janiszewski, A. Karch, B. Robinson and D. Sommer, JHEP 1404 , 163 (2014) [arXiv:1401.6479 [hep-th]].
- [24] F. W. Shu, K. Lin, A. Wang and Q. Wu, JHEP 1404 , 056 (2014) [arXiv:1403.0946 [hep-th]]; K. Lin, F. W. Shu, A. Wang and Q. Wu, arXiv:1404.3413 [hep-th].
- [25] T. P. Sotiriou, I. Vega and D. Vernieri, Phys. Rev. D 90 , 044046 (2014) [arXiv:1405.3715 [gr-qc]].
- [26] C. Eling and Y. Oz, Phys. Rev. Lett. 107 , 201602 (2011) [arXiv:1107.2134 [hep-th]].
- [27] T. Jacobson, Class. Quant. Grav. 28 , 245011 (2011) [arXiv:1108.1496 [gr-qc]].
- [28] D. Blas, O. Pujolas and S. Sibiryakov, JHEP 0910 , 029 (2009) [arXiv:0906.3046 [hep-th]].
- [29] C. Eling, Phys. Rev. D 73 , 084026 (2006) [Erratum-ibid. D 80 , 129905 (2009)] [gr-qc/0507059]; B. Z. Foster, Phys. Rev. D 73 , 104012 (2006) [Erratum-ibid. D 75 , 129904 (2007)] [gr-qc/0602004].
- [30] Jos´ M. Mart´ ın-Garc´ ıa, 'xAct: e Efficient Tensor Computer Algebra,' http://www.xact.es/
- [31] S. Bhattacharyya, R. Loganayagam, I. Mandal, S. Minwalla and A. Sharma, JHEP 0812 , 116 (2008) [arXiv:0809.4272 [hep-th]].
- [32] M. Van Raamsdonk, JHEP 0805 , 106 (2008) [arXiv:0802.3224 [hep-th]].
- [33] P. Kovtun, D. T. Son and A. O. Starinets, Phys. Rev. Lett. 94 , 111601 (2005) [hep-th/0405231].
- [34] S. Chapman, C. Hoyos and Y. Oz, JHEP 1407 , 027 (2014) [arXiv:1402.2981 [hep-th]].
- [35] C. Hoyos, B. S. Kim and Y. Oz, JHEP 1403 , 050 (2014) [arXiv:1312.6380 [hep-th]].
- [36] C. Eling and Y. Oz, JHEP 1106 , 007 (2011) [arXiv:1103.1657 [hep-th]]. | 10.1007/JHEP11(2014)067 | [
"Christopher Eling",
"Yaron Oz"
] | 2014-08-01T19:08:30+00:00 | 2014-11-05T17:08:33+00:00 | [
"hep-th",
"gr-qc"
] | Horava-Lifshitz Black Hole Hydrodynamics | We consider the holographic hydrodynamics of black holes in generally
covariant gravity theories with a preferred time foliation. Gravitational
perturbations in these theories have spin two and spin zero helicity modes with
generically different speeds. The black hole solutions possess a spacelike
causal boundary called the universal horizon. We relate the flux of the spin
zero perturbation across the universal horizon to the new dissipative transport
in Lifshitz field theory hydrodynamics found in arXiv:1304.7481. We construct
in detail the hydrodynamics of one such black hole solution, and calculate the
ratio of the shear viscosity to the entropy density. |
1408.0269v2 | ## UNIRATIONALITY OF DEL PEZZO SURFACES OF DEGREE TWO OVER FINITE FIELDS (EXTENDED VERSION)
## DINO FESTI, RONALD VAN LUIJK
Abstract. We prove that every del Pezzo surface of degree two over a finite field is unirational, building on the work of Manin and an extension by Salgado, Testa, and V´ arilly-Alvarado, who had proved this for all but three surfaces. Over general fields, we state several sufficient conditions for a del Pezzo surface of degree two to be unirational.
## 1. Introduction
We say that a variety X of dimension n over a field k is unirational if there exists a dominant rational map P n /axisshort/axisshort/arrowaxisright X , defined over k . A del Pezzo surface is a smooth, projective, geometrically integral variety X of which the anticanonical divisor -K X is ample. We define the degree of a del Pezzo surface X as the self intersection number of K X , that is, deg X = K 2 X . If k is an algebraically closed field, then every del Pezzo surface of degree d over k is isomorphic to P 1 × P 1 (with d = 8), or to P 2 blown up in 9 -d points in general position.
Over arbitrary fields, the situation is more complicated. Works of B. Segre, Yu. Manin, J. Koll´r, a M. Pieropan, and A. Knecht prove that every del Pezzo surface of degree d ≥ 3, defined over any field k , is unirational, provided that the set X k ( ) of rational points is non-empty. For references, see [Seg43, Seg51] for k = Q and d = 3, see [Man86, Theorem 29.4 and 30.1] for d ≥ 3 with the extra assumption for d ∈ { 3 4 , } that k has enough elements. See [Kol02, Theorem 1.1] for d = 3 and a general ground field. The earliest reference we could find for d = 4 and a general ground field is [Pie12, Proposition 5.19]. Independently, for d = 4, [Kne13, Theorem 2.1] covers all finite fields.
Since all del Pezzo surfaces over finite fields have a rational point (see [Man86, Corollary 27.1.1]), this implies that every del Pezzo surface of degree at least 3 over a finite field is unirational.
Building on work by Manin (see [Man86, Theorem 29.4]), C. Salgado, D. Testa, and A. V´rillya Alvarado prove that all del Pezzo surfaces of degree 2 over a finite field are unirational as well, except possibly for three isomorphism classes of surfaces (see [STVA14, Theorem 1]). In this paper, we show that these remaining three cases are also unirational, thus proving our first main theorem.
Theorem 1.1. Every del Pezzo surface of degree 2 over a finite field is unirational.
In Section 2 we will recall the basics about del Pezzo surfaces of degree 2, including the fact that the linear system associated to the anti-canonical divisor induces a finite morphism to P 2 of degree 2. This allows us to state the second main theorem.
Theorem 1.2. Suppose k is a field of characteristic not equal to 2 , and let k be an algebraic closure of k . Let X be a del Pezzo surface of degree 2 over k . Let B ⊂ P 2 be the branch locus of the anti-canonical morphism π : X → P 2 . Let C ⊂ P 2 be a projective curve that is birationally equivalent with P 1 over k . Assume that all singular points of C that are contained in B are ordinary singular points. Then the following statements hold.
- (1) Suppose that there is a point P ∈ X k ( ) such that we have Q = π P ( ) ∈ C -B . Suppose that B contains no singular points of C and that all intersection points of B and C have even intersection multiplicity. Then the surface X is unirational.
- (2) Suppose that one of the following two conditions hold.
- (a) There is a point Q ∈ C k ( ) ∩ B k ( ) that is a double or a triple point of C . The curve B contains no other singular points of C , and all intersection points of B and C have even intersection multiplicity.
- (b) There exist two distinct points Q , Q 1 2 ∈ C k ( ) ∩ B k ( ) such that B and C intersect with odd multiplicity at Q 1 and Q 2 and with even intersection multiplicity at all other intersection points. Furthermore, the points Q 1 and Q 2 are smooth points or double points on the curve C , and B contains no other singular points of C .
Then there exists a field extension /lscript of k of degree at most 2 for which the preimage π -1 ( C /lscript ) is birationally equivalent with P 1 /lscript ; for each such field /lscript , the surface X /lscript is unirational.
Corollary 1.3. Suppose k is a field of characteristic not equal to 2 . Let X be a del Pezzo surface of degree 2 over k . Assume that X has a k -rational point, say P . Let C ⊂ P 2 be a geometrically integral curve over k of degree d ≥ 2 and suppose that π P ( ) is a point of multiplicity d -1 on C . Suppose, moreover, that C intersects the branch locus B of the anti-canonical morphism π : X → P 2 with even multiplicity everywhere. Then the following statements hold.
- (1) If π P ( ) is not contained in B , then X is unirational.
- (2) If π P ( ) is contained in B , and it is an ordinary singular point on C and we have d ∈ { 3 4 , } , then there exists a field extension /lscript of k of degree at most 2 for which the preimage π -1 ( C /lscript ) is birationally equivalent with P 1 /lscript ; for each such field /lscript , the surface X /lscript is unirational.
After a quick summary of some basic results on del Pezzo surfaces of degree 2 in the next section, we will present the three difficult surfaces and prove Theorem 1.1 in Section 3. The main tool is Lemma 3.2 (that is, [STVA14, Theorem 17]), which states that it suffices to construct a rational curve on each of the three del Pezzo surfaces.
In Section 4 we give an alternative description of Manin's construction of a rational curve. If a point P on a del Pezzo surface X does not lie on four of the 56 exceptional curves, and not on the ramification locus of the anti-canonical morphism, then Manin's construction, extended by C. Salgado, D. Testa, and A. V´rilly-Alvarado, yields a rational curve that satisfies the assumptions a of case (1) of Corollary 1.3 with the degree d being such that there are 4 -d exceptional curves through P (cf. Proposition 4.14, Remark 4.15, and Example 5.14).
The three difficult surfaces do not contain such a point. The proofs of unirationality of these three cases use a rational curve that is an example of case (2) of Corollary 1.3 instead (cf. Remark 3.4 and Example 5.16). Here we benefit from the fact that if k is a finite field, then any curve that becomes birationally equivalent with P 1 over an extension of k , already is birationally equivalent with P 1 over k itself. For two of the three cases, the rational curve we use has degree 4. For the last case, the curve we use has degree 3, but there also exist quartic curves satisfying the hypotheses of case (2) of Corollary 1.3. This raises the following question (cf. Question 6.6, Remark 6.8, and Example 6.9), which together with case (2) of Corollary 1.3 could help proving unirationality of del Pezzo surfaces of degree 2 over any field of characteristic not equal to 2.
Question 1.4. Let d ∈ { 3 4 , } be an integer. Let X be a del Pezzo surface of degree two over a field of characteristic not equal to 2 , and let P ∈ X k ( ) be a point on the ramification locus of the anti-canonical map π : X → P 2 . Does there exist a geometrically integral curve of degree d in P 2 over k that has an ordinary singular point of multiplicity d -1 at π P ( ) , and that intersects the branch locus of π with even multiplicity everywhere?
For some d , X , and P , the answer to this question is negative (see Example 6.9), but in all cases we know of (all over finite fields), there do exist degenerate curves of degree d with a point of multiplicity at least d -1 at π P ( ). Hence, it may be true that the answer to Question 1.4 is positive for X and P general enough.
In line with case (1) of Corollary 1.3, we can ask, in fact for any integer d ≥ 1, an analogous question for points P that do not lie on the ramification locus, where we do not require the singular point to be ordinary. In this case, if P lies on r ≤ 3 exceptional curves, then Manin's construction shows that the answer is positive for degree d = 4 -r . Therefore, this analogous question is especially interesting when P lies on four exceptional curves (cf. Remark 5.12 and Example 5.15).
In Section 5 we prove Theorem 1.2 and a generalisation, Corollary 5.6, as well as Corollary 1.3. We do this by showing that, under the hypotheses of Theorem 1.2, the pull-back of the curve C to X contains a rational component, so that we may again use Lemma 3.2.
In Section 6 we discuss how to search for curves satisfying the assumptions of Theorem 1.2 and in particular of Corollary 1.3.
The authors would like to thank David Holmes, Ariyan Javanpeykar, Bjorn Poonen, Marco Streng, Damiano Testa, and Anthony V´rilly-Alvarado for useful conversations. a
## 2. Del Pezzo surfaces of degree two
The statements in this section are well known and we will use them freely. Let X be a del Pezzo surface of degree 2 over a field k with canonical divisor K X . The Riemann-Roch spaces L -( K X ) and L -( 2 K X ) have dimension 3 and 7, respectively. Let x, y, z be generators of L -( K X ) and choose an element w ∈ L -( 2 K X ) that is not contained in the image of the natural map Sym 2 L -( K X ) →L -( 2 K X ). Then X embeds into the weighted projective space P = P (1 , 1 1 2) , , with coordinates x, y, z , and w . We will identify X with its image in P , which is a smooth surface of degree 4. Conversely, every smooth surface of degree 4 in P is a del Pezzo surface of degree 2. There are homogeneous polynomials f, g ∈ k x, y, z [ ] of degrees 2 and 4, respectively, such that X ⊂ P is given by
$$w ^ { 2 } + f w = g.$$
If the characteristic of k is not 2, then after completing the square on the left-hand side, we may assume f = 0. For more details and proofs of these facts, see [Kol96, Section III.3, Theorem III.3.5] and [Man86, Section IV.24].
The restriction to X of the 2-uple embedding P → P 6 corresponds to the complete linear system | -2 K X | . Every hyperplane section of X ⊂ P is linearly equivalent with -K X . The projection P /axisshort/axisshort/arrowaxisright P 2 onto the first three coordinates restricts to a finite, separable morphism π X : X → P 2 of degree 2, which corresponds to the complete linear system | -K X | .
The morphism π X is ramified above the branch locus B X ⊂ P 2 given by f 2 + 4 g = 0. If the characteristic of k is not 2, then B X is a smooth curve. We denote the ramification locus π -1 ( B X ) of π X by R X . As for every double cover, the morphism π X induces an involution ι X : X → X that sends a point P ∈ X to the unique second point in the fiber π -1 X ( π X ( P )), or to P itself if π X is ramified at P . If X is clear from the context, then we sometimes leave out the subscript and write π, ι , B , and R for π X , ι X , B X , and R X , respectively.
As in [STVA14], we define a bitangent to B to be a line /lscript ⊂ P 2 for which π -1 ( /lscript ) is reducible. If the characteristic of k is not 2, then this is equivalent to a more intuitive definition: a bitangent line is a line in P 2 whose intersection points with B all have even intersection multiplicity. For each bitangent /lscript to B , the two components of π -1 ( /lscript ) are exceptional curves, say E and ι X ( E ), whose sum E + ι X ( E ) is linearly equivalent with -K X (see [STVA14, Lemma 4]). There are exactly 28 bitangent lines to B , and the irreducible components above them are exactly the 56 exceptional curves on X . Any point of X lies on at most four exceptional curves (see [TVAV09, proof of Lemma 4.1]); a generalised Eckardt point is a point on X that lies on exactly four exceptional curves.
## 3. Proof of the first main theorem
Set k 1 = k 2 = F 3 and k 3 = F 9 . Let γ ∈ k 3 denote an element satisfying γ 2 = γ + 1. Note that γ is not a square in k 3 . For i ∈ { 1 2 3 , , } , we define the surface X i in P = P (1 , 1 1 2) with , , coordinates x, y, z, w over k i by
$$\ddots \cdots \ddots \ddots & \ddots \\ X _ { 1 } \colon & - w ^ { 2 } \, = \, ( x ^ { 2 } + y ^ { 2 } ) ^ { 2 } + y ^ { 3 } z - y z ^ { 3 }, \\ X _ { 2 } \colon & - w ^ { 2 } \, = \, x ^ { 4 } + y ^ { 3 } z - y z ^ { 3 }, \\ X _ { 3 } \colon & \gamma w ^ { 2 } \, = \, x ^ { 4 } + y ^ { 4 } + z ^ { 4 }. \\ \cdots & \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \end{array}$$
These surfaces are smooth, so they are del Pezzo surfaces of degree 2. C. Salgado, D. Testa, and A. V´ arilly-Alvarado proved the following result.
Theorem 3.1. Let X be a del Pezzo surface of degree 2 over a finite field. If X is not isomorphic to X ,X 1 2 , and X 3 , then X is unirational.
Proof. See [STVA14, Theorem 1].
/square
We will use the following lemma to prove the complementary statement, namely that X ,X 1 2 , and X 3 are unirational as well.
Lemma 3.2. Let X be a del Pezzo surface of degree 2 over a field k . Suppose that ρ : P 1 → X is a nonconstant morphism; if the characteristic of k is 2 and the image of ρ is contained in the ramification divisor R X , then assume also that the field k is perfect. Then X is unirational.
Proof. See [STVA14, Theorem 17].
/square
For i ∈ { 1 2 3 , , } , we define a morphism ρ i : P 1 → X i by extending the map A 1 ( ) t → X i given by
↦
$$t \mapsto ( x _ { i } ( t ) \colon y _ { i } ( t ) \colon z _ { i } ( t ) \colon w _ { i } ( t ) ),$$
where
$$\cdots \cdots \cdots & x _ { 1 } ( t ) & = t ^ { 2 } ( t ^ { 2 } - 1 ), & x _ { 2 } ( t ) & = t ( t ^ { 2 } + 1 ) ( t ^ { 4 } - 1 ), & x _ { 3 } ( t ) & = ( t ^ { 4 } + 1 ) ( t ^ { 2 } - \gamma ^ { 3 } ), \\ y _ { 1 } ( t ) & = t ^ { 2 } ( t ^ { 2 } - 1 ) ^ { 2 }, & y _ { 2 } ( t ) & = - t ^ { 4 }, & y _ { 3 } ( t ) & = ( t ^ { 4 } - 1 ) ( t ^ { 2 } + \gamma ^ { 3 } ), \\ z _ { 1 } ( t ) & = t ^ { 8 } - t ^ { 2 } + 1, & z _ { 2 } ( t ) & = t ^ { 8 } + 1, & z _ { 3 } ( t ) & = ( t ^ { 4 } + \gamma ^ { 2 } ) ( t ^ { 2 } - \gamma ), \\ w _ { 1 } ( t ) & = t ( t ^ { 2 } - 1 ) ( t ^ { 4 } + 1 ) ( t ^ { 8 } + 1 ), & w _ { 2 } ( t ) & = t ^ { 2 } ( t ^ { 2 } + 1 ) ( t ^ { 1 0 } - 1 ), & w _ { 3 } ( t ) & = \gamma ^ { 2 } t ( t ^ { 8 } - 1 ) ( t ^ { 2 } + \gamma ). \\ & t \tt t \tt i e \tt o a v e t o \tt c h o b k \tt f o c h \tt i t h o m h i m \tt a. \tt i e \tt u a l d o f i n d \tt i e \tt t h o n l o m m i a l e$$
It is easy to check for each i that the morphism ρ i is well defined, that is, the polynomials x , y , z i i i , and w i satisfy the equation of X i , and that ρ i is non-constant.
Theorem 3.3. The del Pezzo surfaces X ,X 1 2 , and X 3 are unirational.
Proof. By Lemma 3.2, the existence of ρ , ρ 1 2 , and ρ 3 implies that X ,X 1 2 , and X 3 are unirational.
/square
Proof of Theorem 1.1. This follows from Theorems 3.1 and 3.3.
$$\Box$$
Remark 3.4 . Take any i ∈ { 1 2 3 , , } . Set A i = ρ i ( P 1 ) and C i = π i ( A i ), where π i = π X i : X i → P 2 is as described in the previous section. By Remark 2 of [STVA14], the surface X i is minimal, and the Picard group Pic X i is generated by the class of the anticanonical divisor -K X i . The same remark states that the linear system | -nK X i | does not contain a geometrically integral curve of geometric genus zero for n ≤ 3 if i ∈ { 1 2 , } , nor for n ≤ 2 if i = 3. For i ∈ { 1 2 , } , the curve A i has degree 8, so it is contained in the linear system | -4 K X i | . The curve A 3 has degree 6, so it is contained in the linear system | -3 K X i | . This means that the curve A i has minimal degree among all rational curves on X i . The restriction of π i to A i is a double cover A i → C i . The curve C i ⊂ P 2 has degree 4 for i ∈ { 1 2 , } and degree 3 for i = 3, and C i is given by the vanishing of h i , with
$$h _ { 1 } = & x ^ { 4 } + x y ^ { 3 } + y ^ { 4 } - x ^ { 2 } y z - x y ^ { 2 } z, \\ h _ { 2 } = & x ^ { 4 } - x ^ { 2 } y ^ { 2 } - y ^ { 4 } + x ^ { 2 } y z + y z ^ { 3 }, \\ h _ { 3 } = & x ^ { 2 } y + x y ^ { 2 } + x ^ { 2 } z - x y z + y ^ { 2 } z - x z ^ { 2 } - y z ^ { 2 } - z ^ { 3 }. \\ \cdot \quad & \sim \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot$$
For i ∈ { 1 2 , } , the curve C i has an ordinary triple point Q i , with Q 1 = (0 : 0 : 1), Q 2 = (0 : 1 : 1). The curve C 3 has an ordinary double point at Q 3 = (1 : 1 : 1). For all i , the point Q i lies on the branch locus B i = B X i .
We will see later that the curve C i intersects the branch locus B i with even multiplicity everywhere. Of course, one could check this directly as well using the polynomial h i . In fact, had we defined C i by the vanishing of h i , then one would easily check that C i satisfies the conditions of part (2) of Corollary 1.3, which gives an alternative proof of unirationality of X i without the need of the explicit morphism ρ i (see Example 5.16). Indeed, in practice we first found the curves C 1 , C 2 , and C 3 , and then constructed the parametrisations ρ , ρ 1 2 , ρ 3 , which allow for the more direct proof that we gave of Theorem 3.3.
The following section gives an alternative description of Yu. Manin's original construction of curves on del Pezzo surfaces of degree 2 and the generalisation of C. Salgado, D. Testa, and A. V´ arilly-Alvarado. The benefit of this description is that it allows us to write down the curves explicitly, in full generality. This will be used in Section 6. It also shows that quartic plane curves with a triple point and cubic points with a double point play an essential role in Manin's construction as well.
## 4. Manin's construction
Let k be a field and X a del Pezzo surface of degree 2 over k . Let π : X → P 2 be the associated double covering map, with ramification locus R ⊂ X , and associated involution ι : X → X . Let P ∈ X k ( ) be a point. For this entire section, we will assume that P does not lie on R . Yu. Manin proves unirationality of X under the extra condition that P does not lie on any of the 56 exceptional curves of X (see [Man86, Theorem 29.4]). His proof relies on the following construction of a non-constant morphism ρ : P 1 → X . Let bl P : ˜ X → X be the blow-up of X at P . Assume for the remainder of this paragraph that P does not lie on an exceptional curve of X . Then ˜ X is a del Pezzo surface of degree 1 (see [STVA14, Corollary 14]), on which there are 240 exceptional curves, splitting up naturally in 120 pairs ( E,E ′ ) with E + E ′ linearly equivalent to -2 K ˜ X , where K ˜ X is a canonical divisor on ˜ . X Let E ′ 0 ⊂ ˜ X be the exceptional curve on ˜ X that is paired with the exceptional curve E 0 above P . Then any isomorphism P 1 → E ′ 0 composed with the restriction of bl P to E ′ 0 yields a morphism ρ : P 1 → X . Unirationality of X then follows from applying the same construction to the generic point of bl P ( E ′ 0 ) instead of P (see [Man86, Theorem 29.4]).
- C. Salgado, D. Testa, and A. V´ arilly-Alvarado generalise Manin's construction to the case that P is not a generalised Eckardt point, that is, P lies on at most three exceptional curves. They show that in this case there is still a unique effective divisor on ˜ that is linearly equivalent X to -2 K ˜ X -E 0 ∼ -2 bl ∗ P K X -3 E 0 , and this divisor contains a component that is birational to P 1 (see [STVA14, Theorem 16 and its proof]).
We give a different description of this rational component by generalising a well-known alternative description of Manin's construction in terms of an elliptic fibration associated to the anti-canonical map ˜ X /axisshort/axisshort/arrowaxisright P 1 . We drop all assumptions on the number of exceptional curves going through P . Let bl P,ι P ( ) : Bl P,ι P ( ) X → X denote the blow-up of X at the points P and ι ( P ), which factors as the composition of bl P and the blow-up bl ι ( P ) of ˜ X at the point corresponding to ι ( P ). Let E P and E ι ( P ) be the exceptional curves above P and ι ( P ), respectively. Let bl Q : Bl Q P 2 → P 2 denote the blow-up of P 2 at Q . The map π : X → P 2 induces a morphism π ′ : Bl P,ι P ( ) X → Bl Q P 2 . Let L Q denote the line in the dual of P 2 consisting of all lines L ⊂ P 2 going through Q , and note that L Q is isomorphic to P 1 . We can identify the blow-up Bl Q P 2 with the subvariety of P 2 × L Q consisting of pairs ( R,L ) with R ∈ L . Let ψ : Bl P,ι P ( ) X → L Q be the composition of π ′ with the projection Bl Q P 2 → L Q . The fiber of ψ above a line L through Q is isomorphic to the pull back π ∗ ( L ) on X , which is linearly equivalent to -K X . By the adjunction formula we have
$$2 g ( \pi ^ { * } ( L ) ) - 2 = \pi ^ { * } ( L ) \cdot ( \pi ^ { * } ( L ) + K _ { X } ) = 0,$$
from which it follows that g π ( ∗ ( L )) = 1, so ψ is an elliptic fibration. The morphism π ′ maps the exceptional curves E P and E ι ( P ) isomorphically to the exceptional curve on Bl Q P 2 above Q , which in turn projects isomorphically to L Q . Hence, we obtain two sections σ P and σ ι ( P ) of ψ , mapping to E P and E ι ( P ) , respectively. The rational map ψ ◦ bl -1 ι ( P ) : ˜ X /axisshort/axisshort/arrowaxisright L Q ∼ = P 1 is associated to the linear system | -K ˜ X | on ˜ , X which has a unique base point, namely the point associated to ι ( P ). It also factors as the composition of π ◦ bl P and the projection p Q : P 2 /axisshort/axisshort/arrowaxisright L Q away from Q . These maps are summarised in the following commutative diagram.
We fix the section σ ι ( P ) of ψ as the neutral element of the Mordell-Weil group (Bl P,ι P ( ) X )( L Q ) of sections of ψ and we set
$$\rho = \text{bl} _ { P, \iota ( P ) } \circ ( - \sigma _ { P } ) \colon \mathfrak { L } _ { Q } \to X.$$
The following Proposition is well known among experts and implies that there is no conflict with the definition of ρ in Manin's construction above, at least after an appropriate identification of L Q with P 1 .
Proposition 4.1. If P does not lie on an exceptional curve, then bl ι ( P ) ◦ -( σ P ): L Q → ˜ X induces an isomorphism from L Q to the exceptional curve E ′ 0 ⊂ ˜ X of Manin's construction.
Proof. Given that P does not lie on the ramification curve R , the assumption that P does not lie on an exceptional curve implies that ˜ is a del Pezzo surface of degree 1 (see [STVA14, Corollary X 13]). Therefore, the lemma follows from the following general statement. For any del Pezzo surface Y of degree 1, the anti-canonical map Y /axisshort/axisshort/arrowaxisright P 1 induces an elliptic fibration of the blow-up ˜ Y of Y in the unique base point of the linear system | -K Y | , with the exceptional curve as the zero section. Multiplication by -1 induces an involution of Y that switches any two exceptional curves E and E ′ for which E + E ′ is linearly equivalent to -2 K Y . /square
We continue to investigate the general case, in which P may lie on any number of exceptional curves.
Proposition 4.2. The following statements are equivalent.
- (1) The map ρ : L Q → X is not constant.
- (2) The section σ P does not have order 2 .
- (3) The composition p Q ◦ π ◦ ρ : L Q → X → P 2 /axisshort/axisshort/arrowaxisright L Q is a well-defined rational map that extends to the identity on L Q .
Proof. The map ρ is constant if and only if the image of -σ P is contained in E P or E ι ( P ) . Since σ P , σ ι ( P ) , and -σ P are all sections of ϕ , this happens if and only if -σ P = σ P or -σ P = σ ι ( P ) . The latter case does not happen as σ ι ( P ) = 0 in the group (Bl P,ι P ( ) X )( L Q ). Hence, the map ρ is constant if and only if σ P = -σ P , which proves the equivalence of (1) and (2). Moreover, if ρ is constant, then it sends L Q to P and the image of π ◦ ρ is then Q . Since π is finite, the map ρ is non-constant if and only if π ◦ ρ is non-constant. As the base locus of p Q consists of just the one point Q , the map π ◦ ρ is non-constant if and and only if the composition p Q ◦ π ◦ ρ is a well-defined rational map. If this is the case, then by the commutivity of the diagram above, this composition coincides with ψ ◦ ( -σ P ) = id L Q . This proves that (3) is equivalent as well. /square
We will see later, in Theorem 4.13, that the conditions in Proposition 4.2 are satisfied if and only if P is not a generalised Eckardt point.
We set B = π R ( ), so B is the branch curve of π . Recall that the fiber of ψ above a line L ∈ L Q is isomorphic to the pull back π ∗ ( L ), which is singular if and only if L is tangent to B , and reducible if and only if L is a bitangent to B . In the latter case, the fiber has Kodaira type I 2 , with the two components being the strict transforms of the two exceptional curves of X above L . We now state the first alternative description of ρ .
Proposition 4.3. Let L ∈ L Q be a line that is not tangent to the branch curve B . Then ρ L ( ) is the unique point on π -1 ( L ) that is linearly equivalent with the divisor 2( ( ι P )) -( P ) on π -1 ( L ) .
Proof. Since L is not tangent to B , the curve C = π -1 ( L ) is smooth and has genus 1. By the definition of ρ , we find ρ L ( ) = -P in the group C k ( ) with ι ( P ) as the neutral element. The statement therefore follows from a standard result on elliptic curves (see [Sil09, Corollary III.3.5]). /square
The following lemma will allow us to describe ρ L ( ) more explicitly, also in the more general case that L is not a bitangent to B .
Lemma 4.4. Let x, y, w denote the coordinates of the weighted projective space P (1 , 1 2) , over k . Set Ω = (0 : 0 : 1) and U = P (1 , 1 2) , -{ } Ω , and let α : U → P 1 be the projection away from Ω . Let C ⊂ P (1 , 1 2) , be a reduced, not-necessarily-irreducible curve given by w 2 + fw = g for some homogeneous polynomials f, g ∈ k x, y [ ] of degree 2 and 4 , respectively. Let S ∈ C k ( ) be a rational point where the restriction α | C : C → P 1 does not ramify. Then there is a unique section τ : P 1 → U of α of which the image intersects C at S with multiplicity at least 3 . The following hold.
- (1) The image τ ( P 1 ) has degree 2 in P (1 , 1 2) , .
- (2) If C is reducible, then τ induces an isomorphism from P 1 to the irreducible component of C containing S .
- (3) If C is irreducible, then τ ( P 1 ) intersects C in a unique fourth point. If this point is different from S , then the corresponding intersection multiplicity is 1 .
/negationslash
/negationslash
Proof. Let τ : P 1 → U be a section of α . Then there are homogeneous polynomials q, r ∈ k x, y [ ] with no factor of q dividing r doubly, such that τ sends ( x : y ) to ( xq x, y ( ) : yq x, y ( ) : r x, y ( )). Any point of P 1 on which q vanishes maps to Ω ∈ U , so we conclude that q is a constant and thus r has degree 2. Without loss of generality we assume q = 1. The image τ ( P 1 ) is given by w = ( r x, y ), so τ ( P 1 ) has degree 2 as well, which proves (1). After switching x and y if necessary, we may assume that S lies in the affine part given by y = 0. Set v = w/y 2 and t = x/y and let a, b, c ∈ k be such that r x, y ( ) = ax 2 + bxy + cy 2 . Then on the curve τ ( P 1 ), given by v = ( r t, 1), we have
$$v ( S ) = r ( t, 1 ) ( S ), \quad \frac { d v } { d t } ( S ) = \frac { d r ( t, 1 ) } { d t } ( S ), \quad \text{and} \quad \frac { d ^ { 2 } v } { d t ^ { 2 } } ( S ) = \frac { d ^ { 2 } r ( t, 1 ) } { d t ^ { 2 } } ( S ).$$
Hence, the condition that τ ( P 1 ) intersects C with multiplicity at least 3 at S is equivalent with these equalities holding on C as well. This gives three linearly independent linear equations in a, b , and c , so there are unique a, b, c ∈ k for which τ ( P 1 ) intersects C at S with multiplicity at least 3. Suppose that τ is this section. Given that α | C does not ramify at S , there is a unique irreducible component C 0 of C containing S . If C 0 and τ ( P 1 ) do not coincide, then the total number of their intersection points, counted with multiplicities, is (deg C 0 ) · (deg τ ( P 1 )) / (1 · 1 · 2) = deg C 0 . Since C is a double cover of P 1 , either C is irreducible, in which case C 0 = C has degree 4, so there is a unique fourth intersection point (proving (3)), or C has two isomorphic irreducible components of degree 2. In the latter case, the intersection multiplicity at S exceeds the total intersection number C 0 · τ ( P 1 ). From this contradiction, we conclude that C 0 and τ ( P 1 ) do coincide; as τ is a section of α | C 0 , we find that τ induces an isomorphism from P 1 to C 0 , which proves (2). /square
Remark 4.5 . In the case that C in Lemma 4.4 is smooth, the Riemann-Roch Theorem can be used to give a more efficient proof.
Remark 4.6 . Assume the same setting as in Lemma 4.4. We will make the section τ explicit. Write f and g as
$$f = f _ { 2 } x ^ { 2 } + f _ { 1 } x y + f _ { 0 } y ^ { 2 } \quad \text{and} \quad g = g _ { 4 } x ^ { 4 } + g _ { 3 } x ^ { 3 } y + g _ { 2 } x ^ { 2 } y ^ { 2 } + g _ { 1 } x y ^ { 3 } + g _ { 0 } y ^ { 4 }.$$
/negationslash
Without loss of generality, we assume S = (0 : 1 : w 0 ) for some w 0 ∈ k with w 2 0 + f w 0 0 = g 0 . Write δ = f 0 +2 w 0 . The fact that α | C does not ramify at S is equivalent with δ = 0. As in the proof of Lemma 4.4, we set v = w/y 2 and t = x/y , and we let a, b, c ∈ k be such that τ sends ( x : y ) to ( x : y : r x, y ( )) with r x, y ( ) = ax 2 + bxy + cy 2 . From τ ((0 : 1)) = S , we find c = w 0 . In
terms of the affine coordinates t, v , the intersection points of τ ( P 1 ) and C correspond to the roots of F t ( ) = r t, ( 1) 2 + f ( t, 1) r t, ( 1) -g t, ( 1). Write
$$F ( t ) = F _ { 4 } t ^ { 4 } + F _ { 3 } t ^ { 3 } + F _ { 2 } t ^ { 2 } + F _ { 1 } t + F _ { 0 },$$
with F i ∈ k . Then we have
$$F _ { 0 } & = 0, \\ F _ { 1 } & = \delta b + w _ { 0 } f _ { 1 } - g _ { 1 }, \\ F _ { 2 } & = \delta a + b ^ { 2 } + f _ { 1 } b + w _ { 0 } f _ { 2 } - g _ { 2 }, \\ F _ { 3 } & = 2 a b + f _ { 1 } a + f _ { 2 } b - g _ { 3 }, \\ F _ { 4 } & = a ^ { 2 } + f _ { 2 } a - g _ { 4 }. \\ \intertext { v \cap f \cap f \Phi 1 \, \sum \, \text{and} \, C \text{ to some multi-lit} }$$
/negationslash
The intersection multiplicity of τ ( P 1 ) and C at S equals the multiplicity of t = 0 as a root of F t ( ). Since this multiplicity is at least 3, we find F 1 = F 2 = 0. This determines a and b uniquely, as we have δ = 0. If, for these a, b , we have F 3 = F 4 = 0 as well, then τ ( P 1 ) is contained in C , so C consists of two components. Otherwise, the fourth intersection point corresponds to t = -F /F 3 4 , so it is ( -F 3 : F 4 : aF 2 3 -bF F 3 4 + w F 0 2 4 ).
We identify X with a model in P given by w 2 + fw = g for some homogeneous polynomials f, g ∈ k x, y, z [ ] of degrees 2 and 4, respectively. Set Ω = (0 : 0 : 0 : 1) ∈ P and U = P - { Ω . } The projection ϕ : U → P 2 onto the first three coordinates, that is, the projection away from Ω, restricts to the map π : X → P 2 .
Proposition 4.7. Let L ∈ L Q be a line through Q . Then there is a unique morphism τ : L → U with ϕ ◦ τ = id L of which the image intersects X at P with multiplicity at least 3 . The following hold.
- (1) The image τ ( L ) has degree 2 in P .
- (2) If L is bitangent to the branch curve B , then τ induces an isomorphism from L to the unique exceptional curve above L that contains P .
- (3) If L is not bitangent to the branch curve B , then τ ( L ) intersects X in a unique fourth point, which is ρ L ( ) . If ρ L ( ) = P , then the corresponding intersection multiplicity is 1 .
/negationslash
Proof. The inverse image X L = π -1 ( L ) is contained in the inverse image U L = ϕ -1 ( L ). Note that every morphism τ : L → U with ϕ ◦ τ = id L factors through U L , and the intersection multiplicity of τ ( P ) with X at P equals the intersection multiplicity of τ ( P ) with X L at P . Since U L ∪ { Ω } is isomorphic to the weighted projective space P (1 , 1 2), the unique existence of , τ follows from Lemma 4.4, applied to C = X L .
/negationslash
The same lemma also implies (1) and (2). Suppose L is not bitangent to B . Then by Lemma 4.4, the image τ ( L ) intersects X L , and thus X , in a unique fourth point, say T , and the corresponding intersection multiplicity is 1 if T = P . By (1) of Lemma 4.4, the divisor ( T ) + 3( P ) is linearly equivalent to the intersection of X L with a curve of degree 2 in U L ∪ { Ω } ∼ = P (1 , 1 2). , Since ( P ) + ( ( ι P )) is the intersection of X L with a curve of degree 1, and all curves in P (1 , 1 2) of the , same degree are linearly equivalent, we find that the divisors ( T )+3( P ) and 2( P )+2( ( ι P )) on X L are linearly equivalent. From Proposition 4.3 we find T = ρ L ( ), which finishes the proof of (3). /square
From now on, for any L ∈ L Q , we let
denote the map associated to L as in Proposition 4.7. Let the map ν : Bl Q P 2 → U be given by sending the pair ( S, L ) ∈ Bl Q P 2 ⊂ P 2 × L Q with S ∈ L to τ L ( S ).
Proposition 4.8. The map ν : Bl Q P 2 → U is a morphism.
/negationslash
Proof. Since ν is well defined globally, it suffices to show that ν is a morphism locally. Let T ∈ P 2 be a point not equal to Q , and let V T ⊂ Bl Q P 2 be the open subset of all ( S, L ) with T ∈ L . Let L T denote the line in the dual of P 2 consisting of all lines L ⊂ P 2 going through T . Let L ◦ Q ⊂ L Q denote the subset of all L ∈ L Q with T ∈ L . For each L ∈ L ◦ Q , there is an isomorphism µ L : L → L T that sends S ∈ L to the line through T and S . The map ν 0 : V T → L T × L ◦ Q sending ( S, L ) to ( µ L ( S , L ) ) is an isomorphism. Let Mor( L T , U ) denote the scheme of morphisms from L T to U . Define α : L ◦ Q → Mor( L T , U ) to be the map that sends L to τ L ◦ ( µ L ) -1 . We may identify L T with P 1 , and write O L T ( n ) for the sheaf on L T corresponding to O P 1 ( n ), for any integer n . For any L ∈ L ◦ Q , it folows from Remark 4.6 that there exist elements ξ, η, ζ ∈ Γ( O L T (1)) and ω ∈ Γ( O L T (2)), such that α L ( ): L T → U ⊂ P is given by S → ( ξ S ( ) : η S ( ) : ζ ( S ) : ω S ( )). Since we can do this for the generic point of L ◦ Q as well, we find that α is a rational map, and as L ◦ Q is a smooth curve, the map α is in fact a morphism. On V T , the map ν factors as
/negationslash
↦
$$V _ { T } \stackrel { \nu _ { 0 } } { \longrightarrow } \mathfrak { L } _ { T } \times \mathfrak { L } _ { Q } ^ { \circ } \stackrel { ( \text{id} _ { \mathfrak { L } _ { T } }, \alpha ) } { \longrightarrow } \mathfrak { L } _ { T } \times \text{Mor} ( \mathfrak { L } _ { T }, U ) \stackrel { \text{ev} } { \longrightarrow } U$$
where the last morphism is the evaluation map given by ev( L, β ) = β L ( ). Hence, the restriction of ν to V T is a morphism. Since T was arbitrary, we conclude that ν is a morphism as well. /square
The following proposition shows that we can combine all the maps τ L : L → U to one map P 2 → U .
Proposition 4.9. The map ν : Bl Q P 2 → U factors through a unique morphism ξ : P 2 → U . Moreover, the following hold.
- (1) For any L ∈ L Q , the restriction ξ | L of ξ to L equals τ L .
- (2) The map ξ is a section of the map ϕ : U → P 2 with ξ Q ( ) = P .
- (3) The image ξ ( P 2 ) has degree 2 .
- (4) The intersection X ∩ ξ ( P 2 ) is the union of ρ ( L Q ) and the exceptional curves on X that contain P .
Proof. By Zariski's Main Theorem (see [Har77, Corollary III.11.4 and its proof]), we have an isomorphism (bl Q ) ∗ O Bl Q P 2 → O P 2 . Note also that the map ν contracts the exceptional curve above Q to the point P . Therefore, it follows from [Gro61, Lemme 8.11.1] that ν factors as the composition of bl Q and a unique morphism ξ : P 2 → U . Let L be a line through Q and S a point on L . Then we have ξ S ( ) = ν S, L ( ) = τ L ( S ), which proves (1). It also follows that we have ϕ ξ S ( ( )) = ϕ τ ( L ( S )) = S , so ξ is a section of ϕ . From τ L ( Q ) = P , we find ξ Q ( ) = P , which proves (2). The line ϕ -1 ( S ) intersects Y = ( ξ P 2 ) in the same number of points as it intersects τ L ( L ), so the degree of Y equals 2 by Proposition 4.7. This proves (3). It follows from the same proposition that the intersection of τ L ( L ) = ξ L ( ) with X is equal to { P, ρ L ( ) } if L is not bitangent to B , and equal to the exceptional curve above L that contains P if L is bitangent to B . As Y = ξ ( P 2 ) is the union of the images ξ L ( ) as L runs through L Q , the last statement follows as well. /square
Remark 4.10 . We can make the morphism ξ : P 2 → U completely explicit. Let X ⊂ P be given by w 2 + fw = g with f, g ∈ k x, y, z [ ] homogeneous of degree 2 and 4, respectively. Write f and g as
$$f = f _ { 2 } + f _ { 1 } z + f _ { 0 } z ^ { 2 } \quad \text{and} \quad g = g _ { 4 } + g _ { 3 } z + g _ { 2 } z ^ { 2 } + g _ { 1 } z ^ { 3 } + g _ { 0 } z ^ { 4 },$$
/negationslash with f , g i i ∈ k x, y [ ] homogeneous of degree i . Without loss of generality, we assume Q = (0 : 0 : 1), that is, P = (0 : 0 : 1 : w 0 ) for some w 0 ∈ k with w 2 0 + f w 0 0 = g 0 . Write δ = 2 w 0 + f 0 . The fact that π does not ramify at P is equivalent with δ = 0. Let ξ ′ : P 2 → U be the morphism that sends ( x : y : z ) to ( x : y : z : w ) with w satisfying
$$\delta ^ { 3 } w = \delta ^ { 2 } ( g _ { 2 } + g _ { 1 } z + 2 g _ { 0 } z ^ { 2 } - w _ { 0 } f ) - g _ { 1 } ^ { 2 } + f _ { 1 } ( g _ { 0 } f _ { 1 } - f _ { 0 } g _ { 1 } ).$$
We claim ξ = ξ ′ . To prove the claim, one first checks that this w equals a + bz + w z 0 2 with a and b satisfying
$$( 4 ) \quad \delta b + w _ { 0 } f _ { 1 } - g _ { 1 } = 0 \quad \text{ and } \quad \delta a + b ^ { 2 } + f _ { 1 } b + w _ { 0 } f _ { 2 } - g _ { 2 } = 0.$$
Now take any λ ∈ k . By projecting away from the y -coordinate, we identify the line L ⊂ P 2 and the hyperplane H ⊂ P , both given by y = λx , with P 1 and P (1 , 1 2), respectively. , The intersection H ∩ X gets identified with a curve C ∈ P (1 , 1 2). , Under these identifications, the equations (4) correspond to the vanishing of F 1 and F 2 as in Remark 4.6. This implies that the restriction ξ ′ | L of ξ ′ to L corresponds to the map τ L : L → U . From part (1) of Proposition 4.9 we therefore conclude ξ ′ = . This explicit description of ξ ξ could have also been used to give a less-enlightening proof of Propositions 4.8 and 4.9.
Remark 4.11 . Assume the characteristic of k is not 2. We have assumed that P does not lie on the ramification curve of π , but we may deform the surface X and the point P such that in the limit, the point does lie on the ramification curve. In the limit we have δ = 0 and f 0 = -2 w 0 and g 0 = w 2 0 + f w 0 0 = -w 2 0 , which implies that the limit of ξ ( P 2 ), which is given by (3), is the non-reduced surface in P given by the vanishing of
$$- g _ { 1 } ^ { 2 } + f _ { 1 } ( g _ { 0 } f _ { 1 } - f _ { 0 } g _ { 1 } ) = - ( g _ { 1 } - w _ { 0 } f _ { 1 } ) ^ { 2 }.$$
The intersection of the surface given by g 1 -w f 0 1 = 0 with U = P -{ } Ω is exactly the pull back under the projection ϕ : U → P 2 of the tangent line L to B at Q , so we find that, as a divisor, the limit of the intersection X ∩ ξ ( P 2 ) is 2 π ∗ ( L ).
It follows from Propositions 4.12 and 4.14 below that the divisor bl ∗ P ( ξ ( P 2 ) ∩ X ) -3 E P on ˜ X corresponds with the effective divisor denoted by D in the proof of [STVA14, Theorem 16]. We will mimic part of this proof, working on X instead of ˜ , to show the divisor X ξ ( P 2 ) ∩ X is reduced.
Proposition 4.12. Set Y = ξ ( P 2 ) ∈ Div P and let D ∈ Div X be the pull back of Y to X . Then D is linearly equivalent with -2 K X . We have D = D 1 + ∑ E E , where the sum runs over all exceptional curves E containing P , and where D 1 = ρ ( L Q ) if ρ is not constant and D 1 = 0 otherwise.
/negationslash
Proof. By Proposition 4.9, the image Y of ξ is a hypersurface in P of degree 2, which implies the first statement. Let E denote the set of all exceptional curves E containing P . By Proposition 4.9, we may write D = D 1 + ∑ E n E E , where the sum runs over all E ∈ E , and where D 1 is effective and supported on the integral curve ρ ( L Q ) if ρ is not constant and D 1 = 0 otherwise. If ρ is not constant, then D 1 = m ρ · ( L Q ) for some positive integer m . If we had m > 1, that is, the intersection Y ∩ X is nonreduced along ρ ( L Q ), then for any line L ⊂ P 2 through Q , we find, after intersecting with the hyperplane H L = ϕ -1 ( L ) ⊂ P , that the intersection of τ L ( L ) = Y ∩ H L with X is not smooth at the point ρ L ( ). This contradicts Proposition 4.7, part (3), when ρ L ( ) = P , so we conclude m = 1 and thus D 1 = ρ ( L Q ) in the case that ρ is not constant.
It remains to prove n E = 1 for all E ∈ E . From Proposition 4.9 we find n E ≥ 1 for all E ∈ E . Suppose there is an E 0 ∈ E with n E 0 > 1. Then D -2 E 0 is an effective divisor that is linearly equivalent with -2 K X -2 E 0 , and thus with 2 ( ι E 0 ). Since these divisors have negative selfintersection, their linear system contains only one effective divisor, so we find D -2 E 0 = 2 ( ι E 0 ). As P is not contained in the ramification locus, the exceptional curve ι ( E 0 ) does not contain P , so it is not contained in E . Hence, we find D 1 = 2 ( ι E 0 ), which implies that ρ is not constant and ρ ( L Q ) = ι ( E 0 ). But then the equality D 1 = 2 ( ι E 0 ) contradicts m = 1, so we conclude that we have n E = 1 for all E ∈ E . /square
We are now ready to prove that the conditions of Proposition 4.2 hold if and only if P is not a generalised Eckardt point (cf. [STVA14, Theorem 16]).
Theorem 4.13. The morphism ρ : L Q → X is constant if and only if P is a generalised Eckardt point.
Proof. Let E denote the set of all exceptional curves containing P , and set n = # . Let E D and D 1 be as in Proposition 4.12, so that D 1 is linearly equivalent with -2 K X -∑ E E , where the sum runs over all E ∈ E . Then we have -K X · D 1 = 4 -n . As -K X is ample and D 1 is effective, we have -K X · D 1 ≥ 0 with equality if and only if D 1 = 0. We conclude that P is a generalised Eckardt point, that is, we have n = 4, if and only if D 1 = 0, that is, ρ is constant. /square
The following proposition describes the image of ρ more precisely in the case that P is not a generalised Eckardt point.
Proposition 4.14. Suppose P is not a generalised Eckardt point. Let n be the number of lines in L Q that are bitangent to the branch curve B . Then π induces an isomorphism from ρ ( L Q ) to its image C . Furthermore, the curve C has degree 4 -n and Q is a point of multiplicity 3 -n on C .
Proof. By Theorem 4.13, the morphism ρ is not constant, so C is an irreducible curve. Since P is not a ramification point of π , the number of exceptional curves through P is n . Let D and D 1 = ρ ( L Q ) be as in Proposition 4.12. Note that the same proposition tells us that D and D 1 are reduced. Since ξ is a section of ϕ : U → P 2 , and ϕ restricts to π : X → P 2 , the map π induces an isomorphism from D = ξ ( P 2 ) ∩ X onto its image. In particular, it induces an isomorphism D 1 → C , so we find π ∗ ( D 1 ) = C . Let M ⊂ P 2 be any line not containing Q . Then the projection formula (see [Kol96, Proposition VI.2.11]) implies
$$\deg C = C \cdot M = \pi _ { * } ( D _ { 1 } ) \cdot M = D _ { 1 } \cdot \pi ^ { * } ( M ) = D _ { 1 } \cdot ( - K _ { X } ) = 4 - n.$$
Let m be the multiplicity of Q on C . By Proposition 4.2, the projection map p Q : P 2 /axisshort/axisshort/arrowaxisright L Q induces a birational map C /axisshort/axisshort/arrowaxisright L Q . This implies m = deg C -1 = 3 -n . /square
Remark 4.15 . Proposition 4.14 shows that if P is not a generalised Eckardt point, then C = π ρ ( ( L Q )) is a curve in P 2 with a point of multiplicity deg C -1 off the branch locus B . Moreover, in the next section, we will see that C intersects B with even intersection multiplicity everywhere. We will also see that outside characteristic 2, all such curves pull back to rational curves on X and imply unirationality of X . See Example 5.14.
## 5. Proof of the second main theorem
From now on, we assume that the characteristic of k is not 2. We continue the notation of the previous section, but we drop the assumption that P does not lie on the ramification locus R . In the previous section, we described the image in P 2 of the rational curve on X that is obtained through Manin's construction. In this section we will conversely describe all curves C ⊂ P 2 of which the pull back to X contains a component of genus 0 (cf. Remark 5.8). Whenever such a component is rational over k , the surface X is unirational by Lemma 3.2. The first subsection gives a description in terms of the normalisation of C , while the second subsection gives a description in terms of C itself, under a mild hypothesis.
- 5.1. The pull back of a plane curve in terms of its normalisation. Let k denote an algebraic closure of k . Given a reduced curve C over k , the normalisation map ϑ : ˜ C → C is unique up to isomorphism; the curve ˜ is regular and both C ϑ and ˜ are called the normalisation of C C . For more details, see [Mum99, Theorem III.8.3] for the case that C is irreducible. For the general case, take the disjoint union of the normalisations of the irreducible components. If X is a variety over k and /lscript is a field extension of k , then we set X /lscript = X × k /lscript .
A variety X over a field k is called rational if X is integral and there is a birational map X /axisshort/axisshort/arrowaxisright P n for some nonnegative integer n ; we say that X is geometrically rational if X /lscript is rational for some field extension /lscript of k .
Recall that the geometric genus g C ( ) of a geometrically integral curve C over k is defined to be the geometric genus of the unique regular projective geometrically integral model of C . If C is itself projective, then this model is the normalisation ˜ C of C . Note that we have g C ( k ) ≤ g C ( ) with equality if and only if ˜ is smooth (see [Tat52]). C In particular, we have g C ( ) = 0 if and only if ˜ is smooth and C C is geometrically rational.
Recall that if C ⊂ P 2 is a curve over k and S ∈ C a closed point with local ring O S,C , while C ′ ⊂ P 2 is a curve that is given in an open neighbourhood V ⊂ P 2 of S by h = 0 for some h ∈ O P 2 ( V ), then the intersection multiplicity µ S ( C, C ′ ) of C and C ′ at S is the length of the O S,C -module O S,C / h ( ), assuming that S does not lie on a common component of C and C ′ . If S is a smooth point of C , then the local ring O S,C is a discrete valuation ring, say with valuation v S , and µ S ( C, C ′ ) equals v S ( h ).
We extend the notion of intersection multiplicity, replacing the point S on the curve C by a 'branch' of C , that is, a point of the normalisation of C .
Definition 5.1. Let C ⊂ P 2 be a curve and let ϑ : ˜ C → C be the normalisation of C . Let T ∈ ˜ C be a closed point with local ring O T,C ˜ . Let C ′ ⊂ P 2 be a curve that is given in an open neighbourhood V ⊂ P 2 of ϑ T ( ) by h = 0 for some h ∈ O P 2 ( V ) . If the curves C and C ′ have no irreducible components in common, then the intersection multiplicity µ T ( ˜ C,C ′ ) of ˜ C and C ′ at T is defined to be the length of the O T,C ˜ -module O T,C ˜ / h ( ) .
With the same notation as above, the intersection multiplicity µ T ( ˜ C,C ′ ) is the same as ord T ( h ) as defined in [Ful98, Section 1.2]. Since ˜ C is regular, the local ring O T,C ˜ is a discrete valuation ring, say with valuation v T , and we have µ T ( ˜ C,C ′ ) = v T ( h ). If k is algebraically closed, then we have µ T ( ˜ C,C ′ ) = dim k O T,C ˜ / h ( ).
Lemma 5.2. Let C, C ′ ⊂ P 2 be curves and let ϑ : ˜ C → C be the normalisation of C . Then for every S ∈ C we have
↦
where the summation runs over all closed points T ∈ ˜ C with ϑ T ( ) = S and where [ k T ( ) : k S ( )] denotes the degree of the residue field extension.
$$\lambda _ { S } ( C, C ^ { \prime } ) = \sum _ { T \mapsto S } \mu _ { T } ( \tilde { C }, C ^ { \prime } ) \cdot [ k ( T ) \colon k ( S ) ], \\ n \ r u n s \ o v e r \ a l l \, & \text{closed points} \, T \in \tilde { C } \, \text{ with } \vartheta ( T ) = S \text{ and}$$
Proof.
$$P r o o f. \text{ This follows immediately from [Ful98, Example 1.2.3]. }$$
Definition 5.3. Let C, C ′ ⊂ P 2 be curves over k that do not have any components in common. Let Γ denote either C or its normalisation ˜ C . Then we define the subset b (Γ , C ′ ) of Γ( k ) as
$$b ( \Gamma, C ^ { \prime } ) = \{ T \in \Gamma ( \overline { k } ) \ \colon \ \mu _ { T } ( \Gamma, C ^ { \prime } ) \ i s \ o d d \}.$$
Recall that we are assuming that the characteristic of k is not 2.
Lemma 5.4. Let D be a geometrically integral curve on X , let C = π D ( ) be its image under π , and assume C is not equal to the branch locus B . The restriction of π to D induces a morphism ˜ : π ˜ D → ˜ C between the normalisations of D and C . The branch locus of the morphism ˜ D k → ˜ C k induced by ˜ π is exactly b C, B ( ˜ ) ⊂ ˜ ( C k ) .
Proof. Without loss of generality, we assume k = k . Let ϑ denote the normalisation map ˜ C → C . Let T ∈ ˜ ( C k ) be a point. Since ˜ is regular, the local ring C O T,C ˜ is a discrete valuation ring, say with valuation v T . As the characteristic of k is not equal to 2, there is an open neighbourhood V ⊂ P 2 of ϑ T ( ) and an element h ∈ O P 2 ( V ) such that the double cover π -1 ( V ) of V is isomorphic to the subvariety of V × A 1 ( u ) given by u 2 = h . We denote the image of h in the local ring O T,C ˜ and the function field k C ( ˜ ) = k C ( ) by h as well. The extension k C ( ) ⊂ k D ( ) of function fields is obtained by adjoining a square root η ∈ k D ( ) of h to k C ( ). Note that the degree of the restriction of π to D is 1 if and only if this extension is trivial, i.e., h is a square in k C ( ). The intersection B ∩ V is given by h = 0, so we have µ T ( ˜ C,B ) = v T ( h ). Suppose T ′ ∈ ˜ ( D k ) is a point with ˜( π T ′ ) = T . Since the characteristic of k is not equal to 2, the extension O T,C ˜ ⊂ O T ,D ′ ˜ of discrete valuation rings of k C ( ) and k D ( ) = k C ( )( η ), respectively, is ramified if and only if v T ( h ) is odd, that is, T is contained in b C, B ( ˜ ), which proves the lemma. /square
Proposition 5.5. Let D be a geometrically integral projective curve on X , let C = π D ( ) be its image under π , and assume g C ( ) = 0 . Assume also that C is not equal to the branch locus B . Let ˜ C denote the normalisation of C and set n = # ( ˜ b C, B ) . The following statements hold.
- (1) If n = 0 , then π restricts to a birational morphism D → C and g D ( ) = 0 .
- (2) If n > 0 , then π restricts to a double cover D → C and g D ( ) = g D ( k ) = 1 2 n -1 .
/negationslash
Proof. From g C ( ) = 0, we find that the normalisation ˜ C is smooth. Since the characteristic of k is not 2 and for any finite separable field extension /lscript of k we have g D ( /lscript ) = g D ( ) (see [Tat52, Corollary 2]), we may (and do) replace k , without loss of generality, by a quadratic extension /lscript for which ˜ ( C /lscript ) = ∅ . Then ˜ is isomorphic to C P 1 . Let ˜ denote the normalisation of D D . The morphism π induces a morphism ˜ : π ˜ D → ˜ = C ∼ P 1 of degree at most 2. We claim that ˜ D is smooth. Indeed, if deg(˜) = 1, then this is clear. π If deg(˜) = 2, then because the characteristic of π k is not 2, the curve ˜ D can be covered by open affine curves that are given by y 2 = f ( x ) for some polynomial f ∈ k x [ ]; the regularity of ˜ D implies that each polynomial f is separable, which implies that ˜ D is smooth. This shows that g D ( ) = g D ( k ), so we may (and do) replace k , without loss of generality, by k .
From g C ( ) = 0, we find that C does not equal the branch locus B , so we may apply Lemma 5.4. The Riemann-Hurwitz formula then yields
$$2 g ( D ) - 2 = 2 g ( \tilde { D } ) - 2 = \deg ( \tilde { \pi } ) \cdot ( 2 g ( \tilde { C } ) - 2 ) + n = n - 2 \deg ( \tilde { \pi } ).$$
If n = 0, then we find deg(˜) = 1 and π g D ( ) = 0. If n > 0, then ˜ is not unramified, so deg(˜) = 2 π π and we obtain g D ( ) = 1 2 n -1. /square
Corollary 5.6. Let C ⊂ P 2 be a geometrically integral projective curve with g C ( ) = 0 that is not equal to the branch locus B . Let ˜ C denote its normalisation and set n = # ( ˜ b C, B ) . The following statements hold.
- (1) If n = 0 , then there exists a field extension /lscript of k of degree at most 2 such that the preimage π -1 ( C /lscript ) consists of two irreducible components that are birationally equivalent with C /lscript . For each such /lscript for which C /lscript is rational, the surface X /lscript is unirational.
- (2) If n = 0 and C is rational and there exists a rational point P ∈ X k ( ) with π P ( ) ∈ C -B , then the preimage π -1 ( C ) consists of two rational components and X is unirational.
- (3) If n > 0 , then the preimage π -1 ( C ) is geometrically integral and has geometric genus 1 2 n -1 .
- (4) If n = 2 and the preimage π -1 ( C ) is rational, then the surface X is unirational.
Proof. As the characteristic of k is not 2, the branch curve B is smooth of genus 3. Set A = π -1 ( C ) and A k = A × k k . The morphism A → C induced by π has degree 2. From g C ( ) = 0, we find that C is not the branch curve B , so A is geometrically reduced. Since A k = A × k k consists of at most two components, there is an extension /lscript of k of degree at most 2 such that the components of A /lscript = A × k /lscript are geometrically irreducible. Let /lscript be such an extension and let D be an irreducible component of A /lscript .
Suppose n = 0. Applying Proposition 5.5 to D /lscript and C /lscript = π D ( /lscript ) shows that the morphism D /lscript → C /lscript induced by π is a birational map. Since A /lscript → C /lscript has degree 2, there is a unique second component of A /lscript , which equals ι ( D /lscript ). If, moreover, C /lscript is rational, then Lemma 3.2 implies the unirationality of X /lscript , proving statement (1) .
/negationslash
Assume C is itself rational and there is a rational point P ∈ X k ( ) such that π P ( ) ∈ C -B . Then we have that P = ( ι P ) and the points P and ι ( P ) lie in different components of A /lscript = D /lscript ∪ ι ( D /lscript ). Since the Galois group G = G /lscript/k ( ) fixes the points P and ι ( P ), it follows that G also fixes D /lscript and ι ( D /lscript ), so these components are defined over k . Then statement (2) follows from (1) taking /lscript = k .
Suppose n > 0. By Proposition 5.5, the morphism D k → C k induced by π has degree 2, so D k is the only component of A k and therefore A is geometrically integral. Its genus follows from Proposition 5.5.
Statement (4) follows immediately from Lemma 3.2.
/square
Remark 5.7 . Let D be a geometrically integral curve over a field k with g D ( ) = 0. Then there exists a field extension /lscript of k of degree at most 2 such that D /lscript is rational. In fact, if k is a
finite field, then D is rational over k . Therefore, if k is finite in Corollary 5.6, then C is rational; moreover, by case (3) and (4) we conclude that if n = 2, then X is unirational over k .
Remark 5.8 . Proposition 5.5 and Corollary 5.6 imply that the geometrically integral projective curves D ⊂ X with g D ( ) = 0 are exactly the geometrically irreducible components above geometrically integral projective curves C ⊂ P 2 with g C ( ) = 0 and # ( ˜ b C, B ) ∈ { 0 2 , } , where ˜ denotes C the normalisation of C .
- 5.2. The pull back of a plane curve in terms of the curve itself. In the previous subsection, we described the preimage π -1 ( C ) ⊂ X of a curve C ⊂ P 2 , by looking at the intersection points of the branch locus B with the normalisation ˜ of C C . In this section we will see that we can give an analogous description of π -1 ( C ) by looking at the intersection points of B and C itself, if we assume that all singular points of C that lie on B are ordinary singular points.
The following proposition describes the integer n used in Proposition 5.5 in terms of C directly.
Proposition 5.9. Let C, C ′ ⊂ P 2 be two projective plane curves with no components in common. Let ˜ C be the normalisation of C . Assume also that C ′ is smooth and that all singular points of C that lie on C ′ are ordinary singularities of C . For each point S ∈ C k ( ) , let m S denote the multiplicity of S on C . Then we have
$$\# b ( \tilde { C }, C ^ { \prime } ) = \sum _ { S \in C ( \overline { k } ) \cap C ^ { \prime } ( \overline { k } ) } c _ { S } ( C, C ^ { \prime } )$$
with
/negationslash
$$c _ { S } ( C, C ^ { \prime } ) = \begin{cases} m _ { S } & \text{if $m_{S}\equiv \mu_{S}(C,C^{\prime})$} \pmod{2}$}, \\ m _ { S } - 1 & \text{if $m_{S}\not\equiv \mu_{S}(C,C^{\prime})$} \pmod{2}. \end{cases}$$
Proof. Let ϑ : ˜ C → C be the normalisation map. Then we may write b C, C ( ˜ ′ ) = ⋃ S b S ( ˜ C,C ′ ) with
$$b _ { S } ( \tilde { C }, C ^ { \prime } ) = \{ T \in \vartheta ^ { - 1 } ( S ) \ \colon \ \mu _ { T } ( \tilde { C }, C ^ { \prime } ) \text{ is odd} \}$$
/negationslash and where the disjoint union runs over all S ∈ C k ( ) ∩ C ′ ( k ). Suppose S ∈ C k ( ) ∩ C ′ ( k ). Since C ′ is smooth and the point S is either smooth or an ordinary singularity on C , at most one of the m S points T ∈ ϑ -1 ( S ) satisfies µ T ( ˜ C,C ′ ) > 1. Hence, there is a point T 0 ∈ ϑ -1 ( S ) such that for all T ∈ ϑ -1 ( S ) with T = T 0 we have µ T ( ˜ C,C ′ ) = 1 and thus T ∈ b S ( ˜ C,C ′ ). Since we are working over an algebraically closed field, Lemma 5.2 yields µ S ( C, C ′ ) = µ T 0 ( ˜ C,C ′ ) + m S -1. Hence, we have T 0 ∈ b S ( ˜ C,C ′ ) if and only if m S and µ S ( C, C ′ ) have the same parity. It follows that # b S ( ˜ C,C ′ ) = c S ( C, C ′ ). The proposition follows. /square
/negationslash
We will continue to use the notation c S ( C, C ′ ) of Proposition 5.9, which we call the contribution of S with respect to C ′ . We set c S ( C, C ′ ) = 0 for S ∈ C k ( ) with S ∈ C ′ .
Remark 5.10 . Let C ⊂ P 2 be a geometrically integral projective curve. The points of contribution 0 with respect to C ′ are the points of C k ( ) that are not on C ′ , together with the smooth points S ∈ C k ( ) for which µ S ( C, C ′ ) is even. The points of contribution 1 are the smooth and double points S of C k ( ) with S ∈ C ′ for which µ S ( C, C ′ ) is odd. The points of type m> 1 are the singular points S of C k ( ) of multiplicity m or m +1 with S ∈ C ′ for which µ S ( C, C ′ ) ≡ m (mod 2).
Remark 5.8 describes all geometrically integral projective curves D of genus 0 on X . In particular, we have # ( ˜ b C, B ) ∈ { 0 2 , } , where ˜ is the normalisation of the image C C = π D ( ). The next proposition can be used to express this condition in terms of C directly, under the assumption that the singular points of C that lie on the branch locus B are ordinary singularities.
Proposition 5.11. Let C and C ′ be two geometrically integral projective curves in P 2 . Let ˜ C denote the normalisation of C and let C s ⊂ C k ( ) denote the set of singular points of C . Assume that C ′ is smooth and that all singular points of C that lie on C ′ are ordinary. Then the following statements hold.
- (1) The set b C, C ( ˜ ′ ) is empty if and only if the sets b C, C ( ′ ) and C s ∩ C ′ are.
- (2) We have # ( ˜ b C, C ′ ) = 2 if and only if either
- (a) b C, C ( ′ ) = ∅ and there exists a point S ∈ C k ( ) such that m S ∈ { 2 3 , } and C s ∩ C ′ = { S } , or
/negationslash
- (b) there exist two points S , S 1 2 ∈ C k ( ) with S 1 = S 2 and b C, C ( ′ ) = { S , S 1 2 } and m S 1 , m S 2 ∈ { 1 2 , } and C s ∩ C ′ ⊂ { S , S 1 2 } .
Proof. Given that the contributions c S ( C, C ′ ) are nonnegative, this follows easily from Proposition 5.9 and Remark 5.10. /square
/negationslash
Remark 5.12 . Suppose P ∈ X k ( ) is a rational point that does not lie on the ramification curve, so π P ( ) ∈ B . Suppose C is a geometrically integral curve of degree d that has a singular point of multiplicity d -1 at π P ( ), and that intersects B with even multiplicity everywhere. Then Proposition 5.11 shows that b C, B ( ˜ ) is empty, so by Corollary 5.6, the pull back π ∗ ( C ) splits into two components.
If X is general enough, then the Picard group Pic X of X is generated by the canonical divisor K X , and the automorphism group of X acts trivially on Pic X , so these two components would be linearly equivalent to the same multiple of K X ; as their union is linearly equivalent to -dK X , we find that d is even. Hence, for odd d , the answer to the analogous question mentioned below Question 1.4 is negative for X general enough.
It is possible, however, that, even for odd d , a variation of this analogous question still has a positive answer. If we forget the del Pezzo surface, and only consider the quartic curve B ⊂ P 2 with a point Q ∈ P 2 that does not lie on B , we could ask for the existence of a curve of degree d that intersects B with even multiplicity everywhere, and on which Q is a point of multiplicity d -1. The argument above merely shows that if such a curve exists for odd d and Q lifts to a rational point on the del Pezzo surface, then the surface does not have Picard number one.
Proof of Theorem 1.2. Assume that the assumptions of statement (1) hold. This implies that C s ∩ B = ∅ and b C, B ( ) = ∅ . Therefore, by Proposition 5.11, we have # ( ˜ b C, B ) = 0. Statement (1) follows from applying part (2) of Corollary 5.6.
Assume statement (2a) holds. This means that C s ∩ B = { Q } and b C, B ( ) = ∅ . Since Q is a double or triple point of C , Proposition 5.11 implies that # ( ˜ b C, B ) = 2. The conclusion of statement (2) follows from applying part (4) of Corollary 5.6 and Remark 5.7.
Assume statement (2b) holds. It means that b C, B ( ) = { Q , Q 1 2 } and C s ∩ B ⊆ { Q , Q 1 2 } . Since the points Q 1 and Q 2 are distinct, Proposition 5.11 implies that # ( ˜ b C, B ) = 2. As before, the conclusion of statement (2) follows from part (4) of Corollary 5.6 and Remark 5.7. This concludes the proof of the theorem. /square
/negationslash
Proof of Corollary 1.3. Set Q = π P ( ). As in Section 4, let L Q denote the line in the dual of P 2 consisting of all lines L ⊂ P 2 going through Q , and note that L Q is isomorphic to P 1 . Since C has degree d and π P ( ) is a point of multiplicity d -1, each line in L Q intersects C in a unique d -th point, counted with multiplicity. It follows that C is smooth at all points T = Q . It also follows that the rational map C → L Q that sends a point T ∈ C to the line through T and Q is birational, so C is birationally equivalent with P 1 . By hypothesis, all intersection points of B and C have even intersection multiplicity.
Assume that Q is not contained in B . Since C is smooth away from Q , the curve B contains no singular points of C . Then X is unirational by part (1) of Theorem 1.2. This proves part (1).
Assume that Q is contained B , that Q is an ordinary singularity of C , and d ∈ { 3 4 , } . Then Q is a double or a triple point of C . Since Q is the only singularity of C , the curve B contains no other singular points of C . Then X is unirational by part (2) of Theorem 1.2. This proves part (2). /square
We now give some examples of curves that satisfy the conditions of Theorem 1.2 or Corollary 1.3.
/negationslash
Example 5.13 . If C is a bitangent to the branch curve B that is defined over k , and C k ( ) contains a point Q ∈ B that lifts to a k -rational point on X , then Theorem 1.2 implies that X is unirational. We can also prove this directly. Indeed, in this case the pull back π -1 ( C ) consists of two exceptional curves that are defined over k , so X is not minimal. Blowing down one of these exceptional curves
yields a del Pezzo surface Y of degree 3 with a rational point. This implies that Y , and therefore also X , is unirational.
Example 5.14 . Suppose the point P ∈ X k ( ) is not a generalised Eckardt point and P is not on the ramification curve. Set Q = π P ( ), let ρ : L Q → X be as in Section 4, and set C = π ρ ( ( L Q )). Then by Proposition 4.14, the map ρ ( L Q ) → C has degree 1, so by Propositions 5.5 and 5.11, the intersection multiplicity of C and the branch curve B is even at all intersection points. Also by Proposition 4.14, the curve C has a point Q off the branch curve B of multiplicity deg C -1, so the curves of Manin's construction are examples of the curves described in Corollary 1.3. See Remark 4.15.
$$E x a m p l e \, 5. 1 5. \, & \text{Consider the surface } X \subset \mathbb { P } ( 1, 1, 1, 2 ) \, \text{over} \, \mathbb { F } _ { 3 }, \, \text{defined by the equation} \\ & w ^ { 2 } = x ^ { 4 } + y ^ { 4 } + z ^ { 4 }.$$
The surface X is a del Pezzo surface of degree 2. All its rational points either are on the ramification curve, or they are generalised Eckardt points. In fact, the surface X has 154 rational points over F 9 , with 28 of those lying on the ramification locus. The remaining 126 are generalised Eckardt points, which is also the maximum number of generalised Eckardt points a del Pezzo surface of degree two can have (see [STVA14, before Example 7]). It follows that Manin's method does not apply to this surface. Let P be the point (0 : 0 : 1 : 1) on X . Then P is a generalised Eckardt point and its image Q = π P ( ) = (0 : 0 : 1) ∈ P 2 does not lie on the branch locus B , which is given by x 4 + y 4 + z 4 = 0. Consider the curve C ⊂ P 2 given by x y 3 + xy 3 = z x ( + y ) 2 ( y -x ). The curve C is a geometrically integral quartic plane curve that has a triple point at Q and that intersects B with even multiplicity everywhere. Therefore, by case (1) of Corollary 1.3, the surface X is unirational.
Of course, unirationality of X was already known: it follows for instance from Lemma 20 in [STVA14] (cf. Example 5.17 below). It is nice to see, though, that, even though Manin's construction and the generalisation in [STVA14] do not produce a curve in P 2 of some degree d with a point of multiplicity d -1 at Q , and even intersection multiplicity with B everywhere, such curves do still exist, and then case (1) of Corollary 1.3 implies unirationality of X . This gives a positive answer to the question below Question 1.4 for d = 4 and this particular surface X and this generalised Eckardt point P .
One might ask whether there are curves of lower degree satisfying the hypotheses of case (1) of Corollary 1.3. Indeed, there are conics that do, for example the one given by y 2 = xz . An exhaustive computer search, based on Corollary 5.6, part (2), and Corollary 6.2, shows that there are no cubic curves with a double point at Q satisfying the hypotheses of Corollary 1.3 and its case (1).
Example 5.16 . Let X ,X ,X 1 2 3 be the three del Pezzo surfaces defined as in Section 3 and let B i be their branch locus, for i = 1 2 3. For , , i = 1 2 3, all rational points of the surface , , X i lie on the ramification locus. Consider the rational points P 1 = (0 : 0 : 1 : 0) ∈ X 1 , P 2 = (0 : 1 : 1 : 0) ∈ X 2 , and P 3 = (1 : 1 : 1 : 0) ∈ X 3 , and set Q i = π P ( i ). Clearly, we have Q i ∈ B i . Set d 1 = d 2 = 4 and d 3 = 3. Let C i ⊂ P 2 be the projective plane curve of degree d i given by the polynomial h i defined as in Remark 3.4. The curve C i is geometrically irreducible and it has an ordinary singular point at Q i of multiplicity d i -1. Given that the curve C i pulls back to the geometrically irreducible rational curve A i of Remark 3.4, we find from Corollary 5.6 and Proposition 5.11 that C i intersects B i with even multiplicity everywhere.
Of course, one could also check directly that C i intersects B i with even multiplicity everywhere. Then Corollary 1.3 and Remark 5.7 give an alternative proof that the surface X i is unirational (cf. Remark 3.4). There is a quartic alternative for C 3 as well. The curve C ′ 3 ⊂ P 2 given by the vanishing of
$$h _ { 3 } ^ { \prime } = & \gamma ^ { 2 } x ^ { 4 } + x ^ { 3 } y + \gamma x ^ { 2 } y ^ { 2 } + \gamma ^ { 3 } x y ^ { 3 } - y ^ { 4 } + x ^ { 3 } z + \gamma x ^ { 2 } y z + x y ^ { 2 } z \\ & - \gamma y ^ { 3 } z + \gamma x ^ { 2 } z ^ { 2 } + x y z ^ { 2 } + \gamma ^ { 3 } y ^ { 2 } z ^ { 2 } + \gamma ^ { 3 } x z ^ { 3 } - \gamma y z ^ { 3 } - z ^ { 4 }$$
is geometrically integral, has an ordinary triple point at ( -1 : 1 : 1), and intersects B with even multiplicity everywhere.
Example 5.17 . Let k be a field with characteristic different from 2. Let a , . . . , a 1 6 ∈ k be such that the variety X in the weighted projective space P = P (1 , 1 1 2) defined by , ,
$$w ^ { 2 } = a _ { 1 } ^ { 2 } x ^ { 4 } + a _ { 2 } ^ { 2 } y ^ { 4 } + a _ { 3 } ^ { 2 } z ^ { 4 } + a _ { 4 } x ^ { 2 } y ^ { 2 } + a _ { 5 } x ^ { 2 } z ^ { 2 } + a _ { 6 } y ^ { 2 } z ^ { 2 }$$
is a del Pezzo surface of degree 2. This is the surface of Lemma 20 in [STVA14], where it is noted that the surface in P given by w = a x 1 2 + a y 2 2 + a z 3 2 intersects the surface X in a curve D , which the anti-canonical map π : X → P 2 sends isomorphically to the plane quartic curve C ⊂ P 2 given by
$$( a _ { 4 } - 2 a _ { 1 } a _ { 2 } ) x ^ { 2 } y ^ { 2 } + ( a _ { 5 } - 2 a _ { 1 } a _ { 3 } ) x ^ { 2 } z ^ { 2 } + ( a _ { 6 } - 2 a _ { 2 } a _ { 3 } ) y ^ { 2 } z ^ { 2 } = 0.$$
They also note that this curve C is birationally equivalent to a conic under the standard Cremona transformation, so C and D are rational over an extension of k of degree at most 2. If they are rational over k , then X is unirational.
Indeed, one checks that the curve C satisfies the conditions of part (1) of Corollary 5.6, and if C is rational over k , then it also satisfies the conditions of part (1) of Theorem 1.2, where one can take P to be any of the points on X above any of the singular points (0 : 0 : 1), (0 : 1 : 0), and (1 : 0 : 0) of C .
## 6. Finding appropriate curves
In this section, we assume that the characteristic of k is not 2, and we give sufficient easilyverifiable conditions for a curve C to satisfy the hypotheses of Corollary 1.3. This is also how we found the three curves, C , C , 1 2 and C 3 of Remark 3.4, whose existence implies unirationality of the three difficult surfaces X 1 , X 2 , X 3 (see Example 5.16 and Remark 6.7).
Let X ⊂ P (1 , 1 1 2) be a del Pezzo surface of degree 2, given by , , w 2 = g with g ∈ k x, y, z [ ] homogeneous of degree 4. Let B ⊂ P 2 ( x, y, z ) be the branch curve of the projection π : X → P 2 . Then B is given by g = 0. Let P ∈ X k ( ) be a rational point and set Q = π P ( ). Without loss of generality, we assume Q = (0 : 0 : 1). Let C ⊂ P 2 be a geometrically irreducible curve of degree d ≥ 2 on which Q is a point of multiplicity d -1.
There are coprime homogeneous polynomials f d -1 , f d ∈ k x, y [ ] of degree d -1 and d , respectively, such that C is given by zf d -1 = f d . The projection away from Q induces a birational map from C to the family L Q of lines in P 2 through Q . Its inverse is a morphism ϑ that sends a line L ∈ L Q to the d -th intersection point of L with C . If we identify L Q with P 1 , where ( s : t ) ∈ P 1 corresponds to the line given by sy = tx , then ϑ : P 1 → C sends ( s : t ) to
$$( s f _ { d - 1 } ( s, t ) \colon t f _ { d - 1 } ( s, t ) \colon f _ { d } ( s, t ) ).$$
The curve C has no singularities outside Q , and we may identify ϑ : P 1 → C with the normalisation of C . The points on P 1 above the point Q are exactly the points where f d -1 ( s, t ) vanishes. The curve C has an ordinary singularity at Q if and only if d > 2 and f d -1 ( s, t ) vanishes at d -1 distinct k -points of P 1 ( s, t ).
The pull back π ∗ ( C ) is birationally equivalent with the curve given by w 2 = G in the weighted projective space P (1 , 1 2 , d ) with coordinates s, t, w , and with
$$G = g ( s f _ { d - 1 } ( s, t ), t f _ { d - 1 } ( s, t ), f _ { d } ( s, t ) ) \in k [ s, t ]. \\ \intertext {. \ F o r \ a n y \ p o i n t \ T \, \in \, \mathbb { P } ^ { 1 } ( \overline { k } ), \ t h e \ i n t e r s e c i o n \ m u l t i p l i c i t y \ \mu _ { T } ($$
Proposition 6.1. For any point T ∈ P 1 ( k ) , the intersection multiplicity µ T ( P 1 , B ) equals the order of vanishing of G at T .
Proof. Since C either has degree 2 or it is singular, it is not equal to B . As C is irreducible, it has no irreducible components in common with B . By symmetry between s and t , we may assume T = ( α : 1) for some α ∈ k . Then the local ring O T, P 1 is isomorphic the the localisation of k s [ ] at the maximal ideal ( s -α ). Let /lscript ∈ k x, y, z [ ] be a linear form that does not vanish at ϑ T ( ). Then locally around ϑ T ( ) ∈ P 2 , the curve B is given by the vanishing of the element g//lscript 4 , whose image in O T, P 1 is G s, ( 1) /L s, ( 1) 4 with L s, t ( ) = /lscript ( sf d -1 ( s, t ) , tf d -1 ( s, t ) , f d ( s, t ) ) . Since L s, ( 1) does not vanish at α , we find that µ T ( P 1 , B ) equals the order of vanishing of G s, ( 1) at α , which equals the order of vanishing of G at T . /square
Proof. By Proposition 6.1, we have b ( P 1 , B ) = ∅ if and only if the order of vanishing of G is even at every point T ∈ P 1 ( k ). This is equivalent with G being a square in k s, t [ ]. /square
If B does not contain the unique singular point Q of C , then ϑ induces a bijection b ( P 1 , B ) → b C, B ( ), so in this case we also have b C, B ( ) = ∅ if and only if G is a square in k s, t [ ]. The following proposition gives an analog of this statement when Q is contained in B .
Proposition 6.3. Suppose that Q is contained in B . Then H = G/f d -1 ( s, t ) is contained in k s, t [ ] . Suppose, furthermore, that the tangent line to B at Q is given by h = 0 with h ∈ k x, y [ ] , and that Q is an ordinary singular point on the curve C . Then the following statements hold.
- (1) Suppose d is odd. Then the set b C, B ( ) is empty if and only if H is a square in k s, t [ ] .
- (2) Suppose d is even. If h divides f d -1 , then H/h s,t ( ) is contained in k s, t [ ] . The set b C, B ( ) is empty if and only if h divides f d -1 and H/h s,t ( ) is a square in k s, t [ ] .
Proof. Write g = ∑ 4 i =0 g z i 4 -i , where g i ∈ k x, y [ ] is homogeneous of degree i for all 0 ≤ i ≤ 4. If g Q ( ) = g 0 vanishes, then each monomial of g is divisible by x or y , which implies that G is divisible by f d -1 , which in turn shows H ∈ k s, t [ ]. Suppose that all hypotheses hold. By g Q ( ) = 0 we find g 0 = 0. The tangent line to B at Q is given by g 1 = 0, so h is a scalar multiple of g 1 . Note that all statements are invariant under the action of GL ( 2 k ) on P 1 and P 2 given on their respective homogeneous coordinate rings k s, t [ ] and k x, y, z [ ] by γ s ( ) = as + bt, γ ( ) = t cs + dt and γ x ( ) = ax + by, γ ( y ) = cx + dy, γ ( z ) = z for
$$\gamma = \begin{pmatrix} a & b \\ c & d \end{pmatrix} \cdot \\ \swarrow \, \mathcal { C } \, \mathfrak { m } \, \mathfrak { m } \, \circ \, \mathfrak { m } \, \cdot$$
After applying an appropriate element γ ∈ GL ( ) and rescaling 2 k h , we assume, without loss of generality, that h = g 1 = y .
If y divides f d -1 , then t divides f d -1 ( s, t ); since all monomials in g besides y are divisible by x 2 , xy , or y 2 , it follows that in this case G is divisible by tf d -1 ( s, t ), so H/t is contained in k s, t [ ]. This does not depend on d being even.
/negationslash
/negationslash
In the open neighbourhood of Q given by z = 0, the curve B is given by the vanishing of g/z 4 = g x/z, y/z, ( 1) = ∑ i g i ( x/z, y/z ). The maximal ideal m of the local ring O Q,C is generated by x/z and y/z , so the image of g/z 4 in O Q,C / m 2 is g 1 ( x/z, y/z ) = y/z . Let T ∈ P 1 ( k ) be a point with ϑ T ( ) = Q , and let n be the maximal ideal of the local ring O T, P 1 . Then the image of g in O T, P 1 / n 2 equals the image of y/z , which is tf d -1 ( s, t ) /f d ( s, t ). The point T corresponds to a linear factor of f d -1 ( s, t ). Since f d ( s, t ) does not vanish at T , we find that the valuation v T ( g ) of g in O T, P 1 is at least 2 if t vanishes at T , that is, µ T ( P 1 , B ) ≥ 2 if T = (1 : 0). We have µ T ( P 1 , B ) = v g ( ) = 1 if T = (1 : 0). From Lemma 5.2 we conclude
$$\mu _ { Q } ( C, B ) = \begin{cases} d - 2 + \mu _ { ( 1 \colon 0 ) } ( \mathbb { P } ^ { 1 }, B ) & \text{if $y$ divides $f_{d-1}$}, \\ d - 1 & \text{otherwise}. \end{cases}$$
$$\begin{matrix} \end{matrix}$$
We now consider the two cases.
/negationslash
- (1) Suppose d is odd. From (5) it follows that µ Q ( C, B ) is even if and only if either y divides f d -1 and µ (1:0) ( P 1 , B ) is odd, or y does not divide f d -1 . This happens if and only if µ T ( P 1 , B ) is odd for all T ∈ P 1 at which f d -1 ( s, t ) vanishes. For all other points R ∈ C with R = Q , the multiplicity µ R ( C, B ) is even if and only if µ ϑ -1 ( R ) ( P 1 , B ) is even. From Proposition 6.1, we conclude that b C, B ( ) is empty if and only if the order of vanishing of G is odd at all points T ∈ P 1 at which f d -1 ( s, t ) vanishes and even at all other points. This is equivalent with H being a square in k s, t [ ].
/negationslash
- (2) Suppose d is even. From (5) it follows that µ Q ( C, B ) is even if and only if y divides f d -1 and µ (1:0) ( P 1 , B ) is even. As in the case for odd d , this implies that b C, B ( ) is empty if and only if the order of vanishing of G is odd at all points T = (1 : 0) at which f d -1 ( s, t ) vanishes, and even at all other points, including (1 : 0). Since the order of vanishing of tf d -1 ( s, t ) at (1 : 0) is 2, this is equivalent to G/ tf ( d -1 ( s, t )) = H/t being a square in k s, t [ ].
This finishes the proof.
/square
We have already seen that the pull back π ∗ ( C ) is birationally equivalent with the curve given by w 2 = G in P (1 , 1 2 , d ). This curve splits into two k -rational components if and only if G is a square in k s, t [ ]. If Q is an ordinary singular point of C that lies on B , then this never happens. However, the curve π ∗ ( C ) may itself be k -rational, in which case G factors as a square times a quadric.
We will now focus on the case d = 4, so Q is a triple point. The following corollary says that if Q is an ordinary triple point, then we do not need to factorise G , as we know exactly which part should be the square, and which the quadric.
Corollary 6.4. Suppose that Q is an ordinary singular point of C that lies on B . If the pull back π ∗ ( C ) ⊂ X is k -rational, then we have d ≤ 4 .
Moreover, suppose d = 4 , and let the tangent line to B at Q be given by h = 0 with h ∈ k x, y [ ] . Then the pull back π ∗ ( C ) ⊂ X is k -rational if and only if there is a constant c ∈ k ∗ such that the following statements hold:
- (1) the polynomial h divides f 3 ;
- (2) the polynomial cH s, t /h s, t ( ) ( ) is a square in k s, t [ ] ;
- (3) the conic given by cw 2 = f 3 ( s, t ) /h s, t ( ) in P 2 ( s, t, w ) is k -rational.
Proof. Suppose π ∗ ( C ) is k -rational. Then π ∗ ( C ) is geometrically integral and has genus g π ( ∗ ( C )) = 0. From Proposition 5.5 we obtain b ( P 1 , B ) = 2. From Proposition 5.9 we conclude that the contribution c Q ( C, B ) is at most 2. Moreover, this proposition also gives c Q ( C, B ) ≥ d -2 with equality if and only if µ Q ( C, B ) is even. We conclude d ≤ 4.
Suppose d = 4. Then we have equality c Q ( C, B ) = 2 = # ( b P 1 , B ), so µ Q ( C, B ) is even, and we find that b C, B ( ) is empty. From Proposition 6.3 we find that h divides f 3 , and m = H s,t /h s, t ( ) ( ) is a square in k s, t [ ]. Let c be the main coefficient of m s, ( 1). Then cm is a square in k s, t [ ]. Therefore, the k -rational curve given by w 2 = G with
$$G & = c m \cdot h ^ { 2 } ( s, t ) \cdot c ^ { - 1 } f _ { 3 } ( s, t ) / h ( s, t )$$
in P (1 , 1 2 , d ) is birationally equivalent with the conic given by cw 2 = f 3 ( s, t ) /h s, t ( ) in P 2 ( s, t, w ), which is therefore also k -rational.
Conversely, if there is a constant c such that cH s, t /h s, t ( ) ( ) is a square in k s, t [ ], then it follows from (6) that the conic given by cw 2 = f 3 ( s, t ) /h s, t ( ) in P 2 ( s, t, w ) is birationally equivalent with the curve in P (1 , 1 2 , d ) given by w 2 = G , which is birationally equivalent with π ∗ ( C ). Hence, if this conic is k -rational, then so is π ∗ ( C ). /square
Remark 6.5 . Corollary 6.4 helps us in finding all curves C of degree d = 4 that satisfy the conditions of case (2) of Corollary 1.3 with /lscript = k . More explicitly, after a linear transformation of P 2 , we may assume that Q = (0 : 0 : 1), and the tangent line to B at Q is given by y = 0. Then we claim that every curve C of degree d = 4 that satisfies the conditions of case (2) of Corollary 1.3 with /lscript = k is given by
$$y z \phi _ { 2 } = x ^ { 4 } + y \phi _ { 3 }$$
for some homogeneous φ , φ 2 3 ∈ k x, y [ ] of degree 2 and 3, respectively, with φ 2 squarefree and not divisible by y . Indeed, we find that f 3 is divisible by y , so there is a φ 2 ∈ k x, y [ ] such that f 3 = yφ 2 ; since C is irreducible, the polynomial f 4 is not divisible by y , so the coefficient of x 4 in f 4 is nonzero, and after scaling φ , f 2 3 , and f 4 , we may assume that there exists a φ 3 ∈ k x, y [ ] such that f 4 = x 4 + yφ 3 . Moreover, Q is an ordinary singularity if and only if φ 2 is squarefree and not divisible by y .
Hence, to find all such curves C , we are looking for all pairs ( φ , φ 2 3 ) with φ i ∈ k x, y [ ] homogeneous of degree i , such that
- (1) the polynomial φ 2 is squarefree and y does not divide φ 2 ,
- (2) the curve given by yzφ 2 = x 4 + yφ 3 is geometrically integral,
- (3) there is a constant c ∈ k ∗ such that polynomial c · G s, t / t ( ) ( 2 φ 2 ( s, t )) with
$$G = g ( s t \phi _ { 2 } ( s, t ), \ t ^ { 2 } \phi _ { 2 } ( s, t ), \ s ^ { 4 } + t \phi _ { 3 } ( s, t ) )$$
is a square,
- (4) the conic given by cw 2 = φ ( s, t ) in P 2 ( s, t, w ), with c as in (3), is k -rational.
2 Note for (3) that, because the characteristic is not 2, a homogeneous polynomial H ∈ k s, t [ ] of even degree is a square in k s, t [ ] if and only if there is a constant c ∈ k ∗ such that cH is a square in k s, t [ ], which happens if and only if γ -1 H s, ( 1) is a square in k s [ ], where γ is the main coefficient of H s, ( 1). This follows from the fact that a monic polynomial in k s [ ] is a square in k s [ ] if and only if it is a square in k s [ ]. Moreover, the c ∈ k ∗ for which cH is a square, form a coset in k /k ∗ ∗ 2 , so whether or not (4) holds does not depends on the choice of c .
Question 1.4 for d = 4 can be rephrased using Remark 6.5. It is equivalent to the following question.
Question 6.6. Let k be a field of characteristic not equal to 2 , and g ∈ k x, y, z [ ] a homogeneous polynomial of degree 4 such that the curve B ⊂ P 2 ( x, y, z ) given by g = 0 is smooth, it contains the point Q = (0 : 0 : 1) , and the tangent line to B at Q is given by y = 0 . Do there exist homogeneous polynomials φ , φ 2 3 ∈ k x, y [ ] of degree 2 and 3 , such that conditions (1)-(3) of Remark 6.5 are satisfied?
Remark 6.7 . If k is a (small) finite field, then we can list all pairs ( φ , φ 2 3 ) with φ i ∈ k x, y [ ] homogeneous of degree i , and check for each whether the conditions (1)-(4) of Remark 6.5 are satisfied. In fact, condition (4) is automatically satisfied over finite fields. Indeed, this is how we found the curves C , C 1 2 given in Remark 3.4, whose existence implies unirationality of the three difficult surfaces X ,X 1 2 (see Example 5.16). Finding the rational cubic curve C 3 on X 3 , as given in Remark 3.4, was easier, based on part (1) of Proposition 6.3.
Remark 6.8 . For any integer i , let k x, y [ ] i denote the ( i + 1)-dimensional space of homogeneous polynomials of degree i . In general, over any field, we can describe the set of pairs ( φ , φ 2 3 ) ∈ k x, y [ ] 2 × k x, y [ ] 3 satisfying condition (3) of Remark 6.5 as follows.
Identify k x, y [ ] 2 × k x, y [ ] 3 with the affine space A 7 and let R denote the coordinate ring of A 7 , that is, R is the polynomial ring in the 3 + 4 = 7 coefficients of φ 2 and φ 3 . Let Z ⊂ A 7 be the locus of all ( φ , φ 2 3 ) that satisfy condition (3).
For generic φ , φ 2 3 , that is, with the variables of R as coefficients, the coefficients of the polynomial
$$G ^ { \prime } = G ( s, t ) / ( t ^ { 2 } \phi _ { 2 } ( s, t ) )$$
/negationslash of condition (3) of Remark 6.5 lie in R . For general enough g , the coefficient c ∈ R of s 12 in the polynomial G ′ ∈ R s, t [ ] is nonzero. On the open set U of A 7 given by c = 0, we may complete G s, ′ ( 1) to a square in the sense that there are polynomials G , G 1 2 ∈ R c [ -1 ][ s ] with G 1 monic of degree 6 in s and G 2 of degree at most 5 in s such that G s, ′ ( 1) = cG 2 1 -G 2 . The vanishing of the six coefficients in R of G 2 determines the locus Z ∩ U inside U of all pairs ( φ , φ 2 3 ) at which cG ′ is a square. Note that we have c = G ′ (1 , 0). For each point ( s 0 : t 0 ) ∈ P 1 , we can use an automorphism of P 1 that sends ( s 0 : t 0 ) to (1 : 0), to similarly describe the intersection of Z with the open subset of A 7 where G s ,t ′ ( 0 0 ) is nonzero; it is also given by the vanishing of six polynomials in R . We can cover A 7 with open subsets of this form, thus describing Z completely.
A naive dimension count suggests that the locus Z has dimension 7 -6 = 1. This is consistent with the following, similarly naive, dimension count. The family of quartic curves in P 2 is 14dimensional, as it is the projective space P ( k x, y, z [ ] 4 ), where k x, y, z [ ] 4 is the 15-dimensional vector space of polynomials of degree 4. The codimension of the subset of those curves having a triple point at Q is 6, and demanding that the intersection multiplicity µ Q ( C, B ) is at least 4 cuts down another dimension. Since B is also a quartic curve, by Bezout's theorem it follows that B and C have 16 intersection points, counted with multiplicity. Hence, generically, the curves in the remaining 7-dimensional family intersect B , besides in Q , in 16 -4 = 12 more points. One might expect the subfamily of those curves where this degenerates to six points with multiplicity 2 to have codimension 6, in which case this would leave a 1-dimensional family of quartic curves with a triple point at Q ∈ B and intersecting B with even multiplicity everywhere.
However, the locus Z also contains some degenerate components that we are not interested in. For example, the locus of all (0 , φ 3 ) for which f 4 = x 4 + yφ 3 is a square is contained in Z and has
dimension 2. Also, for any smooth conic Γ that contains Q , that has its tangent line at Q given by y = 0, and that has even intersection multiplicity with B everywhere, we get a 1-dimensional subset of Z consisting of pairs ( φ , φ 2 3 ) that correspond with the union of Γ with any double line through Q (these lines are parametrised by P 1 ). Note that in all these degenerate cases the curve C is reducible. Another degenerate case is the limit of Manin's construction. By Remark 4.11, this limit curve is the non-reduced curve π π ∗ ∗ (2 L ) = 4 L , where L is the tangent line to B at Q , given by y = 0. Hence, this quartic curve is given by y 4 = 0, which does not correspond to a point on the affine set Z , as the coefficient of x 4 is zero.
Let Z 0 denote the affine subset of Z corresponding to curves C that are geometrically integral and on which Q is an ordinary triple point. Then Questions 1.4 (for d = 4) and 6.6 can be rephrased by asking whether the subset Z 0 contains a k -rational point.
Example 6.9 . Let B ⊂ P 2 be the smooth curve given by
$$y ^ { 4 } - x ^ { 4 } - x ^ { 3 } y - x y ^ { 3 } + y ^ { 3 } z + y z ^ { 3 } = 0$$
over k = F 3 , and let Q be the point (0 : 0 : 1) ∈ B k ( ). The tangent line to B at Q is given by y = 0. We ran through all the homogeneous polynomials in x, y of degree 2 and 3 over k and we did not find any pair ( φ , φ 2 3 ) of polynomials satisfying conditions (1)-(3) of Remark 6.5; we did find pairs satisfying only conditions (2) and (3). This means that Questions 1.4 (for d = 4) and 6.6 have negative answer in this specific case. It could, however, still be true that the answer is positive for X and P general enough.
Notice that the curve B is isomorphic to the Fermat curve x 4 + y 4 + z 4 = 0 via the following linear change of variables
$$( x \colon y \colon z ) \mapsto ( x - z \colon x + y + z \colon z ).$$
↦
## References
[Ful98]
W. Fulton. Intersection theory , volume 2 of Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathematics [Results in Mathematics and Related Areas. 3rd Series. A Series of Modern Surveys in Mathematics] . Springer-Verlag, Berlin, second edition, 1998.
[Gro61] A. Grothendieck. El´ments de g´om´ etrie alg´brique. II. Etude globale ´l´mentaire de quelques classes de ´ e e e ´ e e morphismes. Inst. Hautes Etudes Sci. Publ. Math. ´ , (8):222, 1961.
[Har77]
R. Hartshorne. Algebraic geometry . Springer-Verlag, New York-Heidelberg, 1977. Graduate Texts in Mathematics, No. 52.
[Kne13]
[Kol96]
[Kol02] [Man86]
[Mum99]
[Pie12]
[Seg43]
A. Knecht. Degree of unirationality for del Pezzo surfaces over finite fields. preprint, available at http://www67.homepage.villanova.edu/amanda.knecht/UnirationalDegree.pdf , 2013.
J. Koll´r. a Rational curves on algebraic varieties , volume 32 of Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathematics [Results in Mathematics and Related Areas. 3rd Series. A Series of Modern Surveys in Mathematics] . Springer-Verlag, Berlin, 1996.
J. Koll´r. Unirationality of cubic hypersurfaces. a J. Inst. Math. Jussieu , 1(3):467-476, 2002.
Yu. I. Manin. Cubic forms , volume 4 of North-Holland Mathematical Library . North-Holland Publishing Co., Amsterdam, second edition, 1986. Algebra, geometry, arithmetic, Translated from the Russian by M. Hazewinkel.
D. Mumford. The red book of varieties and schemes , volume 1358 of Lecture Notes in Mathematics . Springer-Verlag, Berlin, expanded edition, 1999. Includes the Michigan lectures (1974) on curves and their Jacobians, With contributions by Enrico Arbarello.
M. Pieropan. On the unirationality of del Pezzo surface over an arbitrary field. Master thesis, available at http://www.algant.eu/documents/theses/pieropan.pdf , 2012.
B. Segre. A note on arithmetical properties of cubic surfaces. Journal of the London Mathematical Society , 1(1):24-31, 1943.
[Seg51]
B. Segre. On the rational solutions of homogeneous cubic equations in four variables. Math. Notae , 11:1-68, 1951.
[Sil09] J.H. Silverman. The arithmetic of elliptic curves , volume 106 of Graduate Texts in Mathematics . Springer, Dordrecht, second edition, 2009.
[STVA14] C. Salgado, D. Testa, and A. V´rilly-Alvarado. On the unirationality of del Pezzo surfaces of degree two. a Journal of the London Mathematical Society , 90:121-139, 2014.
[Tat52] J. Tate. Genus change in inseparable extensions of function fields. Proc. Amer. Math. Soc. , 3:400-406, 1952.
[TVAV09] D. Testa, A. V´ arilly-Alvarado, and M. Velasco. Cox rings of degree one del Pezzo surfaces. Algebra Number Theory , 3(7):729-761, 2009. | null | [
"Dino Festi",
"Ronald van Luijk"
] | 2014-08-01T19:12:17+00:00 | 2015-05-05T20:16:32+00:00 | [
"math.AG",
"14J26, 14G15"
] | Unirationality of del Pezzo surfaces of degree two over finite fields (extended version) | We prove that every del Pezzo surface of degree two over a finite field is
unirational, building on the work of Manin and an extension by Salgado, Testa,
and V\'arilly-Alvarado, who had proved this for all but three surfaces. Over
general fields, we state several sufficient conditions for a del Pezzo surface
of degree two to be unirational. |
1408.0270v2 | ## Anomalous response of superconducting titanium nitride resonators to terahertz radiation
J. Bueno 1 , ∗ , P. C. J. J. Coumou 2 , G. Zheng , P. J. de Visser 2 1 2 , , T. M. Klapwijk 2 3 , , E. F. C. Driessen 4 5 , , S. Doyle , 6 J. J. A Baselmans 1 1 SRON, Netherlands Institute of Space Research, Utrecht, The Netherlands 2 Kavli Institute of Nanoscience, Delft University of Technology, Delft, The Netherlands 3 Physics Department, Moscow State Pedagogical University, 119991 Moscow, Russia 4 Univ. Grenoble Alpes, INAC-SPSMS, F-38000 Grenoble, France 5 CEA, INAC-SPSMS, F-38000 Grenoble, France 6 Cardiff University - Department of Physics and Astronomy, UK ∗
email address: [email protected]
(Dated: August 21, 2014)
We present an experimental study of KIDs fabricated of atomic layer deposited TiN films, and characterized at radiation frequencies of 350 GHz. The responsivity to radiation is measured and found to increase with increasing radiation powers, opposite to what is expected from theory and observed for hybrid niobium titanium nitride / aluminium (NbTiN/Al) and all-aluminium (allAl) KIDs. The noise is found to be independent of the level of the radiation power. The noise equivalent power (NEP) improves with higher radiation powers, also opposite to what is observed and well understood for hybrid NbTiN/Al and all-Al KIDs. We suggest that an inhomogeneous state of these disordered superconductors should be used to explain these observations.
Superconducting resonators have been proposed as kinetic inductance detectors (KIDs) for sensitive multipixel radiation detection [1]. Antenna-coupled hybrid niobium titanium nitride / aluminium (NbTiN/Al) KIDs [2, 3] and all-aluminium (all-Al) [4] have shown generationrecombination noise and photon noise limited performance. KIDs can also be constructed as lumped element kinetic inductance detectors (LEKIDs) [5] in which the KID is arranged as a photon absorbing area matched to free space. Aluminium LEKIDs have also shown generation-recombination and photon noise limited performance [6]. However, the low normal-state resistivity of Al makes the design of the absorber for very high frequency radiation complex. Therefore, superconductors with a high normal state resistivity have recently become of particular interest [9].
Afigure of merit F to optimize the responsivity of KIDs is defined as F = α sc τQ F i res /N (0) V [9], with α sc the kinetic inductance fraction, τ the quasiparticle recombination time, Q i the internal quality factor, F res the resonance frequency, N (0) is the single spin electron density of states at the Fermi level, and V the volume of the KID. For example, Al KIDs have a long quasiparticle recombination time (a few milliseconds) and high internal quality factors (above one million), but their kinetic inductance fraction is low and their volume large.
lowers the KID volume. The high surface impedance also eases matching to free space and optimises the photon absorption. Given the high quality NbTiN resonators pioneered by Barends et al. [7, 8], titanium nitride (TiN) has been proposed, because it has the previously mentioned properties in addition to a tuneable critical temperature, which facilitates a relatively long quasiparticle lifetime[9]. Currently, several groups are studying the implementation of TiN KID devices and instruments [10-14]. However, a material like TiN has also drawn the attention of the condensed matter physics community, interested in the disorder-induced superconductorto-insulator transition [15]. It has been shown that the superconducting state becomes gradually increasingly inhomogeneous for increasing disorder, a feature absent in the traditional description of the absorption of electromagnetic radiation by a superconductor. Consequently, a careful empirical study of the direct absorption of radiation by a strongly disordered superconductor such as TiN is urgently needed.
Superconducting materials with a high resistivity in the normal-state are promising because of their high quality factor and a long enough relaxation time. The high normal resistance implies a large sheet inductance, resulting in a large kinetic inductance fraction, which
In this letter we concentrate our experiments on atomic layer deposited (ALD) TiN exposed to 350 GHz radiation. A number of unanticipated trends were found in these experiments: i) the responsivity to the 350 GHz radiation increases for increasing radiation power, opposite to what we observe in hybrid NbTiN/Al and all-Al KIDs, and what is expected for a conventional, uniform, superconducting state; ii) the optical noise equivalent power (NEP opt ) for 350 GHz radiation is much larger than the photon noise limited NEP at low radiation powers whereas it becomes comparable at higher radiation
powers; iii) the electrical NEP is much lower than the optical NEP, in contrast to what we find for the hybrid NbTiN/Al KIDs.
Several ALD TiN films have been deposited on high resistivity ( > 10 kΩcm) Si (100) substrates covered with a thin surface layer of native silicon oxide (for details see Coumou et al. [17]). The microwave properties of these TiN films have been analysed in detail by Driessen et al. [18] and by Coumou et al. [19]. In the latter case it was compared to a study of the local values of the superconducting density of states. The microwave response deviates from standard Mattis-Bardeen theory [21]. A modified Usadel equation was used, including a disorderdependent pair breaking parameter, which smears the BCS density of states analogous to what is known for magnetic impurities and a pair-breaking current. Physically, it refers to a ground state in which the Cooperpairs have a finite momentum. The pair-breaking parameter modifies the quasiparticle DoS, smoothening the coherence peaks, which enter into the generalized MattisBardeen expressions [18].
The local DoS of these films is independently determined using scanning tunnelling spectroscopy (STS). A clear correspondence between the STS measurements and the model is found for the least-disordered films. In contrast, the measured tunnel curves are no longer uniquely characterizing a film for the most-disordered superconducting films [19], but are found to vary laterally. For the present experiments we use a 45 nm thick film because it has a relatively low disorder, and therefore, is expected to allow the application of a uniform model with a broadened DoS model (although it did not have the best properties as a detector). However, it is to be expected that effects of the tendency to non-uniformity of the superconductor with increasing disorder will depend on the parameters and the techniques used in the experiments.
The 45 nm ALD TiN film has a superconducting transition temperature T c of 3 2 . K, a sheet inductance L s of 16 5 . pH and a sheet resistance R s of 41 5 . Ω. Two KIDs, capacitively coupled to a feedline of ∼ 50 Ω, were fabricated in a single layer process. The KIDs were designed to absorb 350 GHz radiation and consist of three inductors and an interdigitated capacitor, all in parallel, forming a tank circuit with a resonance frequency f res of ∼ 3 4 . GHz. The width of the lines of the inductors is 24 µ m and the separation between adjacent lines is 60 µ m, with a total area of the inductors of 1200 x 986 µ m. The quality factor Q of the KIDs is 7 × 10 , set by a coupling Q of 7 4 × 10 , with an internal 4 Q of 1 8 . × 10 . 6
The experiments are performed in a pulse tube precooled adiabatic demagnetisation refrigerator (ADR) with a temperature of 100 mK, well below the superconducting transition temperature of the film. The sample is mounted in a sample holder inside a light-tight box, in a box-in-a-box configuration [20]. The 350 GHz radiation illuminating the KIDs is coming from a cryogenic black
3K stage
FIG. 1. (Color online) Schematic picture of the experimental setup. Left: The KIDs are mounted in a sample box thermally anchored to the base temperature plate of the ADR ( ∼ 100 mK). The sample box is surrounded by a light-tight box to prevent stray radiation from the 3 K stage getting into the detector. A cryogenic black body source is placed at the 3 K stage and used to illuminate the sample. The optical access to the sample box is formed by different stacks of filters placed at the black body source at the light-tight box. Right: Image of the device with the two KIDs used in these experiments, and a detailed view of one of them. The sample dimensions are 20 x 4 mm, whereas each KID was roughly 1 x 1 mm and designed to absorb radiation at 350 GHz.
body placed inside the cooler at the 3 K stage, which temperature can be varied from 3 up to 35 K. Two stacks of metal mesh IR filters placed at the black body source and at the light-tight box determine the band pass of the radiation coupled to the KIDs, which is 50 GHz centered around 350 GHz. The same experimental setup is described in more detail by Janssen et al. [3] and by de Visser et al. [4]. A schematic picture of the experimental setup and device is shown in Fig. 1.
The radiation power absorbed by the KIDs P opt is calculated by integrating the black body spectral radiance over the solid angle illuminating the KIDs, taking into account the filter transmission. Following, the conventional approach we use the number of excess quasiparticles N qp in a KID as the leading quantity. It is parametrized as N qp = η pb P opt τ/ ∆, with η pb the so-called quasiparticle creation efficiency i.e. the number of excess quasiparticles per incoming photon, and ∆ the superconducting energy gap. The absorption of the superconductor is assumed to be given by the Mattis-Bardeen theory [21] and the electron-phonon controlled recombination time as given in Kaplan et al. [22]. For a photon noise limited KID, the quasiparticle lifetime at a given radiation power is expected to be inversely proportional to the square root of the absorbed power τ ∝ P -1 / 2 opt . Therefore the 350 GHz radiation responsivity δx/δP opt (where x is either amplitude or phase response) is expected to be proportional to the square root of the absorbed power δx/δP opt ∝ P -1 / 2 opt . These dependences were recently verified both for hybrid NbTiN/Al as well as and for the all-Al KIDs [3, 4]. A
FIG. 2. (Color online) Amplitude responsivity measured with a VNA as a function of radiation power at different bath temperatures. The amplitude responsivity increases when the radiation power increases, opposite to what is observed in hybrid NbTiN/Al [3] and all-Al [4] KIDs, and opposite to the usual assumptions for quasiparticle recombination. The inset shows the phase response as a function of radiation power at different bath temperatures. The phase response is presumably dominated by the TLS contribution at the lowest bath temperatures.
spurious contribution in the phase response (shown in the inset in Fig. 2), is found 1 . Therefore, we only present and discuss the amplitude data. In the conventional response the responsivity of a KID decreases with the increase of radiation power due to the reduction in life time of quasiparticles at higher quasiparticle densities. This has been clearly shown for both hybrid NbTiN/Al [3] and all-Al [4] KIDs. In contrast, the TiN KIDs reported in this paper become more responsive at higher radiation powers, as shown in Fig. 2.
In Fig. 3, we show the optical NEP for 350 GHz radiation as a function of radiation power from the black-body source. The optical NEP is defined as the power spectral density divided by the responsivity NEP opt ( ω ) = √ S A ( ω / δA/δP ) ( opt ), where S A is the noise in the amplitude signal and A the amplitude response. The noise in the amplitude signal at different radiation powers is measured at 250 mK, still well below T c / 10, and found to be independent of radiation power. Since the noise is constant and the responsivity increases with the radiation power, we find a radiation power dependent optical NEP. Electromagnetic analytical model [23] is used to evaluate the optical efficiency η for these KIDs, showing a value of 45%. The measured optical NEP has been compared to the photon shot noise NEP, which is calculated from √ 2 hνP opt (1 + ηB ) + 2∆ P opt /η pb /η , in which h is Planck's constant, ν the radiation frequency, P opt the ra-
1 At temperatures below 250 mK, the response in the phase has an initial strong negative response followed by a gradually rising positive response, which is tentatively attributed to the temperature dependence of the dielectric material.
FIG. 3. (Color online) Optical NEP for 350 GHz radiation in the amplitude signal defined as the power spectral density divided by the responsivity measured at a modulation frequency of 100 Hz (blue) compared to the electrical NEP calculated using the disordered-dependent pair breaking parameter model (black dot). Solid symbols were measured with a VNA whereas open symbols were measured with an IQ mixer. The green line is the expected photon noise limited NEP.
diation power, and (1 -ηB ) the correction to Poissonian statistics due to photon bunching for a single mode, with B the mode occupation [4]. The result clearly shows that the detectors are more sensitive at higher radiation powers, as is shown in Fig. 3. No sign of photon noise was observed in these experiments in contrast to recent experiments in which photon noise was reported [24], which may be related to different photon-power densities [25].
We also calculated the electrical NEP, determined using the standard technique described by Baselmans et al. [26], from
$$\stackrel { \iota \text{-body} } { \text{spec-} } & \quad N E P ^ { 2 } ( \omega ) = S _ { A } ( \omega ) \left [ \frac { \eta \eta _ { p b } \tau } { \Delta } \frac { \delta A } { \delta N _ { q p } } \right ] ^ { - 2 } ( 1 + \omega ^ { 2 } \tau ^ { 2 } ) ( 1 + \omega ^ { 2 } \tau _ { r e s } ^ { 2 } ) \\ \stackrel { \iota } { \text{e am-} } & \quad \text{with $w$ to documentation for common and $w$ to documentation}$$
with ω the resonator-frequency and τ res the resonator ring time. The noise and quasiparticle lifetime can be measured directly, but not the electrical responsivity δA/δN qp . Instead, the temperature dependence of the resonator δA/δT has been measured (Fig. 4) with T the temperature of the KID. The electrical response has been compared to Mattis-Bardeen theory [21]. The electrical response was converted into responsivity calculating the number of thermally excited quasiparticles present at the given temperature. The number of quasiparticles was calculated using the relation shown in Eq. 2 with a standard BCS DoS
$$& n _ { q p } = 4 N ( 0 ) \int _ { 0 } ^ { \infty } N ( E ) f ( E ) \mathrm d E \approx \quad \quad ( 2 ) \\ & \approx 2 N ( 0 ) \sqrt { 2 \pi k _ { B } T \Delta ( 0 ) } e ^ { - \Delta ( 0 ) / k _ { B } T },$$
where N E ( ) is the normalised quasiparticle density of
states, f ( E ) the quasiparticle energy distribution, k B the Boltzmann constant, and ∆(0) the superconducting energy gap at zero temperature. At this point we emphasize that for our material the correct conceptual framework and theoretical model is unclear. It is not possible to describe the temperature dependence of the resonancefrequency with standard formulas, taking the measured T c and a fixed relationship between k B T c and ∆(0) of 1 76 valid for weak-coupling superconductors. This leads . to the blue dotted curve in Fig. 4, and clearly signals a deviation. One can also take the energy-gap as a free adjustable parameter. This leads to an excellent fit and consequently to a DoS represented by the black dashed curve in the inset. In the more realistic model including the disorder [18] which leads to the red solid line, we obtain an excellent fit as well, but with a broadened DoS, as also observed with local tunnelling spectroscopy, and shown in the inset. However, as we will argue below, such an approach is potentially misleading because it suggests that one is dealing with a uniform superconducting state.
We determine the quasiparticle lifetime by measuring the change of the amplitude signal, while equilibrium is being restored, after applying a short high power microwave pulse. The pulse decay time which we interpret as the quasiparticle lifetime is found to be ∼ 40 µ sec at 250 mK (same temperature at which the optical NEP was measured). We find that it does not follow the exponential pattern expected from Kaplan et al. [22], as reported before by Coumou et al. [17]. The noise is determined with the black body source present at 3 5 K. We . find that the electrical NEP equals to 7 × 10 -17 W/Hz 1 / 2 , which is in agreement with the electrical NEP reported by Leduc et al. [9], taking into account the different superconducting transition temperatures and volumes. The difference between the electrical NEP and the measured optical NEP at the lowest radiation power is a factor of 100. Although TiN is a good absorber , we cannot under2 stand its behaviour using standard BCS theory [16]. No better understanding is achieved when a more realistic model including the disorder [18] is used for the analysis.
We have calculated the electrical NEP using our current understanding of strongly disordered superconductors. However, in our analysis we continue to assume an homogeneous superconducting state, whereas it has been clearly demonstrated that at a high enough degree of disorder, the superconducting energy gap ∆ varies spatially [15, 19]. In such an inhomogeneous system, excited quasiparticles might get trapped in low-gap puddles. In addition, the film may start to act as a random array of superconducting islands coupled by weak-link type Josephson junctions [27], which will have its own typical optical response.
Figure 4 B shows a cartoon of such a spatially varying
2 Room temperature transmission measurements with a vector network analyser and cryogenic FTS measurements showed an absorption efficiency of ∼ 75 -90% respectively.
FIG. 4. (Color online) A) Resonance frequency as a function of the bath temperature. The (grey) circles are the measured resonance frequencies, the dotted (blue) line is the prediction of the temperature dependence using Mattis-Bardeen theory [21] with the standard BCS DoS, the (black) dashed line is a fit to the Mattis-Bardeen theory with the standard BCS DoS and using the energy gap as a fit parameter, and the (red) solid line is a fit to the broadened DoS model in which both the energy gap and the broadening of the DoS are fitting parameters. The inset shows the quasiparticle DoS used for each fit. B) Cartoon of the disorder dependent landscaped energy gap conjecture. The quasiparticles get trapped in the valleys but can escape and move freely when their thermal energy becomes comparable to the roughness of the gap.
energy gap ∆( ). In this figure, glyph[vector] r µ p denotes the chemical potential, and the grey shade indicates the thermal distribution of quasiparticles. We estimate the root mean square variation of the gap ∆ 1 = √ 〈 (∆( ) glyph[vector] r -〈 (∆( )) ) glyph[vector] r 〉 2 〉 from the broadening parameter used to describe the electrodynamic response of the resonator: ∆ 1 ≈ 44 µe V. At a temperature of 100 mK, most quasiparticles will be trapped in the low-gap areas, since the thermal energy k B T = 9 µe V glyph[lessmuch] ∆ . 1 At higher temperatures, the thermal energy and the variations of the energy gap become comparable and any (photon-induced) excess quasiparticles will be free to move. A similar effect will occur with increasing radiation power: at first excess quasiparticles will be trapped in the low-gap regions, but at increasing power, the excess quasiparticles will be free to move, and will not be sensitive to the underlying gap inhomogeneity anymore. The superconductor will increasingly behave as an electronically homogeneous material. Note that in this qualitative model, there is still a coherent superconducting condensate present, which will determine the phase response of the KID, whereas the dissipative (amplitude) response will involve a combination of mobile and trapped quasiparticles. This model might therefore also explain the difference in phase and amplitude relaxation time that was observed by Gao et al. in TiN KIDs [12].
In summary, we have studied ALD TiN KIDs at a radiation frequency of 350 GHz using the same setup as used for hybrid NbTiN/Al [3] and all-Al KIDs [4]. A radiation power dependent responsivity was found, in contrast to the well understood hybrid NbTiN/Al and all-Al KIDs. Since the noise is independent of radiation power, the optical NEP is also a strong function of the radiation power. At high radiation powers, comparable to the incident radiation in ground-based telescopes, the optical NEP approaches the requirements for ground-based instruments (recently achieved by increasing the received power per unit volume of the KIDs [24]). Despite the fact that the TiN is a good absorber, the difference between the electrical NEP and the measured optical NEP at the lowest radiation power is a factor of 100.
The authors thank M. Spirito for room temperature
- [1] P. K. Day, H. G. LeDuc, B. A. Mazin, A. Vayonakis, and J. Zmuidzinas, Nature 293 , 817 (2003)
- [2] S. J. C. Yates, J. J. A. Baselmans, A. Endo, R. M. J. Janssen, L. Ferrari, P. Diener, and A. M. Baryshev, Appl. Phys. Lett. 99 , 073505 (2011)
- [3] R. M. J. Janssen, J. J. A. Baselmans, A. Endo, L. Ferrari, S. J. C. Yates, A. M. Baryshev, and T. M. Klapwijk, Appl. Phys. Lett. 103 , 203503 (2013)
- [4] P. J. de Visser, J. J. A. Baselmans, J. Bueno, N. Llombart, and T. M. Klapwijk, Nature communications 5 , 3130 (2014)
- [5] S. Doyle, P. Mauskopf, J. Naylon, A. Porch, and C. Duncombe, J. Low Temp. Phys. 151 , 530 (2008)
- [6] P. Mauskopf, S. Doyle, P. Barry, S. Rowe, A. Bidead, P. Ade, C. Tucker, E. Castillo, A. Monfardini, J. Goupy, and M. Calvo, J. Low Temp. Phys. 176 , 545 (2014)
- [7] R. Barends, H. L. Hortensius, T. Zijlstra, J. J. A. Baselmans, S. J. C. Yates, J. R. Gao, and T. M. Klapwijk, Appl. Phys. Lett. 92 , 223502 (2008)
- [8] R. Barends, N. Vercruyssen, A. Endo, P. J. de Visser, T. Zijlstra, T. M. Klapwijk, P. Diener, S. J. C. Yates, and J. J. A. Baselmans, Appl. Phys. Lett. 97 , 023508 (2010)
- [9] H. G. Leduc, B. Bumble, P. K. Day, A. D. Turner, B. H. Eom, S. Golwala, D. C.Moore, O. Noroozian, J. Zmuidzinas, J. Gao, B. A. Mazin, S. McHugh, A. Merrill, Appl. Phys. Lett. 97 , 102509 (2010)
- [10] M. R. Vissers, J. Gao, D. S. Wisbey, D. A. Hite, C. C. Tsuei, A. D. Corcoles, M. Steffen, and D. P. Pappas, Appl. Phys. Lett. 97 , 232509 (2010)
- [11] M. Sandberg, M. R. Vissers, J. S. Kline, M. Weides, J. Gao, D. S. Wisbey. and D. P. Pappas, Appl. Phys. Lett. 100 , 262605 (2012)
- [12] J. Gao, M. R. Vissers, M. O. Sandberg, F. C. S. da Silva, S. W. Nam, D. P. Pappas, D. S. Wisbey, E. C. Langman, S. R. Meeker, B. A. Mazin, H. G. Leduc, J. Zmuidzinas, and K. D. Irwin, Appl. Phys. Lett. 101 , 142602 (2012)
- [13] C. M. McKenney, H. G. Leduc, L. J. Swenson, P. K. Day,
transmission measurements, P. A. R. Ade for cryogenic FTS measurements, B. Blazquez and N. Llombart for electromagnetic modelling, and R. M. J. Janssen for stimulating discussions. This work was supported as part of a collaborative project, SPACEKIDS, funded via grant 313320 provided by the European Commission under Theme SPA.2012 2 2-01 of Framework Programme . . 7. This work is part of the research programme of the Foundation for Fundamental Research on Matter (FOM), which is part of the Netherlands Organisation for Scientific Research (NWO). T. M. K. also acknowledges financial support from the Ministry of Science and Education of Russia under contract No. 14.B25.31.0007 and from the European Research Council Advanced grant No. 339306 (METIQUM). E. F. C. D. was financially supported by the CEA-Eurotalents program.
- B. H. Eom, J. Zmuidzinas, Proc. SPIE 8452 , 84520S (2012)
- [14] B. A. Mazin, B. Bumble, S. R. Meeker, K. O Brien, ′ S. McHugh, and E. Langman, Optics Express 20 , 1503 (2012)
- [15] B. Sac´p´ e, C. Chapelier, T. I. Baturina, V. M. Vinokur, e M. R. Baklanov, and M. Sanquer, Phys. Rev. Lett. 101 , 157006 (2008)
- [16] J. Bardeen, L. N. Cooper, and J. R. Schrieffer, Phys. Rev. 108 , 1175 (1957)
- [17] P. C. J. J. Coumou, M. R. Zuiddam, E. F. C. Driessen, P. J. de Visser, J. J. A. Baselmans, and T. M. Klapwijk, IEEE Transactions on Applied Superconductivity 23 , 7500404 (2013)
- [18] E. F. C. Driessen, P. C. J. J. Coumou, R. R. Tromp, P. J. de Visser, and T. M. Klapwijk, Phys. Rev. Lett. 109 , 107003 (2012)
- [19] P. C. J. J. Coumou, E. F. C. Driessen, J. Bueno, and T. M. Klapwijk, Phys. Rev. B 88 , 180505(R) (2013)
- [20] J. J. Baselmans, S. Yates, P. Diener, and P. J. de Visser, J. Low Temp. Phys. 167 , 360 (2012)
- [21] D. C. Mattis, and J. Bardeen, Phys. Rev. 111 , 412 (1958)
- [22] S. B. Kaplan, C. C. Chi, D. N. Langenberg, J. J. Chang, S. Jafarey, and D. J. Scalapino, Phys. Rev. B 14 , 4854 (1976)
- [23] B. Blazquez, N. Llombart, D. Cavallo, A. Freni, and A. Neto, Antennas and Propagation, IEEE Transactions on , 99 , 1 (2014)
- [24] J.Hubmayr, J. Beall, D. Becker, H.-M. Cho, M. Devlin, B. Dober, C. Groppi, G.C. Hilton, K.D. Irwin, D. Li, P. Mauskopf, D.P. Pappas, J. Van Lanen, M.R. Vissers, and J. Gao, arXiv:1406.4010 (2014)
- [25] C. M. Bradford, private communication
- [26] J. Baselmans, S. J. C. Yates, R. Barends, Y. J. Y. Lankwarden, J. R. Gao, H. Hoevers, and T.M. Klapwijk J. Low Temp. Phys. 151 , 524 (2008)
- [27] T. Cea, D. Bucheli, G. Seibold, L. Benfatto, J. Lorenzana, C. Castellani, Phys. Rev. B 89 , 174506 (2014) | 10.1063/1.4901536 | [
"J. Bueno",
"P. C. J. J. Coumou",
"G. Zheng",
"P. J. de Visser",
"T. M. Klapwijk",
"E. F. C. Driessen",
"S. Doyle",
"J. J. A Baselmans"
] | 2014-08-01T19:14:52+00:00 | 2014-08-20T09:24:25+00:00 | [
"cond-mat.supr-con"
] | Anomalous response of superconducting titanium nitride resonators to terahertz radiation | We present an experimental study of KIDs fabricated of atomic layer deposited
TiN films, and characterized at radiation frequencies of $350$~GHz. The
responsivity to radiation is measured and found to increase with increasing
radiation powers, opposite to what is expected from theory and observed for
hybrid niobium titanium nitride / aluminium (NbTiN/Al) and all-aluminium
(all-Al) KIDs. The noise is found to be independent of the level of the
radiation power. The noise equivalent power (NEP) improves with higher
radiation powers, also opposite to what is observed and well understood for
hybrid NbTiN/Al and all-Al KIDs. We suggest that an inhomogeneous state of
these disordered superconductors should be used to explain these observations. |
1408.0271v1 | ## Nonmetallic Low-Temperature Normal State of K 0 70 . Fe 1 46 . Se 1 85 . Te 0 15 .
Kefeng Wang, 1, ∗ Hyejin Ryu, 1, 2 Erik Kampert, 3 M. Uhlarz, 3 J. Warren, 4 J. Wosnitza, 3, 5 and C. Petrovic 1, 2, †
1 Condensed Matter Physics and Materials Science Department, Brookhaven National Laboratory, Upton, New York 11973, USA 2 Department of Physics and Astronomy, Stony Brook University, Stony Brook, New York 11794-3800, USA 3 Hochfeld-Magnetlabor Dresden (HLD), Helmholtz-Zentrum Dresden-Rossendorf, D-01314 Dresden, Germany 4 Instrument Division, Brookhaven National Laboratory, Upton, New York 11973, USA 5 Institut f¨r u Festk¨rperphysik, o TU Dresden, D-01062 Dresden, Germany
(Dated: August 4, 2014)
The normal-state in-plane resistivity below the zero-field superconducting transition temperature T c and the upper critical field µ H 0 c 2 ( T ) were measured by suppressing superconductivity in pulsed magnetic fields for K 0 70 . Fe 1 46 . Se 1 85 . Te 0 15 . . The normal-state resistivity ρ ab is found to increase logarithmically with decrasing temperature as T T c → 0. Similar to granular metals, our results suggest that a superconductor - insulator transition below zero-field T c may be induced in high magnetic fields. This is related to the intrinsic real-space phase-separated states common to all inhomogeneous superconductors.
PACS numbers: 72.80.Ga,72.20.Pa,75.47.Np
## INTRODUCTION
The normal-state properties (including pseudogap at the Fermi surface and proximity of magnetism at the phase diagrams) of unconventional superconductors such as highT c copper oxides, iron-pnictides/chalcogenides and heavy-fermions have received much attention [1]. The linear temperature dependence of the in-plane resistivity above T c [2-4] is considered to be related to magnetic quantum criticality and antiferromagnetic fluctuations [5]. Since the normal (nonsuperconducting) states of interest appear under extremely high magnetic fields and at low temperatures ( T → 0), only a few studies were performed investigating the normal-state properties of highT c superconductors below T c so far. In copper oxides examples include underdoped La 2 -x Sr x CuO 4 and Bi Sr 2 2 -x La CuO , where x 6 ρ T ( ) diverges logarithmically below T c and where a superconductor - insulator transition (SIT) is induced in pulsed magnetic fields [6-9]. This came as a surprise since insulating state revealed by the magnetic-field induced breakdown of superconductivity was not expected in a conventional metal. Studies (Ba,K)Fe As 2 2 and LaFeAsO 1 -x F x resistivity suggested a slight upturn below zero-field T c after the partial suppression of superconductivity up to 61 T pulsed field [10, 11]. In SmFeAsO 1 -x F x magnetic-field induced logarithmic insulating behavior in the resistivity spans one decade in temperature above the zero-field T , c stems from the large magnetoresistance that extends to temperatures well above the log-T regime and is not linked to the superconducting state [12].
K Fe x 2 -y Se 2 exhibits a high-temperature insulatormetal crossover at T s ∼ 100 -200 K before becoming superconducting at T c = 30 K [13-15]. Insulating ground states can also be found by tuning parameters such as Fe occupancy or isovalent S substitution on the
Se atomic site [16, 17]. As opposed to copper oxides, a dual description including both itinerant and localized carriers/magnetic moments may be required [18, 19]. But similar to the copper oxides, it is imperative to probe the normal state below zero-field T c . This is even more so due to the argued granular-like superconductivity arising in nanoscale superconducting grains separated by non-superconducting islands, with a macroscopic superconducting condensate being established via Josephson tunneling among the grains [20, 21]. In the underdoped cuprates electronic granularity stems from the nature of the chemical bonds; hole-rich superconducting regions may form for low level of hole doping even though the crystal structure is not granular [22]. In K Fe x 2 -y Se 2 crystals, the phase separation is an intrinsic feature of the crystal structure [23, 24].
Here, we measured the normal-state in-plane resistivity below T c for K 0 70 . Fe 1 46 . Se 1 85 . Te 0 15 . by suppressing superconductivity in pulsed magnetic fields. The normalstate resistivity ρ ab is found to increase logarithmically as T T c -→ 0. Similar to granular superconductors and underdoped La 2 -x Sr x CuO 4 and Bi Sr 2 2 -x La CuO x 6 [6-9], our results suggest for the first time that magnetic-fielddriven SIT is induced in high magnetic field in Fe-based superconductors.
## EXPERIMENTAL
Single crystals of pure K x Fe 2 -y Se 2 and nominally 10% Te-doped K Fe x 2 -y Se 2 used in this study were grown and characterized as described previously [17]. The powder X-ray diffraction (XRD) spectra were taken with Cu K α radiation ( λ = 0 15418 nm) by a Rigaku Miniflex X-ray . machine, confirming phase purity. The cleaved surface image was determined in a JEOL JSM-6500 scanning
/s40/s51/s48/s48/s75/s41
/s32
/s32
/s40/s48/s75/s41/s47
FIG. 1. (color online) Nominal 7 5% Te substitution of Se in . K Fe x 2 -y Se 2 induces a decrease of the superconducting transition temperature T c and of the semiconductor-metal transition temperature T s . (a) The powder x-ray diffraction patterns for Te-doped and pure K x Fe 2 -y Se 2 crystals. The shift of the diffraction peaks toward lower angle shows the expanded lattice in Te-doped crystals. (b) The energy-dispersive X-ray spectroscopy of a typical K 0 70 . Fe 1 46 . Se 1 85 . Te 0 15 . single crystal, which clearly shows the Te peak. (c) The SEM image of a typical single crystal showing stripe phases. (d) Temperature dependence of the resistivity for pure and Te-doped K Fe x 2 -y Se . 2 The inset shows the magnified part around the superconducting transition temperature. The T c onset ( ) decreases from 30 to 29.5 K and T s decreases from 130 to 100 K.
electron microscope (SEM). The average stoichiometry was determined by energy-dispersive X-ray spectroscopy (EDX) in the same SEM. Four-probe resistivity was measured in pulsed magnetic fields up to 65 T and in Quantum Design PPMS-14 and PPMS-9. Contacts were attached to the ab plane of crystals in a glove box. Sample dimensions were measured by an optical microscope Nikon SMZ-800 with 10 µ m resolution. The electric current used in the pulsed field measurement was always 2 mA and the corresponding electric current density in the sample was ∼ 1 39 . × 10 4 A/m . 2
## RESULTS AND DISCUSSIONS
Fig. 1(a) shows the diffraction patterns of both compounds and all reflections can be indexed in the I4/mmm space group. The lattice parameters for K 0 70 . Fe 1 46 . Se 1 85 . Te 0 15 . from the refinement are a = b = 3 910(3) . nm and c = 14 268(6) nm, . which are larger than for K Fe x 2 -y Se 2 ( a = b = 3 881(1) nm and . c = 14 151(6) nm). The expanded crystal lattice implies that . Te enters the lattice, which is further confirmed by the EDX large-area elemental analysis which shows the average stoichiometry K 0 70(6) . Fe 1 46(3) . Se 1 84(8) . Te 0 15(2) . [Fig. 1(b)]. The SEM image [Fig. 1(c)] shows a typical phase-separation pattern. Compared to undoped K Fe x 2 -y Se 2 [25-27], the pattern is inhomogeneous: Stripes can be up to 100 nm large whereas their lower limit is below 50 nm. The local-point analysis on the background (dark) areas gives the stoichiometry K 0 78(4) . Fe 1 70(3) . Se 1 84(8) . Te 0 15(2) . , which is close to the 245 insulating phase [28]. About 7 5% Te substitution in-. duces a T c onset ( ) decrease from ∼ 30 K to ∼ 29 5 . K, and metal-semiconductor transition temperature, T s , decrease from ∼ 130 K to ∼ 100 K [Fig. 1(d)].
Near T c , the upper critical field slopes, d µ H 0 c 2 ( T )/d T | T = T c in K Fe x 2 -y Se 2 and other ternary/quaternary iron arsenide and selenide superconductors, are about 5 to 12 TK -1 for H ab ‖ and 1 to 3 TK -1 for H c ‖ . Using the WerthamerHelfand-Hohenberg (WHH) estimate [29], results in µ H 0 c 2 ( T ) | T =0 = (125-275) T and (30-60) T for H ab ‖ and H c ‖ , respectively [30-32]. The upper critical field for H c ‖ presents almost linear temperature dependence up to 60 T instead of the saturation predicted by WHH. This is ascribed to multiband effect [30, 31]. Although the upper critical field for H c ‖ is larger than the WHH estimate, it is smaller than µ H 0 c 2 ( T ) for H ab ‖ which is well explained within the WHH theory [31, 32]. For K 0 70 . Fe 1 46 . Se 1 85 . Te 0 15 . , we observe similar slopes of µ H 0 c 2 ( T ) with d µ H 0 c 2 ( T )/d T | T = T c ∼ 2 1(4) TK . -1 for H c ‖ [Fig. 2(a-c)] and d µ H 0 c 2 ( T )/d T | T = T c ∼ 4 9(3) . TK -1 for H ab ‖ . However, in K 0 70 . Fe 1 46 . Se 1 85 . Te 0 15 . , at lower temperatures µ H 0 c 2 ( T ) is significantly reduced. A magnetic field of 61 T suppresses the superconductivity at 1.7 K for H c ‖ , as opposed to H a ‖ [Fig. 3(a)] . This is ascribed to a change in the temperature dependence of µ H 0 c 2 ( T ).
Effects of Pauli spin paramagnetism and spin-orbit scattering can be included in the WHH theory for a single-band s-wave weak-coupling type-II superconductor in the dirty limit by adding the Maki parameters α and λ so [29, 33]. Then, µ H 0 c 2 ( T ) is given by
$$\mathbb { e } _ { 2 } ^ { \mathbb { e } } \ln ( \frac { 1 } { t } ) & = ( \frac { 1 } { 2 } + \frac { i \lambda _ { s o } } { 4 \gamma } ) \psi ( \frac { 1 } { 2 } + \frac { h + \lambda / 2 + i \gamma } { 2 t } ) \\ & + ( \frac { 1 } { 2 } - \frac { i \lambda _ { s o } } { 4 \gamma } ) \psi ( \frac { 1 } { 2 } + \frac { h + \lambda / 2 - i \gamma } { 2 t } ) - \psi ( \frac { 1 } { 2 } ),$$
/s32/s99/s109/s41
/s32/s40/s109
FIG. 2. (color online) Field dependence of the in-plane resistivity in a K 0 70 . Fe 1 46 . Se 1 85 . Te 0 15 . single crystal with magnetic field applied parallel to the c axis at fixed temperatures. (a) The resistivity data for temperatures between 22 and ∼ 33 K up to 14 T. (b) Temperature dependence of µ H 0 c 2 ( T ) near zero-field T c for the Te-doped crystal with H c ‖ using different criteria in the resistive transition shown in (a). (c) Resistivity data for temperatures between 1.75 and 33 K in pulsed fields up to 60 T.
where ψ x ( ) is the digamma function, γ = [( αh ) 2 -( λ so / 2) ] 2 1 / 2 , and
$$h = \frac { 4 \mu _ { 0 } H _ { c 2 } ( T ) } { \pi ^ { 2 } T _ { c } [ - \text{d} \mu _ { 0 } H _ { c 2 } ( T ) / \text{d} T ] _ { T = T _ { c } } }. \quad \begin{pmatrix} & \lim i \\ & = \ i \\ & \xi ( 0 ) \\ & \end{pmatrix}.$$
In our data, we observe a convex curvature and a tendency to saturation for µ H 0 c 2 ( T ) with H c ‖ at low temperatures [Fig. 3(b)] which can be well understood by the WHH theory with α = 0 and λ = 0, i.e., with orbitally limited H c 2 : µ H 0 ∗ c 2 (0) = -0.693 T c ( dH c 2 /dT ) T c [the blue curve in Fig. 3(b)]. Hence, the isovalent Te substitution suppresses the multiband character and promotes a single-band orbital-limited behavior.
The in-plane upper critical field is limited due to spin paragmagnetism since α = 0 28 . and λ so = 0 [red curve in Fig. 3(b)]. When compared to α H ab ‖ = 5 6 and . α H ab ‖ = 1 9 in Tl . 0 58 . Rb 0 42 . Fe 1 72 . Se 2 and K Fe x 2 -y Se 2 , respectively [30, 32], the spinparamagnetic effect in the Te-doped system is very
/s99
/s99
FIG. 3. (color online) The upper critical fields in K 0 70 . Fe 1 46 . Se 1 85 . Te 0 15 . follow the temperature dependence as predicted by the WHH theory. (a) Field dependence of the in-plane resistivity of K 0 70 . Fe 1 46 . Se 1 85 . Te 0 15 . for pulsed magnetic fields applied perpendicular to the c -axis. (b) Temperature dependence of the upper critical field H c 2 ( T ) (plotted as µ H 0 c 2 T c [ -dµ H 0 c 2 /dT ] Tc ) for magnetic fields parallel (circles) and perpendicular (balls) to the c axis, respectively. The solid lines are the fitting results using WHH theory with different parameters.
small ( α < 1), but the high anisotropy of µ H 0 c 2 ( T ) for the different principal crystallographic directions remains, suggesting that an anistropic Fermi surface is preserved. Open electron orbits along the c axis render an orbital-limited field unlikely and spin-paramagnetic effects should play an important role for H ab ‖ . The paramagnetically limited field µ H 0 p c 2 (0) is given by µ H 0 p c 2 (0)= µ H 0 ∗ c 2 (0)/ √ 1 + α 2 and α = √ 2 H ∗ c 2 (0)/ H p (0), where µ H 0 p (0) is zero-temperature Pauli limited field and the orbitally limited µ H 0 ∗ c 2 given above [33]. This gives µ H 0 p,ab c 2 (0) = 100 T whereas µ H 0 p,c c 2 (0) = 45 7 T. . The zero-temperature coherence length, ξ (0), can be estimated using the Ginzburg-Landau formula µ H 0 c 2 (0) = Φ /2 0 πξ 2 (0), where Φ =2.07 0 × 10 -15 Wb, resulting ξ (0) ab = 2 7(1) nm and . ξ (0) c = 1 8(3) nm, respectively. .
The reduction of the Maki parameter, α , in K 0 70 . Fe 1 46 . Se 1 85 . Te 0 15 . when compared to K Fe x 2 -y Se 2 cannot be explained by disorder effects since for electronic systems with more disorder one expects a larger α [34]. Therefore, another effect must compete and prevail over the disorder induced by Te substitution. With strong-coupling correction for electron-boson and electron-electron interactions µ H 0 p (0)=1 86(1+ . λ ) ε η ∆ η ib (1 -I ), where strong-coupling intraband corrections for the gap are described by η ∆ , I is the Stoner factor I = N E ( F ) J , N E ( F ) is the electronic density of states (DOS) per spin at the Fermi level, E F , J is an effective exchange integral, η ib is introduced to describe phenomenologically the effect of the gap
/s32
/s32/s99/s109/s41
/s32/s40/s109
FIG. 4. (color online) (a) Temperature dependence of the in-plane resistivity in several DC (open symbols) and pulsed (filled symbols) magnetic fields for temperature between 1 75 . and 100 K. Pulsed-field data (filled symbols) near T c is consistent with results in DC and zero fields (open symbols). The two half-filled symbols for 61 T at 1.75 and 4 K denote the extrapolated values of the resistivity after complete suppression of the superconductivity. The red and green dashed lines are the linear fitting result for the 61 T data and 50 T data when using logarithmic scale, respectively. Inset: resistivity of the mixed state in 50 T field and the blue dotted line is an exponential fitting result (see text) (b) The schematic temperature-field phase diagram of K 0 70 . Fe 1 46 . Se 1 85 . Te 0 15 . .
anisotropy, λ is the electron-boson coupling constant and ε = 0.5 or 1 [35]. The spin-paramagnetic effect can be less pronounced when the Stoner factor becomes small which can be achieved through suppressing either the DOS at the Fermi level or the effective exchange integral J [35]. The reduction of both, the multiband character in H c 2 ( T ) and the spin paramagnetism when compared to pure K Fe x 2 -y Se 2 , implies that electronic states, removed by Te substitution, carry substantial partial density of states at the Fermi level, i.e., that Te substitution reduces the occupancy of heavily renormalized d xy orbitals [36].
In K 0 70 . Fe 1 46 . Se 1 85 . Te 0 15 . , a magnetic field of 61 T parallel to the c axis completely suppresses superconductivity above 1.75 K [Fig. 2(c) and 3(b)], providing the opportunity to study the normal-state properties as T -→ 0. The application of 61 T leads to a negative dρ/dT resembling insulating rather than metallic behavior [Fig. 4(a)]. It is apparent that a SIT occurs right below T c (filled balls) after the superconductivity is completely suppressed above H c 2 . At 50 T, the resistivity as well shows SIT and is nearly identical to that at 61 T field until superconductivity sets in below about 15 K. We first note that the insulating behavior in high magnetic fields appears only below the zero-field T c . It is also instructive to note that above the zero-field T c the field dependence of the resistivity up to 61 T is minute for H c ‖ , in contrast
/s32
/s52/s48
to SmFeAsO 1 -x F x [12]. The temperature-magnetic-field phase diagram for K 0 70 . Fe 1 46 . Se 1 85 . Te 0 15 . is shown in Fig. 4(b).
A SIT was first reported in ultrathin, disordered and granular type-I superconductors [37, 38]. Then, a SIT with an insulating resistivity, increasing logarithmically below T c , was reported in highT c cuprate superconductors utilizing pulsed magnetic fields [6-8]. Such a temperature dependence is different from variable range hopping (VRH) and thermally activated transport observed in semiconductors. Kondo-type magnetic scattering is also unlikely since the spin-flip scattering should be suppressed in 61 T. Evidence for an intrinsic electronic phase separation in the cuprate superconductors [39] suggests that the logarithmic increase in the resistivity and the SIT originates from an inhomogeneous state, presumably similar to granular Nb. [38, 40].
K Fe x 2 -y Se 2 features a highly defective crystal structure where the magnetic insulating regions are spatially intrinsically separated and coexist with superconducting regions on the a length scale of 100 nm to 1 µ m [25-27]. However, as opposed to two-dimensional alternate stacking of insulating and (super)conducting layers, the stripe-type phase separation in K x Fe 2 -y Se 2 is more three-dimensional [21-23]. The conductivity in such inhomogeneous samples must include contributions from the insulating ( σ i ) and metallic ( σ m ) regions, i.e., σ = σ m + σ i . When T < T c , σ m goes to infinity due to superconductivity and the insulating part of the sample is short-circuited ( σ = σ m = ∞ ). But even around T c and in particular in the normal state at high magnetic field when T -→ 0, σ ∼ σ m holds as well since insulating regions have a nearly two orders of magnitude higher resistivity [41]. This is consistent with the observation that insulating regions do not contribute to the spectral weight in angular resolved photoemission data close to E F [36].
Therefore, superconductivity in K Fe x 2 -y Se 2 is similar to that in granular superconductors, namely a threedimensional array of superconducting grains in an insulating matrix [38, 40]. Strong magnetic fields suppress superconductivity in each grain. The possible scenarios explaining a SIT include Anderson localization [42] and the bosonic scenario where the gap is driven to zero by phase fluctuations [43, 44]. The observed logarithmic divergence of the resistivity in the nonmetallic normal state of K 0 70 . Fe 1 46 . Se 1 85 . Te 0 15 . [Fig. 4(a)] does not follow the VRH conduction which is observed for Anderson localization. In the bosonic scenario the Cooper pairs are localized in granules (similar to defects). For a H > H c 2 , viritual Cooper pairs may form, leading to a reduction of the DOS. When T -→ 0, the pairs cannot travel between grains and the resistivity becomes much larger than for a normal metal without electronelectron interactions [43, 44]. In contrast to the spinparamagnetic critical field, the orbital critical field de-
pends on the grain size as H o c 2 ∼ φ /Rξ 0 where R is the average grain radius and ξ is the coherence length. Since the upper critical field in K 0 70 . Fe 1 46 . Se 1 85 . Te 0 15 . is about 55(3) T and the grain coherence length is 2.7(1) nm, we estimate a lower limit of the stripe grain size to be R = 21(2) nm. Near the SIT on superconducting side it is expected that ρ = ρ 0 exp( T/T 0 ) [40] in high fields due to the destruction of fluctuating quasilocalized Cooper pairs and a concomitant increase of the number of quasiparticles at the Fermi level. Our 50 T resistivity data in the mixed state can be described well by this formula [The blue line in inset of Fig. 4(a)]. Different heat treatment and quenching can bring about changes in the normalstate resistivities, arrangement and connectivity of the superconducting grains [28]. This can influence the tunneling conductance in the granular array and, therefore, it would be of great interest to study superconducting fluctuations in such samples in high magnetic fields in the normal state below T c for further comparison with theory [38, 40].
Even though the logarithmic temperature dependence of the resistivity in copper oxides is consistent with the theory of granular superconductivity if electrons in different CuO planes are incoherent in the low-temperature limit [45, 46], this is still controversial since the SIT in copper oxides occurs on the metallic side of the Mott limit ( k F l > 1, where k F is the Fermi wave vector and l is the mean free path) [47]. Approaches offered to explain insulating states in copper-oxide superconductors postulate spin-charge separation [48], invoke a bipolaron theory of strongly correlated Mott-Hubbard insulators [49], or propose Mott insulating states in antiferromagnetic vortices within the framework of SO(5) theory of antiferromagnetism and superconductivity [50]. In the light of increasing evidence for insulating states in the different kinds of highT c superconductors in pulsed magnetic fields, more theoretical work may be needed to determine which of these approaches is compatible with real-space and/or electronic phase separation in materials of different structure and bonding types.
## CONCLUSION
In summary, the normal-state in-plane resistivity and the upper critical fields of K 0 70 . Fe 1 46 . Se 1 85 . Te 0 15 . are measured by suppressing superconductivity in pulsed magnetic fields. The temperature dependence of µ H 0 c 2 ( T ) can be well described using the WHH theory, in contrast to K x Fe 2 -y Se 2 where the linear temperature dependence signals multiband effects. After suppressing superconductivity, the normal-state resistivity, ρ ab , is found to be insulating and to diverge logarithmically as T T c -→ 0. Similar to the SIT in granular superconductors, our results suggest that the mechanism for the SIT in high magnetic fields is related to intrinsic phase-separated states in this kind of materials.
## ACKNOWLEDGMENTS
We thank Myron Strongin and Dragana Popovic for useful discussions. Work at Brookhaven is supported by the U.S. DOE under contract No. DE-AC02-98CH10886. We acknowledge the support of the HLD at HZDR, member of the European Magnet Field Laboratory. CP acknowledges support by the Alexander von Humboldt Foundation.
- ∗ †
- [email protected]
- [email protected]
- [1] D. J. Scalapino, A common thread: The pairing interaction for unconventional superconductors. , Rev. Mod. Phys. 84 , 1383 (2012).
- [2] H. Takagi, Systematic evolution of temperaturedependent resistivity in La 2 -x Sr x CuO 4 , Phys. Rev. Lett. 69 , 2975 (1992).
- [3] R. Daou, David LeBoeuf, N. Doiron-Leyraud, S.Y. Li, F. Lalibert´, e O. Cyr-Choini`re, Y.J. Jo, L. Balicas, e J.Q. Yan, J.-S. Zhou, J.B. Goodenough, and L. Taillefer, Linear temperature dependence of resistivity and change in the Fermi surface at the pseudogap critical point of a high-T c superconductor , Nat. Phys. 5 , 31 (2009).
- [4] P. Gegenwart, Qimiao Si and F. Steglich, Quantum criticality in heavy-fermion metals , Nat. Physics 4 , 186 (2008).
- [5] H. v. L¨ ohneysen, A. Rosch, M. Vojta and P. W¨ olfle, Fermi-liquid instabilities at magnetic quantum phase transitions , Rev. Mod. Phys. 79 , 1015 (2009).
- [6] Y. Ando, G. S. Boebinger, A. Passner, T. Kimura, and K. Kishio, Logarithmic Divergence of both In-Plane and Out-of-Plane Normal-State Resistivities of Superconducting La 2 -x Sr x CuO 4 in the Zero-Temperature Limit. , Phys. Rev. Lett. 75 , 4662 (1995).
- [7] G. S. Boebinger, Y. Ando, A. Passner, T. Kimura, M. Okuya, J. Shimoyama, K. Kishio, K. Tamasaku, N. Ichikawa, and S. Uchida, Insulator-to-Metal Crossover in the Normal State of La 2? x Sr x CuO 4 Near Optimum Doping , Phys. Rev. Lett. 77 , 5417 (1996).
- [8] S. Ono, Y. Ando, T.Murayama, F. F. Balakirev, J. B. Betts, and G. S. Boebinger, Metal-to-Insulator Crossover in the Low-Temperature Normal State of Bi Sr 2 2 -x La CuO x 6+ δ , Phys. Rev. Lett. 85 , 638 (2000).
- [9] Xiaoyan Shi, G. Logvenov, A. T. Bollinger, I. Boˇovi´, z c C. Panagopoulos, and Dragana Popovi´, c Emergence of superconductivity from the dynamically heterogeneous insulating state in La 2 -x Sr x CuO 4 , Nature Mater. 12 , 47 (2013).
- [10] H. Q. Yuan, J. Singleton, F. F. Balakirev, S. A. Baily, G. F. Chen, J. L. Luo, and N. L. Wang, Nearly isotropic superconductivity in (Ba,K)Fe As , 2 2 Nature 457 , 565 (2009).
- [11] Y. Kohama, Y. Kamihara, A. Baily, L. Civale, S. C. Riggs, F. F. Balakirev, T. Atake, M. Jaime, M. Hirano
- and H. Hosono, Nearly isotropic superconductivity in (Ba,K)Fe As , Nature 2 2 457 , 565 (2009).
- [12] S. C. Riggs, J. B. Kemper, Y. Jo, Z. Stegen, L. Balicas, G. S. Boebinger, F. F. Balakirev, A. Migliori, H. Chen, R. H. Liu and X. H. Chen, Magnetic field-induced logT insulating behavior in the resistivity of fluorine-doped SmFeAsO 1 -x F , Phys. Rev. B x 79 , 212510 (2009).
- [13] J. Guo, S. Jin, G. Wang, S. Wang, K. Zhu, T. Zhou, M. He, and X. L. Chen, Superconductivity in the iron selenide K x Fe Se 2 2 ( 0 ≤ x ≤ 1 0 , Phys. Rev. B . ) 82 , 180520 (2010).
- [14] E. Dagotto, The unexpected properties of alkali metal iron selenide superconductors , Rev. Mod. Phys. 85 , 849 (2013).
- [15] H.-H. Wen, Overview on the physics and materials of the new superconductor K Fe x 2 -y Se 2 , Rep. Prog. Phys. 75 , 112501 (2012).
- [16] Y. J. Yan, M. Zhang, A. F. Wang, J. Ying, Z. Y. Li, W. Qin, X. G. Guo, J. Q. Li, Jianping Hu and X. H. Chen, Electronic and magnetic phase diagram in K x Fe 2 -y Se 2 superconductors , Sci. Rep. 2, 212 (2012).
- [17] H. C. Lei, M. Abeykoon, E. S. Bozin, K. F. Wang, J. B. Warren, and C. Petrovic, Phase Diagram of K Fe x 2 -y Se 2 -z S z and the Suppression of its Superconducting State by an Fe -Se/S 2 Tetrahedron Distortion , Phys. Rev. Lett. 107 , 137002 (2011).
- [18] P. C. Dai, J. P. Hu, and E. Dagotto, Magnetism and its microscopic origin in iron-based high-temperature superconductors , Nat. Phys. 8 , 709 (2012).
- [19] Rong Yu and Qimiao Si, Orbital-Selective Mott Phase in Multiorbital Models for Alkaline Iron Selenides K 1 -x Fe 2 -y Se 2 , Phys. Rev. Lett. 110 , 146402 (2013).
- [20] M. Sigrist and T. M. Rice, Paramagnetic Effect in High T c Superconductors - A Hint for d-wave Superconductivity , J. Phys. Soc. Jpn. 61 , 4283 (1992).
- [21] R. H. Yuan, et al., Nanoscale phase separation of antiferromagnetic order and superconductivity in K 0 75 . Se 1 75 . Se 2 , Sci. Rep. 2 , 221 (2012).
- [22] V. J. Emery and S.A. Kivelson, Frustrated electronic phase separation and high-temperature superconductors , Physica C 209 , 597 (1993).
- [23] Despina Louca, K. Park, B. Li, J. Neuefeind and JiaqiangYan, The hybrid lattice of K Fe x 2 -y Se 2 : where superconductivity and magnetism coexist. Sci. Rep. 3, 2047 (2013).
- [24] F. Chen, M. Xu, Q. Q. Ge, Y. Zhang, Z. R. Ye, L. X. Yang, J. Jiang, B. P. Xie, R. C. Che, M. Zhang, A. F. Wang, X. H. Chen, D. W. Shen, J. P. Hu, and D. L. Feng, Electronic Identification of the Parental Phases and Mesoscopic Phase Separation of K x Fe 2 -y Se 2 Superconductors , Phys. Rev. X 1 , 021020 (2011).
- [25] D. H. Ryan, W. N. Rowan-Weetaluktuk, J. M. Cadogan, R. Hu, W. E. Straszheim, S. L. Budko, and P. C. Canfield, 57 Fe M ssbauer study of magnetic ordering in ¨ o superconducting K 0 80 . Fe 1 76 . Se 2 00 . single crystals , Phys. Rev. B 83 , 104526 (2011).
- [26] Z. Wang, Y. J. Song, H. L. Shi, Z. W. Wang, Z. Chen, H. F. Tian, G. F. Chen, J. G. Guo, H. X. Yang, and J. Q. Li, Microstructure and ordering of iron vacancies in the superconductor system K y Fe x Se 2 as seen via transmission electron microscopy , Phys. Rev. B 83 , 140505 (2011).
- [27] Wei Li, H. Ding, P. Deng, K. Chang, C. Song, Ke He, L. Wang, X. Ma, J. P. Hu, X. Chen and Q.K. Xue, Phase separation and magnetic order in K-doped iron selenide
superconductor , Nature 8 , 126 (2012).
- [28] Xiaxin Ding, Delong Fang, ZhenyuWang, Huan Yang, Jianzhong Liu, Qiang Deng, Guobin Ma, Chong Meng, Yuhui Hu and Hai-Hu Wen, Influence of microstructure on superconductivity in KxFe2-ySe2 and evidence for a new parent phase K2Fe7Se8 , Nat. Communications 4 , 1897 (2013).
- [29] N. R. Werthamer, E. Helfand and P. C. Hohenberg, Temperature and purity dependence of the superconducting critical field, Hc2.III. Electron spin and spin-orbit effects , Phys. Rev. B 147 , 295 (1966).
- [30] L. Jiao, Y. Kohama, J. L. Zhang, H. D. Wang, B. Maiorov,F. F. Balakirev, Y. Chen, L. N. Wang, T. Shang, M. H. Fang, and H. Q. Yuan, Phys. Rev. B 85 , 064513 (2012).
- [31] E. D. Mun, M. M. Altarawneh, C. H. Mielke, V. S.Zapf, Rongwei Hu, S. L. Bud'ko and P.C. Canfield, Anisotropic H c 2 of K 0 8 . Fe 1 76 . Se 2 determined up to 60 T , Phys. Rev. B 83 , 100514 (2011).
- [32] V. A. Gasparov, A. Audouard, L. Drigo, A. I. Rodigin, C. T. Lin, W. P. Liu, M. Zhang, A. F. Wang, X. H. Chen, H. S. Jeevan, J. Maiwald and P. Gegenwart, Upper critical magnetic field in K 0 83 . Fe 1 83 . Se 2 and Eu 0 5 . K 0 5 . Fe As 2 2 single crystals , Phys. Rev. B 87 , 094508 (2013).
- [33] K. Maki, Effect of Pauli Paramagnetism on Magnetic Properties of High-Field Superconductors , Phys. Rev. 148 , 362 (1966).
- [34] G. Fuchs, S.-L. Drechsler, N. Kozlova, G. Behr, A. K¨ ohler, J. Werner, K. Nenkov, C. Hess, R. Klingeler, J. E. Hamann-Borrero, A. Kondrat, M. Grobosch, A. Narduzzo, M. Knupfer, J. Freudenberger, B. B¨chner, u and L. Schultz, High-Field Pauli-Limiting Behavior and Strongly Enhanced Upper Critical Magnetic Fields near the Transition Temperature of an Arsenic-Deficient LaO 0 9 . F 0 1 . FeAs 1 -δ Superconductor. , Phys. Rev. Lett. 101 , 237003 (2008).
- [35] T. P. Orlando, E. J. McNiff, S. Foner, and M. R. Beasley, Critical fields, Pauli paramagnetic limiting, and material parameters of Nb Sn and V Si 3 3 , Phys. Rev. B 19 , 4545 (1979).
- [36] M. Yi, D. H. Lu, R. Yu, S. C. Riggs, J. H. Chu, B. Lv, Z. K. Liu, M. Lu, Y. T. Cui, M. Hashimoto, S.-K. Mo, Z. Hussain, C. W. Chu, I. R. Fisher, Q. Si and Z.X. Shen, Observation of Temperature-Induced Crossover to an Orbital-Selective Mott Phase in A x Fe 2 -y Se 2 (A = K, Rb) Superconductors , Phys. Rev. Lett. 110 , 067003 (2013).
- [37] M. Strongin, R. S. Thompson, O. F. Kammerer, and J. E. Crow, Destruction of Superconductivity in Disordered Near-Monolayer Films , Phys. Rev. B 1 , 1078 (1970).
- [38] I. S. Beloborodov, A. V. Lopatin, V. M. Vinokur and K. B. Efetov, Granular electronic systems , Rev. Mod. Phys. 79 , 469 (2007).
- [39] K. M. Lang, V. Madhavan, J. E. Hoffman, H. Eisaki, S. Uchida, and J. C. Davis, Imaging the granular structure of high-Tc superconductivity in underdoped Bi Sr CaCu O 2 2 2 8+ δ , Nature 415 , 412 (2002).
- [40] V. F. Gantmakher and V. T. Dolgopolov, Superconductor-insulator quantum phase transition , Physics-Uspekhi 53 , 1 (2010).
- [41] D. P. Shoemaker, D. Y. Chung, H. Claus, M. C. Francisco, S. Avci, A. Llobet, and M. G. Kanatzidis, Phase relations in K x Fe 2 -y Se 2 and the structure of superconducting K x Fe Se 2 2 via high-resolution synchrotron diffraction ,
Phys. Rev. B 86 , 184511 (2012).
- [42] A. M. Finkelstein, Suppression of superconductivity in homogeneously disordered systems , Physica B 197 , 636 (1994).
- [43] I. S. Beloborodov, K. B. Efetov, and A. I. Larkin, Magnetoresistance of granular superconducting metals in a strong magnetic field , Phys. Rev. B 61 , 9145 (2000).
- [44] I. S. Beloborodov and K. B. Efetov, Negative Magnetoresistance of Granular Metals in a Strong Magnetic Field , Phys. Rev. Lett. 82 , 3332 (1999).
- [45] I. S. Beloborodov, K. B. Efetov, A. V. Lopatin, and V. M. Vinokur, Transport Properties of Granular Metals at Low Temperatures , Phys. Rev. Lett. 91 , 246801 (2003).
- [46] Z.-H. Pan, P. Richard, Y. M. Xu, M. Neupane, P. Bishay, A. V. Fedorov, H. Luo, L. Fang, H. H. Wen,
- Z. Wang, and H. Ding, Evolution of Fermi surface and normal-state gap in the chemically substituted cuprates Bi Sr 2 2 -x Bi x CuO 6+ δ , Phys. Rev. B 79 , 092507 (2009).
- [47] Y. Ando, Implication of the Mott-limit violation in highT c cuprates , J. Phys. Chem. Solids 69 , 3195 (2008).
- [48] V. J. Emery, S. A. Kivelson, and O. Zachar, Spin-gap proximity mechanism of high-temperature superconductivity , Phys. Rev. B 56 , 6120 (1997).
- [49] A. S. Alexandrov, V. V. Kabanov, and N. F. Mott, Coherent ab and c transport theory of high-T c cuprates , Phys. Rev. Lett. 77 , 4796 (1996).
- [50] E. Demler, W. Hanker, and S.-C. Zhang, SO (5) theory of antiferromagnetism and superconductivity , Rev. Mod. Phys. 76 , 909 (2004). | 10.1103/PhysRevX.4.031018 | [
"Kefeng Wang",
"Hyejin Ryu",
"Erik Kampert",
"M. Uhlarz",
"J. Warren",
"J. Wosnitza",
"C. Petrovic"
] | 2014-08-01T19:17:09+00:00 | 2014-08-01T19:17:09+00:00 | [
"cond-mat.supr-con"
] | Nonmetallic Low-Temperature Normal State of K0.70Fe1.46Se1.85Te0.15 | The normal-state in-plane resistivity below the zero-field superconducting
transition temperature $T_c$ and the upper critical field Hc2 were measured by
suppressing superconductivity in pulsed magnetic fields for
K0.70Fe1.46Se1.85Te0.15. The normal-state resistivity $\rho_{ab}$ is found to
increase logarithmically with decrasing temperature as
$\frac{T}{T_c}\rightarrow 0$. Similar to granular metals, our results suggest
that a superconductor - insulator transition below zero-field T$_{c}$ may be
induced in high magnetic fields. This is related to the intrinsic real-space
phase-separated states common to all inhomogeneous superconductors. |
1408.0272v2 | ## STOCHASTIC SHORTEST PATHS AND RISK MEASURES
## AXEL PARMENTIER AND FR ´ D ´ RIC MEUNIER E E
Abstract. We consider three shortest path problems in directed graphs with random arc lengths.
For the first and the second problems, a risk measure is involved. While the first problem consists in finding a path minimizing this risk measure, the second one consists in finding a path minimizing a deterministic cost, while satisfying a constraint on the risk measure. We propose algorithms solving these problems for a wide range of risk measures, which includes among several others the CVaR and the probability of being late. Their performances are evaluated through experiments.
One of the key elements in these algorithms is the use of stochastic lower bounds that allow to discard partial solutions. Good stochastic lower bounds are provided by the so-called Stochastic On Time Arrival Problem . This latter problem is the third one studied in this paper and we propose a new and very efficient algorithm solving it.
Complementary discussions on the complexity of the problems are also provided.
## 1. Introduction
- 1.1. Context. Finding shortest paths in networks has numerous applications. There are direct applications, e.g. organizing the routes of a fleet in logistics or finding the best trajectory for an airplane, but there are also less obvious applications, such as the ones arising as subproblems in column generation approaches.
As for other optimization problems, there is currently a need for taking into account uncertainty in shortest path problems, both from theoretical and practical point of views. Taking into account the uncertainties allows a better optimization. For instance, congestion in road networks has a strong influence on travel times, and modeling delay is crucial when on time arrival is required. Several works have already been carried out on this topic - see our literature review in Section 2 - but as far as we know, none deal with the problems expressed in the general form we address in the present paper.
- 1.2. Problems. We consider various shortest path problems with random arc lengths. They all have in common the following elements: a directed graph D = ( V, A ) without loops, two vertices o, d ∈ V that are respectively the origin and the destination, and independent random variables ( X a ) a ∈ A that model random travel times. The objective is always to find the best o d -path. 'Best' can take various meanings and we adopt a versatile approach.
The most natural case consists probably in taking no additional constraints into account. In order to be as general as possible, we study the problem of minimizing a given risk measure of the sum of the X a 's along the path. A risk measure is a mapping from a set of random variables to the set of real numbers. The mean, the variance, or the probability of being above some threshold are examples of risk measures. Another risk measure is the conditional value at risk and is well known in finance but maybe less in operations research. It has many interesting features. Details are given later in the paper.
Given a risk measure ρ , we define the following problem.
Key words and phrases. On time arrival, random travel time, risk measure, resource constraint, shortest path, usual stochastic order.
Stochastic ρ -Shortest Path Problem Instance. A directed graph D = ( V, A ), two vertices o, d ∈ V , independent random variables ( X a ) a ∈ A . Solution. An elementary o d -path P . Measure. The quantity ρ ( ∑ a ∈ P X a ).
When the X a are deterministic and ρ is linear, we obtain the usual (deterministic) shortest path problem.
Another way to take into account stochasticity when modeling an optimization problem consists in introducing probabilistic constraints. Still assuming given a risk measure ρ , we can define the following problem.
Stochastic ρ -Resource Constrained Shortest Path Problem Instance. A directed graph D = ( V, A ), two vertices o, d ∈ V , independent random variables ( X a ) a ∈ A , a nonnegative number ρ 0 , and nonnegative numbers ( c a ) a ∈ A . Solution. An elementary o d -path P such that ρ ( a ∈ P X a ) ≤ ρ 0 . Measure. The quantity c a .
When the X a are deterministic and ρ is linear, we obtain the usual (deterministic) resource constrained shortest path problem.
∑ ∑ a ∈ P
- 1.3. Assumption on the travel times X a . All random variables X a , which model the time needed to traverse an arc a , are assumed to take their values in Z + and to have finite supports. In a computational perspective, it is a natural assumption. When complexity features are discussed, the variable X a is described by its distribution, which is assumed to be encoded as its support and the probability of each value in the support.
- 1.4. Contributions. We propose an algorithm for each of the aforementioned problems when the arc lengths X a satisfy the assumption above and when the risk measure ρ penalizes delay. The mean, the probability of being above some threshold, and the conditional value at risk are examples of such risk measures, but not the variance. 'Penalizing delay' will be mathematically specified with the help of the so-called usual stochastic order, which is an order defined on the set of random variables - see Section 3. However, our approach remains completely natural. The efficiency of our algorithms, which turns out to be high, is proved by an extensive experimental study.
One of the main ideas used in these algorithms consists in attaching to each vertex v of the graph a random variable Z v that is a lower bound of the length of a v d -path. 'Lower bound' is here to be understood in the sense of the usual stochastic order. As usual with lower bounds in shortest paths algorithms, these stochastic lower bounds allow to discard a priori some partial paths and to reduce the space of possible solutions.
It turns out that good lower bounds are solutions of another problem, namely the Stochastic On Time Arrival Problem . There are already efficient algorithms solving this problem, but we propose a new one, which seems to be very efficient.
The complexity of the problems dealt with in the paper are also discussed.
- 1.5. Plan. We start with a literature review (Section 2). We describe in Section 3 the usual stochastic order and the various risk measures covered by our work. The stochastic lower bounds used in our algorithms and their links with the Stochastic On Time Arrival Problem are described in Section 4, as well as the new algorithm we propose for this latter problem. The algorithms for the Stochastic ρ -Shortest Path Problem and the Stochastic ρ -Resource Constrained Shortest Path Problem are described in Sections 5 and 6 respectively, as well as their complexity analysis. The paper ends with an extensive experimental study (Section 7).
## 2. Literature review
- 2.1. Deterministic shortest path. The shortest path problem with deterministic travel times is one of the most studied problems in Operations Research. Efficient polynomial algorithms are known since the end of the fifties. Recently, research has focused on improving algorithm efficiency thanks to pre-processing, and state of the art techniques compute shortest paths on continental road networks in fractions of a microsecond. An exhaustive survey on shortest path algorithms for route planning can be found in (Bast et al. , 2014).
- 2.2. Deterministic constrained and multicriteria shortest path. Deterministic shortest paths with resource constraints have many direct applications in logistics and transportation. Moreover, many operations research problems require to find such shortest paths as a subroutine (e.g. vehicle routing problems). Resource constrained shortest path have therefore been extensively studied, and a detailed literature review is out of the scope of this paper. A review of state of the art techniques can be found in (Irnich & Desaulniers, 2005).
Several problems close to the resource constrained shortest path problem belong to multicriteria optimization (Ehrgott, 2005). Many algorithms have been developed for the bicriteria (Raith & Ehrgott, 2009) and multicriteria shortest path problem (Tarapata, 2007). State of the art solution methods for resource constrained or multicriteria shortest path problems often rely on labeling algorithms with pre-processing (Boland et al. , 2006; Dumitrescu & Boland, 2003).
- 2.3. Stochastic shortest path. Stochastic shortest path problems have been extensively studied since the seminal work of Frank (1969). Models differ by the probability distributions used to model delay on arcs, and by the risk measure optimized.
A first line of papers considers the probability of on time arrival: a path maximizing the probability of on time arrival, or analogously, a path with minimum quantile of given order is searched. Approaches have been developed for both continuous (Chen & Ji, 2005; Frank, 1969; Nikolova, 2010; Nikolova et al. , 2006) and discrete distributions (Mirchandani, 1976). An efficient labeling algorithm has recently been described when arc distributions are normal (Chen et al. , 2013). In a recent preprint, Niknami et al. (2014) proposes independently an idea that bears similarities with ours by using the solution of the Stochastic On Time Arrival Problem to obtain lower bounds for the Stochastic ρ -Shortest Path Problem with ρ ( ) · = P ( · ≥ τ ). Note that our approach covers actually more general situations, even in this special case.
Asecond line of papers defines a shortest path as a path minimizing the expectation of a cost function (Loui, 1983). Dynamic programming can be used when cost functions are affine or exponential (Eiger et al. , 1985). Murthy & Sarkar (1996, 1998) present an efficient labeling algorithm when arc distributions are normal and cost functions are piecewise-linear and concave. Instead of considering the expectation of a cost functions, Nikolova (2010); Nikolova et al. (2006); Sivakumar & Batta (1994) search the path that minimizes a positive linear combination of mean and variance.
Finally, Miller-Hooks & Mahmassani (2003) suggest to use stochastic dominance to compare paths. Algorithms to generate all non-dominated paths are proposed in Miller-Hooks & Mahmassani (2003, 1998); Miller-Hooks (1997); Nie & Wu (2009); Nie et al. (2012).
- 2.4. Stochastic on time arrival. The Stochastic on Time Arrival Problem searches to maximize the probability of arrival before a given thresholds on adaptive paths (Fu, 2001; Fu & Rilett, 1998; Hall, 1986). An adaptive path is an application which, given a vertex reached and the time it took to get to this vertex indicates the next arc to choose to maximize the probability of on time arrival at destination. Fan & Nie (2006) provide an algorithm for continuous distributions, and Nie & Fan (2006) provide a pseudo-polynomial algorithm for discrete distributions. Samaranayake et al. (2012) develop a faster algorithm for discrete distributions, and Sabran et al. (2014) provide pre-processing techniques to improve algorithm speed.
2.5. Shortest path under probability constraint. Finally, the problem of finding a minimum cost path for deterministic arc costs under stochastic resource constraints have been introduced in Kosuch & Lisser (2010), which proposes a solution algorithm based on linear programming is derived.
## 3. Usual stochastic order and risk measures
3.1. Usual stochastic order. For any random variable X , let F X ( ) = t P ( X ≤ t ) be its cumulative distribution. The usual stochastic order ≤ st is a partial order defined on the set of cumulative distributions as follows: F X ≤ st F Y if F X ( ) t ≥ F Y ( ) for all t t . Endowed with the usual stochastic order, the set of cumulative distributions turns out to be a lattice . It means that each collection of cumulative distributions has a unique least upper bound ( join ) and a unique greatest lower bound ( meet ).
We use the same notation ≤ st to extend the usual stochastic order to random variables. Two random variables X and Y are such that X ≤ st Y if F X ≤ st F Y . Note that in the context of random variables, it is a quasiorder : it is reflexive and transitive like a partial order, but not antisymmetric. The usual stochastic order is a classical notion in probability theory and enjoys several properties. Missing details and a more complete introduction may be found in M¨ller & Stoyan (2002). u
One property is particularly useful in our work and is used in the paper without further mention.
Let X 1 , X 2 , Y 1 , and Y 2 be random variables such that X 1 and Y 1 are independent, and X 2 and Y 2 are independent. If X 1 ≤ st X 2 and Y 1 ≤ st Y 2 , then X 1 + Y 1 ≤ st X 2 + Y 2 .
By a slight abuse, we define the meet (resp. the join) of a collection ( X i ) of random variables as a random variable with cumulative distribution equal to the meet (resp. the join) of the F X i 's. We use the classical notation ∧ for the meet and ∨ for the join. We have in particular for any collection ( X i ) of random variables the following equality
$$F _ { \wedge _ { i } X _ { i } } ( t ) = \max _ { i } ( F _ { X _ { i } } ( t ) ) \text{ \ for all } t \in \mathbb { Z } _ { + }.$$
The meet of the empty collection is as usual equal to + ∞ and its join is equal to -∞ .
Since the supports of the random variables are assumed to be integer valued and of finite cardinality, their meet (resp. the join) can easily be computed: for each t in the union of their supports, we simply keep the largest (resp. smallest) value among those taken by the cumulative distributions at t . In the paper, we only use the meet operation.
3.2. Risk measures. A risk measure ρ is said to be consistent with the usual stochastic order if X ≤ st Y implies ρ X ( ) ≤ st ρ Y ( ). The algorithms for the Stochastic ρ -Shortest Path Problem and the Stochastic ρ -Resource Constrained Shortest Path Problem introduced in this paper only require the consistency of the risk measure with the usual stochastic order. It is not a strong assumption since being consistent with the risk measure simply means penalizing delay. Note however that the variance does not satisfy this condition.
For examples, the following risk measures are covered by our work.
- · ρ ( ) = · E [ f ( )] · for any increasing function f .
- · ρ ( ) = · P ( · ≥ τ ) for any τ ∈ Z + .
- · ρ ( ) = · V aR β ( ) for any · β ∈ [0 , 1].
- · ρ ( ) = · CVaR β ( ) for any · β ∈ [0 , 1].
The first risk measure among the ones listed can be used for instance when the cost of being late is modeled. We can cite as an illustration the case of goods delivery, an increasing step function can model the successive penalties of a late delivery.
The second one is typically used to measure the probability of going beyond some deadline. It corresponds to the risk measures of a traveler going to the airport: the only onjective is to arrive before the time τ when boarding gate is closed.
The third and fourth risk measures are widely used in finance.
V aR β ( ) is · the value at risk (with confidence level β ) and is defined by
$$V a R _ { \beta } ( X ) = \min \{ t \in \mathbb { Z } _ { + } \, \colon \, \mathbb { P } ( X \leq t ) \geq \beta \}.$$
Given a confidence level β , a random variable X is smaller than its value at risk V aR β ( X ) with probability at least β , and V aR β ( X ) is the threshold value for which this inequality holds. For instance, if a bus company advertises the policy 'our buses are on time for β % of the trips', then the shortest path with respect to V AR β gives the smallest travel time for which this policy holds.
CVaR β ( ) is the · conditional value at risk (with confidence level β ) and is defined by
$$C V a R _ { \beta } ( X ) = \mathbb { E } \left [ X | X \geq V a R _ { \beta } ( X ) \right ].$$
Given a confidence level β , the conditional value at risk of a random variable X can be described in a somehow informal way as its average value in the 1 -β worst cases. In the context of shortest paths with random travel times, a β equal to 0 corresponds to a shortest path problem where expectation is minimized. A β equal to 1 corresponds to a robust approach: the optimal solution is the solution with minimal worst case behavior. As a consequence, CVaR β enables to obtain a tradeoff between expectation and robustness. This cost function suits well most traveler behavior: a small path with small CVaR gives a small average traveling time with a low risk of very large travel time.
## 4. Stochastic lower bounds and On Time Arrival
4.1. Stochastic lower bounds. Efficient shortest path algorithms are often based on lower bounds that allow to discard partial paths and to reduce the space of solutions that is explored. We do not depart from it and introduce stochastic lower bounds, which, instead of being single values, are random variables.
We consider the following equation, where δ + ( v ) denotes the set of arcs of the form ( v, u ).
$$\begin{cases} \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdots \quad \cdot \end{cases}, \\ ( 1 ) \quad \begin{cases} Z _ { d } = 0, \\ Z _ { v } \stackrel { ( d ) } { = } \bigwedge _ { ( v, u ) \in \delta ^ { + } ( v ) } ( X _ { ( v, u ) } + Z _ { u } ) \quad \text{for all } v \in V \, \langle d \}. \end{cases} \\ \text{Proposition } 3 \text{ below shows that this equation has always a solution which can be calcu} \\ \text{polynomial time. Before stating and proving this proposition, we explain the interest}$$
Proposition 3 below shows that this equation has always a solution which can be calculated in polynomial time. Before stating and proving this proposition, we explain the interest of this equation for stochastic shortest path problems.
Proposition 1. Let ( Z v ) v ∈ V be a solution of Equation (1) . For any vertex v ∈ V and any v d -path P , the inequality Z v ≤ st ∑ a ∈ P X a holds. Proof. An induction on the number of arcs in P allows to conclude. /square
4.2. On Time Arrival. The Stochastic On Time Arrival Problem aims at finding the path in D that maximizes the probability of reaching d before some time limit τ ∈ Z + . If the solution has to be computed a priori, we are exactly in the case of the Stochastic ρ -Shortest Path Problem , with ρ X ( ) = P ( X ≥ τ ), which is a risk measure covered by the present work. In the Stochastic On Time Arrival Problem , the path can be built during the trip. Concretely, an optimal policy has to computed. A policy π for this problem associates to each pair ( v, t ) ∈ V × Z + an arc a ∈ δ + ( v ). Such a policy π is optimal if it maximizes the probability of reaching d before τ . If there is in D a circuit C with P ( X a = 0) = 1 for all a ∈ C , the problem may not be well-defined. We suppose therefore that there are no such circuits. Under this assumption, and using elementary properties of Markov chains, we get that solving Stochastic On Time Arrival Problem is
equivalent to finding a solution of the following equation for all t ∈ Z + (see also Fan et al. (2005) for a first formulation of this kind in a continuous setting):
$$( 2 ) & \quad \begin{cases} \ F _ { d } ( t ) = 1 \\ \ F _ { v } ( t ) = \max _ { ( v, u ) \in \delta ^ { + } ( v ) } \sum _ { k = 0 } ^ { t } \mathbb { P } ( X _ { ( v, u ) } = k ) F _ { u } ( t - k ) \quad \text{for all } v \in V \, \langle \, \{ d \}. \end{cases} \\ \text{The quantity } F _ { v } ( t ) \, \text{is then the probability to reach } d \, \text{from } v \, \text{ in less than time } t \, \text{ under an optima} \\ \text{policy. The decision to be taken at time } 0 \, \text{ on vertex } o \, \text{ is determined by the } ( o, u ) \in A \, \text{ for which}$$
The quantity F v ( ) is then the probability to reach t d from v in less than time t under an optimal policy. The decision to be taken at time 0 on vertex o is determined by the ( o, u ) ∈ A for which the maximum is attained when v = o and t = τ in Equation (2).
Given a realization of the random variables X a , the path obtained following the policy of Equation (2) is not necessary elementary. Samaranayake et al. (2012) provides a simple example of network in which an event realization leads to a non-simple path. We therefore adopt the following hypothesis: when a path crosses several times the same arc a , then we consider that delays encountered on the arc correspond to independent realizations of the random variable X a .
Without loss of generality, we can assume that there exists in D at least one v d -path (otherwise, the vertex v can simply be removed from D ). We have lim t → ∞ + F v ( ) = 1 for all t v ∈ V and F v ( ) is · non-decreasing. Thus, F v ( ) can be considered as a cumulative distribution, and we can introduce · a random variable U v whose cumulative distribution function is F v ( ). ·
This construction shows that any algorithm solving Equation (1) solves the Stochastic On Time Arrival Problem as well.
Proposition 2. The U v 's defined above satisfy Equation (1) and conversely, any solution of Equation (1) has cumulative distributions that satisfy Equation (2) .
Proof. The equivalence between (1) and (2) follows directly from the equality
$$F _ { X \wedge Y } ( t ) = \max ( F _ { X } ( t ), F _ { Y } ( t ) ).$$
Remark 1 . If all X a 's are deterministic, the Stochastic On Time Arrival Problem coincides with the usual deterministic shortest problem, for which by the way the solution computed a priori coincides with the solution computed during the trip.
- 4.3. An algorithm. We describe now a new polynomial algorithm computing a solution of Equation (1), and thus solving also the Stochastic On Time Arrival Problem .
/negationslash
A random variable Z ′ v and an integer t ′ v are attached to each vertex v and updated during the algorithm. Initially, Z ′ v = + ∞ and t ′ v = + ∞ for each vertex v = d , while Z ′ d = 0 and t ′ d = 0. During the algorithm, a queue L of vertices 'to be expanded' is maintained. Initially, the queue L contains only d .
The algorithm ends when L is empty. While L is not empty, the following operations are repeated:
- · Extract from L the vertex u with minimum t ′ u . In case there are several such vertices, choose one with maximal value of F Z ′ u ( t ′ u ).
- · For each arc ( v, u ) in δ -( u ), expand u along ( v, u ): if Z ′ v /notlessequal st X ( v,u ) + Z ′ u , then
- · Set t ′ u = + ∞ .
- -Update t ′ v = min( t ′ v , min { t : F Z ′ v ( ) t < F X ( v,u ) + Z ′ u ( ) t } ).
- -Update Z ′ v to Z ′ v ∧ ( X ( v,u ) + Z ′ u ).
- -Add v to L (if it is not already present).
Proposition 3. This algorithm terminates in less than T V | | iterations, where T is the maximum length an elementary o d -path can take (over all possible paths and over all possible events). Setting Z v to be the last value of Z ′ v for each v ∈ V provides then a solution of Equation (1) .
Algorithms in Nie & Fan (2006) and in Samaranayake et al. (2012) make a stronger assumption on arc distributions: they suppose the existence of a δ > 0 such that X a ≥ δ for all arcs a . Moreover, the number of iterations of the algorithm in Nie & Fan (2006) is always T V | | and in Samaranayake et al. (2012) it is a Ω( T V | | ), while in our approach, it corresponds to a worst case behavior: In general, we expect our algorithm to terminate after γ V | | iterations for a small γ . The worst γ encountered in the numerical experiments was 3 3, see Section 7 for more details. .
Proof of Proposition 3. At any time during the algorithm, let Z ′′ v have the distribution of Z ′ v the last time v was expanded and let t ′′ v be the value of t ′ v right before being set to + ∞ because of the expansion. If v has never been expanded, let Z ′′ v = + ∞ and t ′′ v = -1. The update rule of t ′ v ensures thus that
$$F _ { Z ^ { \prime \prime } _ { v } } ( t ) = F _ { Z ^ { \prime } _ { v } } ( t ) \text{ \ for all } t \leq t ^ { \prime } _ { v } - 1.$$
Thus, if the algorithm terminates, we have t ′ v = + ∞ and Z ′′ v = Z ′ v for each vertex v . As Z ′ v is only updated when one of its outneighbor u is expanded, we have
$$Z _ { v } ^ { \prime } \stackrel { ( d ) } { = } \bigwedge _ { ( v, u ) \in \delta ^ { + } ( v ) } \left ( X _ { ( v, u ) } + Z _ { u } ^ { \prime \prime } \right ). \\ \intertext { Thoroughness $Z^{\prime}$ is the solution to $E$ation $I$}$$
Therefore Z ′ v is then a solution to Equation (1).
It remains to prove that the algorithm terminates in a finite number of iterations.
We prove that min v ∈ V t ′ v does not decrease along the algorithm, and that if this quantity remains constant for consecutive iterations, then max v ∈ V F Z ′ v ( t ′ v ) does not increase.
Equation (3) implies that Z ′ v ≤ st X ( v,u ) + Z ′′ u for any ( v, u ) ∈ δ + ( v ). Since F Z ′ u and F Z ′′ u coincide up to t ′ u -1, we have
$$F _ { Z ^ { \prime } _ { v } } ( t ) \geq F _ { X _ { ( v, u ) } + Z ^ { \prime } _ { u } } ( t ) \ \text{ for all } t \leq t ^ { \prime } _ { u } - 1.$$
Moreover, just before being updated because of the expansion of a vertex u , the integer t ′ v is equal to t ′′ u because of the way the vertex u has been selected in L . This remark and Equation (3) implies together that after an expansion along an arc ( v, u ), we have t ′ v ≥ t ′′ u .
Consider the algorithm right after a vertex u has been expanded.
Let v be such that ( v, u ) ∈ δ + ( v ). If t ′ v has just been updated to t ′′ u , then
$$F _ { Z ^ { \prime } _ { v } } ( t ^ { \prime } _ { v } ) = \sum _ { k = 0 } ^ { t ^ { \prime \prime } _ { u } } \mathbb { P } ( X _ { ( v, u ) } = k ) F _ { Z ^ { \prime } _ { u } } ( t ^ { \prime \prime } _ { u } - k ), \\ \intertext { n d e c r e a s i n \sigma \text{ man. we have } F _ { \pi ^ { \prime } } ( t ^ { \prime } ) < F _ { \pi ^ { \prime } } ( t ^ { \prime \prime } ). \text{ if } t ^ { \prime }.$$
and since F Z ′ u is a nondecreasing map, we have F Z ′ v ( t ′ v ) ≤ F Z ′ u ( t ′′ u ). If t ′ v has not been updated while being already such that t ′ v = t ′′ u , then F Z ′ v ( t ′ v ) ≤ F Z ′ u ( t ′′ u ) because of the selection rule of u . Therefore, the next vertex w to be expanded will either be such that t ′ w > t ′′ u , or such that t ′ w = t ′′ u and F Z ′ w ( t ′ w ) ≤ F Z ′ u ( t ′′ u ).
To finish the proof of termination, we show now that t ′ v ≥ t ′′ v +1 at any time during the algorithm. If t ′ v = + ∞ , the inequality is clearly satisfied. Otherwise, let u be the last vertex that has been expanded and whose expansion has lead to an update of t ′ v . We consider the algorithm right after this expansion. t ′′ u is thus the value of t ′ u right before having being set to + ∞ , and t ′ v has been updated by the expansion.
Note that t ′′ u ≥ t ′′ v since min w V ∈ t ′′ w does not decrease along the algorithm (it does not decrease because each t ′′ w is the value of min v ′ ∈ V t ′ v ′ at some previous iteration, and this latter quantity is non decreasing, as already noted). We already know that t ′ v ≥ t ′′ u . Thus, if t ′ v > t ′′ u , we have t ′ v ≥ t ′′ v +1 as required. It remains to check whether it is possible to have t ′ v = t ′′ u = t ′′ v simultaneously. In such a case, we would necessarily have F Z ′ v ( t ′′ v ) ≥ F Z ′ u ( t ′′ u ) since max w V ∈ F Z ′ w ( t ′ w ) does not increase when min w V ∈ t ′ w remains constant. Since F Z ′ u is a nondecreasing map, we would have F Z ′ u ( t ′′ u ) ≥ F X ( v,u ) + Z ′ u ( t ′′ u ) and thus
$$F _ { Z _ { v } ^ { \prime } } ( t ) \geq F _ { X _ { ( v, u ) } + Z _ { u } ^ { \prime } } ( t ) \text{ \ for all } t \leq t _ { u } ^ { \prime \prime }$$
with the help of Equation (4). Thus, t ′ v should have been updated to a value greater than t ′′ u . Hence, in any case, we have t ′ v ≥ t ′′ v +1.
Now, denote by T v the maximum length an elementary v d -path can take. We claim that once Z ′ v has been updated for the first time, we have F Z ′ v ( T v ) = 1 all along the algorithm. It can be proved by a direct induction on the number of arcs an elementary path may have with the help of Equation (3). Moreover, since the inequality t ′ v ≥ t ′′ v + 1 holds along the algorithm, we may have F Z ′ v ( ) t < F X ( v,u ) + Z ′ u ( ) t only for t > t ′′ v . Therefore, after at most T v iterations, the test ' Z ′ v /notlessequal st X ( v,u ) + Z ′ u ' is always false. It implies that after at most max v ∈ V T v iterations, the algorithm terminates. /square
Remark 2 . For each elementary path P , define T P to be the maximum length it can takes over all possible realizations. Denote by P vd the set of all elementary v d -paths. Proposition 3 is actually true for T = min P ∈P od T P . Indeed, in the proof above, we show that the algorithms terminates after a finite number of iterations. Denote T v = min P ∈P vd T P . Proposition 1 ensures that F Z v ( T v ) = 1. Thus, as soon as t ′′ v = T v , the test ' Z ′ v /notlessequal st X ( v,u ) + Z ′ u ' is always false. It implies that after at most max v ∈ V T v iterations, the algorithm terminates.
Remark 3 . When all X a are deterministic, this algorithm coincides with Dijkstra's algorithm.
## 5. Stochastic ρ -Shortest Path Problem
5.1. Complexity. When ρ is linear over independent random variables and dealt with as an oracle, the Stochastic ρ -Shortest Path Problem can clearly be solved in polynomial time by any Dijsktra-like algorithm. However, we were not able to decide whether the problem remains polynomial in the general case, still dealing with ρ as an oracle. Nevertheless, we are able to prove the following theorem.
Theorem 4. There is no polynomial algorithms with a complexity function independent of ρ solving the Stochastic ρ -Shortest Path Problem , unless P =NP .
Proof. The proof consists in describing a polynomial reduction of the deterministic resource constraint path problem to some Stochastic ρ -Shortest Path Problem , with an adequate risk measure ρ consistent with the usual stochastic order. The deterministic resource constrained path problem consists in a directed graph D = ( V, A ), two vertices o, d ∈ V , nonnegative integers ( c a ) a ∈ A (the costs), nonnegative integers ( r a ) a ∈ A (the resources), and a nonnegative integer R (the capacity). It aims at finding a path P with ∑ a ∈ P r a ≤ R and with minimal cost ∑ a ∈ P c a . It is an NP-hard problem, see Handler & Zang (1980).
To that purpose, we define two risk measures ρ min and ρ max .
$$\rho _ { \min } ( X ) \ = \ \min \{ t \in \mathbb { Z } _ { + } \, \colon \, F _ { X } ( t ) > 0 ) \}$$
$$\rho _ { \max } ( X ) \ = \ \begin{cases} \ 0 & \text{if $\max\{t\in\mathbb{Z}_{+}\, \colon\,\mathbb{P}(X=t)>0\} \leq M$} \\ 1 & \text{otherwise,} \end{cases}$$
where M = R (1 + max a ∈ A c a ). We define
$$\overline { \rho } = \rho _ { \min } + \left ( \sum _ { a \in A } c _ { a } \right ) \rho _ { \max }. \\ \text{tent with the usual stochastic order}$$
The risk measure ρ min is consistent with the usual stochastic order because F X ( ) t ≥ F Y ( ) for all t t implies min { t ∈ Z + : F X ( ) t > 0) } ≤ min { t ∈ Z + : F Y ( ) t > 0) } . As well as ρ min , the risk measure ρ max is consistent with the usual stochastic order because max { t ∈ Z + : P ( X = ) t > 0 } = min { t ∈ Z + : F X ( ) = 1 t } and F X ( ) t ≥ F Y ( ) t for all t implies min { t ∈ Z + : F X ( ) = 1) t } ≤ min { t ∈ Z + : F Y ( ) = 1) t } . Thus ρ is also consistent with the usual stochastic order.
We describe now the reduction to the Stochastic ρ -Shortest Path Problem . Given an arc a , we define X a as follows:
$$\mathbb { P } ( X _ { a } = c _ { a } ) = \frac { 1 } { 2 } \quad \text{and} \quad \mathbb { P } \left ( X _ { a } = r _ { a } ( 1 + \max _ { b \in A } c _ { b } ) \right ) = \frac { 1 } { 2 }. \\ \text{v take two values}$$
X a can only take two values.
The deterministic resource constrained shortest path problem has a feasible solution if and only if the Stochastic ρ -Shortest Path Problem has a solution of cost less than ∑ a ∈ A c a . In this case, the optimal solutions of both problems coincide, as we show now.
∑ Note the ρ used in the proof being fixed, the Stochastic ρ -Shortest Path Problem becomes polynomially solvable. The proof above works precisely because the complexity function must be independent of the risk measure.
We assume that the deterministic resource constrained shortest path problem has a feasible solution. Let P be an optimal solution of the Stochastic ρ -Shortest Path Problem . We have then ρ ( ∑ a ∈ P X a ) = ∑ a ∈ P c a and ∑ a ∈ P r a ≤ R , which implies that the deterministic resource constraint shortest path problem has a feasible solution of cost ∑ a ∈ P c a . Conversely, an optimal solution P ′ of the deterministic resource constrained shortest path problem provides a feasible solution of the Stochastic ρ -Shortest Path Problem of cost a ∈ P ′ c a . /square
5.2. Algorithm. For any path P , we denote by X P the random variable ∑ a ∈ P X a . Sometimes in the proofs or in the computations, the path P can be non-simple. In this case, independent copies of the X a 's appearing several times in the sum are used. We assume given for each vertex v a random variable Z LB v such that Z LB v ≤ st X P for all v d -paths P . Such a collection of random variables ( Z LB v ) v ∈ V , which play the role of stochastic lower bounds, can be computed for instance by the techniques presented in Section 4.
These stochastic lower bounds are used to discard some partial paths that are not subpaths of an optimal path. It resembles classical techniques used by shortest path algorithms in the deterministic setting, see for instance Bast et al. (2014) for state of the art techniques, in which the lower bound is a real number, while in our setting the lower bound is a random variable.
We present the algorithm, which is a labeling one. A label λ is a pair ( v λ , Y λ ), where v λ is a vertex and Y λ is a random variable. Labels are stored in a queue L . Initially, L contains a unique label λ o = ( o, 0). If there is no o d -paths, the algorithm returns nothing. Otherwise, an elementary o d -path P 0 is computed and ρ UB od is set to ρ X ( P 0 ). The algorithm ends when L is empty. While L is not empty, the following operations are repeated:
- · Extract a label λ of L .
- · If v λ = d and ρ Y ( λ ) < ρ UB od : update ρ UB od to ρ Y ( λ ).
$$Y _ { \lambda } + X _ { ( v _ { \lambda }, v ) }.$$
- · Otherwise: If ρ Y ( λ + Z LB v λ ) < ρ UB od , expand label λ : for each arc ( v λ , v ) in δ + ( v λ ), add a new label ( v, Y ) to L , where Y ( d ) = Y + X .
/negationslash
Theorem 5. Suppose that P ( X a = 0) > 0 for all a ∈ A and that there exists at least one solution. If ρ is consistent with the usual stochastic order, then the algorithm described above terminates after a finite number of iterations and at the end, ρ UB od is the optimal value of a solution of the Stochastic ρ -Shortest Path Problem .
Before giving the proof, we give the two ideas the algorithm relies on. Note that a label λ corresponds actually to some o v -λ path P λ that the algorithm tries to expand in an optimal path (more details are given in the proof below).
The first idea is the following. At any time during the algorithm, ρ UB od is an upper bound on the optimal cost. Suppose that ρ UB od is not tight. If P is an o v -subpath of an optimal solution, then
$$\rho ( X _ { P } + Z _ { v } ^ { L B } ) < \rho _ { o d } ^ { U B }.$$
In an enumeration of all the o d -paths to find an optimal path, any partial path P that does not satisfy Equation (5) can thus be discarded.
The second idea is that, when an o d -path P is such that ρ X ( P ) < ρ UB od , then ρ UB od can be updated to ρ X ( P ). Such an update reduces the number of paths satisfying Equation (5).
As it is described, the algorithm computes only the optimal cost of a solution. The algorithm can easily be adapted to find the optimal path itself, simply by maintaining, for each label λ the label whose expansion has lead to its creation.
Remark 4 . It is easy to deal with the presence of variables X a with X a = 0. If X ( u,v ) = 0, remove ( u, v ) and add arcs ( w,v ) for w ∈ N -( u ) with X ( w,v ) ( d ) = X ( w,u ) as well as arcs ( u, w ) for w ∈ N + ( v ) with X ( u,w ) ( d ) = X ( v,w ) .
Proof of Theorem 5. We associate to each label λ an o v -λ path P λ such that Y λ = X P λ : If P µ is associated to the label µ and if λ is obtained by expanding µ , we define P λ as the path P µ +( v µ , v λ ). It is easy to see that there is at most one label associated to a given path. This remark will be useful in the proof.
Initially, ρ UB od is equal to ρ X ( P 0 ), where P 0 is the elementary o d -path computed at the beginning of the algorithm. Besides, ρ UB od can only decrease during the algorithm. In addition, we have ρ X ( P λ ) ≤ ρ Y ( λ + Z v λ ) by consistency of ρ with the usual stochastic order, and because X P λ ( d ) = Y λ ≤ st Y λ + Z LB v λ . Thus, as only labels such that ρ Y ( λ + Z LB v λ ) < ρ UB od are expanded, and as, given a path Q , there is at most one label such that P λ = Q , the number of labels that are expanded during the algorithm is upper-bounded by the number of paths P such that ρ X ( P ) < ρ X ( P 0 ).
To prove that the number of paths P satisfying ρ X ( P ) < ρ X ( P 0 ) is finite, we need the following claim.
/negationslash
Let X and Y be random variables satisfying the assumption of Section 1.3 and denote by S k be the sum of k independent copies of X . Suppose that P ( X = 0) > 0 . Then there exists an integer n such that Y ≤ st S n .
This claim is proved as follows. Given an integer t , the event S k ≤ t requires that at least k -t copies of X are equal to 0. Using this idea and then the Stirling formula, we obtain F S k ( ) t ≤ ( ) k t P X ( = 0) k -t = O k P X ( t ( = 0) k -t ). For any integer t , we have thus lim k → ∞ + F S k ( ) t = 0. It implies that for any t ≤ t Y max , where t Y max is the maximum value Y can take, there is a k t such that F S k ( ) t < F Y ( ) when t k ≥ k t . The quantity t Y max is finite, because of the finiteness of the support of Y . Defining n = max 0 ≤ ≤ t t Y max k t proves the claim.
As a consequence of the claim, there is only a finite number of paths P such that X P 0 /notlessequal st X P . As X P 0 ≤ st X P implies ρ X ( P 0 ) ≤ ρ X ( P ) by consistency of ρ with the usual stochastic order, there is only a finite number of paths P such that ρ X ( P ) < ρ X ( P 0 ). Hence, only a finite number of labels are expanded and the algorithm terminates after a finite number of iterations.
We prove now that the algorithm returns the correct value. Note that ρ UB od is at any time the cost of some o d -path. Let P ∗ be the o d -path that has updated ρ UB od for the last time. Note that P ∗ is not necessarily elementary.
/negationslash
Suppose that there is at least one o d -path, otherwise the algorithm would return + ∞ . Denote by P any elementary o d -path. There are subpaths Q of P starting at o and for which there exists a label λ such that Q = P λ . The path reduced to the vertex o is one of them. Let P 1 be such a subpath with the largest number of arcs, µ be such that P 1 = P µ , and v be the destination of P 1 . We also denote by P 2 the v d -subpath of P . Since P 1 has been chosen with a largest number of arcs, the label µ has not been expanded. If v = d , we have ρ X ( P ) larger than or equal to the last value of ρ UB od . Otherwise, v = d and we had ρ Y ( µ + Z LB v ) ≥ ρ UB od when µ has been extracted. Since ρ UB od is non-increasing during the algorithm and since we have X P 2 ≥ st Z LB v , the cost ρ X ( P ) is necessarily larger than or equal to the last value of ρ UB od , which is larger or equal to the cost of some elementary o d -path: P ∗ itself if it is elementary, or the elementary subpath obtained by removing the cycles otherwise. Thus the last value of ρ UB od , which is equal to ρ P ( ∗ ), is the optimal cost. /square
## 5.3. Speeding up the algorithm.
- 5.3.1. Keys. We can speed up the algorithm by making L a priority queue and using a key equal to ρ Y ( λ + Z LB v λ ) for the element ( v λ , Y λ ). The idea behind this choice is that we give priority to the label that seems to be the most promising, as ρ Y ( λ + Z LB v λ ) is a lower bound on the cost of any o d -having P λ as subpath. The efficiency of the methods depends on the quality of the stochastic lower bound Z LB v λ . In such a way, we may discard more partial paths, since the upper bound ρ UB od will likely be smaller than without the use of such keys.
5.3.2. Upper bounds. Suppose that, in addition to the lower bounds Z LB v , we maintain for each vertex v a list /lscript v = ( X P 1 , X P 2 , . . . , X P k ), where the P i are elementary v d -paths. Then, these paths can be used to update the minimum known cost during the algorithm. Let λ be a label. The path P λ is an o v -λ path. As a consequence, for any distribution X P i ∈ /lscript v λ , path P λ + P i is an o d -path. Thus, if ρ UB od > ρ X ( λ + X P i ), we have identified a path whose cost is smaller than the current upper bound on the cost of the optimal path, and ρ UB od can be updated to ρ X ( λ + X P i ). Improving the upper bound during the algorithm enables to reduce the number of labels considered, and thus speeds up the algorithm.
The lists /lscript v can be built heuristically during some preprocessing. In our experiments, we built them during the computation of the lower bounds Z LB v with the Stochastic On Time Arrival Problem algorithm.
## 6. Stochastic ρ -Resource Constrained Shortest Path Problem
- 6.1. Complexity. The Stochastic ρ -Resource Constrained Shortest Path Problem is NP-hard, since the deterministic case is a special case and is already NP-hard.
- 6.2. Algorithm. As in Section 5.2, we denote by X P the random variable ∑ a ∈ P X a , where independent copies of X a 's appearing several time in the sum are used, and we assume given for each vertex v a random variable Z LB v such that Z LB v ≤ st X P for each v d -path P . We assume moreover given for each vertex v the optimal cost π v of an unconstrained v d -path. The costs ( π v ) v ∈ V can
be efficiently computed using Dijkstra's algorithm.
The labeling algorithm for the Stochastic ρ -Resource Constrained Shortest Path Problem goes as follows. Alabel λ is a triple ( v λ , Y λ , c λ ) where v λ is a vertex, Y λ is a random variable, and c λ ∈ R is a cost. Labels are stored in a queue L . Initially, L contains a unique label λ o = ( o, 0 0), , and c UB od = 1+ ∑ a ∈ A c a . The algorithm ends when L is empty. While L is not empty, the following operations are repeated:
- · Extract a label λ of L .
- · If v λ = d and if the inequalities ρ Y ( λ ) ≤ ρ 0 and c λ < c UB od hold: update c UB od to c λ .
- · Otherwise: If ρ Y ( λ + Z LB v ) ≤ ρ 0 and c λ + π v λ < c UB od , expand label λ : for each arc ( v λ , v ) in δ + ( v λ ), add a new label ( v, Y λ + X ( v λ ,v ) , c λ + c ( v λ ,v ) ) to L .
In the following theorem, we use the notation 'sup ρ > ρ 0 ', which means that there exists a random variable Y such that ρ Y ( ) > ρ 0 .
/negationslash
Theorem 6. Suppose that for each arc a we have c a = 0 or P ( X a = 0) > 0 and that sup ρ > ρ 0 . If ρ is consistent with the usual stochastic order, then the algorithm described above terminates after a finite number of iterations and at the end, either c UB od = 1 + ∑ a ∈ A c a and the Stochastic ρ -Resource Constrained Shortest Path Problem admits no feasible solutions, or c UB od is finite and equal to the optimal value of a solution of the Stochastic ρ -Resource Constrained Shortest Path Problem .
/negationslash
In the case when sup ρ ≤ ρ 0 , any path is feasible, and the optimal solution of the Stochastic ρ -Resource Constrained Shortest Path Problem is the unconstrained shortest path, which can be computed with Dijkstra's algorithm.
As noted in Section 5.2, this algorithm can easily be adapted to return a shortest path itself, and to deal with arcs a such that c a = 0 and X a = 0 by the same kind as the one presented in Remark 4.
We do not give the proof of Theorem 6 since it is very similar to the one of Theorem 5. We nevertheless explain three ideas on which the algorithm relies. They are very similar to the ones used in Section 5.2 for the Stochastic ρ -Resource Constrained Shortest Path Problem . Again, a label λ corresponds to some o v -λ path P λ that the algorithm tries to expand in an optimal path
The first idea is the following. If an o v -path P is a subpath of a feasible o d -path, then
ρ ( X P + Z LB v ) ≤ ρ . 0 This inequality provides thus a condition that allows to discard partial path that cannot be completed into feasible paths.
The second idea is the following. Assume that c UB od is an upper bound on the optimal cost of the Stochastic ρ -Resource Constrained Shortest Path Problem . Suppose that c UB od is not tight. If P is an o v -path subpath of an optimal solution, then
$$c _ { P } + \pi _ { v } < c _ { o d } ^ { U B }.$$
It can be used to discard partial paths that cannot be completed into optimal paths.
The third idea is the following. Each time a feasible o d -path P satisfying c P < c UB od is encountered, we have met a better feasible path and c UB od can be updated to c P .
6.3. Speeding up the algorithm. The same techniques as the ones described in Section 5 can be used to speed up the algorithm.
6.3.1. Keys. The cost c λ + π v λ , can be used as a key for the element v λ . It represents a lower bound on the best cost that could be obtained with a path extending P λ .
6.3.2. Upper bounds on the complete path. Compared to the Stochastic ρ -Shortest Path Problem of Section 5, the only difference in the context of the Stochastic ρ -Resource Constrained Shortest Path Problem is that a list of pairs ( X P i , c P i ) must be stored for each vertex v instead of a list of sole random variables X P i .
## 7. Experimental results
The algorithms have been coded in C++ . The priority queue was implemented using the datastructures map and multimap of the C++ standard library. Experiments were carried out on a MacBook Pro with a 2,5 GHz Intel Core i5 processor and 4 Go RAM. Fast Convolution products were performed using the BSD licensed Fast Fourier Transform library KissFFT (Bogerding, 2013).
7.1. Instances. The algorithms were tested on square grid networks of various sizes and distribution types. The origin is the upper left corner of the grid and the destination is the lower right corner of the grid. A rough description of these instances is available on Table 1. The name of each instance indicates the width of the grid, and the distribution. Three types of distributions have been considered: randomly generated generic distributions, lognormal distributions with randomly generated parameters, and gamma distributions with different sizes.
Instances can be found on the following webpage:
## http://cermics.enpc.fr/ ∼ parmenta/shortest path/
The remaining of this subsection is devoted to additional information regarding the way the instances were generated.
7.1.1. Distributions of the X a 's. We describe the way the distributions of arc travel times X a are built. For all three distributions, the minimum value X a can take is uniformly drawn at random between 0 and 50. We denote this minimum by t 0 .
The generic distributions are then built as follows. The size of the support is then drawn uniformly at random between 1 and 2 t 0 . The quantity P ( X a = t 0 + ) is set to t r t ∑ t r t , where r t is generated by selecting at random some prefixed intervals, in which we draw uniformly at random the value of r t . The prefixed intervals are chosen in a way that enforces strong variations in the value of the variances among the arcs.
Regarding the lognormal (resp. gamma) distributions, a maximum value for the mean M is first selected. For the lognormal distributions, M is set to 2 t 0 , except for instance g100Ll, for which it is set to 4 t 0 . For the gamma distributions, we set M = 10. Then, a mean µ is uniformly generated between 1 and M , and a standard deviation σ 2 is uniformly generated in [ M -µ, 2 M -µ ]. The quantity P ( X a = t 0 + ) is set to t r t ∑ t r t , where r t is the density of a lognormal (resp. gamma) distributions of mean µ and variance σ 2 at time t . When r t becomes smaller than some threshold ε , it is set to 0. It ensures a finite-size support.
Since we are expecting the complexity of the operations on distributions to be correlated to the size of paths distributions, we provide in Table 1 the size /lscript of the Stochastic On Time Arrival Problem solution distribution support at the origin.
7.1.2. Risk measures. We made the experiments with two risk measures: P ( · ≥ τ ) and CVaR β .
For the first one, we set τ to be equal to min { t : F Z o ( ) t ≥ p } for some parameter p , where Z o is the solution of the Stochastic On Time Arrival Problem at the origin. For the Stochastic ρ -Shortest Path Problem , p was chosen in { 0 5 . , 0 8 . , 0 95 . } . For the Stochastic ρ -Resource Constrained Shortest Path Problem , p was set to 0 95. .
We did so because if τ is too small, then P ( X P ≥ τ ) = 1 for each path P from origin to destination, which implies that all the solutions of the Stochastic ρ -Shortest Path Problem have the same solutions, and that the Stochastic ρ -Resource Constrained Shortest Path
Table 1. Grid instances description.
| Instance | | V | | | A | | /lscript | Dist. type |
|------------|---------|---------|------------|------------------|
| g10R | 100 | 360 | 191 | Generic |
| g40R | 1600 | 6240 | 953 | Generic |
| g100R | 10000 | 39600 | 2449 | Generic |
| g10Ls | 100 | 360 | 138 | Lognormal |
| g40Ls | 1600 | 6240 | 389 | Lognormal |
| g100Ls | 10000 | 39600 | 797 | Lognormal |
| g100Ll | 10000 | 39600 | 2091 | Lognormal - long |
| g10G | 100 | 360 | 157 | Gamma |
| g40G | 1600 | 6240 | 538 | Gamma |
| g100G | 10000 | 39600 | 1301 | Gamma |
Problem is infeasible. On the contrary, when τ is too large, all elementary origin-destination paths have the same risk measure 0, which implies that they are all optimal solutions of the Stochastic ρ -Shortest Path Problem , and that the Stochastic ρ -Resource Constrained Shortest Path Problem is practically an unconstrained deterministic shortest path problem.
For the second risk measure, we made the experiments for the Stochastic ρ -Shortest Path Problem with β ∈ { 0 01 0 05 0 25 . , . , . } . The values 0 01 and 0 05 are usual values of the parameter. . . While the value 0 25 is higher than what is usually used for the . CVaR , we think that in the context of the Stochastic ρ -Shortest Path Problem , such a value corresponds to the expected travel time of the worst day in a week, which sounds reasonable when a commuter searches an optimal path. For the Stochastic ρ -Resource Constrained Shortest Path Problem , β was set to 0 05. .
7.1.3. Additional parameters in the case of the Stochastic ρ -Resource Constrained Shortest Path Problem . The Stochastic ρ -Resource Constrained Shortest Path Problem requires two additional parameters: the resource constraint ρ 0 and the arc costs c a . We set ρ 0 to be equal to αρ Z ( o ) +(1 -α ρ X ) ( Q ) for some α ∈ [0 , 1]. We use ρ Z ( o ) and ρ X ( Q ), where Z o is again the solution of the Stochastic On Time Arrival Problem at the origin, and where Q is the optimal solution of the unconstrained deterministic shortest path problem. For each instance, we solved the Stochastic ρ -Resource Constrained Shortest Path Problem for α ∈ { 0 02 0 1 . , . , 0 5 . } . We did so in order to have feasible but yet nontrivial solutions. The costs c a were generated randomly, using a uniform law between 1 and twice the smallest t such that P ( X a = ) t > 0
7.2. Stochastic On Time Arrival Problem. Table 2 describes the performance of the Stochastic On Time Arrival Problem algorithm. 'Upd.' is the number of updates operations, which can be compared to the number of arcs, and 'Exp.' is the number of expansion of vertices, which should be compared to the number of vertices in the graph. First note that the algorithm is able to solve the Stochastic On Time Arrival Problem on instances with 10 000 vertices in ′ between 1 and 10 seconds. On all the instances we tested, the number of expansions was smaller than 3 3 . | V | , and often close to | V | , which is the number of expansions needed to check that a solution is a feasible solution of the Stochastic On Time Arrival Problem . Finally, the ratio of the number of expansions divided by the number of vertices in the instance increases slowly with the size of the instances, which allows to presume that the performance of the algorithm will remain high on larger instances. The limiting factor in our experiments was the memory available.
Remark 5 . Note that Stochastic On Time Arrival Problem algorithm CPU times to derive bounds to solve Stochastic ρ -Shortest Path Problem and Stochastic ρ -Resource Constrained Shortest Path Problem given in Tables 3 to 6 are larger than the those of Table 2.
Table 2. Stochastic On Time Arrival Problem algorithm performances.
| Instance | Upd. | Exp. | CPU time (s) |
|------------|--------|--------|----------------|
| g10R | 398 | 111 | 0.0037 |
| g40R | 14060 | 3598 | 0.3944 |
| g100R | 103397 | 26095 | 6.0776 |
| g10Ls | 403 | 113 | 0.0061 |
| g40Ls | 11074 | 2838 | 0.2712 |
| g100Ls | 80291 | 20271 | 3.1419 |
| g100Ll | 129797 | 32764 | 14.4251 |
| g10G | 430 | 121 | 0.0053 |
| g40G | 13696 | 3513 | 0.2853 |
| g100G | 107802 | 27214 | 4.1233 |
Indeed, this is due to the computation of non-dominated path distributions in order to obtain upper bounds as explained in Section 5.3.2.
7.3. Stochastic ρ -Shortest Path Problem. Table 3 presents our numerical results when ρ ( ) = · P ( · ≥ τ ) is used, and Table 4 when ρ ( ) · = CVaR β ( ). · In Table 3, the value of the parameter p used to choose τ has been displayed. The next columns provide the parameters τ or β , and the cost of the lower bound obtained by the solution of Stochastic On Time Arrival Problem (SOTA) at the origin and the upper bound obtained by applying ρ ( ) · to the optimal solution of the deterministic shortest path with expectation as cost. The performance of the Stochastic On Time Arrival Problem algorithm used as preprocessing is then given. The next columns indicates the number of labels treated and expanded by the algorithm. Finally, the next columns provide the CPU time for the label algorithm, the cost of the optimal solution, and the total CPU time, which is roughly equal the sum of the CPU time of the Stochastic On Time Arrival Problem algorithm and the CPU time of the labeling algorithm. The additional time is due to memory allocation.
We note that the algorithm is able to solve Stochastic ρ -Shortest Path Problem for all instances in less that one minute. The most time-consuming phase is always the solution of Stochastic On Time Arrival Problem done in preprocessing. The label algorithm runs on most instances in less than one second. The number of labels expanded by the Stochastic ρ -Shortest Path Problem turns out to be small, which indicates that the quality of the lower bounds provided by the Stochastic On Time Arrival Problem algorithm is good. No clear correlations appear between the parameter τ or β and the performance of the algorithm. The cost of the solution of Stochastic On Time Arrival Problem , the cost of the optimal solution of Stochastic ρ -Shortest Path Problem , and the cost of the heuristic solution obtained by taking the path with minimal expectation are within a small range.
We remark that when parameter p is set to be equal to 0 5, the path with minimum expected . length often also minimizes P ( X ≥ τ ). This sounds reasonable, because when p = 0 5, we expect a . path with small P ( X ≥ τ ) to have a small median and thus a small expectation.
7.4. Stochastic ρ -Resource Constrained Shortest Path Problem. Table 5 contains the numerical experiments for the Stochastic ρ -Resource Constrained Shortest Path Problem with ρ ( ) = · P ( · ≥ τ ) as risk measure, and Table 6 contains the results with ρ ( ) = · CVaR β ( ). · Both tables contain the tradeoff α between the risk measure of the solution of Stochastic On Time Arrival Problem , the cost of the optimal unconstrained path and the resource constraint ρ 0 . The next column indicates the time consumed by the Stochastic On Time Arrival Problem
Table 3. Performance of Stochastic ρ -Shortest Path Problem algorithm with P ( · ≥ τ ) as risk measure
| Instance | p | τ | SOTA P ( · ≥ τ ) | ESP P ( · ≥ τ ) | SOTA CPU (s) | λ treat. | λ exp. | SSPP CPU (s) | Opt. sol. P ( · ≥ τ ) | Total CPU (s) |
|------------|------|------|--------------------|-------------------|----------------|------------|----------|----------------|-------------------------|-----------------|
| g40R g40R | 0.5 | 1773 | 0.495 | 0.498 | 0.788 | 304 | 79 | 0.081 | 0.498 | 0.878 |
| | 0.8 | 1817 | 0.197 | 0.218 | 0.772 | 304 | 82 | 0.055 | 0.203 | 0.836 |
| g40R | 0.95 | 1855 | 0.049 | 0.074 | 0.784 | 304 | 83 | 0.059 | 0.051 | 0.852 |
| g100R | 0.5 | 4492 | 0.497 | 0.502 | 12.153 | 794 | 201 | 0.484 | 0.502 | 12.695 |
| g100R | 0.8 | 4565 | 0.197 | 0.203 | 12.11 | 1466 | 370 | 0.913 | 0.2 | 13.08 |
| g100R | 0.95 | 4630 | 0.049 | 0.057 | 12.729 | 794 | 214 | 0.498 | 0.05 | 13.287 |
| g40Ls | 0.5 | 1242 | 0.488 | 0.502 | 0.55 | 304 | 79 | 0.031 | 0.502 | 0.588 |
| g40Ls | 0.8 | 1268 | 0.199 | 0.217 | 0.544 | 452 | 117 | 0.048 | 0.217 | 0.598 |
| g40Ls | 0.95 | 1297 | 0.047 | 0.061 | 0.544 | 304 | 81 | 0.03 | 0.051 | 0.582 |
| g100Ls | 0.5 | 2959 | 0.493 | 0.496 | 6.203 | 794 | 201 | 0.132 | 0.496 | 6.378 |
| g100Ls | 0.8 | 2999 | 0.197 | 0.203 | 6.254 | 794 | 201 | 0.14 | 0.203 | 6.436 |
| g100Ls | 0.95 | 3040 | 0.049 | 0.053 | 6.322 | 1082 | 273 | 0.176 | 0.05 | 6.542 |
| g100Ll | 0.5 | 3971 | 0.498 | 0.503 | 30.104 | 1765 | 444 | 1.339 | 0.503 | 31.512 |
| g100Ll | 0.8 | 4070 | 0.2 | 0.204 | 30.186 | 1260 | 320 | 0.936 | 0.203 | 31.191 |
| g100Ll | 0.95 | 4170 | 0.05 | 0.053 | 30.156 | 1577 | 407 | 1.162 | 0.051 | 31.387 |
| g40G | 0.5 | 658 | 0.485 | 0.485 | 0.539 | 301 | 79 | 0.052 | 0.485 | 0.599 |
| g40G | 0.8 | 680 | 0.19 | 0.192 | 0.535 | 301 | 79 | 0.048 | 0.192 | 0.59 |
| g40G | 0.95 | 701 | 0.05 | 0.052 | 0.539 | 301 | 81 | 0.048 | 0.05 | 0.594 |
| g100G | 0.5 | 1708 | 0.492 | 0.498 | 8.702 | 2679 | 673 | 0.928 | 0.498 | 9.679 |
| g100G | 0.8 | 1746 | 0.195 | 0.204 | 8.601 | 8256 | 2079 | 2.906 | 0.202 | 11.555 |
| g100G | 0.95 | 1782 | 0.049 | 0.058 | 8.624 | 8436 | 2139 | 2.992 | 0.052 | 11.666 |
Table 4. Performance of Stochastic ρ -Shortest Path Problem algorithm with CVaR as risk measure
| Instance | β | SOTA CV aR | ESP CV aR | SOTA CPU (s) | λ treat. | λ exp. | SSPP CPU (s) | Opt. sol. CVaR | Total CPU (s) |
|------------|------|--------------|-------------|----------------|------------|----------|----------------|------------------|-----------------|
| g40R g40R | 0.25 | 1837.19 | 1845.26 | 0.8 | 260 | 83 | 0.052 | 1838.44 | 0.86 |
| | 0.05 | 1873.51 | 1889.45 | 0.776 | 258 | 87 | 0.053 | 1874.78 | 0.837 |
| g40R | 0.01 | 1899.09 | 1922.08 | 0.775 | 253 | 94 | 0.055 | 1901.06 | 0.839 |
| g100R | 0.25 | 4599.41 | 4603.78 | 12.871 | 784 | 211 | 0.495 | 4600.43 | 13.424 |
| g100R | 0.05 | 4663.33 | 4672.6 | 12.481 | 1076 | 303 | 0.715 | 4664.4 | 13.254 |
| g100R | 0.01 | 4711.35 | 4724.45 | 12.156 | 1076 | 315 | 0.715 | 4712.11 | 12.929 |
| g40Ls | 0.25 | 1283.57 | 1287.44 | 0.545 | 282 | 81 | 0.03 | 1285.22 | 0.582 |
| g40Ls | 0.05 | 1312.48 | 1319.5 | 0.543 | 273 | 86 | 0.031 | 1313.44 | 0.581 |
| g40Ls | 0.01 | 1336.08 | 1345.9 | 0.543 | 259 | 93 | 0.037 | 1336.71 | 0.587 |
| g100Ls | 0.25 | 3021 | 3022.7 | 6.274 | 731 | 203 | 0.125 | 3021.97 | 6.441 |
| g100Ls | 0.05 | 3062.44 | 3065.51 | 6.323 | 934 | 276 | 0.179 | 3063.26 | 6.56 |
| g100Ls | 0.01 | 3095.33 | 3099.82 | 6.265 | 612 | 205 | 0.116 | 3095.59 | 6.425 |
| g100Ll | 0.25 | 4124.03 | 4126.8 | 30.856 | 1269 | 335 | 0.959 | 4125.52 | 31.885 |
| g100Ll | 0.05 | 4224.1 | 4228.73 | 30.691 | 755 | 211 | 0.565 | 4225.12 | 31.325 |
| g100Ll | 0.01 | 4302.64 | 4309.32 | 30.432 | 737 | 214 | 0.556 | 4303.22 | 31.057 |
| g40G | 0.25 | 691.308 | 691.725 | 0.554 | 272 | 80 | 0.055 | 691.523 | 0.619 |
| g40G | 0.05 | 713.333 | 714.331 | 0.551 | 252 | 82 | 0.045 | 713.517 | 0.604 |
| g40G | 0.01 | 730.689 | 732.304 | 0.555 | 224 | 85 | 0.044 | 730.875 | 0.607 |
| g100G | 0.25 | 1765.15 | 1768.07 | 8.64 | 8051 | 2083 | 2.904 | 1766.62 | 11.592 |
| g100G | 0.05 | 1801.33 | 1807.01 | 8.651 | 12803 | 3301 | 4.625 | 1803.17 | 13.326 |
| g100G | 0.01 | 1829.4 | 1837.5 | 8.605 | 19187 | 4952 | 6.975 | 1831.5 | 15.628 |
(SOTA) algorithm used in preprocessing. The three next columns provide the number of labels treated and expanded by the Stochastic ρ -Resource Constrained Shortest Path Problem (SRCSP) labeling algorithm and the CPU time of the labeling algorithm. The labeling algorithm was stopped after five minutes. The column ρ P ( ) is the value of the risk measure evaluated on the solution found, LB is a lower bound on its cost (only displayed when the optimal solution was not
| Inst. | τ | α | SP cost | ρ 0 | SOTA CPU (s) | λ treat. | λ exp. | SRCSP CPU (s) | ρ ( P ) | LB | c P | gap % | Total CPU (s) |
|---------|------|------|-----------|-------|----------------|------------|----------|-----------------|-----------|----------|----------|---------|-----------------|
| g10R | 498 | 0.02 | 134.106 | 0.069 | 0.008 | 67 | 19 | 0.003 | 0.050 | | 320.702 | 0.0 | 0.012 |
| g10R | 498 | 0.1 | 134.106 | 0.145 | 0.008 | 79 | 22 | 0.003 | 0.070 | | 219.858 | 0.0 | 0.012 |
| g10R | 498 | 0.5 | 134.106 | 0.525 | 0.007 | 318 | 82 | 0.013 | 0.258 | | 202.102 | 0.0 | 0.021 |
| g40R | 1855 | 0.02 | 584.055 | 0.068 | 0.733 | 1827 | 462 | 0.319 | 0.062 | | 1283.772 | 0.0 | 1.063 |
| g40R | 1855 | 0.1 | 584.055 | 0.144 | 0.724 | 75110 | 19177 | 13.486 | 0.123 | | 1205.852 | 0.0 | 14.22 |
| g40R | 1855 | 0.5 | 584.055 | 0.525 | 0.726 | 1355571 | 429418 | 300 | 0.479 | 856.668 | 1248.012 | 45.7 | 300.735 |
| g10Ls | 418 | 0.02 | 251.844 | 0.069 | 0.012 | 111 | 30 | 0.005 | 0.059 | | 421.254 | 0.0 | 0.018 |
| g10Ls | 418 | 0.1 | 251.844 | 0.145 | 0.012 | 360 | 94 | 0.016 | 0.146 | | 400.475 | 0.0 | 0.029 |
| g10Ls | 418 | 0.5 | 251.844 | 0.525 | 0.013 | 1662 | 427 | 0.078 | 0.516 | | 371.532 | 0.0 | 0.092 |
| g40Ls | 1297 | 0.02 | 833.451 | 0.066 | 0.533 | 1147 | 290 | 0.103 | 0.061 | | 1541.857 | 0.0 | 0.646 |
| g40Ls | 1297 | 0.1 | 833.451 | 0.143 | 0.531 | 119204 | 30376 | 10.991 | 0.120 | | 1490.817 | 0.0 | 11.533 |
| g40Ls | 1297 | 0.5 | 833.451 | 0.524 | 0.547 | 2823376 | 861653 | 300 | 0.480 | 1220.566 | 1433.313 | 17.4 | 300.556 |
| g10G | 185 | 0.02 | 213.27 | 0.062 | 0.009 | 117 | 31 | 0.005 | 0.058 | | 297.191 | 0.0 | 0.014 |
| g10G | 185 | 0.1 | 213.27 | 0.116 | 0.01 | 175 | 46 | 0.007 | 0.076 | | 269.665 | 0.0 | 0.018 |
| g10G | 185 | 0.5 | 213.27 | 0.386 | 0.008 | 149 | 45 | 0.007 | 0.243 | | 231.777 | 0.0 | 0.016 |
| g40G | 701 | 0.02 | 892.028 | 0.069 | 0.535 | 10528 | 2635 | 1.758 | 0.064 | | 1577.910 | 0.0 | 2.305 |
| g40G | 701 | 0.1 | 892.028 | 0.145 | 0.527 | 1231773 | 307952 | 207.613 | 0.143 | | 1473.075 | 0.0 | 208.151 |
| g40G | 701 | 0.5 | 892.028 | 0.525 | 0.553 | 1468578 | 551686 | 300 | - | 999.764 | ∞ | ∞ | 300.56 |
Table 5. Performance of Stochastic ρ -Resource Constrained Shortest
Path Problem algorithm with P ( · ≥ τ ) as risk measure, with p = 0 95 .
found), and c P is its cost. When a cost equal to ∞ is displayed, it means that the algorithm was not able to find any feasible solution. One of the instances was moreover proved to be not feasible (g10G of Table 6). Finally, we provide the total CPU time, which is roughly equal to the sum of the preprocessing CPU time and the labeling algorithm CPU time.
The performance of the algorithm is strongly influenced by the parameter α , which models the hardness of the resource constraint. A small α corresponds to a hard resource constraint and a large α corresponds to an easy resource constraint. The algorithm performance relies on the use of stochastic lower bound to cut potentially infeasible path. Thus, when the resource constraint is hard, which corresponds to a small α , the number of feasible paths is small, and the algorithm is efficient. The labeling algorithm was able to solve all the instances with P ( · ≥ τ ) and most instances with ρ ( ) = · CVaR β ( ) · when α = 0 02 in the allocated time. . On the contrary, when α is larger, the number of feasible paths is much bigger, making the algorithm less efficient.
## References
Bast, Hannah, Delling, Daniel, Goldberg, Andrew, M¨ller-Hannemann, Matthias, Pajor, Thomas, u Sanders, Peter, Wagner, Dorothea, & Werneck, Renato. 2014. Route Planning in Transportation Networks .
Bogerding, Mark. 2013. KissFFT library .
Boland, Natashia, Dethridge, John, & Dumitrescu, Irina. 2006. Accelerated label setting algorithms for the elementary resource constrained shortest path problem. Operations Research Letters , 34 (1), 58-68.
Chen, Anthony, & Ji, Zhaowang. 2005. Path finding under uncertainty. Journal of advanced transportation , 39 (1), 19-37.
Chen, Bi Yu, Lam, William HK, Sumalee, Agachai, Li, Qingquan, Shao, Hu, & Fang, Zhixiang. 2013. Finding reliable shortest paths in road networks under uncertainty. Networks and spatial economics , 13 (2), 123-148.
Dumitrescu, Irina, & Boland, Natashia. 2003. Improved preprocessing, labeling and scaling algorithms for the Weight-Constrained Shortest Path Problem. Networks , 42 (3), 135-153.
| Inst. | α | SP cost | ρ 0 | SOTA CPU (s) | λ treat. | λ exp. | SRCSP CPU (s) | ρ ( P ) | LB | c P | gap % | Total CPU (s) |
|---------|------|-----------|----------|----------------|------------|----------|-----------------|-----------|----------|----------|---------|-----------------|
| g10R | 0.02 | 134.106 | 509.474 | 0.007 | 67 | 19 | 0.002 | 505.811 | | 219.858 | 0.0 | 0.01 |
| g10R | 0.1 | 134.106 | 533.22 | 0.007 | 382 | 98 | 0.016 | 513.358 | | 202.102 | 0.0 | 0.025 |
| g10R | 0.5 | 134.106 | 651.949 | 0.008 | 612 | 244 | 0.035 | 645.537 | | 150.467 | 0.0 | 0.044 |
| g40R | 0.02 | 584.055 | 1896.34 | 0.765 | 28474 | 7235 | 4.913 | 1889.475 | | 1205.852 | 0.0 | 5.691 |
| g40R | 0.1 | 584.055 | 1987.67 | 0.727 | 1256561 | 427909 | 300 | 1987.238 | 793.711 | 1050.238 | 32.3 | 300.736 |
| g40R | 0.5 | 584.055 | 2444.32 | 0.731 | 670441 | 514178 | 300 | 2439.891 | 613.004 | 742.644 | 21.1 | 300.741 |
| g10Ls | 0.02 | 251.844 | 432.411 | 0.012 | 111 | 30 | 0.005 | 429.405 | | 421.254 | 0.0 | 0.018 |
| g10Ls | 0.1 | 251.844 | 453.657 | 0.013 | 1050 | 268 | 0.049 | 445.882 | | 374.680 | 0.0 | 0.064 |
| g10Ls | 0.5 | 251.844 | 559.885 | 0.012 | 1357 | 353 | 0.076 | 513.708 | | 283.292 | 0.0 | 0.089 |
| g40Ls | 0.02 | 833.451 | 1334.62 | 0.521 | 94912 | 24051 | 8.796 | 1333.675 | | 1490.817 | 0.0 | 9.328 |
| g40Ls | 0.1 | 833.451 | 1423.18 | 0.558 | 2520108 | 1025174 | 300 | 1417.246 | 1028.961 | 1336.761 | 29.9 | 300.569 |
| g40Ls | 0.5 | 833.451 | 1865.97 | 0.538 | 1202558 | 836889 | 300.001 | 1860.527 | 880.748 | 1007.203 | 14.4 | 300.548 |
| g10G | 0.02 | 213.27 | 192.069 | 0.009 | 101 | 26 | 0.004 | | | INFEAS. | | 0.014 |
| g10G | 0.1 | 213.27 | 195.236 | 0.009 | 118 | 31 | 0.005 | 194.351 | | 269.665 | 0.0 | 0.015 |
| g10G | 0.5 | 213.27 | 211.069 | 0.009 | 173 | 50 | 0.008 | 203.028 | | 231.777 | 0.0 | 0.018 |
| g40G | 0.02 | 892.028 | 719.158 | 0.547 | 18664 | 4669 | 3.133 | 718.039 | | 1562.206 | 0.0 | 3.69 |
| g40G | 0.1 | 892.028 | 742.458 | 0.562 | 1631484 | 505573 | 300 | - | 1121.193 | ∞ | ∞ | 300.573 |
| g40G | 0.5 | 892.028 | 858.958 | 0.573 | 839101 | 632986 | 300 | 857.531 | 923.304 | 1163.971 | 26.1 | 300.583 |
Table 6. Performance of Stochastic ρ -Resource Constrained Shortest
Path Problem algorithm with CVaR β with β = 0 05 as risk measure .
Ehrgott, Matthias. 2005. Multicriteria optimization . Vol. 2. Springer.
Eiger, Amir, Mirchandani, Pitu B, & Soroush, Hossein. 1985. Path preferences and optimal paths in probabilistic networks. Transportation Science , 19 (1), 75-84.
Fan, Yueyue, & Nie, Yu. 2006. Optimal routing for maximizing the travel time reliability. Networks and Spatial Economics , 6 (3-4), 333-344.
- Fan, YY, Kalaba, RE, & Moore II, JE. 2005. Arriving on time. Journal of Optimization Theory and Applications , 127 (3), 497-513.
- Frank, H. 1969. Shortest paths in probabilistic graphs. Operations Research , 17 (4), 583-599.
- Fu, Liping. 2001. An adaptive routing algorithm for in-vehicle route guidance systems with real-time information. Transportation Research Part B: Methodological , 35 (8), 749-765.
Fu, Liping, & Rilett, Larry R. 1998. Expected shortest paths in dynamic and stochastic traffic networks. Transportation Research Part B: Methodological , 32 (7), 499-516.
Hall, Randolph W. 1986. The fastest path through a network with random time-dependent travel times. Transportation science , 20 (3), 182-188.
Handler, Gabriel Y, & Zang, Israel. 1980. A dual algorithm for the constrained shortest path problem. Networks , 10 (4), 293-309.
Irnich, Stefan, & Desaulniers, Guy. 2005. Shortest path problems with resource constraints . Springer. Kosuch, Stefanie, & Lisser, Abdel. 2010. Stochastic shortest path problem with delay excess penalty. Electronic Notes in Discrete Mathematics , 36 , 511-518.
Loui, Ronald Prescott. 1983. Optimal paths in graphs with stochastic or multidimensional weights. Communications of the ACM , 26 (9), 670-676.
Miller-Hooks, Elise, & Mahmassani, Hani. 2003. Path comparisons for a priori and time-adaptive decisions in stochastic, time-varying networks. European Journal of Operational Research , 146 (1), 67-82.
Miller-Hooks, Elise D, & Mahmassani, Hani S. 1998. Least possible time paths in stochastic, time-varying networks. Computers & operations research , 25 (12), 1107-1125.
Miller-Hooks, Elise Deborah. 1997. Optimal routing in time-varying, stochastic networks: algorithms and implementations . The University of Texas at Austin.
Mirchandani, Pitu B. 1976. Shortest distance and reliability of probabilistic networks. Computers & Operations Research , 3 (4), 347-355.
M¨ uller, Alfred, & Stoyan, Dietrich. 2002. Comparison methods for stochastic models and risks . Vol. 389. Wiley.
Murthy, Ishwar, & Sarkar, Sumit. 1996. A relaxation-based pruning technique for a class of stochastic shortest path problems. Transportation Science , 30 (3), 220-236.
Murthy, Ishwar, & Sarkar, Sumit. 1998. Stochastic shortest path problems with piecewise-linear concave utility functions. Management Science , 44 (11-part-2), S125-S136.
Nie, Yu, & Fan, Yueyue. 2006. Arriving-on-time problem: discrete algorithm that ensures convergence. Transportation Research Record: Journal of the Transportation Research Board , 1964 (1), 193-200.
Nie, Yu Marco, & Wu, Xing. 2009. Shortest path problem considering on-time arrival probability. Transportation Research Part B: Methodological , 43 (6), 597-613.
Nie, Yu Marco, Wu, Xing, & Homem-de Mello, Tito. 2012. Optimal path problems with secondorder stochastic dominance constraints. Networks and Spatial Economics , 12 (4), 561-587.
Niknami, Mehrdad, Samaranayake, Samitha, & Bayen, Alexandre. 2014. Tractable Pathfinding for the Stochastic On-Time Arrival Problem. arXiv preprint arXiv:1408.4490 .
Nikolova, Evdokia. 2010. High-performance heuristics for optimization in stochastic traffic engineering problems. Pages 352-360 of: Large-Scale Scientific Computing . Springer.
Nikolova, Evdokia, Kelner, Jonathan A, Brand, Matthew, & Mitzenmacher, Michael. 2006. Stochastic shortest paths via quasi-convex maximization. Pages 552-563 of: Algorithms-ESA 2006 . Springer.
Raith, Andrea, & Ehrgott, Matthias. 2009. A comparison of solution strategies for biobjective shortest path problems. Computers & Operations Research , 36 (4), 1299-1331.
Sabran, Guillaume, Samaranayake, Samitha, & Bayen, Alexandre M. 2014. Precomputation techniques for the stochastic on-time arrival problem. Pages 138-146 of: ALENEX . SIAM.
Samaranayake, Samitha, Blandin, Sebastien, & Bayen, A. 2012. A tractable class of algorithms for reliable routing in stochastic networks. Transportation Research Part C: Emerging Technologies , 20 (1), 199-217.
Sivakumar, Raj A, & Batta, Rajan. 1994. The variance-constrained shortest path problem. Transportation Science , 28 (4), 309-316.
Tarapata, Zbigniew. 2007. Selected multicriteria shortest path problems: An analysis of complexity, models and adaptation of standard algorithms. International Journal of Applied Mathematics and Computer Science , 17 (2), 269-287.
A. Parmentier, Universit´ Paris Est, CERMICS, 6-8 avenue Blaise Pascal, Cit´ Descartes, 77455 e e Marne-la-Vall´ ee, Cedex 2, France
E-mail address : [email protected]
F. Meunier, Universit´ e Paris Est, CERMICS, 6-8 avenue Blaise Pascal, Cit´ e Descartes, 77455 Marne-la-Vall´ ee, Cedex 2, France
E-mail address : [email protected] | null | [
"Axel Parmentier",
"Frédéric Meunier"
] | 2014-08-01T19:20:58+00:00 | 2014-09-26T16:38:46+00:00 | [
"cs.DS",
"90B99"
] | Stochastic Shortest Paths and Risk Measures | We consider three shortest path problems in directed graphs with random arc
lengths. For the first and the second problems, a risk measure is involved.
While the first problem consists in finding a path minimizing this risk
measure, the second one consists in finding a path minimizing a deterministic
cost, while satisfying a constraint on the risk measure. We propose algorithms
solving these problems for a wide range of risk measures, which includes among
several others the $CVaR$ and the probability of being late. Their performances
are evaluated through experiments. One of the key elements in these algorithms
is the use of stochastic lower bounds that allow to discard partial solutions.
Good stochastic lower bounds are provided by the so-called Stochastic Ontime
Arrival Problem. This latter problem is the third one studied in this paper and
we propose a new and very efficient algorithm solving it. Complementary
discussions on the complexity of the problems are also provided. |
1408.0274v1 | Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
THERMAL AND THERMOELECTRIC TRANSPORT IN NANOSTRUCTURES AND LOW-
## DIMENSIONAL SYSTEMS
## Li Shi
Department of Mechanical Engineering and Texas Materials Institute, The University of Texas at
Austin, Austin, TX 78712
## [email protected]
## Abstract
Significant progress has been made in recent studies of thermal and thermoelectric transport phenomena in nanostructures and low-dimensional systems. This article reviews several intriguing quantum and classical size effects on thermal and thermoelectric properties that have been predicted by theoretical calculations or observed in experiments. Attention is focused on the Casimir limit in phonon boundary scattering and the effect of phonon confinement on the lattice thermal conductivity of semiconductor nanowires (NWs) and nanomeshes; the effects of thickness, lateral size, and interface interaction on the lattice thermal conductivity of carbon nanotubes (CNTs) and graphene; and the phonon-drag thermopower and quantum size effects on the thermoelectric power factor in semiconductor NWs. Further experimental and theoretical investigations are suggested for better understanding of some of these nanoscale transport phenomena.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
## INTRODUCTION
In recent years, one focus of thermophysical engineering research has been to address the increasingly severe heat dissipation problem in microelectronic and other functional devices. As the critical feature size of electronic devices is reduced to increase the packing density and to reduce cost, the power density is increased and can exceed 1000 W/cm at local hot spots [1]. 2 Understanding thermal transport processes around these hot spots and the size effects in the thermal properties of materials has been a topic of both scientific interest and practical importance [2]. In addition, the increasing awareness of the global challenge in energy and sustainability has recently rekindled the interest in efficient and environmentally friendly thermal energy conversion and storage technologies, including solid-state thermoelectric (TE) power generation and refrigeration based on the Seebeck and Peltier effects [3]. Various size effects have been investigated as possible new approaches to enhancing the transport properties of materials for these applications. These properties include the thermal conductivity ( ), the thermopower or Seebeck coefficient ( S ), and the electrical conductivity ( ).
The room-temperature of dense bulk inorganic solids span the range between about 1 W m -1 K -1 in amorphous materials such as silica to 2230 W m -1 K -1 reported for natural diamond, although a higher value of 3320 W m -1 K -1 has been suggested for isotopically enriched C-12 diamond [4, 5]. High- materials are used for electronic cooling and thermal energy storage. In comparison, low- materials such as the recently reported disordered, layered thin films [6] of a ultralow value not much higher than that of air, are desirable for thermal insulation and TE waste heat recovery and refrigeration. The energy efficiency of TE devices is governed by the materials figure of merit, which is defined as
$$Z \equiv S ^ { 2 } \sigma / \kappa,$$
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
Figure 1 shows that the dimensionless ZT product reaches the peak value close to unity within a narrow temperature ( T ) range near room, intermediate, and high temperature, respectively, for Bi2Te3, PbTe, and SiGe alloys [7-13]. For TE conversion to become sufficiently efficient and economical for vehicle waste heat recovery, the ZT value needs to be increased above 2 over a large temperature range. A larger average ZT value may be needed to make solarthermal-electricity conversion and TE refrigeration a commercially viable option beyond the current niche markets including cooling car seats and optoelectronic devices [14-16]. Although there is no established theoretical limit in ZT , increasing ZT to above unity has proven difficult because the interdependence of the three properties. For example, increasing by increasing the charge carrier concentration can lead to a decrease in S and an increase of the electronic contribution ( e ) to [17]. In addition, because traditional TE materials contain elements such as Te and Ge that are not abundant on earth and become increasingly costly due to their use in other technologies, alternative highZT materials containing only earth-abundant elements need to be discovered before TE conversion can play a role in addressing the worldwide challenge in energy and sustainability.
Considerable advancements have been made in controlling materials structure down to the nanometer scale. When the characteristic dimension of the materials structure becomes comparable to the wavelength or mean free path of electrons and phonons that are the energy carriers in solids, quantum confinement and classical interface scattering effects can emerge and modify the transport properties.
The de Broglie wavelength of an electron is
$$\lambda _ { e } = 2 \pi \hbar { / } \sqrt { 2 m ^ { * } E }$$
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
where ħ is the reduced Planck constant, m* is the electron effective mass, and E is the electron energy. When E measured from the band edge is of the order of the thermal energy kBT , where kB is the Boltzmann constant, e can be as long as tens of nanometers in Bi, Bi0.95Sb0.5 and InSb that all have a very small m* .
In comparison, the spectral energy density of phonons at low temperatures peaks at a wavelength ( 0 , p ) given as
$$\lambda _ { p, 0 } T \approx 2 \pi \hbar { v } _ { s } / x k _ { B } \,,$$
where x takes a value close to 5, 4, and 3 for three-, two-, and one-dimensional (3D, 2D, and 1D) systems. For a typical sound velocity vs of the order of 10 m/s, 4 0 , p T ≈ 100 nm. This expression breaks down at high temperature as the obtained 0 , p value becomes smaller than the minimum allowable wavelength min = 2 a , where a is the lattice constant. In this case, the spectral energy density peaks at min , corresponding to the edge of the Brillouin zone of the reciprocal lattice. Hence, except at very low temperatures or in crystals with a very large a , the spectral phonon density near room temperature peaks at a wavelength of the order of 1 nm. However, because of the lack of umklapp scattering for long-wavelength phonons, the spectral phonon thermal conductivity can peak at a longer wavelength than 0 , p or min .
The mean free path of electrons and phonons can range from as short as close to the atomic spacing to as long as more than microns in high quality crystals or at low temperatures. A long electron mean free path ( l e ) is usually found in smallm * semiconductors or semimetals, such as InSb, Bi, and graphene and its derivatives including carbon nanotubes (CNTs). Here, the term nanostructures is broadly employed for those with a characteristic size in the sub-100 nm scale, which can be comparable to the mean free paths of the energy carriers. In comparison, the
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
term low-dimensional (low-D) systems is used exclusively for those with the characteristic size of one or more dimensions being comparable to or smaller than the wavelength of electrons or phonons.
A number of quantum and classical size effects on thermal and TE properties in low-D systems and nanostructures are reviewed in this article. It is not the intention of this review to provide an extensive coverage of the vast amount of experimental and theoretical reports, some of which have been reviewed in several recent publications [3, 5, 18-21]. Instead, besides firmly established knowledge in this rapidly expanding field, this review pays attention to several unresolved issues in thermal and TE transport in nanomaterials, so as to stimulate further investigations into these intriguing phenomena.
## PHONON CONFINEMENT, INTERFACE SCATTERING, AND THE CASIMIR LIMIT IN SEMICONDUCTOR NANOSTRUCTURES
## Lattice Thermal Conductivity of Nanowires
The thermal conductivity is dominated by the phonon or lattice contribution ( l ) except in metals and degenerately doped, small-band gap semiconductors. This contribution can be interpreted using
$$\kappa _ { l } = \sum _ { s } \int v _ { z } ( s, q ) ^ { 2 } \tau ( s, q ) \frac { k _ { B } ( \hbar { \omega } ( s, q ) / k _ { B } T ) ^ { 2 } e ^ { \hbar { \omega } ( s, q ) / k _ { B } T } } { \left ( e ^ { \hbar { \omega } ( s, q ) / k _ { B } T } - 1 \right ) ^ { 2 } } \frac { d q ^ { 3 } } { 8 \pi ^ { 3 } }$$
where for each polarization and wave vector s q , vz is the group velocity ( v ) component along the transport direction, = / l v is the relaxation time, l is the phonon mean free path, is the frequency, and dq 3 is an infinitesimal volume in the Brillouin zone of the reciprocal lattice. For
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
an isotropic bulk crystal, the integration in the q space can be carried out over the phonon
## frequency domain as
$$\kappa _ { l } = \sum _ { s } \frac { 1 } { 3 } k _ { B } \int _ { 0 } ^ { \omega _ { Z B } ( s ) } v ( s, \omega ) ^ { 2 } \tau ( s, \omega ) \frac { ( \hbar { \omega } / k _ { B } T ) ^ { 2 } e ^ { \hbar { \omega } / k _ { B } T } } { \left ( e ^ { \hbar { \omega } / k _ { B } T } - 1 \right ) ^ { 2 } } \, D ( s, \omega ) d \omega$$
where D s ( , ) is the density of states of polarization s , ZB is the zone boundary frequency, and the factor of 1/3 originates from the replacement of vz 2 in Eq. (4) with v 2 in Eq. (5). Under the Debye approximation of a constant group velocity vs , D s ( , ) = 2 /2 2 vs 3 . Phenomenological models have been reported in the literature [22-24] for calculating various scattering rates and for differentiating between the normal and umklapp processes of three-phonon scattering events.
Surface scattering of phonons in a nanowire (NW) is expected to lower l and to values below those for bulk crystals. Casimir [25] derived that the average boundary scattering mean free path ( l b ) of quasi-particles equals the diameter ( d ) for a circular rod, and 1.12 W for a rod with a square cross section of a width W , for the case that the quasi-particles are scattered diffusely by the surface. The results for other cross section shapes have been reported in a recent theoretical work [26]. Ziman extended Casimir's theory by defining a specularity parameter ( p ) as the probability that the quasi-particle undergoes specular reflection at the surface [27], with the lower limit p = 0 for diffuse scattering, and the upper limit p = 1 for specular scattering. For random surface roughness, Ziman obtained the following expression for a plane wave at normal incidence,
$$p ( \lambda ) = \exp \left ( - \frac { 1 6 \pi ^ { 3 } \eta ^ { 2 } } { \lambda ^ { 2 } } \right ),$$
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
where is the wavelength of the energy carriers, and is the roughness height. Ziman suggested that this expression can underestimate p for incident waves at a large or glancing angle. In this framework, the boundary scattering mean free path is obtained as
$$l _ { b } = \alpha \frac { 1 + p } { 1 - p } l _ { b, c }$$
where l b,c is Casimir's result for diffuse scattering in an isotropic material with = 1 and p = 0. For anisotropic materials, can be larger than and smaller than 1 for transport along the crystallographic direction of a high and low group velocity, respectively, because of a phonon focusing effect [28].
The mean free path ( l i ) of other intrinsic scattering processes can be combined with l b via Matthiessen's rule to obtain the overall mean free path as
$$l ^ { - 1 } = l _ { b } ^ { - 1 } + l _ { i } ^ { - 1 } \.$$
It is worth noting that Matthiessen's rule is based on the assumption that different scattering processes are independent of each other.
With the use of a phenomenological model of Morelli et al. [23] that is expressed in a more detailed form than Eqs. (4-5), A. L. Moore has calculated the room-temperature thermal conductivity of Si wires as a function or the wire diameter. The result is shown in Fig. 2. In addition to the use of Casimir's result of l b = D and the same frequency-dependent umklapp scattering rate as that found in bulk Si, the calculation is based on the bulk Si phonon dispersion.
The phonon dispersion can become different from that in the bulk as the NW diameter is reduced to approach the dominant phonon wavelength. This issue has been examined by a number of studies. Some of the initial works suggested that the transition from bulk to quasi-one dimensional (1D) phonon dispersion in a 20-nm-diameter Si NWs can reduce the phonon group
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
velocity component ( ) ) along the NW axis [29, 30], to be about half of the phonon group , ( n V velocity v in the bulk [30]. In this approach [29, 30], the bulk phonon group velocity v in the expressions for the bulk scattering rates and thermal conductivity was replaced by ) for , ( n V calculating the corresponding values for the NW. Besides the enhanced boundary scattering rate because of the small NW diameter, this approach also resulted in enhanced isotope and umklapp scattering rates because of the reduced velocity value used in the calculation. In a different approach [31], the v term in the thermal conductivity and scattering rate expressions was also replaced with the group velocity component along the NW axis, and the quasi-1D density of states was used in place of the D ( ) term in the bulk thermal conductivity expression of the type of Eq. (5). For the case of a 15-nm-diameter Ge NW [32], this latter approach yielded a factor of two lower thermal conductivity than that calculated with the bulk phonon dispersion.
In comparison, Mingo [33] suggested that the bulk umklapp and impurity scattering expressions are valid for the NW diameter d > 10 nm and at temperatures T > 40 K, because the energy spacing between consecutive subbands, that is ħ ~(2 / d ħv ) , is larger than the thermal energy kBT , or the 0 , p value of 2 nm or less is still considerably smaller than d . In this case, the total phonon group velocity instead of just the axial component in the NW should be used in the bulk scattering expressions, according to [33]. In fact, the axial component of the phonon group velocity in a NW is that along the transport direction, and is thus equivalent to the bulk vz in Eq. (4) instead of the bulk v in Eq. (5). As mentioned in the paragraph following Eq. (5), the vz 2 value averaged over the entire q space is one third of the average v 2 for the case of an isotropic bulk crystal. Recognizing that the phonon group velocity component along the NW axis is essentially vz instead of v , Mingo has derived the thermal conductivity expression for a quasi-1D NW, i.e. Eqs. 4-6 of ref. [33], based on the actual phonon dispersion in the NW. It is worth noting that the
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
thermal conductivity expression obtained by Mingo [33] does not contain the 1/3 factor in Eq.
(6) of ref. [31]. For Si NWs of a diffuse surface and a diameter of 20 nm and larger, the roomtemperature thermal conductivity results calculated by Mingo [33] based on the actual dispersion are close to those calculated based on the bulk phonon dispersion, and are comparable to the calculation results shown in Fig. 2.
When the NW diameter is reduced further to be the order of the dominant phonon wavelength, the scattering rates need to be calculated based on explicit integration of scattering amplitudes amongst the confined phonon dispersions. Moreover, the frequency dependence of the scattering rates can be different between 3D, 2D and 1D systems. For example, the relaxation time approximation can be used to derive that the isotope scattering rate is proportional to 4 , 3 , and 2 for a linear acoustic branch in 3D, 2D, and 1D systems, respectively [34].
Several measurements of Si thin films [35], Si NWs grown by a vapor-liquid-solid (VLS) process [36], SnO2 nanobelts [37], and InAs NWs [38] have yielded values that can be attributed to diffuse surface scattering alone. However, the of a 22 nm diameter Si NW [36] fell somewhat below the calculation result in Fig. 2 based on diffuse surface scattering and bulk phonon dispersion. In addition, the thermal conductivity of this NW shows unusual linear temperature dependence from 20 K to room temperature. In comparison, the low-temperature thermal conductivity is expected to be dominated by boundary scattering to follow the T and T 3 dependence of specific heat, respectively, for a 1D and 3D system. Because the 22 nm diameter of the NW is still much larger than the dominant phonon wavelength in most of the temperature range of the measurement, 1D behavior is not expected for the NW. Such linear behavior was also observed in another measurement of thin Si NWs, although only the thermal conductance but not the thermal conductivity was reported in that work because of uncertainty in the cross
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
section [39]. One theory was able to explain the linear behavior by considering specular scattering of long-wavelength phonons at the NW surface, combined with incoherent, diffuse scattering of short-wavelength phonons [40]. However, it is unclear whether the model can also match the measured thermal conductivity magnitude. When the mean free path of shortwavelength phonons are assumed to be the same as the NW diameter, specular surface scattering of long-wavelength phonons is expected to increase the thermal conductivity.
Moreover, two recent measurements of Si NWs fabricated by electrochemical etching [41] or reactive ion etching (RIE) [42] have yielded values one order of magnitude lower than the calculation result shown in Fig. 2 for diffuse surface and bulk phonon dispersion. The value reported in ref. [42] for an array of ~10 nm wide etched Si NWs supported on SiO2 is even lower than the thermal conductivity of amorphous Si, that is, the amorphous limit. However, in a recent work by Hippalgaonkar et al. [43], the measured thermal conductivity of Si NWs etched by RIE was just slightly lower than the calculated result shown in Fig. 2, and considerably higher than those reported earlier for electrochemically etched Si NWs.
These surprising results have led to theoretical investigations of whether the Casimir limit of diffuse surface scattering can be violated. Moore et al. [44] employed Monte Carlo (MC) phonon particle transport simulation to investigate the effect of adding periodic sawtooth roughness to a Si NW of a 22 x 22 nm square cross section. For a NW with diffuse and flat surfaces, the MC simulation yielded similar thermal conductivity as the calculation result of Fig. 2.However, the MC simulation obtained lower thermal conductance ( Gth ) for NWs with largeangle sawtooth roughness than the value for flat, diffuse NW of a cross section the same as the smallest cross section of the sawtooth NW. Based on an analysis of photon scattering by a single sawtooth, Moore et al. [44] suspected that the added periodic sawtooth roughness can yield
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
strong phonon backscattering, which is equivalent to a negative p value. Although the backscattering effect by periodic sawtooth roughness can be used to explain the roomtemperature value measured for the 22 nm diameter Si NW by Li et al. [36], the effective calculated by the MC method is still considerably higher than those reported for etched Si NWs of a much larger diameter and smaller roughness to diameter ratio.
Kazan et al. [45] developed a model to fit the values reported for both VLS Si NWs and electrochemically etched rough Si NWs. One emphasis of their model is on scattering between optical and acoustic phonons. However, their l b term is not smaller than the Casimir result for diffuse surface scattering.
Martin et al. [46] developed a surface roughness scattering model based on the first-order perturbative approximation, or the Born approximation. When adding this surface roughness scattering term on top of a diffuse surface scattering term using Matthiessen's rule, Martin et al. [46] were able to fit the reported of rough Si NWs by adjusting the surface roughness to be 33.25 nm. By setting the surface roughness to be close to atomic corrugation of 0.1-0.3 nm, the model can also fit the reported of VLS Si NWs.
However, a recent report by Carrete et al. [47] suggested that the Born approximation is invalid when the surface roughness is comparable to the wavelength of phonons. Based on a Green's function calculation, these authors [47] reported that the Born approximation can overestimate the surface scattering by one order of magnitude, and that the Casimir formula yields a good approximation to the phonon mean free path. Carrete et al. [47] suggested that Casimir's limit in the mean free path can be overcome using large defects that penetrate deep in the NWs, but it is difficult to achieve a reduction in of more than one order of magnitude below that calculated based on diffuse surface scattering and the actual phonon dispersion
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
without compromising other properties including the structure stability. Based on the finding,
Carrete et al. [47] emphasized that a plausible explanation for the experimental results reported for rough Si NWs remain an open question for theory.
In addition, Sadhu and Sinha [48] suggested that surface scattering has been counted twice in Martin et al. [46]'s calculation that included the frequency-independent diffuse surface scattering along with the frequency-dependent scattering. Sadhu and Sinha [48] proposed that coherent effects, specifically, interference of scattered waves by multiple correlated points at the NW surface of a Gaussian-type roughness distribution, can result in mean free paths for highfrequency phonons well below the Casimir limit. Moreover, Sadhu and Sinha [48] emphasized that the surface roughness correlation length needs to be less than the NW diameter in order for the surface scattering mean free path to go below the Casimir limit. Based on this theory, these authors [48] attributed the much lower thermal conductivity observed in the electrochemically etched NWs [41] to a larger ratio between the roughness amplitude and the correlation length than that of the RIE etched NWs of Hippalgaonkar et al. [43]. Sadhu and Sinha [48] suggested that the room-temperature umklapp scattering mean free paths of the high-frequency phonons are much longer than the attenuation length due to surface scattering in ~50-nm-diameter electrochemically etched Si NWs. In addition, it was assumed in their work [48] that the thin oxide layer at the NW surface does not play a dominant role in surface scattering, or phonon transmission into and out of the amorphous oxide shell does not greatly diminish wave interference, so that the coherent scattering effect can be observed. It was suggested that the Bourret approximation used in their calculation [48] imposes less restrictive condition on the relative magnitudes of the phonon wavelength and the surface roughness compared to the Born approximation. Nevertheless, it was pointed out in that work [48] that there is no guarantee that
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
the Bourret solution will converge to the actual solution even if the approximation is valid, and that further studies are needed to examine scattering by non-Gaussian roughness statistics in actual NWs in the experiments.
Most recently, He and Galli [49] employed equilibrium molecular dynamics (MD) simulation to calculate the thermal conductivity of 15-nm-diameter Si NWs of different surface terminations, surface roughness, and core defects. High-temperature annealing and subsequent cooling was used in the MD simulation to reconstruct the NW surface so that the resulting surface can better match the experimental condition. For a 15-nm-diameter Si NW without surface amorphous layer or disorder, the calculated thermal conductivity was only a factor of three smaller than the bulk value. This reduction was attributed to the suppression in both the phonon lifetime and group velocity. Adding an amorphous Si or amorphous surface layer of 0.5 and 1 nm thickness leads to an additional reduction of the thermal conductivity by a factor of 3 and 4, respectively, so that the MD thermal conductivity is about one order of magnitude lower than the bulk value. These results reveal the important effect of the thin amorphous layer on phonon scattering by the surface, contracting the assumption made by Sadhu and Sinha [48]. Adding ripples of 0.5 and 1.5 nm depth to the NW with a 0.5 nm amorphous Si layer results in a reduction in the MD thermal conductivity by another factor of 1.5 [49], so that the MD thermal conductivity value would be similar to the calculation result shown in Fig. 2. In comparison, adding 1.5 nm ripples to a NW with a 1-nm-thick SiO2 layer leads to a thermal conductivity that is a factor of 30 lower than the bulk value. This value would be somewhat lower than the calculation result shown in Fig. 2, and fall between the value of the value obtained by Li et al. [36] and that by Boukai et al. [42]. However, only in the NWs with grain boundaries in the NW core, the MD thermal conductivity value can be a factor of 100 lower than the bulk value and
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
approach the data of Boukai et al. [42] for the ~10-nm-diameter Si NW. Based on these calculation results, He and Galli [49] proposed that core defects or large contact thermal resistance might have been present in those Si NWs where the measured thermal conductivity value was unusually low. These results suggest the need for additional experimental and theoretical studies for clarifying the relative contributions from the core defects and surface scattering.
## Nanomeshes and Phononic Crystals
Introduction of artificial spatial periodicity in the elastic properties of a material can in principle result in the appearance of bandgaps in the phonon frequency spectrum, and flattened folded optical branches of a low group velocity in a reduced Brillouin zone [50]. These effects on the phonon dispersion in the long-wavelength, low-frequency range may be observed directly in experiments. For example, a recent work reported Brillouin light scattering (BLS) measurements of low-frequency phonon dispersion even in a superlattice structures consisting of alternating porous Si layers of two different porosity values and with ~ 500 nm superlattice period [51]. Low-lying flat optical phonon branches of ~40 GHz frequency were observed by the BLS measurements. Moreover, thermal conductivity suppression measured in epitaxial superlattice structures was attributed to the reduced phonon group velocity of the folded phonon branches and the formation of phononic bandgaps in several works including ref. [52].
Recently, very low in-plane thermal conductivity values approaching the amorphous limit have been reported for periodic thin nanomesh Si membranes [53, 54]. The values are even lower than those measured for etched NW arrays with a similar critical dimension. In one work [54], the thermal conductivity reduction was attributed to phonon-surface scattering at the 23 nm wide necks in the periodic holey Si membrane structure of 55 nm pitch. The measured thermal
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
conductivity is lowered than the result of a MC phonon transport calculation [55] that considers diffuse surface scattering at the pore surface. However, the geometry considered in the MC simulation is different from that in the measurement. Hence, further MC calculations of the actual experimental structures would be helpful to verify whether the experimental values are indeed well below that caused by diffuse surface scattering alone. In another work [53], the low thermal conductivity is attributed to phononic bandgaps and reduced group velocity in the periodic nanomesh phononic crystal structure, which has a neck width of about 18 or 23 nm and a period of 34 nm. In a more recent work [56], the cross-plane thermal conductivity of 500-nmthick Si membranes with patterned periodic pores of 300 nm pore diameter and 500 nm period was measured with the use of a transient thermal reflectance method. The room-temperature value is 5.84 ± 1.3 W m -1 K -1 , about one order of magnitude lower than the result from a MC calculation [57] that assumes diffuse phonon-surface scattering. The measurement result was attributed to modification of the phonon dispersion and group velocity by the periodic phononic crystal structure.
However, the phonon wavelength spectrum is very broad. Although the spectral phonon thermal conductivity peaks at a somewhat longer wavelength than 0 , p given in Eq. (3), this wavelength is still much smaller than the period or pore size in the reported Si nanomesh structures, and may be comparable to the surface roughness of the pores. Diffuse scattering by random interface roughness can potentially destroy the phase coherence required for phononic bandgap formation. In fact, even for thin film epitaxial superlattice structures, atomic mixing at the interface is believed to result in partial diffuse scattering of short-wavelength phonons according to some theories [58]. This group of theories [58] treats phonons as incoherent particles, and considers interface scattering as the classical size effect. These models can explain
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
the experimental data reported for superlattices of periods thicker than ~5 monatomic layers. In addition, the ~200 nm wide spacing between pores in ref. [56] can be comparable to or shorter than the room-temperature umklapp phonon scattering mean free path. Thus, umklapp scattering can destroy phase coherence and phononic bandgap formation in the high frequency range important for thermal transport at room temperature. Hence, there exist opportunities to carry out further theoretical and experimental works to investigate whether phononic bandgap formation or another effect has played a dominant role in the very low thermal conductivity observed in these Si nanomesh structures.
## Progresses and Challenges in Thermal Measurements of Individual NWs
Most of the aforementioned measurements of individual NWs and nanomeshes were made using suspended resistance thermometer (RT) devices similar to or modified from the method reported earlier by Shi [59], Kim et al. [60], and Shi et al. [61]. Figure 3 shows a scanning electron microscopy (SEM) image of one of such devices and the thermal resistance circuit. Moore and Shi [62] recently provided a numerical heat transfer analysis of the systematic measurement errors of this method for suspended and supported NWs at different temperatures. Besides the systematic errors that can be large for supported NW samples, a major measurement uncertainty originates from the contact thermal resistance ( Rc ) between the NW sample and the two micro-thermometers. The Rc can be large. For example, Zhou et al. [63, 64] found that individual InAs NWs suspended between two micro-thermometers tended to crack near one membrane, possibly due to vibration. Such a crack was visible only in high resolution SEM images. With the presence of such a crack, one InAs NW essentially made a point contact to one of the two suspended thermometers, resulting in very large Rc .
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
In an effort to determine the Rc , Mavrokefalos et al. [65] developed a four-probe TE measurement procedure to obtain the intrinsic , S , and of the same individual NW suspended between the two micro-thermometers. In this procedure, the electrical contact between the four Pt electrodes and the NW sample are used as thermocouple junctions to determine the temperature drops at the two contacts, as illustrated in Fig. 3. This method has been successfully applied for measuring Bi2Te3 NWs [66], where clean electrical contact between the NW and the underlying Pt electrodes can be made by annealing of the sample in a forming gas. However, for some NWs such as Si NWs with a stable native oxide on the surface, making clean electrical contact between the NW and the Pt electrodes on the suspended device has remained difficult, preventing the use of the four-probe TE measurement procedure.
Recently, Hippalgaonkar et al. [43] have patterned Si NWs together with the microthermometers without an interface between the Si NWs and the Si membrane support under the Pt resistance thermometers. These authors [43] measured different patterned Si NW samples of different lengths, and determined that the junction resistance between the Si NW and the Si membrane from the measured thermal resistance versus length curve. The obtained of the etched Si NWs was slightly lower than the Casimir limit shown in Fig. 2, and considerably higher than those reported earlier for electrochemically etched Si NWs, as discussed above.
Besides the uncertainty due to the Rc , the cross section and of rough NWs are nonuniform along the NW and are not well defined. The narrower sections of the NWs make a larger contribution to the measured thermal resistance than the wider sections. If the NW diameter d is taken to be the peak-to-peak diameter, it can overestimate the cross section that scales with d 2 and underestimate . In addition, the Si NWs found with unusually low are all degenerately doped. Hence, it will be of value to characterize the doping and defect concentrations of the
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
measured NWs in future works and account for these factors in theoretical analysis. As suggested in a recent review [18], due to the sensitivity of single NW thermal conductivity measurements, additional independent characterization will be extremely helpful.
## PHONON TRANSPORT IN ISOLATED LOW DIMENSIONAL SYSTEMS
Theoretical Analysis of the Thickness Dependence of the Thermal Conductivity in Low-Dimensional Systems
As the thickness is reduced below the dominant phonon wavelength, such as in singelayer graphene (SLG) and single-walled CNTs (SWCNTs), the transition from 3D to 2D or 1D transport can reverse the thickness dependence of . In 2D or 1D systems, the allowable wave vector states lie on well separated planes or lines in the reciprocal space, or only one plane for the case of SLG and one line for a 1D monatomic lattice chain. Such quantization strongly restricts the phase space that can satisfy the energy and momentum conservation requirements for phonon-phonon scattering. In addition, the atomically smooth surfaces of suspended flat SLG or straight SWCNTs cannot scatter phonons diffusely because of the absence of states in the SLG or SWCNT with a velocity component perpendicular to the surface. Moreover, inter-layer scattering that reduces the basal plane of graphite is eliminated in the SLG and SWCNT. Hence, it has been suggested [67, 68] that these atomically thin carbon allotropes with strong sp 2 bonding can possess ultrahigh exceeding the previous record values found in diamond and graphite.
Because of the reflection symmetry in 2D suspended SLG, Lindsay et al. [69] has suggested an additional selection rule for phonon-phonon scattering processes involving the outof-plane transverse acoustic phonons, i.e. ZA or flexural phonons. This selection rule requires
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
that such processes need to involve an even number of ZA phonons, resulting in a very restricted phase space for umklapp scattering of the ZA phonons.
The ZA branch shows a quadratic dependence of on q , with a lower group velocity than the linear longitudinal acoustic (LA) and in-plane transverse acoustic (TA) branch. It is worth noting that the phonon branch with the lower group velocity can make a larger contribution to the specific heat and at low temperatures because of higher occupation of these phonon modes. For graphene, the thermal energy kBT at room temperature is still much lower than those of LA and ZA phonons of energy higher than the zone-boundary value of the ZA branch. When the phonon mean free path is constant and the same for all phonon modes, Equation (4) can be used to show that the ZA contribution to dominates the LA and TA contributions at room temperature and below in suspended flat SLG [70, 71]. The additional restrictive selection rule for ZA phonons serve to reduce ZA phonon scattering and increases its contribution to in flat suspended SLG [69, 70]. This theory conflicts with another calculation based on the relaxation time approximation (RTA) [68], which does not calculate the scattering phase space and yields a large umklapp scattering rate for ZA phonons.
The calculation of Lindsay et al. [72] shows that the of SWCNTs approaches that of flat 2D graphene at the large diameter limit. When the SWCNT diameter is reduced, the increased curvature breaks the reflection symmetry, leading to enhanced anharmonic scattering of ZA phonons and decreased . As the diameter is reduced further, the energy separation between adjacent 1D phonon energy subbands is increased, and optical phonons are pushed to higher frequencies. This feature lowers the scattering rate for low- acoustic phonons and increases their contribution to . These two competing effects result in a minimum in the d
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
curve, which occurs at dmin ≈ 1.5 nm for a given CNT length L = 3 m based on the calculation of Lindsay et al. [72].
## Theoretical Analysis of the Length Dependence of the Thermal Conductivity in Low-Dimensional Systems
Several theoretical studies have predicted that the value increases and diverges with increasing lateral size of 2D and 1D solids [73-75]. Under the Debye approximation, the phonon density of states ( D ( )) is proportional to 2 , , and 0 for the 3D, 2D, and 1D cases. Hence, as the dimension is reduced, the D ( ) for low- phonons is increased compared to that of high- phonons. Low- phonons with enhanced D ( ) in 1D and 2D are not subject to anharmonic scattering and are essentially ballistic carriers of heat. During a thermal measurement, when the two ends of the 1D or 2D system are in contact with two thermal reservoirs at different temperatures, the finite interface transmission coefficient ( ≤ 1) at the contact limits the contribution of the low- phonons to a finite value of G0 , where G0 is the ballistic thermal conductance [76] of these low-frequency phonons and is independent of the sample length ( L ). Hence, as L increases, the measured in-plane or axial is expected to increase. However, Mingo and Broido have suggested that both higher-order umklapp phonon scattering and defect scattering can be effective to prevent the divergence [77].
It is worth noting that the long umklapp scattering mean free path of long-wavelength phonons in clean, suspended graphene and CNTs creates a challenge for classical MD calculation of thermal transport in these low-D systems, because currently MD calculation can only handle short graphene ribbons or CNTs with a length smaller than the long phonon mean free path. As pointed by Mingo and Broido [78], the thermal conductance values obtained by some non-equilibrium MD calculations of thermal transport in short, suspended CNTs violate the
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
ballistic thermal conductance limit. A common practice to circumvent this problem is to extrapolate the MD-calculated thermal conductivity versus length data to the long length limit. However, the accuracy of this approach remains to be investigated. More importantly, classical MD calculations do not obey the Bose-Einstein statistics of phonon quasi-particles. This and other issues are expected to lead to future research opportunities in MD simulations.
## Thermal Conductivity Measurements of Suspended Carbon Nanotubes
For verifying the ultrahigh and intriguing transport phenomena suggested for CNTs, different methods have been reported for thermal transport measurements of individual CNTs. Kim et al. [60] fabricated an earlier version of the suspended micro-RT device shown in Fig. 3 for thermal measurements of multi-walled (MW) CNTs, and reported a room temperature value of about 3000 W m -1 K -1 for a 14-nm-diameter MWCNT, as shown in Fig 4. In this device [60] as well as a T-junction sensor [79], the thermal conductance Gth of the CNT sample assembled in the device was obtained from a steady state measurement. The contact thermal resistance problem can be a much more serious problem for these direct G measurements of high- CNTs than for low- NWs. Prasher [80] has shown that the lower effective measured by Kim et al. [60] at low temperatures in the MWCNT than in graphite was caused by a large Rc .
In a different approach, Lu et at. [81] demonstrated a 3 method based on modulated electrical heating of the sample for measuring the thermal conductivity and the specific heat of an individual Pt wire and a MWCNT bundle. This method was used by Choi et al. [82] for a 1.4 m-long MWCNT. However, the 3 and other self electrical heating methods cannot be applied readily to SWCNTs, because selective coupling between electrons and optical phonons can result in non-equilibrium phonon populations [83]. To account for non-equilibrium transport in a current-carrying SWCNT, Pop et al. [84] developed a coupled electron-phonon transport model
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
to fit the current-voltage characteristics of a SWCNT by adjusting the coupling constant between optical and acoustic phonons and contact thermal and electrical resistances. This approach was used to extract that the Gth of a 2.6 m-long SWCNT is inversely proportional to T at above 300 K [84].
Raman spectroscopy has been explored as a non-contact technique for probing thermal transport in carbon nanomaterials. The anti-Stokes to Stokes intensity ratio of a Raman peak directly yields the temperature of the Raman-active optical phonons in the laser spot [85] . This method can be used when the temperature of the optical phonons is as high as about 600 K, below which the anti-Stokes peak is too small to be measured accurately. Alternatively, the shift in the Raman peak position can be used to probe the equivalent local temperature of phonons that are involved in anharmonic scattering of the Raman-active optical phonons [86, 87] . However, the Raman peak position also depends on strain and impurity doping [86, 88] , which needs to remain constant during the thermal measurement in order to attribute the observed Raman peak shift to the temperature change alone. In addition, long-wavelength acoustic phonons can have a very long mean free path in clean, suspended CNTs and graphene. If these phonons do not interact with the photons or electrons effectively and are not involved in the anharmonic scattering processes of Raman-active optical phonons, they would be at a lower local temperature than the optical phonon temperature in optically heated or electrically biased CNTs and graphene. Indeed, non-equilibrium between different phonon populations has been observed by Raman spectroscopy in electrically biased CNTs [89-91].
There have been efforts of measuring the thermal conductance of CNTs with the use of micro-Raman spectroscopy. Hsu et al. [92] used the Raman peak shift to probe the local temperature and the ratio between the contact thermal resistance and the intrinsic thermal
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
resistance in optically heated individual suspended CNTs and bundles, where phonon transport was diffusive because of defect scattering. Deshpande et al. [93] employed this approach to profile temperature distribution along electrically biased, high-quality suspended SWCNTs. For a 2- m-long SWCNT, the Raman measurement revealed non-equilibrium between different phonon populations, and the Raman measurement result was used to determine the thermal conductivity contribution from the hot optical phonons with a non-negligible group velocity in the CNT. For a 5 m-long SWCNT, Deshpande et al. [93] reported that different phonon populations were approximately at local equilibrium and the thermal conductance was proportional to T at above 300 K. In another work, Li et al. [94] employed Raman spectroscopy to measure the T and G of current-carrying SW- and MW- CNTs where local thermal equilibrium among different phonon branches was established in the >30 m-long CNTs. In these Raman measurements of thermal transport in CNTs [92, 93], Rc could be determined from the obtained T profile. However, the Gth T dependence has not been obtained partly because the temperature resolution of the Raman measurements was 50 - 100 K and the temperature rise used in the experiments usually exceeded 100 K [93, 95]. In addition, it is important to conduct the Raman measurements in vacuum so as to eliminate heat loss from electrically or optically heated CNTs to surrounding gas molecules [96]. Moreover, the radiation loss needs to be properly accounted for in Raman measurements of long suspended CNTs.
These measurements of individual SWCNTs and MWCNTs have yielded values that vary by one order of magnitude, as shown in Fig. 4. The variation can be attributed to different sample qualities as well as the uncertainties in Rc and the cross section. T he cross section of the CNT sample was not characterized in detail in many of these measurements. The 14-nm diameter of the MWCNT sample reported by Kim et al. [60] was determined from a SEM
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
measurement, which can consist of an uncertainty of several nanometers and cannot determine the number of walls of the sample. MWCNTs and double-walled CNTs (DWCNTs) appear to be much brighter than SWCNTs in SEM images because of a larger scattering cross section [97]. Based on this feature, Yu et al. [98] thought that their CNT sample grown by chemical vapor deposition (CVD) was a SWCNT, and estimated that the diameter was in a large range between 1 and 3 nm. The CNT diameter of several other measurements was determined with the use of atomic force microscopy (AFM) measurement, which cannot determine the number of walls.
In two experiments, Fujii et al. [79] and Chang et al. [99] employed transmission electron microscopy (TEM) to characterize the cross section of individual MWCNTs assembled on the suspended micro- RT device or T-junction sensor for thermal measurement. In these works, the effective of the MWCNTs grown by arc discharge was found to increase with decreasing diameter, as shown in Fig. 5 This trend was attributed to reduced inter-layer scattering in . MWCNTs with decreasing numbers of walls. However, it is unclear whether Rc could have played a role on the measurement results. In particular, Rc is expected to make a larger contribution to the measured thermal resistance for a MWCNT with a larger diameter and smaller intrinsic thermal resistance. In a more recent combined TEM and G measurement of CNTs grown by CVD between two suspended RTs, Pettes and Shi [97] determined the Rc by comparing the measured G before and after Pt-C was deposited at the contact areas to the MWCNT samples. The obtained intrinsic values of the CVD MWCNT samples were also found to decrease with increasing diameter. However, the diameter dependence observed by Pettes and Shi [97] correlates well with the TEM observation of increasing defect density with increasing diameter of the CVD MWCNT samples.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
Despite the uncertainty in Rc , the values reported for the several MWCNT and SWCNT samples characterized by TEM are expected to be more accurate than those without. These combined TEM and G measurements yielded room-temperature in the range of 300-2900 W m -1 K -1 for MWCNTs grown by arc discharge [79, 99], of 45-310 W m -1 K -1 for CVD MWCNTs [97] , and about 600 W m -1 K -1 for CVD SWCNTs and DWCNTs [97]. Except for the CVD MWCNTs where the Rc was determined [97] , these values are the lower bound of the intrinsic because of the presence of Rc .
In addition, there have been experimental efforts for verifying theoretical prediction of length-dependent in CNTs. Based on micro-RT measurements, Change et al. [100] reported length-dependent in defective MWCNTs, which had a phonon mean free path shorter than 100 nm. However, Pettes and Shi [97, 101] were able to fit their data with the use of a lengthindependent and Rc given by a fin resistance model. In addition, using the self-heating 3 method, Wang et al. [102] reported length-dependent of SWCNTs supported on SiO2. However, the scanning thermal microscopy (SThM) measurement reported by Shi et al. [103] has revealed that most of the electrical heating in a supported SWCNT is dissipated through the SiO2 substrate instead of along the SWCNT. Consequently, the temperature rise in the currentcarrying SWCNT is much more sensitive to the unknown CNT-SiO2 interface thermal resistance ( Ri ) than the of the SWCNT. Hence, the values reported by Wang et al. [102] could have considerable uncertainty due to the unknown Ri as well as the uncertainty regarding whether local equilibrium was established in the electrically biased SWCNTs [104].
Hence, further development of the experimental capability for thermal measurements of individual CNTs is required to verify the theoretical predictions of length- or diameterdependence of the CNT thermal conductivity. In particular, while even the chiral index of one
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
measured CVD SWCNT has been obtained from selected area electron diffraction analysis [97], measuring Rc of individual CNTs remains an important area for future study.
## Thermal Conductivity Measurements of Suspended Graphene
For measuring thermal transport in graphene, Balandin et al. [105] used a Raman laser beam to heat a rectangular SLG flake suspended across a trench. The temperature at the laser spot of the graphene was determined from the Raman G peak shift. In the initial measurement [105] , the thermal conductivity was assumed to be the same for the suspended region and the supported region of the same graphene monolayer. The thermal coupling between the graphene monolayer and the underlying SiO2 support at the two ends of the trench was ignored. The thicker graphite regions at the two ends of the graphene monolayer were assumed to be perfect heat sinks. The optical absorbance of the graphene monolayer was determined by comparing the Raman peak intensity value measured on the SLG and that on graphite. This method obtained an optical absorbance of about 6.5% per pass of the laser beam through the SLG. Based on the measured graphene temperature and optical absorbance, the near-room temperature of suspended SLG was obtained to be (4.84 ± 0.44 to 5.30 ± 0.48) × 10 Wm 3 -1 K -1 [105-107], exceeding those of diamond and graphite.
In comparison, Nair et al. [108] used d irect transmission measurement to determine the optical absorbance of SLG to be 2.3%. In the experiment, the reflectance was ignored because it is expected to be much smaller than the absorbance based on a theory. Using this 2.3% value, Faugeras et al. [109] measured a value of about 600 Wm -1 K -1 for a SLG sample suspended over a 44- m-diameter hole when the Raman laser beam raised the SLG temperature at the laser spot to 660 K, which was determined from the anti-Stokes to Stokes ratio. Their circular Corbino membrane geometry matches the laser beam intensity profile and heat flow direction better than
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
the rectangular flake geometry. Because of a larger contact area with the support, in addition, the contact thermal resistance is expected to be less an issue for the circular membrane geometry than for the rectangular flake geometry. Faugeras et al. [109] suggested that the discrepancy between their values and those reported by Balandin et al. [105] was caused by the different assumptions in the optical absorbance value.
In two other Raman measurements by Cai et al. [110] and Chen et al. [111], the optical absorbance of SLG samples grown by CVD and suspended over holes of 2.9-9.7 diameters was obtained to be 3.3 ± 1.1% based on the measured transmission. Based on this optical absorbance, Cai et al. [110] reported that the measured of the suspended graphene decreases from (2.5 ±1.1) × 10 Wm 3 -1 K -1 to (1.4 ± 0.5) × 10 Wm 3 -1 K -1 when the temperature of the optically heated SLG determined from the Raman G peak shift increased from about 350 K to about 500 K. The results of Chen et al. [111] are similar and are shown in Fig. 6 together with the basal plane thermal conductivity recommended in ref. [112] for pyrolytic graphite (PG). In addition, with the optical absorbance assumed to be 2.3% for their SLG samples exfoliated from natural graphite directly onto holes of 2.6-6.6 m in diameter, a recent Raman measurement by Lee et al. [113] yielded values similar to the reported values for PG.
Hence, an unresolved issue in these Raman measurements of suspended SLG is the different optical absorbance values used in different works. The recent result by Mak et al. [114] indicates that the optical absorbance can be larger than 2.3% for a certain wavelength range. If the relationship between the optical conductivity and the optical absorbance still obeys the theoretical prediction, the optical absorbance extracted from their result would be close to 3.3% for the 480-515 nm wavelength range of the Raman laser. In addition, polymer or tape residues on the suspended graphene can increase both the absorbance and reflectance of the sample.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
However, it is unlikely that tape residues can considerably increase the Raman G band intensity used in the earlier work [105] to determine the optical absorbance. Moreover, the results obtained based on the Raman peak shift can be complicated by the effect of strain and impurity on the Raman peak position, as discussed in the preceding section. In addition, if the phonons probed by the Raman laser are thermalized locally within a scale comparable to the laser spot, the temperature distribution in the graphene can be solved from the heat diffusion equation with a Gaussian heat source term [109, 110]. The solution relates the contribution from these phonons with the measured thermal resistance ( Rm ), defined as the ratio between the Ramanmeasured temperature rise and the optical absorbance of the graphene. If the mean free paths of these phonons are much larger than the laser beam size so that they are not thermalized locally within the laser spot and thus do not contribute fully to thermal transport , there would be an additional ballistic resistance component ( Rb ) in the measured Rm [110].
Hence, the reported graphene thermal conductivity obtained by the Raman technique contains rather large uncertainty caused by the unresolved issue in the optical absorbance, the effects of strain and impurity doping, low temperature sensitivity of more than 20-50 K of the Raman thermometry technique, and the ballistic thermal resistance. Moreover, these Raman measurements have not been employed to probe the graphene thermal conductivity at low temperatures. It is worth noting that the temperature in the Raman data of Fig. 6 is the hot side temperature in the graphene probed by the Raman data, whereas the cold side temperature was kept at near room temperature. This issue makes it nontrivial to compare the obtained graphene thermal conductivity from the Raman technique with those reported for graphite, where a small temperature drop across the sample was used during the thermal measurement and the temperature reported is usually the average temperature of the sample. These issues suggest that
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
there are abundant opportunities to further develop the Raman and other experimental techniques for thermal conductivity measurement of clean suspended graphene samples.
## INTERACTION BETWEEN PHONONS IN LOW DIMENSIONAL SYSTEMS AND THE ENVIRONMENT
## Effect of Interface Interaction on Thermal Conductivity of Nanotubes and Graphene
Low-dimensional materials such as CNTs and graphene are often supported on a substrate for device applications or embedded in a matrix to form functional nanocomposites. Electron transport in low-D materials such as graphene and CNTs is known to be highly sensitive to the environment. For example, the electrical conductance and Seebeck coefficient of CNTs were found to be very sensitive to adsorbed gas molecules [115]. The electron mobility in graphene supported on an oxidized Si wafer is considerably lower than those for clean suspended graphene [116] and graphene supported on a clean hexagonal boron nitride (h-BN) layer [117].
Several theoretical studies [118, 119] have suggested that the of CNTs embedded in a medium can be rather different from those of free-standing and clean CNTs because of scattering of phonons and even modificiation of the phonon dispersion in the CNTs. Experimental verification of these theoretical predicitons for CNTs has remained a challenge. However, the Raman measurement of Cai et al. [110] has obtained that the near room-temperature of the gold-supported region of their CVD SLG sample was about (370 + 650/-320) W m -1 K -1 , considerably lower than the (2500 + 1100/-1050) W m -1 K -1 value measured in the suspended region of the same sample. Moreover, Seol et al. [70] fabricated a suspended resistance thermometer micro-device for measurements of SLG supported on amorphous SiO2, and found
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
a room-temperature value of about 600 W m -1 K -1 , which is considerably lower than those reported for PG and suspended SLG, as shown in Fig. 6. Raman spectra measured on the supported SLG samples show no presence of the D peak caused by defects. The mobility extracted from the measured S and is comparable to the highest value reported for SLG supported on SiO2. Because these results suggest high-quality SLG samples used in their thermal measurement, Seol et al. [70] attributed the lower of the supported SLG to SLG-SiO2 interaction.
The interface interaction can impact phonon transport in supported graphene in different ways. The first one is to modify the phonon dispersion of the supported graphene. Although the phonon dispersion of graphene grown on Ni and Ru is altered considerably by the strong bonding to the metal substrate, the acoustic phonon dispersion is similar for suspended graphene, graphene physisorbed on Pt, and for graphene grown on Ni after Cu, Ag, or Au is intercalated under the graphene [120-125]. In addition, inelastic neutron scattering of ordered pyrolytic graphite [126] and electron energy loss spectroscopy of thin randomly stacked graphite [127] have shown the phonon dispersion of these materials is similar to 3D Bernal graphite [128] and 2D graphene, respectively, with the observation of the ZO , TO , and LO modes only in the ordered pyrolytic graphite sample. Thus, weakly interacting graphitic layers had been modeled in the literature with dispersions calculated for 2D graphene [129, 130]. Therefore, modification of the phonon dispersion is not expected to play a major role in the observed reduced thermal conductivity in supported graphene, compared to scattering of graphene phonons by the support.
The support can scatter phonons in graphene via several mechanisms. The interface coupling can lead to local perturbation in the diagonal force constant on the graphene atoms. In addition, the atoms in the SiO2 support are not static. The interaction between vibrating atoms at
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
two sides of the interface results in phonon scattering across the interface, although there is no net energy transfer across the interface when the temperature gradient is parallel to the interface.
Perturbation theory was employed in ref. [70] to obtain the total scattering rate due to both scattering across the interface and the local perturbation of the diagonal force constant on the graphene atoms. The obtained substrate scattering rate increases rapidly with decreasing phonon frequency. This frequency dependence is similar to that for th e phonon transmission coefficient across a weak van der Waals interface [131]. In addition, the interface force constant is stronger for the ZA modes than for the in-plane modes when the support is amorphous [70]. When this factor and the rather conformal morphology of supported graphene are taken into account , the substrate scattering model of ref. [70] can explain the measured versus T relation for the supported SLG, and suggests much stronger substrate damping of the ZA phonon lifetime than for the in-plane modes. Consequently, the ZA contribution to the thermal conductivity is smaller than those from the LA and TA branches in the supported SLG [70]. The dominant role of the substrate in damping the ZA modes in graphene instead of the in-plane modes is also observed in a recent MD simulation [132]. In a follow-up MD calculation of thermal transport in a CNT supported on SiO2 [133], the low-frequency phonon lifetime was found to be suppressed most noticeably by the support. This finding is in qualitative agreement with the frequency dependence given by the substrate scattering model of ref. [70]. Moreover, artificially freezing the atoms in the SiO2 support in the MD simulation was found to have negligible effect on the CNT phonon lifetime. This latter result suggests that substrate scattering of phonons in CNTs is dominated by static perturbation of the potential in the CNT. However, the calculation with frozen SiO2 atoms was not included in the earlier reported MD simulation of supported graphene [132]. It remains to be investigated whether the conformal morphology and high interface
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
thermal conductance [134] of supported graphene may increase the role of phonon scattering across the interface compared to the case of supported CNTs. In addition, interaction with the support also breaks the reflection symmetry of graphene, and thus relaxes the restrictive selection rule for three-phonon scattering involving the ZA phonons. This effect has not been accounted for by the substrate scattering model of ref. [70].
It is of considerable interest to employ the sensitive micro-resistance thermometer devices to obtain the accurate versus T relation of suspended graphene. Recently, some successes have been reported in the assembly of a 0.5 μm-long suspended SLG [135] and a 1 μm-long suspended 5-layer graphene [136] between two suspended micro-thermometers. The measured exhibits a T 1.5 dependence at temperatures lower than 150 K for these two suspended samples, and was attributed to a dominant contribution from the ZA modes. However, the obtained room-temperature of these samples is lower than 225 W m -1 K -1 , considerably lower than the theoretical prediction for the ZA contribution.
Meanwhile, Pettes et al. [71] developed a method to assemble suspended bi-layer graphene (BLG) samples on suspended resistance thermometer devices. The measured values of two suspended BLG samples are slightly higher than the results of Seol et al. [70] for SLG supported on SiO2, and considerably lower than the Raman measurement results and theoretical predictions for clean, flat, suspended graphene of similar dimensions, as shown in Fig. 6. An approximately T 1.5 dependence can be observed in the data of the BLG samples at temperatures below 125 K.
In 2D and the low T limit, if the phonon scattering rate 1 p is proportional to ω α , Equation 4 can be used to show that the contribution to is proportional to T 2α for both the quadratic ZA branch and the nearly linear LA and TA branches in graphene. If the phonon mean
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
free path is constant because edge or grain boundary scattering is the dominant phonon scattering process, the LA and TA contribution to is proportional to T 2 , while the ZA contribution is proportional to T 1.5 at low temperatures. Although a dominant ZA contribution and dominant phonon-edge scattering can lead to the T 1.5 dependence, Pettes et al. [71] calculated that the phonon mean free path needs to be much smaller than the sample lateral dimension or reported grain size of similar graphene samples in order to match their measured or those reported in [135, 136] . Hence, instead of this mechanism, Pettes et al. [71] attribute their result to phononscattering by a residual polymeric layer that can be clearly observed by TEM on the suspended BLG. These authors [71] noted that t he umklapp scattering with a positive α on the LA and TA modes together with stronger substrate scattering on the ZA modes give rise to an approximately T 1.5 behavior at low temperatures in the calculation result of Lindsay et al. [70] for the SLG supported on SiO2. Hence, the T 1.5 behavior observed in their BLG samples with polymer residues could be caused by a similarly complex scattering mechanism. Nevertheless, it is also possible that the polymer residue was distributed on the graphene surface as separated patches of lateral size larger than the dominant phonon wavelength, so as to give rise to geometric scattering with a constant mean free path. In this case, the ZA contribution can be larger than the LA and TA contribution, resulting in the T 1.5 behavior at low temperatures. In either case, while the finding of Pettes et al. [71] is useful for the design of graphene-polymer nanocomposites for thermal management, it also points to the need of obtaining ultra-clean suspended graphene samples for fundamental studies of the intrinsic thermal properties of 2D graphene.
In addition, Jang et al. [137] used a different resistance thermometry method to measure the in-plane of few-layer graphene (FLG) embedded in SiO2. These authors [137] reported that
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
the thermal conductivity of the encased FLG increases with the number of layers, as shown in
Fig. 7. This result suggests stronger suppression of the graphene-SiO2 interface interaction than inter-layer interaction in FLG. In comparison, Ghosh et al. [138] reported that the of suspended FLG decreases with increasing number of layers because of interlayer interaction, and approach the graphite limit when the number of layers increases to about ten layers.
## Interface Thermal Conductance of Nanotubes and Graphene
Besides the effects of interface interaction on phonon transport in low-dimensional materials, another property of importance for practical applications is the interface thermal conductance ( gi ) between the low-D materials and the environment. Transient laser reflectance has been employed to determine that the g at CNT-micelles interface is about 12 x 10 W m 6 -2 K -1 and rather small [139]. Electrical breakdown measurements [140, 141], SThM [103], and suspended thermometer devices [97] have been used to determine this property for CNT-solid interfaces. The values obtained from different studies vary over a wide range because of variation in the surface roughness and interface adhesion energy [80, 131], which makes it difficult to determine the contact area accurately.
Interface thermal transport across graphene has been probed with the use of different methods. For graphene-SiO2 interface, the measured gi value is (83-178) x 10 W m 6 -2 K -1 based on a steady state resistance thermometry measurement of encased graphene [134], and (20-110) x 10 W m 6 -2 K -1 for supported graphene exfoliated on SiO2 measured by a transient layer reflectance technique [142]. A Raman measurement yielded gi of (28 + 16/-9.2) x 10 W m 6 -2 K -1 for graphene-Au interface [110] . Another transient laser reflectance measurement obtained 25 x 10 W m 6 -2 K -1 for the Au/Ti/graphene/SiO2 interface [143]. Although these values are somewhat different, the variation is small compared to those reported for CNT-solid interfaces.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
## INTERFACE SCATTERING OF ELECTRONS
## Effect of Surface Scattering on Charge Mobility of Nanowires
Interface scattering of electrons is a complicated process that depends on the interface roughness and charge states. Although NWs provide a model system to investigate the classical size effect of interface scattering of electrons, measuring the electron mean free path in the NW geometry has remained a challenge. Hall measurement has been established for measuring the mobility of thin films. However, this method cannot be applied readily to the NW and CNT geometry. Instead, electron mobility has been extracted from field effect measurement results of CNT transistor devices [144]. However, this method has limited use for degenerately doped NWs explored for TE applications because of the screening of the field effect by the high electron concentration in the NWs. The thermopower itself has been used to probe the electronic structure in bulk semiconductors [145] and in semiconductor junctions [146]. Based on a similar principle, Moore and co-workers [66, 147-149] have employed a two-band or one-band model to extract the Fermi level, carrier concentration and mobility from the measured S and of individual NWs. The as-obtained mobility for some InSb NWs [149] and Bi2Te3 NWs [66] was found to be reduced considerably compared to the bulk values. The mobility reduction is comparable to the l suppression measured on the same Bi2Te3 NWs [66]. For InSb NWs [149] the mobility was reduced more than for the lattice thermal conductivity. These results may be caused by surface charge states or surface depletion in these NWs, where the small m* results in a small concentration of charge carrier that is insufficient to screen the surface charges.
For largem* semiconductor NWs, the charge carrier concentration is large, and the bulk electron mean free path is often very short and smaller than the typical NW diameters in the 30-
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
100 nm range. Surface charges may also be effectively screened by the large carrier concentration in largem* NWs. Consequently, little mobility reduction has been reported in CrSi2 NWs of m*≈ 5m0 [147]. Similar reasons could have led to much less reduction of charge carrier mobility than l suppression in Si NWs [41, 42].
## Effect of Surface Scattering on Seebeck Coefficient of Nanowires
Besides impact on charge mobility, increased interface scattering can influence the Seebeck coefficient due to effects such as that on the r exponent in the energy dependence ( E ) of the charge carrier scattering relaxation time, E r . The Seebeck coefficient consists of a diffusion component ( Sd ) and a phonon drag contribution ( Sp ) [17], i.e.
$$S = S _ { d } + S _ { p }$$
The diffusion component for a single parabolic band in 3D systems is given as
$$S _ { d } = \frac { 1 } { q } \left ( k _ { B } \frac { \left ( r + \frac { 5 } { 2 } \right ) F _ { r + 3 / 2 } ( \xi ) } { \left ( r + \frac { 3 } { 2 } \right ) F _ { r + 1 / 2 } ( \xi ) } - \frac { \zeta } { T } \right )$$
where q equals e for holes and e for electrons, e is the elemental charge, equals ( Ef -Ec ) for electrons and ( Ev-Ef ) for holes, Ef , Ec , and Ev are the Fermi level, conduction band edge, and valence band edge, respectively, and F n ( x ) is the Fermi-Dirac integral of order n [17] . I n the metallic limit of a large , the expression can be simplified to be a linear function of r and T according to:
$$S _ { d } = \frac { \pi ^ { 2 } } { 3 } \frac { ( r + \frac { 3 } { 2 } ) } { q \zeta } T$$
For non-degenerate bulk semiconductors with k , the diffusion component of a single carrier type is reduced to a relatively weak function of the r exponent as
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
$$S _ { d } = - \frac { 1 } { q } \left ( \frac { \xi } { T } - k _ { B } \left ( r + \frac { 5 } { 2 } \right ) \right )$$
Herring [150] considered preferential momentum transfer from electrons to phonons along the direction of the current flow in an isothermal semiconductor, and obtained an additional phonon heat flux besides the electronic heat flux that was considered in the derivation of Eq. (12). These two heat flux terms are normalized by the current density to obtain the phonon drag and electronic contributions to the Peltier coefficient ( ), namely p and e . Based on the Kelvin relation, Herring [150] obtained:
$$S _ { p } \approx \frac { e \beta \nu _ { p } l _ { p } } { q \mu T }$$
where vp and l p are the group velocity and mean free path of those long-wavelength phonons with a wave vector of the order of the Fermi wave vector that scatter charge carriers, β is a parameter that characterizes the strength of the charge carrier-phonon scattering, and is the mobility of the charge carriers.
As shown in Fig. 8, the Sp component can be large and give rise to a large peak in the S-T curve for temperatures between 30 and 150 K in p-type Si with doping concentration of 1 x 10 18 cm -3 or lower [151], where l p is long and is limited by charge carrier-phonon scattering. At lower temperatures, the phonon population is small and l p is limited by boundary scattering, so that Sp is small. At higher temperatures, l p and Sp are reduced by increasing phonon-phonon scattering with increasing T . In highly doped Si, l p is reduced considerably by impurity scattering and charge carrier scattering, the latter of which also leads to momentum transfer back from phonons to charge carriers. These effects result in a small phonon drag contribution observed in p-type Si crystals with a doping concentration of 1 x 10 19 cm -3 [151], where S is dominated by Sd . In addition, decreasing grain size is known to reduces l p and Sp [150].
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
Hence, it is surprising that Boukai et al. [42] observed a pronounced S peak due to phonon drag at T near 200 K in a 20 nm Si NW doped with a high boron concentration of 3 x 10 19 cm -3 . These authors [42] attributed this result to a dimension crossover from 3D to 1D for those long-wavelength phonons that scatter charge carriers. It was suggested that the wavelengths of these phonons are longer than about 3 nm and comparable to the NW diameter, so that there are few transverse wave vector states into which the NW boundary can scatter these phonons. In comparison, the spectral energy density peaks at a phonon wavelength much smaller than the 20 nm diameter of the NWs, based on Eq. (3). Boukai et al. [42] suggested that the dimension crossover of the long-wavelength phonons responsible for the phonon drag still permits the very low measured values because of surface scattering of the shorter-wavelength phonons.
These findings are indeed very intriguing, and have stimulated further theoretical and experimental studies to better understand the phonon drag effect in nanostructures. A recent report [152] has suggested observable phonon drag contribution in the measured S of the inversion layer of a 20-nm-thick, 420-nm-wide Si nanoribbon where the carrier density was tuned by a back gate voltage applied to the Si substrate underneath the nanoribbon. However, the measured S at temperature 300 K, 200 K, and 100 K decreases with decreasing temperature, and does not show a pronounced phonon drag peak at low temperature. Based on the difference between the measured S and the calculated Sd , the relative contribution of the phonon drag was determined to increase from 8.3% at 300 K to 46% at 100 K. These values are much smaller than those reported for the etched Si NW of a 20 nm x 20 nm cross section [42], as well as those in Si crystals with a comparable doping concentration of order 10 18 cm -3 or 1ower [151].
For n-type bulk Si of a doping concentration of 2.7 x 10 19 cm -3 [151] and 1.7 x 10 19 cm -3 [153], respectively, the measured S data were found to decrease linearly with T and extrapolates
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
to zero at T = 0 K, resembling the metallic behavior given by Eq. (11). Boukai et al. [42] observed a similar linear metallic behavior for patterned p-type Si NWs of doping concentration higher than 7 x 10 19 cm -3 . In comparison, although a linear S-T behavior was reported by Hochbaum et al. [41] for a highly-doped 48 nm electrochemically etched p-type Si NW, the S T -curve extrapolates to a negative value of a rather large magnitude at T = 0 K, as shown in Fig. 8.
In an attempt to understand the different S T -trends observed in the bulk Si and Si NWs, the Sd for p-type Si of different boron doping concentrations has been calculated and shown in Fig. 8. The calculation is based on a two band model that accounts for the contributions from both electrons and holes. For different doping concentrations and temperatures, the Fermi energy is determined based on charge neutrality and Fermi-Dirac statistics without approximating it as the Boltzmann distribution. The value suggested by Herring [150] has been used, along with a r mobility ratio of two between electrons and holes. The model reproduces the reported Sd results of a single band model for non-degenerate Ge [150] and Si [151]. The calculated Sd T curves in Fig. 8 at T > 200 K show a smaller slope than the NW measurement data, and a divergence at the low T limit. The divergence is caused a Fermi level settled near the middle between the band edge and the dopant energy level in the carrier freeze-out regime. In comparison, Geballe and Hull's [151] measurement result for a Si crystal of 1.8 x 10 18 cm -3 boron doping decreases with T and shows a sign change at low T , as shown in Fig. 8. This behavior in the carrier-feezeout regime has been attributed to impurity band conduction [151], which yields an additional contribution to Sd of an opposite sign. Neither the impurity conduction band mechanism nor the phonon drag contribution has been accounted by the calculation results in Fig. 8. It remain to be investigated whether the large slope shown in the reported S T -curve of the highly-doped 48 nm electrochemically etched p-type Si NW is caused by these intriguing transport effects or other
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
experimental factors. Similarly, further theoretical studies are also needed for better understanding the measured S T data of the polycrystalline Si samples [154] shown in Fig. 8.
In addition, although the S of individual CNTs and NWs can be conveniently measured with electrodes and resistance thermometer lines patterned on the CNT or NW [155, 156], it is worth noting that the average temperature measured by the thermometer lines can be different from the temperature at the contact point between the thermometer line and the NW. It would be necessary to use a numerical heat transfer analysis to evaluate and compensate for this error [148]. Moreover, for measuring the S of an array of NWs between two thermometer lines [42], the excitation current used to measure the resistance of one of the thermometer lines can leak through the NW array to the other thermometer line and return to form an unwanted current loop. This issue can cause uncertainty in the measured thermometer line temperature, especially when the electrical resistance of the NW arrays is not much larger than that of the thermometer line.
## THERMOELECTRIC TRANSPORT IN LOW DIMENSIONAL CONDUCTORS
In 1993, Hicks and Dresselhaus [157, 158] proposed that the sharply peaked electronic density of states (DOS) in 2D quantum wells and 1D quantum wires can be used to increase S 2 and ZT . These works stimulated a number of subsequent theoretical studies. In one of them, Broido and Reinecke [159] suggested that the ZT enhancement in realistic quantum well thin film superlattice systems can be limited by several factors, including electron tunneling through the barrier layers that modifies the DOS and limits the power factor, increased charge carrierphonon scattering rates and reduced charge carrier mobility, and parasitic thermal conduction in the barrier layers. Broido and Reinecke [159] further suggested that freestanding structures such as freestanding NWs may be needed to overcome these limitations in order to obtain a large ZT .
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
Several other calculations by Lin et al. [160] and Mingo [161] predicted that ZT larger than 2 at 77 K and 1.5 at 300 K, respectively, is possible in Bi and InSb NWs with diameter smaller than 10 nm. In addition, the calculation by Rabina et al. [162] suggested that ZT larger than 2 at 77 K could be achieved in Bi1-xSbxNWs of diameter less than 20 nm. These materials all possess small m* and thus long e , giving rise to quasi-1D DOS in NWs of diameter smaller than 20 nm. Most recently, Kim et al. [163] suggested that the power factor contribution from each conduction channel in 1D systems can be only 12% and 40% higher than the corresponding ones in 2D and 3D, respectively, so that the thickness of quantum wells and wires must be very small to achieve a higher power factor in lower dimensions after the power factor per channel is renormalized into power factor per cross section.
Many of these calculations are based on the assumption that there is no mobility reduction due to diffuse scattering of electrons. Without accounting for surface charge states, Ziman's expression can yield a larger p for long- e electrons in smallm* semiconductors than for phonons with a short p . Hence, it is in principle possible to suppress the phonon mean free path more than the electron mean free path in NWs of semiconductors with a smallm *.
These theoretical works have motivated a number of experimental studies of thermoelectric transport in thin film and NW structures. There have been several experimental reports of ZT enhancement in thin film superlattice structures [164, 165] and NWs [41, 42]. However, the enhancement has been attributed mainly to suppressed l instead of enhanced S 2 . In addition, for the case of thin film superlattices [164, 165], the ZT enhancement was reported for the cross-plane transport direction instead of the in-plane direction considered in earlier theoretical studies. As mentioned in the section on phononic crystals, interface scattering of phonon quasi-particles is believed to be the main cause of the reduced l in the cross-plane
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
direction of thin film superlattices, although other intriguing wave behaviors such as phonon mini bands and phonon localization in thin film superlattices have also been discussed [166, 167].
In addition, Heremans et al. [168] measured large S in Bi NWs, and attributed the result to semi-metal to semi-conductor transition due to the lifting of the band edges of 1D energy subbands. However power factor enhancement has not been observed in these existing experimental works on quantum wire or well structures, likely because interface charge states and interface scattering suppress and . One notable exception is the extraordinary electron mobility reported by Shim et al. [169] for a 120-nm-diameter single-crystal Bi NW, where a reduced carrier concentration was reported and could be caused by semi-metal to semiconductor transition. Another recent report [152] suggested a significantly enhanced power factor in the inversion layer of a 20-nm-thick Si nanoribbon where a gate voltage was applied to the Si substrate under the ribbon to tune the carrier concentration in the ribbon. In that work [152], no mobility suppression was found in a wider ribbon of the same thickness because of the lack of surface disorders and interface trapped charges compared to etched Si NWs. The power factor enhancement was attributed to confinement of the essentially two-dimensional electronic gas (2DEG) in the inversion layer, as well as an appreciable phonon drag contribution discussed above. However, because the inversion layer is one order of magnitude smaller than the thickness of the nanoribbon, the effective power factor of the entire ribbon would be still rather low. Nevertheless, this work [152] has suggested the feasibility of employing the field effect to tune the Fermi energy of electrons confined in nanometer or atomic layers for experimentally verifying the different theories on quantum enhancement of the thermoelectric power factor.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
## SUMMARY
Significant progresses have been made in both theoretical and experimental studies of the quantum and classical size effects on thermal and TE properties. However, a number of fundamental questions on thermal and TE transport in nanostructures and low-D systems have remained elusive. One of the outstanding questions is whether and how much the l of NWs and nanomeshes can be reduced below that limited by the Casimir result of diffuse surface scattering. In addition, experimental results are still lacking for verifying the predicted l dependence on the length of clean suspended CNTs and graphene and on the diameter of the CNTs. For TE transport, detailed experimental and theoretical studies are needed to better understand whether and how much the power factor can be enhanced in quantum wells and wires, and whether large phonon-drag thermopower can indeed be observed in NWs and other low-D systems.
High-fidelity measurement techniques are critical for addressing these outstanding questions. It is worth noting that the measurement of the three properties entering the ZT expression is more challenging for individual nanostructures and thin films than for bulk crystals, which already requires care to address a number of issues. This challenge can lead to large uncertainty and discrepancy in the measured properties. For example, recent measurements have found neither S 2 enhancement of the power factor [170] nor l suppression [171] in PbSeTe/PbTe nanodot thin film superlattices, for which enhanced ZT was reported in an earlier report [165]. Hence, independent verification of reported measurement properties of individual nanostructures and thin films would benefit a broad research community.
For measurements, techniques for eliminating the contact thermal resistance and for correcting the parasitic radiation loss at high temperatures are particularly important. Although
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
and S measurements are not as complicated as for , sufficient cares are still necessary. As a simple example, even for four-probe electrical measurement of individual NWs, the width of the two voltage electrodes can introduce considerable uncertainty in the NW length and the obtained electrical conductivity when the width is not small compared to the NW length between the two voltage electrodes. Moreover, although electric field effect has proved to be a promising approach to tuning the carrier concentration in the nanostructure sample during thermoelectric measurements, new methods are needed to control the impurity concentration and surface charges and surface roughness so that the delicate quantum confinement effects are not obscured by these non-ideal factors.
Besides challenges in experimental study of thermal and TE transport in the nanoscale, current theoretical models can only deal with ideal systems with small number of atoms or simple band structures, and cannot account for the larger size or more complex lattice or electronic structures that have been investigated in experiments. The disconnect between current experimental and theoretical capabilities suggests abundant future research opportunities to reach down to clean low-dimensional samples in experiments and handle large and complex nanostructures in theoretical models.
## Acknowledgement
The author acknowledges collaborations and helpful discussions with a number of colleagues for related research. A partial list includes Alex Balandin, David Broido, David Cahill, Gang Chen, Steve Cronin, Chris Dames, Mildred Dresselhaus, I-Kai Hsu, Rui Huang, Song Jing, Insun Jo, Philip Kim, Deyu Li, Xiaoguang Li, Lucas Lindsay, Arun Majumdar, Anastassios Mavrokefalos, Paul McEuen, Alan McGaughey, Natalio Mingo, Arden Moore, Eric
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
Pop, Michael Pettes, Ravi Prasher, Rodney Ruoff, Jaehun Seol, G. P. Srivastava, Zhen Yao, Laura Ye, Choongho Yu, and Feng Zhou. The author would like to thank Kenneth Goodson for his encouragement and editorial support during the preparation of this manuscript. The author's work in this area has been supported by the Thermal Transport Processes Program of National Science Foundation, Office of Naval Research, and Department of Energy Office of Basic Energy Science. Opinions and views expressed in this review do not necessarily reflect those of the funding agencies or the aforementioned colleagues. Although it can be expected that mistakes will be found in this review in the future, the objective of this review is to stimulate further discussions of several intriguing fundamental questions, as supposed to diminish the different viewpoints that are either included or not included in this review.
## NOMENCLATURE
- E energy [J]
- gi thermal interface conductance [W m -2 K -1 ]
- Gth thermal conductance [W K -1 ]
- G0 ballistic thermal conductance [W K -1 ]
- kB Boltzmann constant
- l mean free path [m]
- m * effective mass [kg]
- m 0 electron rest mass [kg]
- N number of modes
- p surface specularity parameter
- Rc contact thermal resistance
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
- S Seebeck coefficient [V K -1 ]
- T temperature [K]
- v velocity [m/s]
- Z thermoelectric figure or merit [K -1 ]
## Greek
- thermal conductivity [W m -1 K -1 ]
- electrical conductivity [ -1 m -1 ]
- wavelength [m]
- ħ reduced Planck constant
- relaxation time [s]
- chemical potential measured from the band edge [J]
- transmission coefficient
- Peltier coefficient [V]
## Subscripts
- b boundary scattering
- c contact or conduction band edge
- d diffusion
- f Fermi level
- i intrinsic scattering
- l lattice
- p phonon or phonon drag
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
- u umklapp scattering
- v valence band edge
## REFERENCES
- [1] A. Shakouri, "Nanoscale Thermal Transport and Microrefrigerators on a Chip," Proceedings of the IEEE, vol. 94, pp. 1613-1638, 2006.
- [2] D. G. Cahill, W. K. Ford, K. E. Goodson, G. D. Mahan, A. Majumdar, H. J. Maris, R. Merlin, and S. R. Phillpot, "Nanoscale Thermal Transport," Journal of Applied Physics, vol. 93, pp. 793-818, 2003.
- [3] J. Baxter, Z. X. Bian, G. Chen, D. Danielson, M. S. Dresselhaus, A. G. Fedorov, T. S. Fisher, C. W. Jones, E. Maginn, U. Kortshagen, A. Manthiram, A. Nozik, D. R. Rolison, T. Sands, L. Shi, D. Sholl, and Y. Y. Wu, "Nanoscale design to enable the revolution in renewable energy," Energy & Environmental Science, vol. 2, pp. 559-588, 2009.
- [4] T. R. Anthony, W. F. Banholzer, J. F. Fleischer, L. H. Wei, P. K. Kuo, R. L. Thomas, and R. W. Pryor, "Thermal-Diffusivity of Isotopically Enriched C-12 Diamond," Physical Review B, vol. 42, pp. 1104-1111, 1990.
- [5] V. P. Carey, G. Chen, C. Grigoropoulos, M. Kaviany, and A. Majumdar, "A review of Heat Transfer Physics," Nanoscale and Microscale Thermophysical Engineering, vol. 12, pp. 160, 2008.
- [6] C. Chiritescu, D. G. Cahill, N. Nguyen, D. Johnson, A. Bodapati, P. Keblinski, and P. Zschack, "Ultralow thermal conductivity in disordered, layered WSe2 crystals," Science, vol. 315, pp. 351-353, 2007.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
- [7] D. M. Rowe, Ed., CRC Handbook of Thermoelectrics . Boca Raton, FL: CRC Press, 1995.
- [8] Y. Ma, Q. Hao, B. Poudel, Y. C. Lan, B. Yu, D. Z. Wang, G. Chen, and Z. F. Ren, "Enhanced Thermoelectric Figure-of-merit in p-type Nanostructured Bismuth Antimony Tellurium Alloys Made from Elemental Chunks," Nano Letters, vol. 8, pp. 2580-2584, 2008.
- [9] J. P. Heremans, V. Jovovic, E. S. Toberer, A. Saramat, K. Kurosaki, A. Charoenphakdee, S. Yamanaka, and G. J. Snyder, "Enhancement of thermoelectric efficiency in PbTe by distortion of the electronic density of states," Science, vol. 321, pp. 554-557, 2008.
- [10] V. K. Zaitsev, M. I. Fedorov, E. A. Gurieva, I. S. Eremin, P. P. Konstantinov, A. Y. Samunin, and M. V. Vedernikov, "Highly effective Mg2Si1-xSnx Thermoelectrics," Phys. Rev. B, vol. 74, pp. 045207/1-045207/5, 2006.
- [11] X. F. Tang, Q. J. Zhang, L. D. Chen, T. Goto, and T. Hirai, "Synthesis and Thermoelectric Properties of p-type- and n-type-filled Skutterudite RyMxCo4-xSb12(R : Ce,Ba,Y; M : Fe,Ni)," Journal of Applied Physics, vol. 97, 2005.
- [12] M. G. Kanatzidis, "Nanostructured Thermoelectrics: The New Paradigm?," Chemistry of Materials, vol. 22, pp. 648-659, 2010.
- [13] Q. Zhang, J. He, T. J. Zhu, S. N. Zhang, X. B. Zhao, and T. M. Tritt, "High Figures of Merit and Natural Nanostructures in Mg2Si0.4Sn0.6 based Thermoelectric Materials," Applied Physics Letters, vol. 93, 2008.
- [14] F. J. DiSalvo, "Thermoelectric Cooling and Power Generation," Science, vol. 285, pp. 703-706, 1999.
- [15] L. E. Bell, "Cooling, heating, generating power, and recovering waste heat with thermoelectric systems," Science, vol. 321, pp. 1457-1461, 2008.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
- [16] C. B. Vining, "An inconvenient truth about thermoelectrics," Nat Mater, vol. 8, pp. 8385, 2009.
- [17] H. J. Goldsmid, Thermoelectric refrigeration . New York,: Plenum Press, 1964.
- [18] C. J. Vineis, A. Shakouri, A. Majumdar, and M. G. Kanatzidis, "Nanostructured Thermoelectrics: Big Efficiency Gains from Small Features," Advanced Materials, vol. 22, pp.
3970-3980, 2010.
- [19] A. A. Balandin, "Thermal properties of graphene and nanostructured carbon materials," Nature Materials, vol. 10, pp. 569-581, 2011.
- [20] A. J. Minnich, M. S. Dresselhaus, Z. F. Ren, and G. Chen, "Bulk nanostructured thermoelectric materials: current research and future prospects," Energy & Environmental Science, vol. 2, pp. 466-479, 2009.
- [21] W. Kim, R. Wang, and A. Majumdar, "Nanostructuring expands thermal limits," Nano Today, vol. 2, pp. 40-47, 2007.
- [22] J. Callaway, "Model for Lattice Thermal Conductivity at Low Temperatures," Physical Review, vol. 113, p. 1046, 1959.
- [23] D. T. Morelli, J. P. Heremans, and G. A. Slack, "Estimation of the Isotope Effect on the Lattice Thermal Conductivity of Group IV and Group III-V Semiconductors," Physical Review B, vol. 66, p. 195304, 2002.
[24] M. G. Holland, "Analysis of Lattice Thermal Conductivity," Physical Review, vol. 132, p. 2461, 1963.
- [25] H. B. G. Casimir, "Note on the Conduction of Heat in Crystals," Physica, vol. 5, pp. 495500, 1938.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
- [26] Z. Wang and N. Mingo, "Absence of Casimir regime in two-dimensional nanoribbon phonon conduction," Applied Physics Letters, vol. 99, 2011.
- [27] J. M. Ziman, Electrons and phonons : the theory of transport phenomena in solids . Oxford: Clarendon Press, 1962.
- [28] A. K. McCurdy, H. J. Maris, and C. Elbaum, "Anisotropic Heat Conduction in Cubic Crystals in the Boundary Scattering Regime," Physical Review B-Solid State, vol. 2, pp. 40774083, 1970.
- [29] A. Khitun, A. Balandin, and K. L. Wang, "Modification of the lattice thermal conductivity in silicon quantum wires due to spatial confinement of acoustic phonons," Superlattices and Microstructures, vol. 26, pp. 181-193, 1999.
- [30] J. Zou and A. Balandin, "Phonon heat conduction in a semiconductor nanowire," Journal of Applied Physics, vol. 89, pp. 2932-2938, 2001.
- [31] X. Lu, "Lattice thermal conductivity of Si nanowires: Effect of modified phonon density of states," Journal of Applied Physics, vol. 104, pp. 054314-5, 2008.
- [32] M. C. Wingert, Z. C. Y. Chen, E. Dechaumphai, J. Moon, J. H. Kim, J. Xiang, and R. K. Chen, "Thermal Conductivity of Ge and Ge-Si Core-Shell Nanowires in the Phonon Confinement Regime," Nano Letters, vol. 11, pp. 5507-5513, 2011.
- [33] N. Mingo, "Calculation of Si nanowire thermal conductivity using complete phonon dispersion relations," Physical Review B, vol. 68, 2003.
- [34] P. G. Klemens and D. F. Pedraza, "Thermal-conductivity of graphite in the basal-plane," Carbon, vol. 32, pp. 735-741, 1994.
- [35] Y. S. Ju and K. E. Goodson, "Phonon Scattering in Silicon Films with Thickness of Order 100 nm," Applied Physics Letters, vol. 74, pp. 3005-3007, 1999.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
- [36] D. Y. Li, Y. Y. Wu, P. Kim, L. Shi, P. D. Yang, and A. Majumdar, "Thermal Conductivity of Individual Silicon Nanowires," Applied Physics Letters, vol. 83, pp. 2934-2936, 2003.
- [37] L. Shi, Q. Hao, C. H. Yu, N. Mingo, X. Y. Kong, and Z. L. Wang, "Thermal conductivities of individual tin dioxide nanobelts," Applied Physics Letters, vol. 84, pp. 26382640, 2004.
- [38] F. Zhou, A. L. Moore, J. Bolinsson, A. Persson, L. Fröberg, M. T. Pettes, H. Kong, L. Rabenberg, P. Caroff, D. A. Stewart, N. Mingo, K. A. Dick, L. Samuelson, H. Linke, and L. Shi, "Thermal Conductivity of Indium Arsenide Nanowires with Wurtzite and Zinc Blende Phases " Phys. Rev. B, vol. in print, 2011.
- [39] R. Chen, A. I. Hochbaum, P. Murphy, J. Moore, P. D. Yang, and A. Majumdar, "Thermal conductance of thin silicon nanowires," Physical Review Letters, vol. 101, 2008.
- [40] P. G. Murphy and J. E. Moore, "Coherent phonon scattering effects on thermal transport in thin semiconductor nanowires," Physical Review B, vol. 76, 2007.
- [41] A. I. Hochbaum, R. K. Chen, R. D. Delgado, W. J. Liang, E. C. Garnett, M. Najarian, A. Majumdar, and P. D. Yang, "Enhanced Thermoelectric Performance of Rough Silicon Nanowires," Nature, vol. 451, pp. 163-167, 2008.
- [42] A. I. Boukai, Y. Bunimovich, J. Tahir-Kheli, J. K. Yu, W. A. Goddard, and J. R. Heath, "Silicon Nanowires as Efficient Thermoelectric Materials," Nature, vol. 451, pp. 168-171, 2008.
- [43] K. Hippalgaonkar, B. Huang, R. Chen, K. Sawyer, P. Ercius, and A. Majumdar, "Fabrication of Microdevices with Integrated Nanowires for Investigating Low-Dimensional Phonon Transport," Nano Letters, vol. 10, pp. 4341-4348, 2010.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
- [44] A. L. Moore, S. K. Saha, R. S. Prasher, and L. Shi, "Phonon Backscattering and Thermal Conductivity Suppression in Sawtooth Nanowires," Applied Physics Letters, vol. 93, p. 083112, 2008.
- [45] M. Kazan, G. Guisbiers, S. Pereira, M. R. Correia, P. Masri, A. Bruyant, S. Volz, and P. Royer, "Thermal Conductivity of Silicon Bulk and Nanowires: Effects of Isotopic Composition, Phonon Confinement, and Surface Roughness," Journal of Applied Physics, vol. 107, p. 14,
2010.
- [46] P. Martin, Z. Aksamija, E. Pop, and U. Ravaioli, "Impact of Phonon-Surface Roughness Scattering on Thermal Conductivity of Thin Si Nanowires," Physical Review Letters, vol. 102, p. 125503, 2009.
- [47] J. Carrete, L. J. Gallego, L. M. Varela, and N. Mingo, "Surface roughness and thermal conductivity of semiconductor nanowires: Going below the Casimir limit," Physical Review B, vol. 84, p. 075403, 2011.
- [48] J. Sadhu and S. Sinha, "Room-temperature phonon boundary scattering below the Casimir limit," Physical Review B, vol. 84, 2011.
- [49] Y. P. He and G. Galli, "Microscopic Origin of the Reduced Thermal Conductivity of Silicon Nanowires," submitted, 2012.
- [50] S. P. Hepplestone and G. P. Srivastava, "Lattice dynamics and thermal properties of phononic semiconductors," Physical Review B, vol. 84, 2011.
- [51] L. C. Parsons and G. T. Andrews, Observation of hypersonic phononic crystal effects in porous silicon superlattices vol. 95: AIP, 2009.
- [52] S.-i. Tamura, Y. Tanaka, and H. J. Maris, "Phonon group velocity and thermal conduction in superlattices," Physical Review B, vol. 60, p. 2627, 1999.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
- [53] J. K. Yu, S. Mitrovic, D. Tham, J. Varghese, and J. R. Heath, "Reduction of Thermal Conductivity in Phononic Nanomesh Structures," Nature Nanotechnology, vol. 5, pp. 718-721, 2010.
- [54] J. Y. Tang, H. T. Wang, D. H. Lee, M. Fardy, Z. Y. Huo, T. P. Russell, and P. D. Yang, "Holey Silicon as an Efficient Thermoelectric Material," Nano Letters, vol. 10, pp. 4279-4283, 2010.
- [55] Q. Hao, G. Chen, and M.-S. Jeng, "Frequency-dependent Monte Carlo simulations of phonon transport in two-dimensional porous silicon with aligned pores," Journal of Applied Physics, vol. 106, pp. 114321-10, 2009.
- [56] P. E. Hopkins, C. M. Reinke, M. F. Su, R. H. Olsson, E. A. Shaner, Z. C. Leseman, J. R. Serrano, L. M. Phinney, and I. El-Kady, "Reduction in the Thermal Conductivity of Single Crystalline Silicon by Phononic Crystal Patterning," Nano Letters, vol. 11, pp. 107-112, 2011.
- [57] A. Jian and A. J. H. McGaughey, "Effect of Phonon Free Path Distribution on Nanostructure Thermal Conductivity," in ASME 2012 Summer Heat Transfer Conference , Puerto Rico, USA, 2011.
- [58] R. G. Yang and C. Chen, "Nanostructured Thermoelectric Materials: From Superlattices to Nanocomposites," Materials Integration, vol. 18, 2005.
- [59] L. Shi, "Mesoscopic Thermophysical Measurements of Microstructures and Carbon Nanotubes," Ph.D. Dissertation, Mechanical Engineering, University of California, Berkeley, 2001.
- [60] P. Kim, L. Shi, A. Majumdar, and P. L. McEuen, "Thermal Transport Measurements of Individual Multiwalled Nanotubes," Physical Review Letters, vol. 8721, p. 215502, 2001.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
- [61] L. Shi, D. Y. Li, C. H. Yu, W. Y. Jang, D. Kim, Z. Yao, P. Kim, and A. Majumdar, "Measuring Thermal and Thermoelectric Properties of One-dimensional Nanostructures using a Microfabricated Device," Journal of Heat Transfer-Transactions of the Asme, vol. 125, pp. 881888, 2003.
- [62] A. L. Moore and L. Shi, "On Errors in Thermal Conductivity Measurements of Suspended and Supported Nanowires using Micro-thermometer Devices from Low to High Temperatures," Measurement Science and Technology, vol. 22, p. 015103, 2011.
- [63] F. Zhou, "Thermoelectric Transport in Semiconducting Nanowires," Ph.D. Dissertation, Mechanical Engineering, The University of Texas at Austin, Austin, 2009.
- [64] F. Zhou, A. Persson, L. Samuelson, H. Linke, and L. Shi, "Thermal resistance of a nanoscale point contact to an indium arsenide nanowire," Applied Physics Letters, vol. 99, 2011.
- [65] A. Mavrokefalos, M. T. Pettes, F. Zhou, and L. Shi, "Four-probe Measurements of the Inplane Thermoelectric Properties of Nanofilms," Review of Scientific Instruments, vol. 78, p. 034901, 2007.
- [66] A. Mavrokefalos, A. L. Moore, M. T. Pettes, L. Shi, W. Wang, and X. G. Li, "Thermoelectric and Structural Characterizations of Individual Electrodeposited Bismuth Telluride Nanowires," Journal of Applied Physics, vol. 105, p. 104318, 2009.
- [67] S. Berber, Y. K. Kwon, and D. Tomanek, "Unusually High Thermal Conductivity of Carbon Nanotubes," Physical Review Letters, vol. 84, pp. 4613-4616, 2000.
- [68] D. L. Nika, S. Ghosh, E. P. Pokatilov, and A. A. Balandin, "Lattice Thermal Conductivity of Graphene Flakes: Comparison with Bulk Graphite," Applied Physics Letters, vol. 94, p. 203103, 2009.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
- [69] L. Lindsay, D. A. Broido, and N. Mingo, "Flexural Phonons and Thermal Transport in Graphene," Physical Review B, vol. 82, p. 6, 2010.
- [70] J. H. Seol, I. Jo, A. L. Moore, L. Lindsay, Z. H. Aitken, M. T. Pettes, X. S. Li, Z. Yao, R. Huang, D. Broido, N. Mingo, R. S. Ruoff, and L. Shi, "Two-Dimensional Phonon Transport in Supported Graphene," Science, vol. 328, pp. 213-216, 2010.
- [71] M. T. Pettes, I. S. Jo, Z. Yao, and L. Shi, "Influence of Polymeric Residue on the Thermal Conductivity of Suspended Bilayer Graphene," Nano Letters, vol. 11, pp. 1195-1200, 2011.
- [72] L. Lindsay, D. A. Broido, and N. Mingo, "Diameter Dependence of Carbon Nanotube Thermal Conductivity and Extension to the Graphene Limit," Physical Review B, vol. 82, p. 4,
2010.
- [73] P. G. Klemens, "Theory of thermal conduction in thin ceramic films," International Journal of Thermophysics, vol. 22, pp. 265-275, 2001.
- [74] S. Lepri, R. Livi, and A. Politi, "Heat Conduction in Chains of Nonlinear Oscillators," Physical Review Letters, vol. 78, pp. 1896-1899, 1997.
- [75] B. B. Hu, B. W. Li, and H. Zhao, "Heat conduction in one-dimensional chains," Physical Review E, vol. 57, pp. 2992-2995, 1998.
- [76] K. Schwab, E. A. Henriksen, J. M. Worlock, and M. L. Roukes, "Measurement of the quantum of thermal conductance," Nature, vol. 404, pp. 974-977, 2000.
- [77] N. Mingo and D. A. Broido, "Length Dependence of Carbon Nanotube Thermal Conductivity and the "Problem of Long Waves"," Nano Letters, vol. 5, pp. 1221-1225, 2005.
- [78] N. Mingo and D. A. Broido, "Carbon Nanotube Ballistic Thermal Conductance and its Limits," Physical Review Letters, vol. 95, 2005.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
- [79] M. Fujii, X. Zhang, H. Q. Xie, H. Ago, K. Takahashi, T. Ikuta, H. Abe, and T. Shimizu, "Measuring the Thermal Conductivity of a Single Carbon Nanotube," Physical Review Letters,
- vol. 95, 2005.
- [80] R. Prasher, "Thermal boundary resistance and thermal conductivity of multiwalled carbon nanotubes," Physical Review B, vol. 77, pp. 075424:1-4, 2008.
- [81] L. Lu, W. Yi, and D. L. Zhang, "3 omega method for specific heat and thermal conductivity measurements," Review of Scientific Instruments, vol. 72, pp. 2996-3003, 2001.
- [82] T. Y. Choi, D. Poulikakos, J. Tharian, and U. Sennhauser, "Measurement of the Thermal Conductivity of Individual Carbon Nanotubes by the Four-point Three-omega Method," Nano Letters, vol. 6, pp. 1589-1593, 2006.
- [83] Z. Yao, C. L. Kane, and C. Dekker, "High-field Electrical Transport in Single-wall Carbon Nanotubes," Physical Review Letters, vol. 84, pp. 2941-2944, 2000.
- [84] E. Pop, D. Mann, Q. Wang, K. Goodson, and H. J. Dai, "Thermal Conductance of an Individual Single-wall Carbon Nanotube above Room Temperature," Nano Letters, vol. 6, pp. 96-100, 2006.
- [85] D. H. Chae, B. Krauss, K. von Klitzing, and J. H. Smet, "Hot Phonons in an Electrically Biased Graphene Constriction," Nano Letters, vol. 10, pp. 466-471, 2010.
- [86] D. M. Basko, S. Piscanec, and A. C. Ferrari, "Electron-electron interactions and doping dependence of the two-phonon Raman intensity in graphene," Physical Review B, vol. 80, p. 165413, 2009.
- [87] N. Bonini, M. Lazzeri, N. Marzari, and F. Mauri, "Phonon anharmonicities in graphite and graphene," Physical Review Letters, vol. 99, p. 176802, 2007.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
[88] A. Das, S. Pisana, B. Chakraborty, S. Piscanec, S. K. Saha, U. V. Waghmare, K. S. Novoselov, H. R. Krishnamurthy, A. K. Geim, A. C. Ferrari, and A. K. Sood, "Monitoring dopants by Raman scattering in an electrochemically top-gated graphene transistor," Nature Nanotechnology, vol. 3, pp. 210-215, 2008.
[89] A. W. Bushmaker, V. V. Deshpande, M. W. Bockrath, and S. B. Cronin, "Direct observation of mode selective electron-phonon coupling in suspended carbon nanotubes," Nano Letters, vol. 7, pp. 3618-3622, 2007.
[90] M. Steiner, M. Freitag, V. Perebeinos, J. C. Tsang, J. P. Small, M. Kinoshita, D. Yuan, J. Liu, and P. Avouris, "Phonon Populations and Electrical Power Dissipation in Carbon Nanotube Transistors," Nature Nanotechnology, vol. 4, pp. 320-324, 2009.
- [91] M. Oron-Carl and R. Krupke, "Raman Spectroscopic Evidence for Hot-Phonon Generation in Electrically Biased Carbon Nanotubes," Physical Review Letters, vol. 100, p. 127401, 2008.
[92] I. K. Hsu, R. Kumar, A. Bushmaker, S. B. Cronin, M. T. Pettes, L. Shi, T. Brintlinger, M. S. Fuhrer, and J. Cumings, "Optical Measurement of Thermal Transport in Suspended Carbon Nanotubes," Applied Physics Letters, vol. 92, pp. 063119:1-3, 2008.
[93] V. V. Deshpande, S. Hsieh, A. W. Bushmaker, M. Bockrath, and S. B. Cronin, "Spatially Resolved Temperature Measurements of Electrically Heated Carbon Nanotubes," Physical Review Letters, vol. 102, pp. 105501:1-4, 2009.
[94] Q. W. Li, C. H. Liu, X. S. Wang, and S. S. Fan, "Measuring the Thermal Conductivity of Individual Carbon Nanotubes by the Raman Shift Method," Nanotechnology, vol. 20, pp. 145702:1-5, 2009.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
[95] I. K. Hsu, M. T. Pows, A. Bushmaker, M. Aykol, L. Shi, and S. B. Cronin, "Optical Absorption and Thermal Transport of Individual Suspended Carbon Nanotube Bundles," Nano Letters, vol. 9, pp. 590-594, 2009.
[96] I. K. Hsu, M. T. Pettes, M. Aykol, L. Shi, and S. B. Cronin, "The Effect of Gas Environment on Electrical Heating in Suspended Carbon Nanotubes," Journal of Applied Physics, vol. 108, p. 084307, 2010.
[97] M. T. Pettes and L. Shi, "Thermal and Structural Characterizations of Individual Single-, Double-, and Multi-Walled Carbon Nanotubes," Advanced Functional Materials, vol. 19, pp. 3918-3925, 2009.
[98] C. H. Yu, L. Shi, Z. Yao, D. Y. Li, and A. Majumdar, "Thermal Conductance and Thermopower of an Individual Single-wall Carbon Nanotube," Nano Letters, vol. 5, pp. 18421846, 2005.
[99] C. W. Chang, A. M. Fennimore, A. Afanasiev, D. Okawa, T. Ikuno, H. Garcia, D. Y. Li, A. Majumdar, and A. Zettl, "Isotope Effect on the Thermal Conductivity of Boron Nitride Nanotubes," Physical Review Letters, vol. 97, p. 085901, 2006.
[100] C. W. Chang, D. Okawa, H. Garcia, A. Majumdar, and A. Zettl, "Breakdown of Fourier's Law in Nanotube Thermal Conductors," Physical Review Letters, vol. 101, p. 075903, 2008.
[101] M. T. Pettes, "Experimental Investigations of Thermal Transport in Carbon Nanotubes, Graphene and Nanoscale Point Contacts," Ph.D. Ph.D. thesis, Mechanical Engineering, The University of Texas at Austin, Austin, 2011.
[102] Z. L. Wang, D. W. Tang, X. B. Li, X. H. Zheng, W. G. Zhang, L. X. Zheng, Y. T. T. Zhu, A. Z. Jin, H. F. Yang, and C. Z. Gu, "Length-dependent Thermal Conductivity of an Individual Single-wall Carbon Nanotube," Applied Physics Letters, vol. 91, 2007.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
[103] L. Shi, J. Zhou, P. Kim, A. Bachtold, A. Majumdar, and P. L. McEuen, "Thermal Probing of Energy Dissipation in Current-carrying Carbon Nanotubes," Journal of Applied Physics, vol. 105, p. 104306, 2009.
[104] L. Shi, "Comment on ``Length-Dependant Thermal Conductivity of an Individual Singlewall Carbon Nanotube'' [Appl. Phys. Lett. 91, 123119 (2007)]," Applied Physics Letters, vol. 92, pp. 206103:1-2, 2008.
[105] A. A. Balandin, S. Ghosh, W. Z. Bao, I. Calizo, D. Teweldebrhan, F. Miao, and C. N. Lau, "Superior thermal conductivity of single-layer graphene," Nano Letters, vol. 8, pp. 902-907, 2008.
[106] S. Ghosh, D. L. Nika, E. P. Pokatilov, and A. A. Balandin, "Heat Conduction in Graphene: Experimental Study and Theoretical Interpretation," New Journal of Physics, vol. 11, p. 095012, 2009.
[107] S. Ghosh, I. Calizo, D. Teweldebrhan, E. P. Pokatilov, D. L. Nika, A. A. Balandin, W.
Bao, F. Miao, and C. N. Lau, "Extremely High Thermal Conductivity of Graphene: Prospects for Thermal Management Applications in Nanoelectronic Circuits," Applied Physics Letters, vol. 92, p. 151911, 2008.
[108] R. R. Nair, P. Blake, A. N. Grigorenko, K. S. Novoselov, T. J. Booth, T. Stauber, N. M. R. Peres, and A. K. Geim, "Fine Structure Constant Defines Visual Transparency of Graphene," Science, vol. 320, p. 1308, 2008.
[109] C. Faugeras, B. Faugeras, M. Orlita, M. Potemski, R. R. Nair, and A. K. Geim, "Thermal Conductivity of Graphene in Corbino Membrane Geometry," Acs Nano, vol. 4, pp. 1889-1892, 2010.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
[110]
W. W. Cai, A. L. Moore, Y. W. Zhu, X. S. Li, S. S. Chen, L. Shi, and R. S. Ruoff,
"Thermal Transport in Suspended and Supported Monolayer Graphene Grown by Chemical Vapor Deposition," Nano Letters, vol. 10, pp. 1645-1651, 2010.
[111] S. S. Chen, A. L. Moore, W. W. Cai, J. W. Suk, J. H. An, C. Mishra, C. Amos, C. W.
Magnuson, J. Y. Kang, L. Shi, and R. S. Ruoff, "Raman Measurements of Thermal Transport in Suspended Monolayer Graphene of Variable Sizes in Vacuum and Gaseous Environments," Acs Nano, vol. 5, pp. 321-328, 2011.
[112] Y. S. Touloukian, Ed., Thermophysical Properties of Matter . New York: IFI/Plenum, 1970.
[113] J. U. Lee, D. Yoon, H. Kim, S. W. Lee, and H. Cheong, "Thermal Conductivity of Suspended Pristine Graphene Measured by Raman Spectroscopy," Physical Review B, vol. 83, p. 081419, 2011.
[114] K. F. Mak, J. Shan, and T. F. Heinz, "Seeing Many-Body Effects in Single- and FewLayer Graphene: Observation of Two-Dimensional Saddle-Point Excitons," Physical Review Letters, vol. 106, pp. 1-4, 2011.
[115] P. G. Collins, K. Bradley, M. Ishigami, and A. Zettl, "Extreme Oxygen Sensitivity of Electronic Properties of Carbon Nanotubes," Science, vol. 287, pp. 1801-1804, 2000.
[116] K. I. Bolotin, K. J. Sikes, Z. Jiang, M. Klima, G. Fudenberg, J. Hone, P. Kim, and H. L. Stormer, "Ultrahigh Electron Mobility in Suspended Graphene," Solid State Communications, vol. 146, pp. 351-355, 2008.
[117] C. R. Dean, A. F. Young, I. Meric, C. Lee, L. Wang, S. Sorgenfrei, K. Watanabe, T. Taniguchi, P. Kim, K. L. Shepard, and J. Hone, "Boron nitride substrates for high-quality graphene electronics," Nature Nanotechnology, vol. 5, pp. 722-726, 2010.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
[118] C. W. Padgett and D. W. Brenner, "Influence of Chemisorption on the Thermal Conductivity of Single-wall carbon Nanotubes," Nano Letters, vol. 4, pp. 1051-1053, 2004.
[119] R. Prasher, "Thermal Conductance of Single-walled Carbon Nanotube Embedded in an Elastic Half-space," Applied Physics Letters, vol. 90, p. 143110, 2007.
[120] T. Aizawa, Y. Hwang, W. Hayami, R. Souda, S. Otani, and Y. Ishizawa, "PhononDispersion of Monolayer Graphite on Pt(111) and Nbc Surfaces - Bond Softening and Interface Structures," Surface Science, vol. 260, pp. 311-318, 1992.
[121] T. Aizawa, R. Souda, Y. Ishizawa, H. Hirano, T. Yamada, K. I. Tanaka, and C. Oshima, "Phonon-Dispersion in Monolayer Graphite Formed on Ni(111) and Ni(001)," Surface Science, vol. 237, pp. 194-202, 1990.
[122] A. Allard and L. Wirtz, "Graphene on Metallic Substrates: Suppression of the Kohn Anomalies in the Phonon Dispersion," Nano Letters, vol. 10, pp. 4335-4340, 2010.
[123] D. Farias, K. H. Rieder, A. M. Shikin, V. K. Adamchuk, T. Tanaka, and C. Oshima, "Modification of the surface phonon dispersion of a graphite monolayer adsorbed on Ni(111) caused by intercalation of Yb, Cu and Ag," Surface Science, vol. 454, pp. 437-441, 2000.
[124] N. Mounet and N. Marzari, "First-principles determination of the structural, vibrational and thermodynamic properties of diamond, graphite, and derivatives," Physical Review B, vol. 71, 2005.
[125] A. M. Shikin, D. Farias, V. K. Adamchuk, and K. H. Rieder, "Surface phonon dispersion of a graphite monolayer adsorbed on Ni(111) and its modification caused by intercalation of Yb, La and Cu layers," Surface Science, vol. 424, pp. 155-167, 1999.
[126] R. Nicklow, H. G. Smith, and Wakabaya.N, "Lattice-dynamics of pyrolytic-graphite," Physical Review B, vol. 5, pp. 4951-4962, 1972.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
- [127] T. Aizawa, R. Souda, S. Otani, Y. Ishizawa, and C. Oshima, "Bond softening in monolayer graphite formed on transition-metal carbide surfaces," Physical Review B, vol. 42, pp. 11469-11478, 1990.
[128] J. D. Bernal, "The structure of graphite," Proceedings of the Royal Society of London. Series A, Mathematical and Physical Sciences, vol. 106, pp. 749-773, 1924.
[129] R. A. Jishi, L. Venkataraman, M. S. Dresselhaus, and G. Dresselhaus, "Phonon modes in carbon nanotubules," Chemical Physics Letters, vol. 209, pp. 77-82, 1993.
[130] D. L. Nika, E. P. Pokatilov, A. S. Askerov, and A. A. Balandin, "Phonon thermal conduction in graphene: Role of umklapp and edge roughness scattering," Physical Review B, vol. 79, pp. 155413-1-12, 2009.
[131] R. Prasher, "Acoustic Mismatch Model for Thermal Contact Resistance of van der Waals Contacts," Applied Physics Letters, vol. 94, pp. 041905:1-3, 2009.
[132] Z. Y. Ong and E. Pop, "Effect of substrate modes on thermal transport in supported graphene," Physical Review B, vol. 84, 2011.
[133] Z. Y. Ong, E. Pop, and J. Shiomi, "Reduction of phonon lifetimes and thermal conductivity of a carbon nanotube on amorphous silica," Physical Review B, vol. 84, 2011.
[134] Z. Chen, W. Jang, W. Bao, C. N. Lau, and C. Dames, "Thermal Contact Resistance between Graphene and Silicon Dioxide," Applied Physics Letters, vol. 95, p. 161910, 2009.
[135] X. Xu, Y. Wang, K. Zhang, X. Zhao, S. Bae, M. Heinrich, C. T. Bui, R. Xie, J. T. L.
Thong, B. H. Hong, K. P. Loh, B. Li, and B. Oezyilmaz, "Phonon Transport in Suspended Single Layer Graphene," arXiv:1012.2937v1, 2010.
[136] Z. Wang, R. Xie, C. T. Bui, D. Liu, X. Ni, B. Li, and J. T. L. Thong, "Thermal Transport in Suspended and Supported Few-Layer Graphene," Nano Letters, vol. 11, pp. 113-118, 2011.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
[137] W. Y. Jang, Z. Chen, W. Z. Bao, C. N. Lau, and C. Dames, "Thickness-Dependent Thermal Conductivity of Encased Graphene and Ultrathin Graphite," Nano Letters, vol. 10, pp. 3909-3913, 2010.
[138] S. Ghosh, W. Bao, D. L. Nika, S. Subrina, E. P. Pokatilov, C. N. Lau, and A. A. Balandin, "Dimensional Crossover of Thermal Transport in Few-layer Graphene," Nat Mater, vol. 9, pp. 555-558, 2010.
[139] S. T. Huxtable, D. G. Cahill, S. Shenogin, L. P. Xue, R. Ozisik, P. Barone, M. Usrey, M.
- S. Strano, G. Siddons, M. Shim, and P. Keblinski, "Interfacial Heat Flow in Carbon Nanotube Suspensions," Nature Materials, vol. 2, pp. 731-734, 2003.
[140] H. Maune, H. Y. Chiu, and M. Bockrath, "Thermal Resistance of the Nanoscale Constrictions between Carbon Nanotubes and Solid Substrates," Applied Physics Letters, vol. 89, p. 013109, 2006.
[141] E. Pop, D. A. Mann, K. E. Goodson, and H. J. Dai, "Electrical and Thermal Transport in Metallic Single-wall Carbon Nanotubes on Insulating Substrates," Journal of Applied Physics, vol. 101, p. 093710, 2007.
[142] K. F. Mak, C. H. Lui, and T. F. Heinz, "Measurement of the Thermal Conductance of the Graphene/SiO2 Interface," Applied Physics Letters, vol. 97, p. 3, 2010.
[143] Y. K. Koh, M. H. Bae, D. G. Cahill, and E. Pop, "Heat Conduction across Monolayer and Few-Layer Graphenes," Nano Letters, vol. 10, pp. 4363-4368, 2010.
[144] R. Martel, T. Schmidt, H. R. Shea, T. Hertel, and P. Avouris, "Single- and Multi-wall Carbon Nanotube Field-Effect Transistors," Applied Physics Letters, vol. 73, pp. 2447-2449, 1998.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
[145] J. P. Heremans, C. M. Thrush, and D. T. Morelli, "Thermopower Enhancement in Lead Telluride Nanostructures," Physical Review B, vol. 70, p. 115334, 2004.
[146] H. K. Lyeo, A. A. Khajetoorians, L. Shi, K. P. Pipe, R. J. Ram, A. Shakouri, and C. K. Shih, "Profiling the Thermoelectric Power of Semiconductor Junctions with Nanometer Resolution," Science, vol. 303, pp. 816-818, 2004.
[147] F. Zhou, J. Szczech, M. T. Pettes, A. L. Moore, S. Jin, and L. Shi, "Determination of Transport Properties in Chromium Disilicide Nanowires via Combined Thermoelectric and Structural Characterizations," Nano Letters, vol. 7, pp. 1649-1654, 2007.
[148] J. H. Seol, A. L. Moore, S. K. Saha, F. Zhou, L. Shi, Q. L. Ye, R. Scheffler, N. Mingo, and T. Yamada, "Measurement and Analysis of Thermopower and Electrical Conductivity of an Indium Antimonide Nanowire from a Vapor-liquid-solid Method," Journal of Applied Physics, vol. 101, p. 023706, 2007.
[149] F. Zhou, A. L. Moore, M. T. Pettes, Y. Lee, J. H. Seol, Q. L. Ye, L. Rabenberg, and L. Shi, "Effect of Growth Base Pressure on the Thermoelectric Properties of Indium Antimonide Nanowires," Journal of Physics D-Applied Physics, vol. 43, p. 9, 2010.
[150] C. Herring, "Theory of the Thermoelectric Power of Semiconductors," Physical Review, vol. 96, p. 1163, 1954.
[151] T. H. Geballe and G. W. Hull, "Seebeck Effect in Silicon," Physical Review, vol. 98, p. 940, 1955.
[152] H. J. Ryu, Z. Aksamija, D. M. Paskiewicz, S. A. Scott, M. G. Lagally, I. Knezevic, and M. A. Eriksson, "Quantitative Determination of Contributions to the Thermoelectric Power Factor in Si Nanostructures," Physical Review Letters, vol. 105, p. 256601, 2010.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
[153] L. Weber and E. Gmelin, "Transport-Properties of Silicon," Applied Physics a-Materials Science & Processing, vol. 53, pp. 136-140, 1991.
[154] O. Yamashita and N. Sadatomi, "Dependence of Seebeck Coefficient on Carrier Concentration in Heavily B- and P-doped Si1-xGex (x <= 0.05) system," Japanese Journal of Applied Physics Part 1-Regular Papers Short Notes & Review Papers, vol. 38, pp. 6394-6400, 1999.
[155] J. P. Small and P. Kim, "Thermopower Measurement of Individual Single Walled Carbon Nanotubes," Microscale Thermophysical Engineering, vol. 8, pp. 1-5, 2004.
[156] J. P. Small, K. M. Perez, and P. Kim, "Modulation of Thermoelectric Power of Individual Carbon Nanotubes," Physical Review Letters, vol. 91, p. 4, 2003.
[157] L. D. Hicks and M. S. Dresselhaus, "Thermoelectric Figure of Merit of a OneDimensional Conductor," Physical Review B, vol. 47, pp. 16631-16634, 1993.
[158] L. D. Hicks and M. S. Dresselhaus, "Effect of Quantum-Well Structures on the Thermoelectric Figure of Merit," Physical Review B, vol. 47, pp. 12727-12731, 1993.
[159] D. A. Broido and T. L. Reinecke, "Effect of Superlattice Structure on the Thermoelectric Figure of Merit," Physical Review B, vol. 51, pp. 13797-13800, 1995.
[160] Y. M. Lin, X. Z. Sun, and M. S. Dresselhaus, "Theoretical investigation of thermoelectric transport properties of cylindrical Bi nanowires," Physical Review B, vol. 62, pp. 4610-4623, 2000.
[161] N. Mingo, "Thermoelectric figure of merit and maximum power factor in III-V semiconductor nanowires," Applied Physics Letters, vol. 84, pp. 2652-2654, 2004.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
- [162] O. Rabina, Y. M. Lin, and M. S. Dresselhaus, "Anomalously high thermoelectric figure of merit in Bi1-xSbx nanowires by carrier pocket alignment," Applied Physics Letters, vol. 79, pp. 81-83, 2001.
- [163] R. Kim, S. Datta, and M. S. Lundstrom, "Influence of dimensionality on thermoelectric device performance," Journal of Applied Physics, vol. 105, p. 6, 2009.
- [164] R. Venkatasubramanian, E. Siivola, T. Colpitts, and B. O'Quinn, "Thin-film thermoelectric devices with high room-temperature figures of merit," Nature, vol. 413, pp. 597602, 2001.
- [165] T. C. Harman, P. J. Taylor, M. P. Walsh, and B. E. LaForge, "Quantum dot superlattice thermoelectric materials and devices," Science, vol. 297, pp. 2229-2232, 2002.
- [166] R. Venkatasubramanian, "Lattice thermal conductivity reduction and phonon localizationlike behavior in superlattice structures," Physical Review B, vol. 61, pp. 3091-3097, 2000.
- [167] Y. F. Chen, D. Y. Li, J. R. Lukes, Z. H. Ni, and M. H. Chen, "Minimum superlattice thermal conductivity from molecular dynamics," Physical Review B, vol. 72, p. 6, 2005.
- [168] J. P. Heremans, C. M. Thrush, D. T. Morelli, and M. C. Wu, "Thermoelectric Power of Bismuth Nanocomposites," Physical Review Letters, vol. 88, p. 216801, 2002.
[169] W. Shim, J. Ham, K. I. Lee, W. Y. Jeung, M. Johnson, and W. Lee, "On-Film Formation of Bi Nanowires with Extraordinary Electron Mobility," Nano Letters, vol. 9, pp. 18-22, 2009.
- [170] C. J. Vineis, T. C. Harman, S. D. Calawa, M. P. Walsh, R. E. Reeder, R. Singh, and A. Shakouri, "Carrier concentration and temperature dependence of the electronic transport properties of epitaxial PbTe and PbTe/PbSe nanodot superlattices," Physical Review B, vol. 77, p. 14, 2008.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
[171] Y. K. Koh, C. J. Vineis, S. D. Calawa, M. P. Walsh, and D. G. Cahill, "Lattice Thermal
Conductivity of Nanostructured Thermoelectric Materials based on PbTe," Applied Physics
Letters, vol. 94, p. 153101, 2009.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
## Figure captions :
Figure 1. ZT as a function of temperature for bulk TE materials reported in [7-13].
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
Figure 2. Experimental room-temperature thermal conductivity data of Si NWs of different equivalent diameters reported by Li et al. [36] for VLS NWs, Hippalgaonkar et al. [43] for patterned NWs, Hochbaum et al. [41] for electrochemically etched NWs, and Boukai et al. [42] for patterned NWs. The lines are the thermal conductivity of Si wires of different diameters calculated by A. L. Moore in the Casimir diffuse surface limit and with a boron doping concentration of 1 x 10 13 and 1 x 10 19 cm -3 , respectively.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
Figure 3. (a) Scanning electron micrograph (SEM) of a suspended resistance thermometer microdevice. (b) SEM of a Si NW assembled between the two central SiNx membranes of the microdevice. The thermal resistance circuit and the temperature distribution along the NW sample are shown in the schematics at the left and right sides of the SEM, respectively. Th and T s are the temperatures of the heating (upper) and sensing (lower) membranes. Th ' and T s ' are the temperatures at the two ends of the suspended segment of the film sample. T1 , T 2 , T3 , and T 4 are the temperatures at the four contacts to the NW. T0 is the temperature of the substrate. Rs and RB are the thermal resistances of the NW sample and the six beams supporting one membrane. RC1 and RC2 are the contact thermal resistances between the sample and the heating and sensing membranes. V 14 and V 23 are the TE voltages ( V TE) measured between the two outer and between the two inner electrodes.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
Figure 4. Experimental thermal conductivity values of CNTs reported by Kim et al. [60] for a ~14-nm-outer diameter and 2.5- m-long MWCNT grown by arc discharge, by Fujii et al. [79] for a 16.1-nm-outer diameter, 4.9-nm-inner diameter, and 1.89- m-long MWCNT grown by arc discharge, by Chang et al. [99] for a 33-nm-outer diameter MWCNT grown by arc discharge, by Choi et al. [82] for a ~20-nm-outer diameter, ~10-nm-inner diameter, and 1.4- m-long MWCNT grown by plasma-enhanced CVD, by Yu et al. [98] for a 1 3 nm outer diameter and 2.76- mlong SWCNT grown by thermal CVD, by Pop et al. [84] for a 1.7-nm-outer diameter and 2.6- m-long SWCNT grown by thermal CVD, by Li et al. [94] for a 1.8-nm-diameter SWCNT and a and a 8.2-nm-diameter MWCNT grown by thermal CVD, by Pettes et al. [97] for a 1.5-nmdiameter, 2.03- m-long SWCNT, 2.7-nm-outer diameter, 2.3-nm-inner diameter, and 4.02- mlong DWCNT, and a 11.7-nm-outer diameter, 6.7-nm-inner diameter, 1.97 m-long, 7-wall MWCNT, all grown by thermal CVD. Shown for comparison are the basal plane thermal conductivity recommended by Touloukian [112] for pyrolytic graphite.
Figure 5. Thermal conductivity of MWCNTs of different diameters reported by Fujii et al. for MWCNTs grown by arc discharge, by Chang et al. [99] for MWCNTs grown by arc discharge, and by Pettes and Shi [97] for MWCNTs grown by thermal CVD.
[79]
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
Figure 6. Experimental thermal conductivity values of graphene reported by Balandin et al. [105] for SLG samples exfoliated from pyrolytic graphite and suspended across a 2-5 m wide trench, Faugeras et al. [109] for a SLG sample exfoliated from natural graphite and suspended over a 44 m diameter hole, Chen et al. [111] for a SLG sample grown by CVD and suspended over a 9.7- m-diameter hole, Lee et al. [113] for a SLG sample exfoliated from natural graphite and suspended over a 6- m-diameter hole, Seol et al. [70] for a 9.5- m-long, 2.4- m-wide SLG sample exfoliated from natural graphite and supported on SiO2, Wang et al. [136] for 5, 3, or 1- m long three-layer graphene samples exfoliated from graphite and supported onto SiO2, and Pettes et al. [71] for a 5- m-long, 1.8- m-wide suspended BLG sample exfoliated from natural graphite and with polymer residual. Shown for comparison are the basal plane thermal conductivity recommended by Touloukian [112] for pyrolytic graphite, the calculation results of Lindsay et al . [70] for 10- m-long suspended (dashed line) and supported (solid line) SLG samples. For Raman measurements, the temperature is the temperature measured by the Raman laser beam while the substrate temperature is kept at room temperature.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
Figure 7. Thermal conductivity of graphene samples of different layer thickness reported by Ghosh et al. [138] for suspended graphene samples and by Jang et al. [137] for graphene samples encased in SiO2. In the work of Ghosh et al. [138], experimental data obtained for 5-16 m wide graphene samples suspended over 1-5 m wide trenches were normalized with the use of a theoretical model to yield the reported data expected for 5- m-wide suspended graphene samples.
Cite this paper as L. Shi, 'Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems', Nanoscale and Microscale Thermophysical Engineering, vol. 16, p. 79-116, 2012. DOI:10.1080/15567265.2012.667514
Figure 8. Measured Seebeck coefficient (symbols) of boron-doped Si reported by Geballe and
Hull [151] for single-crystal (SC) samples of 1 x 10 18 and 1.5 x 10 19 cm -3 acceptor concentrations, by Yamashita and Sadatomi [154] for polycrystalline (Poly) samples of 1.7 x 10 18 , 3.8 x 10 19 , 7.4 x 10 19 and 5 x 10 20 cm -3 carrier concentration, by Boukai et al. [42] for a 20 x 20 nm NW doped to 3 2 x 10 19 cm -3 and a 10 x 20 nm NW doped to 7 2 x 10 19 cm -3 , and by Hochbaum et al. [41] for a highly doped 48-nm-diameter NW. The lines are calculated Sd results of a two-band model for boron doping concentration of 1 x 10 18 , 1.5 x 10 19 , 3 x 10 19 , 7 x 10 19 , and 5 x 10 20 , respectively, arranged in the top to bottom order in the figure.
Graphic Submitted for Consideration as a Cover Art: A false color SEM image of a resistance thermometer device used for thermal measurement of supported SLG by Seol et al . [70]. | 10.1080/15567265.2012.667514 | [
"Li Shi"
] | 2014-08-01T19:26:48+00:00 | 2014-08-01T19:26:48+00:00 | [
"cond-mat.mes-hall",
"cond-mat.mtrl-sci"
] | Thermal and Thermoelectric Transport in Nanostructures and Low-Dimensional Systems | Significant progress has been made in recent studies of thermal and
thermoelectric transport phenomena in nanostructures and low-dimensional
systems. This article reviews several intriguing quantum and classical size
effects on thermal and thermoelectric properties that have been predicted by
theoretical calculations or observed in experiments. Attention is focused on
the Casimir limit in phonon boundary scattering and the effect of phonon
confinement on the lattice thermal conductivity of semiconductor nanowires
(NWs) and nanomeshes; the effects of thickness, lateral size, and interface
interaction on the lattice thermal conductivity of carbon nanotubes (CNTs) and
graphene; and the phonon-drag thermopower and quantum size effects on the
thermoelectric power factor in semiconductor NWs. Further experimental and
theoretical investigations are suggested for better understanding of some of
these nanoscale transport phenomena. |
1408.0275v2 | ## EUROPEAN ORGANIZATION FOR NUCLEAR RESEARCH (CERN)
CERN-PH-EP-2014-181 LHCb-PAPER-2014-042 July 27, 2014
## Measurement of the B 0 -B 0 and Bs 0 -Bs 0 production asymmetries in pp collisions at √ s = 7TeV
The LHCb collaboration †
## Abstract
The B 0 -B 0 and B 0 s -B 0 s production asymmetries, A P ( B 0 ) and A P ( B 0 s ), are measured by means of a time-dependent analysis of B 0 → J/ψK ∗ 0 , B 0 → D π -+ and B 0 s → D π -s + decays, using a data sample corresponding to an integrated luminosity of 1.0 fb -1 , collected by LHCb in pp collisions at a centre-of-mass energy of 7 TeV. The measurements are performed as a function of transverse momentum and pseudorapidity of the B 0 and B 0 s mesons within the LHCb acceptance. The production asymmetries, integrated over p T and η in the range 4 < p T < 30 GeV /c and 2 5 . < η < 4 5, . are determined to be A P ( B 0 ) = ( -0 35 . ± 0 76 . ± 0 28)% and . A P ( B 0 s ) = (1 09 . ± 2 61 . ± 0 66)%, where the first uncertainties are statistical and the . second systematic.
Submitted to Phys. Lett. B
c © CERN on behalf of the LHCb collaboration, license CC-BY-4.0.
† Authors are listed at the end of this letter.
## 1 Introduction
The production rates of b and ¯ hadrons in b pp collisions are not expected to be identical. This phenomenon, commonly referred to as the production asymmetry, is related to the fact that there can be coalescence between a perturbatively produced b or ¯ quark and the b u and d valence quarks in the beam remnant. Therefore, one can expect a slight excess in the production of B + and B 0 mesons with respect to B -and B 0 mesons, and e.g. of Λ 0 b baryons with respect to Λ 0 b baryons. As b and ¯ quarks are almost entirely produced in b pairs via strong interactions, the existence of B + and B 0 production asymmetries must be compensated by opposite production asymmetries for other B -meson and b -baryon species. These asymmetries are roughly estimated to be at the 1% level for pp collisions at LHC energies, and are expected to be enhanced at forward rapidities and small transverse momenta. Other subtle effects of quantum chromodynamics, beyond the coalescence between beauty quarks and light valence quarks, may also contribute [1-3].
The production asymmetry is one of the key ingredients to perform measurements of CP violation in b -hadron decays at the LHC, since CP asymmetries must be disentangled from other sources. The production asymmetry for B 0 and B 0 s mesons is defined as
$$A _ { \text{P} } \left ( B _ { ( s ) } ^ { 0 } \right ) & \equiv \frac { \sigma \left ( \bar { B } _ { ( s ) } ^ { 0 } \right ) - \sigma \left ( B _ { ( s ) } ^ { 0 } \right ) } { \sigma \left ( \bar { B } _ { ( s ) } ^ { 0 } \right ) + \sigma \left ( B _ { ( s ) } ^ { 0 } \right ) },$$
where σ denotes the production cross-section. Similar asymmetries are also expected when producing charmed hadrons. LHCb has already performed measurements of D + -D -and D + s -D -s production asymmetries, finding values around the 1% level or less [4,5].
In this paper, the values of A P ( B 0 ) and A P ( B 0 s ) are constrained by measuring the oscillations of B 0 and B 0 s mesons with a time-dependent analysis of the B 0 → J/ψ µ ( + µ -) K ∗ 0 ( K π + -), B 0 → D -( K π π + --) π + and B 0 s → D -s ( K K π + --) π + decay rates, without tagging the initial flavour of the decaying B 0 ( s ) meson. The inclusion of chargeconjugate decay modes is implied throughout. The measurements are performed as a function of transverse momentum, p T , and pseudorapidity, η , of the B 0 ( s ) meson within the LHCb acceptance, and then integrated over the range 4 < p T < 30 GeV /c and 2 5 . < η < 4 5. .
## 2 Detector, trigger and simulation
The LHCb detector [6] is a single-arm forward spectrometer covering the pseudorapidity range 2 < η < 5, designed for the study of particles containing b or c quarks. The detector includes a high-precision tracking system consisting of a silicon-strip vertex detector surrounding the pp interaction region, a large-area silicon-strip detector located upstream of a dipole magnet with a bending power of about 4 Tm, and three stations of silicon-strip detectors and straw drift tubes placed downstream of the magnet. The tracking system provides a measurement of momentum with a relative uncertainty that varies from 0.4% at
low momentum to 0.6% at 100 GeV /c . The minimum distance of a track to a primary vertex (PV), the impact parameter, is measured with a resolution of (15 + 29 /p T ) µ m, where p T is in GeV /c . Different types of charged hadrons are distinguished using information from two ring-imaging Cherenkov detectors. Photon, electron and hadron candidates are identified by a calorimeter system consisting of scintillating-pad and preshower detectors, an electromagnetic calorimeter and a hadronic calorimeter. Muons are identified by a system composed of alternating layers of iron and multiwire proportional chambers. The trigger consists of a hardware stage, based on information from the calorimeter and muon systems, followed by a software stage, which applies a full event reconstruction.
In the case of the B 0 → J/ψK ∗ 0 decay, events are first selected by a hardware trigger that requires muon candidates with p T > 1 48 . GeV /c . The subsequent software trigger is composed of two stages. The first stage performs a partial event reconstruction and requires events to have two well identified oppositely charged muons, with invariant mass larger than 2.7 GeV /c 2 . The second stage performs a full event reconstruction and only retains events containing a µ + µ -pair that has invariant mass within 120 MeV /c 2 of the known J/ψ mass [7] and forms a vertex that is significantly displaced from the nearest PV.
In the case of B 0 → D π -+ and B 0 s → D π -s + decays, events are first selected by a hardware trigger requiring a high transverse energy cluster in the calorimeter system. Events passing the hardware trigger are further filtered by a software trigger which requires a two-, three- or four-track secondary vertex with a large sum of p T of the tracks and a significant displacement from the PVs. Subsequently, a multivariate algorithm [8] is applied, aimed at identifying secondary vertices, consistent with the decay of a b hadron.
Simulated events are used to determine the signal selection efficiency, acceptance as function of decay time, decay time resolution, and to model the background. In the simulation, pp collisions are generated using Pythia 6.4 [9] with a specific LHCb configuration [10]. The interaction of the generated particles with the detector, and its response, are implemented using the Geant4 toolkit [11] as described in Ref. [12].
## 3 Data set and selection
The selection of B 0 → J/ψK ∗ 0 candidates is based on the reconstruction of J/ψ → µ + µ -and K ∗ 0 → K π + -decays. The J/ψ candidates are formed from two oppositely charged tracks, identified as muons, having p T > 500 MeV /c and originating from a common vertex. The invariant mass of this pair of muons must lie in the range 3030 -3150 MeV /c 2 . The K ∗ 0 candidates are formed from two oppositely charged tracks, one identified as a kaon and the other as a pion, originating from a common vertex. It is required that the K ∗ 0 candidate has p T > 1 GeV /c and that the invariant mass lies in the range 826 -966 MeV /c 2 .
The B 0 candidates are reconstructed from the J/ψ and K ∗ 0 candidates, with the invariant mass of the µ + µ -pair constrained to the known J/ψ mass. They are required to have an invariant mass in the range 5150 -5400 MeV /c 2 . The decay time of the B 0 candidate is calculated from a vertex and kinematic fit that constrains the candidate to originate from its associated PV [13]. The χ 2 per degree of freedom of the fit is required
to be less than 10. Only B 0 candidates with a decay time greater than 0.2 ps are retained. This lower bound on the decay time rejects a large fraction of the prompt combinatorial background.
In the case of B 0 → D π -+ and B 0 s → D π -s + decays, the selection of the B -meson candidate is based on the reconstruction of D -→ K π π + --and D -s → K K π + --decays, respectively. Requirements are made on the D -( s ) decay products before combining them to form a common vertex. The scalar p T sum of the tracks must exceed 1.8 GeV /c and the maximal distance of closest approach between all possible pairs of tracks must be less than 0.5 mm. The D -( s ) candidate is required to have a significant flight distance with respect to the associated PV, by requiring a χ 2 greater than 36 compared to the zero distance hypothesis. The masses of the D -and D -s candidates must lie within 1850 -1890 MeV /c 2 and 1949 -1989 MeV /c 2 , respectively. They are subsequently combined with a fourth particle, the bachelor pion, to form the B -meson decay vertices. The sum of the D -( s ) and bachelor pion p T values must be larger than 5 GeV /c and the decay time of B -meson candidates must be greater than 0 2 ps. . The cosine of the angle between the B -meson candidate momentum vector and the line segment between the PV and B -meson candidate vertex is required to be larger than 0.999. Particle identification (PID) selection criteria are applied to the kaons and pions from the D -( s ) candidate, and to the bachelor pion, in order to reduce the background from other B -meson decays with a misidentified kaon or pion and from Λ 0 b decays with a misidentified proton to a negligible level.
A final selection is applied to the candidates that satisfy the criteria described above. It uses a multivariate analysis method [14,15], optimized separately for each of the three decay modes, to reject the combinatorial background. The variables used in the selection for the B decay products are the transverse momentum and the impact parameter. For the B candidates the variables employed are the transverse momentum, the distance of flight and the impact parameter.
## 4 Fit model
For each signal and background component, the distributions of invariant mass and decay time of B -meson candidates are modelled by appropriate probability density functions (PDFs). We consider two categories of background: the combinatorial background, due to the random association of tracks, and the partially reconstructed background, due to decays with a topology similar to that of the signal, but with one or more particles not reconstructed. The latter is present only for B 0 ( s ) → D -( s ) π + decays.
## 4.1 Mass model
The signal component for each decay is modelled convolving a double Gaussian function with a function parameterizing the final state radiation. The PDF of the B invariant mass, m , is given by
$$\lambda _ { \nu } ^ { \nu } g ( m ) = A \left [ \Theta ( \mu - m ) \ ( \mu - m ) \right ] ^ { s } \otimes G ( m ),$$
where A is a normalization factor, Θ is the Heaviside function, G is the sum of two Gaussian functions with different widths and zero mean, and µ is the B meson mass. The parameter s glyph[similarequal] -0 99 governs the amount of final state radiation, and is determined . using simulated events for each of the three decay modes. The combinatorial background is modelled by an exponential function for all final states. In the case of B 0 → D π -+ and B 0 s → D π -s + decays, a background component due to partially reconstructed B 0 and B 0 s decays is also present in the low invariant mass region. The main contributions are expected to come from decays with a missing γ or π 0 : B 0 → D ∗-( D γ,D π --0 ) π + decays with D -→ K π π + --; B 0 → D -( K π π + --) ρ + ( π + π 0 ) decays; B 0 s → D ∗-s ( D γ,D π -s -s 0 ) π + decays with D -s → K K π + --; B 0 s → D -s ( K K π + --) ρ + ( π + π 0 ) decays.
We parameterize the partially reconstructed components by means of a kernel estimation technique [16] based on invariant mass distributions obtained from full simulation, using the same selection as for data. In the case of B 0 s → D π -s + decays, there is also a background component due to B 0 → D π + s -decays. We account for this component in the fits using the same parameterization adopted for the signal. The B 0 → D π + s -yield is fixed using the ratio between hadronization fractions measured by LHCb [17,18] and the world average of branching fractions [7].
## 4.2 Decay time model
The time-dependent decay rate of a neutral B 0 ( s ) or B 0 ( s ) meson to a flavour-specific f or ¯ f final state is given by the PDF
$$h \left ( t, \psi \right ) \ = \ & K \left ( 1 - \psi A _ { C P } \right ) \left ( 1 - \psi A _ { f } \right ) \\ & \quad \left \{ e ^ { - \Gamma t } \left [ \Lambda _ { + } \cosh \left ( \frac { \Delta \Gamma t } { 2 } \right ) + \psi \Lambda _ { - } \cos \left ( \Delta m t \right ) \right ] \right \} \otimes R \left ( t \right ) \epsilon \left ( t \right ),$$
where K is a normalization factor, glyph[epsilon1] ( ) is the acceptance as a function of the decay time, t R t ( ) is the decay time resolution function, ∆ m ≡ m H -m L and ∆Γ ≡ Γ L -Γ H are the mass and decay-width differences of the B 0 ( s ) -B 0 ( s ) system mass eigenstates and Γ is the average decay width. The subscripts H and L denote the heavy and light eigenstates, respectively. The two observables are the decay time t and the tag of the final state ψ , which assumes the values ψ = 1 if the final state is f and ψ = -1 if the final state is the CP conjugate ¯ . f The terms Λ + and Λ -are defined as
$$\Lambda _ { \pm } \equiv ( 1 - A _ { P } ) \left | \frac { q } { p } \right | ^ { 1 - \psi } \pm ( 1 + A _ { P } ) \left | \frac { q } { p } \right | ^ { - 1 - \psi },$$
where p and q are complex parameters entering the definition of the two mass eigenstates of the effective Hamiltonian in the B 0 ( s ) system, p B | 0 ( s ) 〉 ± q B | 0 ( s ) 〉 . The symbol A P denotes the production asymmetry of the given B meson, and A f is the detection asymmetry of the final state, defined in terms of the f and ¯ detection efficiencies as f
$$A _ { f } \equiv \frac { \epsilon _ { \bar { f } } - \epsilon _ { f } } { \epsilon _ { \bar { f } } + \epsilon _ { f } }.$$
The direct CP asymmetry A CP is defined as
$$A _ { C P } \equiv \frac { \mathcal { B } \left ( \bar { B } ^ { 0 } _ { ( s ) } \to \bar { f } \right ) - \mathcal { B } \left ( B ^ { 0 } _ { ( s ) } \to f \right ) } { \mathcal { B } \left ( \bar { B } ^ { 0 } _ { ( s ) } \to \bar { f } \right ) + \mathcal { B } \left ( B ^ { 0 } _ { ( s ) } \to f \right ) }.$$
Trigger and event selections lead to distortions in the shapes of the decay time distributions. The signal decay time acceptances are determined from simulated events. For each simulated decay we apply trigger and selection algorithms as in real data.
Concerning the combinatorial and the partially reconstructed backgrounds, empirical parameterizations of the decay time spectra are determined by studying the low and high invariant mass sidebands from data. Partially reconstructed backgrounds are only present in the case of B 0 → D π -+ and B 0 s → D π -s + decays. In the case of B 0 s → D π -s + decays, the additional background component due to B 0 → D π + s -decays is modelled using the same functional form as that of the B 0 s → D π -s + signal, and the value of the production asymmetry is fixed to that obtained from the B 0 → D π -+ fit.
## 4.3 Decay time resolution
The strategy adopted to study the decay time resolution of the detector consists of reconstructing the decay time of fake B candidates formed from a D -decaying to K π π + --and a pion track, both coming from the same PV. The bachelor pion must be selected without introducing biases on the decay time, hence only requirements on momentum and transverse momentum are applied, avoiding the use of impact parameter variables. The decay time distribution of these fake B candidates yields an estimate of the decay time resolution of a real decay. In order to validate the method, simulated events are used for both signals and fake B decays. The resolution is found to be overestimated by about 4 fs. This difference is taken into account as a systematic effect. The simulation also indicates that a dependence of the resolution on the decay time must be considered. Taking this into account, an average decay time resolution of 49 ± 8 fs is estimated. A resolution model, R t ( ), consisting of a triple Gaussian function with zero mean and three different widths, characterized by an average width of 49 fs, is used. The uncertainty of 8 fs on the average width is taken into account as a systematic uncertainty. It is estimated from simulation that the measurement of the decay time is biased by no more than 2 fs, and the effect is accounted for as a systematic uncertainty.
## 5 Determination of the production asymmetries
The production asymmetry for each of the three decay modes is determined by means of a simultaneous fit to the invariant mass and decay time spectra. To account for the dependence of the production asymmetries on the kinematics of the B 0 and B 0 s mesons, each data sample must be divided into bins of ( p T , η ), performing the same fit for each bin.
Table 1: Values of the various physical inputs used in the fits.
| Parameter | Value | Reference |
|-----------------|---------------------|-------------|
| ∆ m d [ps - 1 ] | 0 . 510 ± 0 . 004 | [7] |
| ∆ m s [ps - 1 ] | 17 . 768 ± 0 . 024 | [19] |
| Γ d [ps - 1 ] | 0 . 6583 ± 0 . 0030 | [7] |
| Γ s [ps - 1 ] | 0 . 6596 ± 0 . 0046 | [7] |
| ∆Γ s [ps - 1 ] | 0 . 081 ± 0 . 011 | [7] |
| | q/p | B 0 | 0 . 9997 ± 0 . 0013 | [20] |
| | q/p | B 0 s | 1 . 0003 ± 0 . 0030 | [21] |
In order to validate the fit model, a series of fits to the distributions of events obtained from fast simulations is used to verify the accuracy of the central values and the reliability of the uncertainties. No evidence of biases on central values nor of uncertainty misestimations is found. Furthermore, a global fit to the total sample of selected events is performed for each of the three decay modes. The mass differences ∆ m d and ∆ m s , the mixing parameters | q/p | B 0 and | q/p | B 0 s , the average decay widths Γ d and Γ , and the width difference ∆Γ s s are fixed to the central values of the measurements reported in Table 1. The width difference ∆Γ d is fixed to zero.
According to Eq. 3, for small values of A CP and A f , to first order the decay rate is only sensitive to the sum of these two quantities. For this reason, we fix A CP to zero and leave A f as a free parameter in the fits. It is empirically verified that the choice of different A CP values, up to the few percent level, leads to negligible variations of A P , as expected.
Figure 1 shows the J/ψK + π -, K π π π + --+ and K K π π + --+ invariant mass and decay time distributions, with the results of the global fits overlaid. Figure 2 shows the raw asymmetries, defined as the ratios between the difference and the sum of the overall decay time distributions, as a function of decay time for candidates in the signal mass region. The signal yields, A P values and detection asymmetries obtained from the global fits are reported in Table 2. The A P values obtained from the global fits are not well defined physical quantities, because efficiency corrections as a function of p T and η need to be applied. They are reported here for illustrative purposes only.
Figure 3 shows the two-dimensional distributions of ( p T , η ) for B 0 → J/ψK ∗ 0 , B 0 → D π -+ and B 0 s → D π -s + decays. The background components are subtracted using the sPlot technique [22] and the chosen definition of the various kinematic bins is overlaid. For the two B 0 decays we use a common set of bins, as reported in Table 3, in order to allow a simple combination of the two independent A P measurements. In the case of the B 0 → J/ψK ∗ 0 , two additional bins at small p T and large η are also defined. An accurate knowledge of the decay time resolution is important for B 0 s → D π -s + decay, due to the fast oscillation of the B 0 s meson. For this reason we determine the decay time resolution using the method previously described, applied to events belonging to each ( p T , η ) bin,
Figure 1: Distributions of (left) invariant mass and (right) decay time for (top) B 0 → J/ψK ∗ 0 , (middle) B 0 → D π -+ and (bottom) B 0 s → D π -s + decays, with the results of the fit overlaid. The contributions of the various background sources are also shown.
where a double Gaussian function with zero mean and values of the widths depending on the given bin is used.
Figure 2: Time-dependent raw asymmetries for candidates in the (a) B 0 → J/ψK ∗ 0 , (b) B 0 → D π -+ and (c) B 0 s → D π -s + signal mass regions with the results of the global fits overlaid. In (c) the asymmetry is obtained by folding the B 0 s and B 0 s decay time distributions into one oscillation period, and the offset t 0 = 0 2 ps corresponds to the selection requirement on the . decay time.
Table 2: Values of signal yields, A P , A f and of the correlations ρ A ( P , A f ) obtained from global fits. The smaller value of the correlation in the B 0 s case is due to the much larger mixing frequency of B 0 s mesons.
| Parameter | B 0 → J/ψK ∗ 0 | B 0 → D - π + | B 0 s → D - s π + |
|-----------------|-----------------------|-----------------------|-----------------------|
| N sig | 93 627 ± 360 | 76 682 ± 308 | 16 887 ± 174 |
| A P | - 0 . 0116 ± 0 . 0063 | - 0 . 0058 ± 0 . 0070 | - 0 . 0032 ± 0 . 0166 |
| A f | - 0 . 0086 ± 0 . 0046 | - 0 . 0151 ± 0 . 0049 | - 0 . 0110 ± 0 . 0086 |
| ρ ( A P , A f ) | - 0 . 65 | - 0 . 64 | - 0 . 01 |
T
Figure 3: Distributions of p T and η , where the background components are subtracted using the sPlot technique [22], for (a) B 0 → J/ψK ∗ 0 , (b) B 0 → D π -+ and (c) B 0 s → D π -s + decays. The definition of the various kinematic bins is superimposed.
## 6 Systematic uncertainties
Several sources of systematic uncertainty that affect the determination of the production asymmetries are considered. For the invariant mass model, the effects of the uncertainty on the shapes of all components (signals, combinatorial and partially reconstructed backgrounds) are investigated. For the decay time model, systematic effects related to the decay time resolution and acceptance are studied. The effects of the uncertainties on the external inputs used in the fits, reported in Table 1, are evaluated by repeating the fits with each parameter varied by ± 1 σ . Alternative parameterizations of the background components are also considered. To estimate the contribution of each single source, we repeat the fit for each ( p T , η ) bin after having modified the baseline fit model. The shifts from the relevant baseline values are taken as the systematic uncertainties. To estimate a systematic uncertainty related to the parameterization of final state radiation effects on the signal mass distributions, the parameter s of Eq. 2 is varied by ± 1 σ of the corresponding value obtained from fits to simulated events. A systematic uncertainty related to the invariant mass resolution model is estimated by repeating the fit using a single Gaussian function. The systematic uncertainty related to the parameterization of the mass shape for the combinatorial background is investigated by replacing the exponential function with a
straight line. Concerning the partially reconstructed background, we assess a systematic uncertainty by repeating the fits while excluding the low mass sideband, i.e. applying the requirements m > 5 20 GeV . /c 2 for the B 0 → D π -+ decays and m > 5 33 GeV . /c 2 for B 0 s → D π -s + decays. To estimate the uncertainty related to the parameterization of signal decay time acceptances, different acceptance functions are considered. Effects of inaccuracies in the knowledge of the decay time resolution are estimated by rescaling the widths of the baseline model to obtain an average resolution width differing by ± 8 fs. Simulation studies also indicate that there is a small bias in the reconstructed decay time. The impact of such a bias is assessed by introducing a corresponding bias of ± 2 fs in the decay time resolution model.
The determination of the systematic uncertainties related to the | q/p | input value needs a special treatment, as A P is correlated with | q/p | . For this reason, any variation of | q/p | turns into the same shift of A P in each of the kinematic bins. Such a correlation is taken into account when averaging A P ( B 0 ) measurements from B 0 → J/ψK ∗ 0 and B 0 → D π -+ decays, or when integrating over p T and η . The values of the systematic uncertainties related to the knowledge of | q/p | are 0.0013 in the case of A P ( B 0 ) and 0.0030 in the case of A P ( B 0 s ). The dominant systematic uncertainties for the B 0 → J/ψK ∗ 0 decay are related to the signal mass shape and to | q/p | . For the B 0 → D π -+ decay, the most relevant systematic uncertainties are related to the signal mass shape and to the partially reconstructed background. Systematic uncertainties associated with the decay time resolution and ∆ m s are the main sources for the B 0 s → D π -s + decay.
## 7 Results
The values of A P ( B 0 ) are determined independently for B 0 → J/ψK ∗ 0 and B 0 → D π -+ decays in each kinematic bin and then averaged. Table 3 reports the final results. The overall bin-by-bin agreement between the two sets of independent A P ( B 0 ) measurements is evaluated by means of a χ 2 test, with a χ 2 = 7 for 14 degrees of freedom. The values of A P ( B 0 s ) determined from the B 0 s → D π -s + fits are reported in Table 4.
The integration over p T and η of the bin-by-bin A P values is performed within the ranges 4 < p T < 30 GeV /c and 2 5 . < η < 4 5. . The integrated value of A P is given by
$$A _ { P } = \frac { \sum _ { i } \frac { N _ { i } } { \varepsilon _ { i } } A _ { P, i } } { \sum _ { i } \frac { N _ { i } } { \varepsilon _ { i } } },$$
where the index i runs over the bins, N i is the number of signal events and ε i is the efficiency, defined as the number of selected events divided by the number of produced events in the i -th bin. The signal yield in each bin can be expressed as
$$N _ { i } = \mathcal { L } \cdot \sigma _ { b \bar { b } } \cdot 2 \cdot f _ { d ( s ) } \cdot \mathcal { B } \cdot f _ { i } \cdot \varepsilon _ { i },$$
where L is the integrated luminosity, σ bb ¯ is the bb ¯ cross section, f d s ( ) is the B 0 ( s ) hadronization fraction, f i is the fraction of B mesons produced in the i -th bin and B is the branching
Table 3: Combined values of A P ( B 0 ) from B 0 → J/ψK ∗ 0 and B 0 → D π -+ decays, corresponding to the various kinematic bins. The first uncertainties are statistical and the second systematic. For completeness, the values obtained either from B 0 → J/ψK ∗ 0 or B 0 → D π -+ decays are also reported in the last two columns, with statistical uncertainties only. The values of the last two bins are obtained from B 0 → J/ψK ∗ 0 decays alone.
| p T ( GeV /c ) | η | A P ( B 0 ) | A P ( B 0 → J/ψK ∗ 0 ) | A P ( B 0 → D - π + ) |
|-------------------|-----------------|----------------------------------|--------------------------|-------------------------|
| ( 1 . 0 , 4 . 0) | (4 . 5 , 5 . 2) | 0 . 0016 ± 0 . 0253 ± 0 . 0016 | 0 . 0037 ± 0 . 0260 | - 0 . 0331 ± 0 . 1044 |
| ( 1 . 0 , 4 . 0) | (3 . 7 , 4 . 5) | - 0 . 0158 ± 0 . 0162 ± 0 . 0015 | - 0 . 0161 ± 0 . 0170 | - 0 . 0130 ± 0 . 0519 |
| ( 2 . 0 , 4 . 0) | (3 . 0 , 3 . 7) | 0 . 0055 ± 0 . 0254 ± 0 . 0016 | 0 . 0078 ± 0 . 0271 | - 0 . 0114 ± 0 . 0738 |
| ( 4 . 0 , 12 . 0) | (4 . 5 , 4 . 7) | 0 . 0160 ± 0 . 0736 ± 0 . 0067 | - 0 . 0489 ± 0 . 0840 | 0 . 2353 ± 0 . 1529 |
| ( 4 . 0 , 7 . 0) | (3 . 7 , 4 . 5) | - 0 . 0189 ± 0 . 0158 ± 0 . 0032 | - 0 . 0221 ± 0 . 0184 | - 0 . 0099 ± 0 . 0310 |
| ( 4 . 0 , 7 . 0) | (3 . 0 , 3 . 7) | - 0 . 0311 ± 0 . 0132 ± 0 . 0014 | - 0 . 0342 ± 0 . 0160 | - 0 . 0245 ± 0 . 0232 |
| ( 4 . 0 , 7 . 0) | (2 . 5 , 3 . 0) | 0 . 0556 ± 0 . 0254 ± 0 . 0020 | 0 . 0703 ± 0 . 0324 | 0 . 0321 ± 0 . 0408 |
| ( 7 . 0 , 12 . 0) | (3 . 7 , 4 . 5) | - 0 . 0145 ± 0 . 0205 ± 0 . 0027 | - 0 . 0364 ± 0 . 0269 | 0 . 0161 ± 0 . 0316 |
| ( 7 . 0 , 12 . 0) | (3 . 0 , 3 . 7) | - 0 . 0142 ± 0 . 0111 ± 0 . 0015 | - 0 . 0067 ± 0 . 0173 | - 0 . 0196 ± 0 . 0146 |
| ( 7 . 0 , 12 . 0) | (2 . 5 , 3 . 0) | - 0 . 0236 ± 0 . 0138 ± 0 . 0014 | - 0 . 0341 ± 0 . 0228 | - 0 . 0175 ± 0 . 0173 |
| ( 7 . 0 , 12 . 0) | (2 . 2 , 2 . 5) | - 0 . 0190 ± 0 . 0348 ± 0 . 0034 | - 0 . 0397 ± 0 . 0623 | - 0 . 0096 ± 0 . 0420 |
| (12 . 0 , 30 . 0) | (3 . 7 , 4 . 5) | - 0 . 0550 ± 0 . 0473 ± 0 . 0020 | - 0 . 0195 ± 0 . 0649 | - 0 . 0951 ± 0 . 0690 |
| (12 . 0 , 30 . 0) | (3 . 0 , 3 . 7) | 0 . 0067 ± 0 . 0180 ± 0 . 0021 | - 0 . 0193 ± 0 . 0311 | 0 . 0199 ± 0 . 0220 |
| (12 . 0 , 30 . 0) | (2 . 5 , 3 . 0) | 0 . 0177 ± 0 . 0162 ± 0 . 0023 | 0 . 0295 ± 0 . 0314 | 0 . 0134 ± 0 . 0190 |
| (12 . 0 , 30 . 0) | (2 . 0 , 2 . 5) | - 0 . 0018 ± 0 . 0236 ± 0 . 0020 | 0 . 0031 ± 0 . 0485 | - 0 . 0033 ± 0 . 0270 |
| ( 0 . 2 , 1 . 0) | (4 . 5 , 6 . 0) | - 0 . 0391 ± 0 . 0501 ± 0 . 0016 | - 0 . 0391 ± 0 . 0501 | - |
| ( 1 . 0 , 2 . 2) | (5 . 2 , 6 . 0) | 0 . 0523 ± 0 . 0684 ± 0 . 0025 | 0 . 0523 ± 0 . 0684 | - |
Table 4: Values of A P ( B 0 s ) from B 0 s → D π -s + decays, corresponding to the various kinematic bins. The first uncertainties are statistical and the second systematic.
| p T ( GeV /c | η | A P ( B 0 s ) |
|----------------|-----------------|----------------------------------|
| ( 2 , 4) | (3 . 0 , 5 . 0) | - 0 . 1475 ± 0 . 0895 ± 0 . 0192 |
| ( 4 , 8) | (3 . 5 , 4 . 5) | - 0 . 0471 ± 0 . 0513 ± 0 . 0112 |
| ( 4 , 8) | (2 . 5 , 3 . 5) | 0 . 0376 ± 0 . 0467 ± 0 . 0083 |
| ( 8 , 12) | (3 . 5 , 4 . 5) | 0 . 0582 ± 0 . 0537 ± 0 . 0053 |
| ( 8 , 12) | (2 . 5 , 3 . 5) | 0 . 0370 ± 0 . 0332 ± 0 . 0051 |
| (12 , 30) | (3 . 5 , 4 . 5) | - 0 . 0339 ± 0 . 0750 ± 0 . 0095 |
| (12 , 30) | (2 . 5 , 3 . 5) | - 0 . 0333 ± 0 . 0309 ± 0 . 0040 |
| ( 8 , 30) | (2 . 2 , 2 . 5) | - 0 . 0351 ± 0 . 0485 ± 0 . 0059 |
fraction of the B decay. By substituting N /ε i i from Eq. 8 into Eq. 7, the integrated value of A P becomes
$$A _ { P } = \sum _ { i } \omega _ { i } A _ { P, i } \,,$$
where ω i = f / i ∑ i f i . The values of ω i are determined using simulated events. The
Table 5: Values of ω i determined from simulation and ω data i extracted from data using B 0 → J/ψK ∗ 0 decays in two different binning schemes. The ω i values and the difference between ω i and ω data i values are used to determine the integrated results and to evaluate the related systematic uncertainties, respectively.
| p T ( GeV /c ) | η | ω i | ω data i |
|------------------|-----------------|---------------------|---------------------|
| ( 4 , 7) | (3 . 7 , 4 . 5) | 0 . 1698 ± 0 . 0008 | 0 . 1946 ± 0 . 0025 |
| ( 4 , 7) | (3 . 0 , 3 . 7) | 0 . 2432 ± 0 . 0009 | 0 . 2396 ± 0 . 0036 |
| ( 4 , 7) | (2 . 5 , 3 . 0) | 0 . 2222 ± 0 . 0009 | 0 . 1976 ± 0 . 0051 |
| ( 7 , 12) | (3 . 7 , 4 . 5) | 0 . 0662 ± 0 . 0006 | 0 . 0789 ± 0 . 0016 |
| ( 7 , 12) | (3 . 0 , 3 . 7) | 0 . 1129 ± 0 . 0007 | 0 . 1129 ± 0 . 0045 |
| ( 7 , 12) | (2 . 5 , 3 . 0) | 0 . 1150 ± 0 . 0007 | 0 . 1002 ± 0 . 0019 |
| (12 , 30) | (3 . 7 , 4 . 5) | 0 . 0113 ± 0 . 0003 | 0 . 0160 ± 0 . 0007 |
| (12 , 30) | (3 . 0 , 3 . 7) | 0 . 0276 ± 0 . 0004 | 0 . 0307 ± 0 . 0028 |
| (12 , 30) | (2 . 5 , 3 . 0) | 0 . 0318 ± 0 . 0004 | 0 . 0296 ± 0 . 0025 |
| ( 4 , 8) | (3 . 5 , 4 . 5) | 0 . 2667 ± 0 . 0009 | 0 . 3064 ± 0 . 0020 |
| ( 4 , 8) | (2 . 5 , 3 . 5) | 0 . 4766 ± 0 . 0009 | 0 . 4644 ± 0 . 0030 |
| ( 8 , 12) | (3 . 5 , 4 . 5) | 0 . 0564 ± 0 . 0005 | 0 . 0640 ± 0 . 0015 |
| ( 8 , 12) | (2 . 5 , 3 . 5) | 0 . 1295 ± 0 . 0008 | 0 . 0873 ± 0 . 0019 |
| (12 , 30) | (3 . 5 , 4 . 5) | 0 . 0175 ± 0 . 0003 | 0 . 0238 ± 0 . 0008 |
| (12 , 30) | (2 . 5 , 3 . 5) | 0 . 0532 ± 0 . 0005 | 0 . 0541 ± 0 . 0027 |
difference between the values of ω i predicted by Pythia for B 0 and B 0 s mesons is found to be negligible, if the same bins in p T and η would be used. These values are also extracted from data using B 0 → J/ψK ∗ 0 decays. In this case ω data i is measured as
$$\omega _ { i } ^ { \text{data} } = \frac { N _ { i } } { \varepsilon _ { i } ^ { \text{rec} } } \, / \, \sum _ { j } \frac { N _ { j } } { \varepsilon _ { j } ^ { \text{rec} } },$$
where N i is the yield in the i -th bin and ε rec i is total reconstruction efficiency. The values of ε rec i are determined using both simulated events and data control samples. The values of ω i and ω data i , summarized in Table 5, exhibit systematic differences at the 10% level. The difference in the central value between A P ( B 0 → J/ψK ∗ 0 ) calculated using either ω i or ω data i is found to be 0.0024 using the B 0 binning scheme, and 0.0034 using the B 0 s binning scheme. These values are assigned as systematic uncertainties for A P ( B 0 ) and A P ( B 0 s ). Table 6 summarizes the systematic uncertainties associated with the integrated measurements. In the first row, the combined systematic uncertainties estimated in each bin, as described in the previous section, are reported.
Using Eq. 9, the integrated measurements of A P ( B 0 ) for B 0 → J/ψK ∗ 0 and B 0 → D π -+ decays are found to be
$$\dots$$
$$A _ { P } ( B ^ { 0 } \to J / \psi K ^ { * 0 } ) & = \ - 0. 0 0 3 3 \pm 0. 0 0 9 6 \, ( \text{stat} ) \pm 0. 0 0 2 8 \, ( \text{syst} ), \\ A _ { P } ( B ^ { 0 } \to D ^ { - } \pi ^ { + } ) & = \ - 0. 0 0 3 8 \pm 0. 0 1 2 4 \, ( \text{stat} ) \pm 0. 0 0 2 9 \, ( \text{syst} ),$$
Table 6: Absolute values of systematic uncertainties. The total systematic uncertainties are obtained by summing the individual contributions in quadrature.
| Source | Uncertainty | Uncertainty |
|----------------------------------------------------|---------------|---------------|
| | A P ( B 0 ) | A P ( B 0 s ) |
| Combined systematic uncertainties from bin studies | 0.0004 | 0.0048 |
| Uncertainty on | q/p | | 0.0013 | 0.0030 |
| Difference between ω i and ω data i | 0.0024 | 0.0034 |
| Total | 0.0028 | 0.0066 |
which lead to the average
$$A _ { P } ( B ^ { 0 } ) = - 0. 0 0 3 5 \pm 0. 0 0 7 6 \, ( \text{stat} ) \pm 0. 0 0 2 8 \, ( \text{syst} ).$$
The integrated value of A P ( B 0 s ) is
$$A _ { \text{P} } ( B _ { s } ^ { 0 } ) = \ 0. 0 1 0 9 \pm 0. 0 2 6 1 \, ( \text{stat} ) \pm 0. 0 0 6 6 \, ( \text{syst} ).$$
Finally, the dependencies of A P ( B 0 ) and A P ( B 0 s ) on p T , obtained by integrating over η , and on η , obtained by integrating over p T , are shown in Fig. 4. The corresponding numerical values are reported in Tables 7 and 8.
T
Figure 4: Dependence of (top) A P ( B 0 ) and (bottom) A P ( B 0 s ) on (left) p T and (right) η . The error bars include both statistical and systematic uncertainties.
Table 7: Values of the production asymmetry A P ( B 0 ) in bins of p T and η from B 0 → J/ψK ∗ 0 and B 0 → D π -+ decays. The first uncertainties are statistical and the second systematic.
| Variable | Bin | A P ( B 0 ) |
|---------------|--------------------|---------------------------------------------------------------------|
| p T (GeV /c ) | ( 4 , 7) ( 7 , 12) | 0 . 0033 ± 0 . 0111 ± 0 . 0028 - 0 . 0167 ± 0 . 0084 ± 0 . 0028 ± ± |
| p T (GeV /c ) | (12 , | 0 . 0001 0 . 0130 0 . 0029 |
| p T (GeV /c ) | 30) | |
| η | (2 . 5 , 3 . 0) | 0 . 0264 ± 0 . 0161 ± 0 . 0030 |
| η | (3 . 0 , 3 . 7) | - 0 . 0232 ± 0 . 0093 ± 0 . 0028 |
| η | (3 . 7 , 4 . 5) | - 0 . 0203 ± 0 . 0125 ± 0 . 0021 |
Table 8: Values of the production asymmetry A P ( B 0 s ) in bins of p T and η from B 0 s → D π -s + decays. The first uncertainties are statistical and the second systematic.
| Variable | Bin | A P ( B 0 s ) |
|---------------|--------------------|---------------------------------------------------------------|
| p T (GeV /c ) | ( 4 , 8) ( 8 , 12) | 0 . 0069 ± 0 . 0351 ± 0 . 0067 0 . 0435 ± 0 . 0283 ± 0 . 0039 |
| p T (GeV /c ) | (12 , 30) | - 0 . 0334 ± 0 . 0296 ± 0 . 0038 |
| η | (2.5, 3.5) | 0 . 0315 ± 0 . 0342 ± 0 . 0060 0 . 0286 ± 0 . 0412 ± 0 . |
| η | (3.5, 4.5) | - 0088 |
## 8 Conclusions
The production asymmetries of B 0 and B 0 s mesons have been measured in pp collisions at √ s = 7 TeV within the acceptance of the LHCb detector, using a data sample corresponding to an integrated luminosity of 1.0 fb -1 . The measurements have been performed in bins of p T and η , and provide constraints that can be used to test different models of B -meson production. Furthermore, once integrated using appropriate weights for any reconstructed B 0 ( s ) decay mode, they can be used to derive effective production asymmetries, as inputs for CP violation measurements with the LHCb detector.
The values of the production asymmetries integrated in the ranges 4 < p T < 30 GeV /c and 2 5 . < η < 4 5 have been determined to be .
$$A _ { \text{P} } ( B ^ { 0 } ) = ( - 0. 3 5 \pm 0. 7 6 \pm 0. 2 8 ) \%,$$
$$A _ { \text{P} } ( B _ { s } ^ { 0 } ) = ( \ 1. 0 9 \pm 2. 6 1 \pm 0. 6 6 ) \%,$$
$$\begin{matrix} \cdot \\ \cdot \end{matrix}$$
where the first uncertainties are statistical and the second systematic. No clear evidence of dependences on the values of p T and η has been observed.
## 9 Acknowledgments
We express our gratitude to our colleagues in the CERN accelerator departments for the excellent performance of the LHC. We thank the technical and administrative staff at the LHCb institutes. We acknowledge support from CERN and from the national agencies: CAPES, CNPq, FAPERJ and FINEP (Brazil); NSFC (China); CNRS/IN2P3 (France); BMBF, DFG, HGF and MPG (Germany); SFI (Ireland); INFN (Italy); FOM and NWO (The Netherlands); MNiSW and NCN (Poland); MEN/IFA (Romania); MinES and FANO (Russia); MinECo (Spain); SNSF and SER (Switzerland); NASU (Ukraine); STFC (United Kingdom); NSF (USA). The Tier1 computing centres are supported by IN2P3 (France), KIT and BMBF (Germany), INFN (Italy), NWO and SURF (The Netherlands), PIC (Spain), GridPP (United Kingdom). We are indebted to the communities behind the multiple open source software packages on which we depend. We are also thankful for the computing resources and the access to software R&D tools provided by Yandex LLC (Russia). Individual groups or members have received support from EPLANET, Marie Skglyph[suppress]odowska-Curie l Actions and ERC (European Union), Conseil g´ en´ral de Haute-Savoie, e Labex ENIGMASS and OCEVU, R´ egion Auvergne (France), RFBR (Russia), XuntaGal and GENCAT (Spain), Royal Society and Royal Commission for the Exhibition of 1851 (United Kingdom).
## References
- [1] M. Chaichian and A. Fridman, On a possibility for measuring effects of CP violation at pp colliders , Phys. Lett. B298 (1993) 218.
- [2] E. Norrbin and R. Vogt, Bottom production asymmetries at the LHC , arXiv:hep-ph/0003056 .
- [3] E. Norrbin and T. Sj¨strand, o Production and hadronization of heavy quarks , Eur. Phys. J. C17 (2000) 137, arXiv:hep-ph/0005110 .
- [4] LHCb collaboration, R. Aaij et al. , Measurement of the D ± production asymmetry in 7 TeV pp collisions , Phys. Lett. B718 (2013) 902, arXiv:1210.4112 .
- [5] LHCb collaboration, R. Aaij et al. , Measurement of the D + s -D -s production asymmetry in 7 TeV pp collisions , Phys. Lett. B713 (2012) 186, arXiv:1205.0897 .
- [6] LHCb collaboration, A. A. Alves Jr. et al. , The LHCb detector at the LHC , JINST 3 (2008) S08005.
- [7] Particle Data Group, J. Beringer et al. , Review of particle physics , Phys. Rev. D86 (2012) 010001, and 2013 partial update for the 2014 edition.
- [8] V. V. Gligorov and M. Williams, Efficient, reliable and fast high-level triggering using a bonsai boosted decision tree , JINST 8 (2013) P02013, arXiv:1210.6861 .
| [9] | T. Sj¨strand, o S. Mrenna, and P. Skands, PYTHIA 6.4 physics and manual , JHEP 05 (2006) 026, arXiv:hep-ph/0603175 . |
|-------|-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [10] | I. Belyaev et al. , Handling of the generation of primary events in Gauss , the LHCb simulation framework , Nuclear Science Symposium Conference Record (NSS/MIC) IEEE (2010) 1155. |
| [11] | GEANT4 collaboration, J. Allison et al. , Geant4 developments and applications , IEEE Trans. Nucl. Sci. 53 (2006) 270; GEANT4 collaboration, S. Agostinelli et al. , GEANT4: A simulation toolkit , Nucl. Instrum. Meth. A506 (2003) 250. |
| [12] | M. Clemencic et al. , The LHCb simulation application, Gauss : design, evolution and experience , J. Phys. Conf. Ser. 331 (2011) 032023. |
| [13] | W. D. Hulsbergen, Decay chain fitting with a Kalman filter , Nucl. Instrum. Meth. A552 (2005) 566, arXiv:physics/0503191 . |
| [14] | L. Breiman, J. H. Friedman, R. A. Olshen, and C. J. Stone, Classification and regression trees , Wadsworth international group, Belmont, California, USA, 1984. |
| [15] | R. E. Schapire and Y. Freund, A decision-theoretic generalization of on-line learning and an application to boosting , Jour. Comp. and Syst. Sc. 55 (1997) 119. |
| [16] | K. S. Cranmer, Kernel estimation in high-energy physics , Comput. Phys. Commun. 136 (2001) 198, arXiv:hep-ex/0011057 . |
| [17] | LHCb collaboration, R. Aaij et al. , Measurement of the fragmentation fraction ratio f s /f d and its dependence on B meson kinematics , JHEP 04 (2013) 001, arXiv:1301.5286 . |
| [18] | LHCb collaboration, R. Aaij et al. , Measurement of b hadron production fractions in 7 TeV pp collisions , Phys. Rev. D85 (2012) 032008, arXiv:1111.2357 . |
| [19] | LHCb collaboration, R. Aaij et al. , Precision measurement of the B 0 s - ¯ B 0 s oscil- lation frequency with the decay B 0 s → D - s π + , New J. Phys. 15 (2013) 053021, arXiv:1304.4741 . |
| [20] | Heavy Flavor Averaging Group, Y. Amhis et al. , Averages of b -hadron, c -hadron, and τ -lepton properties as of early 2012 , arXiv:1207.1158 , update available online at http://www.slac.stanford.edu/xorg/hfag . |
| [21] | LHCb collaboration, R. Aaij et al. , Measurement of the flavour-specific CP -violating asymmetry a s sl in B 0 s decays , Phys. Lett. B728 (2014) 607, arXiv:1308.1048 . |
| [22] | M. Pivk and F. R. Le Diberder, sPlot: a statistical tool to unfold data distributions , Nucl. Instrum. Meth. A555 (2005) 356, arXiv:physics/0402083 . |
## LHCb collaboration
| R. Aaij 41 , B. Adeva 37 , M. Adinolfi 46 , A. Affolder 52 , Z. Ajaltouni 5 , S. Akar 6 , J. Albrecht 9 , F. Alessio 38 , M. Alexander 51 , S. Ali 41 , G. Alkhazov 30 , P. Alvarez Cartelle 37 , A.A. Alves Jr 25 , 38 S. Amato 2 , S. Amerio 22 , Y. Amhis 7 , L. An 3 , L. Anderlini 17 ,g , J. Anderson 40 , R. Andreassen 57 , M. Andreotti 16 ,f , J.E. Andrews 58 , R.B. Appleby 54 , O. Aquines Gutierrez 10 , F. Archilli 38 , A. Artamonov 35 , M. Artuso 59 , E. Aslanides 6 , G. Auriemma 25 ,n , M. Baalouch 5 , S. Bachmann 11 J.J. Back 48 , A. Badalov 36 , W. Baldini 16 , R.J. Barlow 54 , C. Barschel 38 , S. Barsuk 7 , W. Barter 47 V. Batozskaya 28 , V. Battista 39 , A. Bay 39 , L. Beaucourt 4 , J. Beddow 51 , F. Bedeschi 23 , I. Bediaga 1 , S. Belogurov 31 , K. Belous 35 , I. Belyaev 31 , E. Ben-Haim 8 , G. Bencivenni 18 , S. Benson 38 , J. Benton 46 , A. Berezhnoy 32 , R. Bernet 40 , M.-O. Bettler 47 , M. van Beuzekom 41 , A. Bien 11 , S. Bifani 45 , T. Bird 54 , A. Bizzeti 17 ,i , P.M. Bjørnstad 54 , T. Blake 48 , F. Blanc 39 , J. Blouw 10 , S. Blusk 59 , V. Bocci 25 , A. Bondar 34 , N. Bondar 30 , 38 , W. Bonivento 15 , 38 , S. Borghi A. Borgia 59 , M. Borsato 7 , T.J.V. Bowcock 52 , E. Bowen 40 , C. Bozzi 16 , T. Brambach 9 , J. van den Brand 42 , J. Bressieux 39 , D. Brett 54 , M. Britsch 10 , T. Britton 59 , J. Brodzicka 54 , N.H. Brook 46 , H. Brown 52 , A. Bursche 40 , G. Busetto 22 ,r , J. Buytaert 38 , S. Cadeddu 15 , R. Calabrese 16 ,f , M. Calvi 20 ,k , M. Calvo Gomez 36 ,p , P. Campana 18 , 38 , D. Campora Perez 38 , A. Carbone 14 ,d , G. Carboni 24 ,l , R. Cardinale 19 , 38 ,j , A. Cardini 15 , L. Carson 50 , K. Carvalho Akiba 2 , G. Casse 52 , L. Cassina 20 , L. Castillo Garcia 38 , M. Cattaneo 38 , Ch. Cauet R. Cenci 58 , M. Charles 8 , Ph. Charpentier 38 , M. Chefdeville 4 , S. Chen 54 , S.-F. Cheung 55 , N. Chiapolini 40 , M. Chrzaszcz 40 , 26 , K. Ciba 38 , X. Cid Vidal 38 , G. Ciezarek 53 , P.E.L. Clarke 50 , M. Clemencic 38 , H.V. Cliff 47 , J. Closier 38 , V. Coco 38 , J. Cogan 6 , E. Cogneras 5 , L. Cojocariu P. Collins 38 , A. Comerma-Montells 11 , A. Contu 15 , A. Cook 46 , M. Coombes 46 , S. Coquereau 8 , G. Corti 38 , M. Corvo 16 ,f , I. Counts 56 , B. Couturier 38 , G.A. Cowan 50 , D.C. Craik 48 , M. Cruz Torres 60 , S. Cunliffe 53 , R. Currie 50 , C. D'Ambrosio 38 , J. Dalseno 46 , P. David 8 , P.N.Y. David 41 , A. Davis 57 , K. De Bruyn 41 , S. De Capua 54 , M. De Cian 11 , J.M. De Miranda 1 |
|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| 54 9 29 , L. De Paula 2 , W. De Silva 57 , P. De Simone 18 , D. Decamp 4 , M. Deckenhoff 9 , L. Del Buono 8 , N. D´l´age e e 4 , D. Derkach 55 , O. Deschamps 5 , F. Dettori 38 , A. Di Canto 38 , H. Dijkstra 38 , S. Donleavy 52 , F. Dordei 11 , M. Dorigo 39 , A. Dosil Su´rez a 37 , D. Dossett 48 , A. Dovbnya 43 , K. Dreimanis 52 , G. Dujany 54 , F. Dupertuis 39 , P. Durante 38 , R. Dzhelyadin 35 , A. Dziurda 26 , A. Dzyuba 30 , S. Easo 49 , 38 , U. Egede 53 , V. Egorychev 31 , S. Eidelman 34 , S. Eisenhardt 50 , U. Eitschberger 9 , R. Ekelhof 9 , L. Eklund 51 , I. El Rifai 5 , Ch. Elsasser 40 , S. Ely 59 , S. Esen 11 , H.-M. Evans 47 , T. Evans 55 , A. Falabella 14 , C. F¨rber a 11 , C. Farinelli 41 , N. Farley 45 , S. Farry 52 , |
- J. He 38 , T. Head 38 , V. Heijne 41 , K. Hennessy 52 , P. Henrard , L. Henry , 5 8
- J.A. Hernando Morata 37 , E. van Herwijnen 38 , M. Heß 62 , A. Hicheur , D. Hill 1 55 , M. Hoballah , 5
- C. Hombach 54 , W. Hulsbergen 41 , P. Hunt 55 , N. Hussain 55 , D. Hutchcroft 52 , D. Hynds 51 ,
- M. Idzik 27 , P. Ilten 56 , R. Jacobsson 38 , A. Jaeger 11 , J. Jalocha 55 , E. Jans 41 , P. Jaton 39 ,
- A. Jawahery 58 , F. Jing , M. John 3 55 , D. Johnson 55 , C.R. Jones 47 , C. Joram 38 , B. Jost 38 ,
- N. Jurik 59 , M. Kaballo , S. Kandybei 9 43 , W. Kanso , M. Karacson 6 38 , T.M. Karbach 38 ,
- S. Karodia 51 , M. Kelsey 59 , I.R. Kenyon 45 , T. Ketel 42 , B. Khanji 20 , C. Khurewathanakul 39 ,
- S. Klaver 54 , K. Klimaszewski 28 , O. Kochebina , M. Kolpin 7 11 , I. Komarov 39 , R.F. Koopman 42 ,
- P. Koppenburg 41 38 , , M. Korolev 32 , A. Kozlinskiy 41 , L. Kravchuk 33 , K. Kreplin 11 , M. Kreps 48 ,
- G. Krocker 11 , P. Krokovny 34 , F. Kruse , W. Kucewicz 9 26 ,o , M. Kucharczyk 20 26 38 , , ,k ,
- V. Kudryavtsev 34 , K. Kurek 28 , T. Kvaratskheliya 31 , V.N. La Thi 39 , D. Lacarrere 38 ,
- G. Lafferty 54 , A. Lai 15 , D. Lambert 50 , R.W. Lambert 42 , G. Lanfranchi 18 , C. Langenbruch 48 ,
- B. Langhans 38 , T. Latham 48 , C. Lazzeroni 45 , R. Le Gac , J. van Leerdam 6 41 , J.-P. Lees 4 ,
- R. Lef`vre , A. Leflat e 5 32 , J. Lefran¸ois c 7 , S. Leo 23 , O. Leroy , T. Lesiak 6 26 , B. Leverington 11 ,
- Y. Li , T. Likhomanenko 3 63 , M. Liles 52 , R. Lindner 38 , C. Linn 38 , F. Lionetto 40 , B. Liu 15 ,
38
S. Lohn
,
I.
Longstaff
51
,
J.H. Lopes , N. Lopez-March
2
39
,
P. Lowdon
40
,
H. Lu , D. Lucchesi
3
22
,r
,
- H. Luo 50 , A. Lupato 22 , E. Luppi 16 ,f , O. Lupton 55 , F. Machefert , I.V. Machikhiliyan 7 31 ,
- F. Maciuc 29 , O. Maev 30 , S. Malde 55 , A. Malinin 63 , G. Manca 15 ,e , G. Mancinelli , A. Mapelli 6 38 ,
- J. Maratas , J.F. Marchand , U. Marconi 5 4 14 , C. Marin Benito 36 , P. Marino 23 ,t , R. M¨ arki 39 ,
- J. Marks 11 , G. Martellotti 25 , A. Martens , A. Mart´ ın S´nchez , M. Martinelli 8 a 7 39 ,
- D. Martinez Santos , F. Martinez Vidal , D. Martins Tostes , A. Massafferri , R. Matev
- Z. Mathe 38 , C. Matteuzzi 20 , A. Mazurov 16 ,f , M. McCann 53 , J. McCarthy 45 , A. McNab 54 ,
- R. McNulty 12 , B. McSkelly 52 , B. Meadows 57 , F. Meier , M. Meissner 9 11 , M. Merk 41
- 42 64 2 1 38 , ,
- D.A. Milanes , M.-N. Minard , N. Moggi 8 4 14 , J. Molina Rodriguez 60 , S. Monteil , M. Morandin 5 22 ,
- P. Morawski 27 , A. Mord` a 6 , M.J. Morello 23 ,t , J. Moron 27 , A.-B. Morris 50 , R. Mountain 59 ,
- F. Muheim 50 , K. M¨ uller 40 , M. Mussini 14 , B. Muster 39 , P. Naik 46 , T. Nakada 39 ,
- R. Nandakumar 49 , I. Nasteva , M. Needham 2 50 , N. Neri 21 , S. Neubert 38 , N. Neufeld 38 ,
- M. Neuner 11 , A.D. Nguyen 39 , T.D. Nguyen 39 , C. Nguyen-Mau 39 ,q , M. Nicol , V. Niess , 7 5 R. Niet , N. Nikitin 9 32 , T. Nikodem 11 , A. Novoselov 35 , D.P. O'Hanlon 48 ,
- A. Oblakowska-Mucha 27 , V. Obraztsov 35 , S. Oggero 41 , S. Ogilvy 51 , O. Okhrimenko 44 ,
- R. Oldeman 15 ,e , G. Onderwater 65 , M. Orlandea 29 , J.M. Otalora Goicochea , P. Owen 2 53 ,
- A. Oyanguren 64 , B.K. Pal 59 , A. Palano 13 ,c , F. Palombo 21 ,u , M. Palutan 18 , J. Panman 38 ,
- A. Papanestis 49 38 , , M. Pappagallo 51 , L.L. Pappalardo 16 ,f , C. Parkes 54 , C.J. Parkinson 9 45 , ,
- G. Passaleva 17 , G.D. Patel 52 , M. Patel 53 , C. Patrignani 19 ,j , A. Pazos Alvarez 37 , A. Pearce 54 ,
- A. Pellegrino 41 , M. Pepe Altarelli 38 , S. Perazzini 14 ,d , E. Perez Trigo 37 , P. Perret , 5
- M. Perrin-Terrin , L. Pescatore 6 45 , E. Pesen 66 , K. Petridis 53 , A. Petrolini 19 ,j ,
- E. Picatoste Olloqui 36 , B. Pietrzyk , T. Pilaˇ 4 r 48 , D. Pinci 25 , A. Pistone 19 , S. Playfer 50 ,
- M. Plo Casasus 37 , F. Polci 8 , A. Poluektov 48 34 , , E. Polycarpo , A. Popov 2 35 , D. Popov 10 ,
- B. Popovici 29 , C. Potterat , E. Price 2 46 , J. Prisciandaro 39 , A. Pritchard 52 , C. Prouve 46 ,
- V. Pugatch 44 , A. Puig Navarro 39 , G. Punzi 23 ,s , W. Qian , B. Rachwal 4 26 , J.H. Rademacker 46 ,
- B. Rakotomiaramanana 39 , M. Rama 18 , M.S. Rangel , I. Raniuk 2 43 , N. Rauschmayr 38 ,
- G. Raven 42 , S. Reichert 54 , M.M. Reid 48 , A.C. dos Reis , S. Ricciardi 1 49 , S. Richards 46 , M. Rihl 38 ,
- K. Rinnert 52 , V. Rives Molina 36 , D.A. Roa Romero , P. Robbe , A.B. Rodrigues , 5 7 1
- E. Rodrigues 54 , P. Rodriguez Perez 54 , S. Roiser 38 , V. Romanovsky 35 , A. Romero Vidal 37 ,
- M. Rotondo 22 , J. Rouvinet 39 , T. Ruf 38 , H. Ruiz 36 , P. Ruiz Valls 64 , J.J. Saborido Silva 37 ,
- N. Sagidova 30 , P. Sail 51 , B. Saitta 15 ,e , V. Salustino Guimaraes , C. Sanchez Mayordomo 2 64 ,
B. Sanmartin Sedes 37 , R. Santacesaria 25 , C. Santamarina Rios 37 , E. Santovetti 24 ,l , A. Sarti 18 ,m , C. Satriano 25 ,n , A. Satta 24 , D.M. Saunders 46 , M. Savrie 16 ,f , D. Savrina 31 32 , , M. Schiller 42 , H. Schindler 38 , M. Schlupp , M. Schmelling 9 10 , B. Schmidt 38 , O. Schneider 39 , A. Schopper 38 , M.-H. Schune , R. Schwemmer 7 38 , B. Sciascia 18 , A. Sciubba 25 , M. Seco 37 , A. Semennikov 31 , I. Sepp 53 , N. Serra 40 , J. Serrano , L. Sestini 6 22 , P. Seyfert 11 , M. Shapkin 35 , I. Shapoval 16 43 , ,f , Y. Shcheglov 30 , T. Shears 52 , L. Shekhtman 34 , V. Shevchenko 63 , A. Shires , R. Silva Coutinho 9 48 , G. Simi 22 , M. Sirendi 47 , N. Skidmore 46 , T. Skwarnicki 59 , N.A. Smith 52 , E. Smith 55 49 , , E. Smith 53 , J. Smith 47 , M. Smith 54 , H. Snoek 41 , M.D. Sokoloff 57 , F.J.P. Soler 51 , F. Soomro 39 , D. Souza 46 , B. Souza De Paula , B. Spaan , A. Sparkes 2 9 50 , P. Spradlin 51 , S. Sridharan 38 , F. Stagni 38 , M. Stahl 11 , S. Stahl 11 , O. Steinkamp 40 , O. Stenyakin 35 , S. Stevenson 55 , S. Stoica 29 , S. Stone 59 , B. Storaci 40 , S. Stracka 23 38 , , M. Straticiuc 29 , U. Straumann 40 , R. Stroili 22 , V.K. Subbiah 38 , L. Sun 57 , W. Sutcliffe 53 , K. Swientek 27 , S. Swientek , V. Syropoulos 9 42 , M. Szczekowski 28 , P. Szczypka 39 38 , , D. Szilard , T. Szumlak 2 27 , S. T'Jampens , M. Teklishyn , 4 7 G. Tellarini 16 ,f , F. Teubert 38 , C. Thomas 55 , E. Thomas 38 , J. van Tilburg 41 , V. Tisserand , 4 M. Tobin 39 , S. Tolk 42 , L. Tomassetti 16 ,f , D. Tonelli 38 , S. Topp-Joergensen 55 , N. Torr 55 , E. Tournefier , S. Tourneur 4 39 , M.T. Tran 39 , M. Tresch 40 , A. Tsaregorodtsev , P. Tsopelas 6 41 , N. Tuning 41 , M. Ubeda Garcia 38 , A. Ukleja 28 , A. Ustyuzhanin 63 , U. Uwer 11 , V. Vagnoni 14 , G. Valenti 14 , A. Vallier 7 , R. Vazquez Gomez 18 , P. Vazquez Regueiro 37 , C. V´ azquez Sierra 37 , S. Vecchi 16 , J.J. Velthuis 46 , M. Veltri 17 ,h , G. Veneziano 39 , M. Vesterinen 11 , B. Viaud , 7 D. Vieira , M. Vieites Diaz 2 37 , X. Vilasis-Cardona 36 ,p , A. Vollhardt 40 , D. Volyanskyy 10 , D. Voong 46 , A. Vorobyev 30 , V. Vorobyev 34 , C. Voß 62 , H. Voss 10 , J.A. de Vries 41 , R. Waldi 62 , C. Wallace 48 , R. Wallace 12 , J. Walsh 23 , S. Wandernoth 11 , J. Wang 59 , D.R. Ward 47 , N.K. Watson 45 , D. Websdale 53 , M. Whitehead 48 , J. Wicht 38 , D. Wiedner 11 , G. Wilkinson 55 , M.P. Williams 45 , M. Williams 56 , F.F. Wilson 49 , J. Wimberley 58 , J. Wishahi , W. Wislicki 9 28 , M. Witek 26 , G. Wormser , S.A. Wotton 7 47 , S. Wright 47 , S. Wu , K. Wyllie 3 38 , Y. Xie 61 , Z. Xing 59 , Z. Xu 39 , Z. Yang , X. Yuan , O. Yushchenko 3 3 35 , M. Zangoli 14 , M. Zavertyaev 10 ,b , L. Zhang 59 , W.C. Zhang 12 , Y. Zhang , A. Zhelezov 3 11 , A. Zhokhov 31 , L. Zhong , A. Zvyagin 3 38 .
1 Centro Brasileiro de Pesquisas F´ ısicas (CBPF), Rio de Janeiro, Brazil
2 Universidade Federal do Rio de Janeiro (UFRJ), Rio de Janeiro, Brazil
3 Center for High Energy Physics, Tsinghua University, Beijing, China
4 LAPP, Universit´ e de Savoie, CNRS/IN2P3, Annecy-Le-Vieux, France
- 5 Clermont Universit´, Universit´ Blaise Pascal, CNRS/IN2P3, LPC, Clermont-Ferrand, France e e 6
- CPPM, Aix-Marseille Universit´, CNRS/IN2P3, Marseille, France e
- 7 LAL, Universit´ Paris-Sud, CNRS/IN2P3, Orsay, France e
- 8 LPNHE, Universit´ e Pierre et Marie Curie, Universit´ Paris Diderot, CNRS/IN2P3, Paris, France e
- 9 Fakult¨t Physik, Technische Universit¨t Dortmund, Dortmund, Germany a a
- 10 Max-Planck-Institut f¨r Kernphysik (MPIK), Heidelberg, Germany u
- 11 Physikalisches Institut, Ruprecht-Karls-Universit¨t Heidelberg, Heidelberg, Germany a
12 School of Physics, University College Dublin, Dublin, Ireland
13 Sezione INFN di Bari, Bari, Italy
14 Sezione INFN di Bologna, Bologna, Italy
- 15 Sezione INFN di Cagliari, Cagliari, Italy
16 Sezione INFN di Ferrara, Ferrara, Italy
17 Sezione INFN di Firenze, Firenze, Italy
- 18 Laboratori Nazionali dell'INFN di Frascati, Frascati, Italy
19 Sezione INFN di Genova, Genova, Italy
- 20 Sezione INFN di Milano Bicocca, Milano, Italy
21 Sezione INFN di Milano, Milano, Italy
- 22 Sezione INFN di Padova, Padova, Italy
- 23 Sezione INFN di Pisa, Pisa, Italy
- 24 Sezione INFN di Roma Tor Vergata, Roma, Italy
- 25 Sezione INFN di Roma La Sapienza, Roma, Italy
- 26 Henryk Niewodniczanski Institute of Nuclear Physics Polish Academy of Sciences, Krak´w, Poland o
- 27 AGH - University of Science and Technology, Faculty of Physics and Applied Computer Science, Krak´ ow, Poland
- 28 National Center for Nuclear Research (NCBJ), Warsaw, Poland
- 29 Horia Hulubei National Institute of Physics and Nuclear Engineering, Bucharest-Magurele, Romania
- 30 Petersburg Nuclear Physics Institute (PNPI), Gatchina, Russia
- 31 Institute of Theoretical and Experimental Physics (ITEP), Moscow, Russia
- 32 Institute of Nuclear Physics, Moscow State University (SINP MSU), Moscow, Russia
- 33 Institute for Nuclear Research of the Russian Academy of Sciences (INR RAN), Moscow, Russia
- 34 Budker Institute of Nuclear Physics (SB RAS) and Novosibirsk State University, Novosibirsk, Russia
- 35
- Institute for High Energy Physics (IHEP), Protvino, Russia
- 36 Universitat de Barcelona, Barcelona, Spain
- 37 Universidad de Santiago de Compostela, Santiago de Compostela, Spain
- 38 European Organization for Nuclear Research (CERN), Geneva, Switzerland
- 39 Ecole Polytechnique F´d´rale de Lausanne (EPFL), Lausanne, Switzerland e e
- 40 Physik-Institut, Universit¨t Z¨rich, Z¨rich, Switzerland a u u
- 41 Nikhef National Institute for Subatomic Physics, Amsterdam, The Netherlands
- 42 Nikhef National Institute for Subatomic Physics and VU University Amsterdam, Amsterdam, The Netherlands
- 43 NSC Kharkiv Institute of Physics and Technology (NSC KIPT), Kharkiv, Ukraine
- 44 Institute for Nuclear Research of the National Academy of Sciences (KINR), Kyiv, Ukraine
- 45 University of Birmingham, Birmingham, United Kingdom
- 46 H.H. Wills Physics Laboratory, University of Bristol, Bristol, United Kingdom
- 47 Cavendish Laboratory, University of Cambridge, Cambridge, United Kingdom
- 48 Department of Physics, University of Warwick, Coventry, United Kingdom
- 49 STFC Rutherford Appleton Laboratory, Didcot, United Kingdom
- 50 School of Physics and Astronomy, University of Edinburgh, Edinburgh, United Kingdom
- 51 School of Physics and Astronomy, University of Glasgow, Glasgow, United Kingdom
- 52 Oliver Lodge Laboratory, University of Liverpool, Liverpool, United Kingdom
- 53 Imperial College London, London, United Kingdom
- 54 School of Physics and Astronomy, University of Manchester, Manchester, United Kingdom
- 55 Department of Physics, University of Oxford, Oxford, United Kingdom
- 56 Massachusetts Institute of Technology, Cambridge, MA, United States
- 57 University of Cincinnati, Cincinnati, OH, United States
- 58 University of Maryland, College Park, MD, United States
- 59 Syracuse University, Syracuse, NY, United States
- 60 Pontif´ ıcia Universidade Cat´lica do Rio de Janeiro (PUC-Rio), Rio de Janeiro, Brazil, associated to o 2
- 61 Institute of Particle Physics, Central China Normal University, Wuhan, Hubei, China, associated to 3
- 62 Institut f¨r Physik, Universit¨t Rostock, Rostock, Germany, associated to u a
- 11
- 63 National Research Centre Kurchatov Institute, Moscow, Russia, associated to 31
- 64 Instituto de Fisica Corpuscular (IFIC), Universitat de Valencia-CSIC, Valencia, Spain, associated to 36
- 65 KVI - University of Groningen, Groningen, The Netherlands, associated to 41
- 66 Celal Bayar University, Manisa, Turkey, associated to 38
a Universidade Federal do Triˆngulo Mineiro (UFTM), Uberaba-MG, Brazil a
- b P.N. Lebedev Physical Institute, Russian Academy of Science (LPI RAS), Moscow, Russia
c Universit` di Bari, Bari, Italy a
d Universit` di Bologna, Bologna, Italy a
- e Universit` di Cagliari, Cagliari, Italy a
- f Universit` di Ferrara, Ferrara, Italy a
- g Universit` di Firenze, Firenze, Italy a
h Universit` di Urbino, Urbino, Italy a
- i Universit` di Modena e Reggio Emilia, Modena, Italy a
- j Universit` di Genova, Genova, Italy a
- k Universit` di Milano Bicocca, Milano, Italy a
- l Universit` di Roma Tor Vergata, Roma, Italy a
- m Universit` di Roma La Sapienza, Roma, Italy a
- n Universit` della Basilicata, Potenza, Italy a
- o AGH - University of Science and Technology, Faculty of Computer Science, Electronics and Telecommunications, Krak´w, Poland o
- p LIFAELS, La Salle, Universitat Ramon Llull, Barcelona, Spain
- q Hanoi University of Science, Hanoi, Viet Nam
r Universit` di Padova, Padova, Italy a
s Universit` di Pisa, Pisa, Italy a
t Scuola Normale Superiore, Pisa, Italy
- u Universit` degli Studi di Milano, Milano, Italy a
v Politecnico di Milano, Milano, Italy | 10.1016/j.physletb.2014.10.005 | [
"LHCb collaboration",
"R. Aaij",
"B. Adeva",
"M. Adinolfi",
"A. Affolder",
"Z. Ajaltouni",
"S. Akar",
"J. Albrecht",
"F. Alessio",
"M. Alexander",
"S. Ali",
"G. Alkhazov",
"P. Alvarez Cartelle",
"A. A. Alves Jr",
"S. Amato",
"S. Amerio",
"Y. Amhis",
"L. An",
"L. Anderlini",
"J. Anderson",
"R. Andreassen",
"M. Andreotti",
"J. E. Andrews",
"R. B. Appleby",
"O. Aquines Gutierrez",
"F. Archilli",
"A. Artamonov",
"M. Artuso",
"E. Aslanides",
"G. Auriemma",
"M. Baalouch",
"S. Bachmann",
"J. J. Back",
"A. Badalov",
"W. Baldini",
"R. J. Barlow",
"C. Barschel",
"S. Barsuk",
"W. Barter",
"V. Batozskaya",
"V. Battista",
"A. Bay",
"L. Beaucourt",
"J. Beddow",
"F. Bedeschi",
"I. Bediaga",
"S. Belogurov",
"K. Belous",
"I. Belyaev",
"E. Ben-Haim",
"G. Bencivenni",
"S. Benson",
"J. Benton",
"A. Berezhnoy",
"R. Bernet",
"M. -O. Bettler",
"M. van Beuzekom",
"A. Bien",
"S. Bifani",
"T. Bird",
"A. Bizzeti",
"P. M. Bjørnstad",
"T. Blake",
"F. Blanc",
"J. Blouw",
"S. Blusk",
"V. Bocci",
"A. Bondar",
"N. Bondar",
"W. Bonivento",
"S. Borghi",
"A. Borgia",
"M. Borsato",
"T. J. V. Bowcock",
"E. Bowen",
"C. Bozzi",
"T. Brambach",
"J. van den Brand",
"J. Bressieux",
"D. Brett",
"M. Britsch",
"T. Britton",
"J. Brodzicka",
"N. H. Brook",
"H. Brown",
"A. Bursche",
"G. Busetto",
"J. Buytaert",
"S. Cadeddu",
"R. Calabrese",
"M. Calvi",
"M. Calvo Gomez",
"P. Campana",
"D. Campora Perez",
"A. Carbone",
"G. Carboni",
"R. Cardinale",
"A. Cardini",
"L. Carson",
"K. Carvalho Akiba",
"G. Casse",
"L. Cassina",
"L. Castillo Garcia",
"M. Cattaneo",
"Ch. Cauet",
"R. Cenci",
"M. Charles",
"Ph. Charpentier",
"M. Chefdeville",
"S. Chen",
"S. -F. Cheung",
"N. Chiapolini",
"M. Chrzaszcz",
"K. Ciba",
"X. Cid Vidal",
"G. Ciezarek",
"P. E. L. Clarke",
"M. Clemencic",
"H. V. Cliff",
"J. Closier",
"V. Coco",
"J. Cogan",
"E. Cogneras",
"L. Cojocariu",
"P. Collins",
"A. Comerma-Montells",
"A. Contu",
"A. Cook",
"M. Coombes",
"S. Coquereau",
"G. Corti",
"M. Corvo",
"I. Counts",
"B. Couturier",
"G. A. Cowan",
"D. C. Craik",
"M. Cruz Torres",
"S. Cunliffe",
"R. Currie",
"C. D'Ambrosio",
"J. Dalseno",
"P. David",
"P. N. Y. David",
"A. Davis",
"K. De Bruyn",
"S. De Capua",
"M. De Cian",
"J. M. De Miranda",
"L. De Paula",
"W. De Silva",
"P. De Simone",
"D. Decamp",
"M. Deckenhoff",
"L. Del Buono",
"N. Déléage",
"D. Derkach",
"O. Deschamps",
"F. Dettori",
"A. Di Canto",
"H. Dijkstra",
"S. Donleavy",
"F. Dordei",
"M. Dorigo",
"A. Dosil Suárez",
"D. Dossett",
"A. Dovbnya",
"K. Dreimanis",
"G. Dujany",
"F. Dupertuis",
"P. Durante",
"R. Dzhelyadin",
"A. Dziurda",
"A. Dzyuba",
"S. Easo",
"U. Egede",
"V. Egorychev",
"S. Eidelman",
"S. Eisenhardt",
"U. Eitschberger",
"R. Ekelhof",
"L. Eklund",
"I. El Rifai",
"Ch. Elsasser",
"S. Ely",
"S. Esen",
"H. -M. Evans",
"T. Evans",
"A. Falabella",
"C. Färber",
"C. Farinelli",
"N. Farley",
"S. Farry",
"RF Fay",
"D. Ferguson",
"V. Fernandez Albor",
"F. Ferreira Rodrigues",
"M. Ferro-Luzzi",
"S. Filippov",
"M. Fiore",
"M. Fiorini",
"M. Firlej",
"C. Fitzpatrick",
"T. Fiutowski",
"M. Fontana",
"F. Fontanelli",
"R. Forty",
"O. Francisco",
"M. Frank",
"C. Frei",
"M. Frosini",
"J. Fu",
"E. Furfaro",
"A. Gallas Torreira",
"D. Galli",
"S. Gallorini",
"S. Gambetta",
"M. Gandelman",
"P. Gandini",
"Y. Gao",
"J. García Pardiñas",
"J. Garofoli",
"J. Garra Tico",
"L. Garrido",
"C. Gaspar",
"R. Gauld",
"L. Gavardi",
"G. Gavrilov",
"A. Geraci",
"E. Gersabeck",
"M. Gersabeck",
"T. Gershon",
"Ph. Ghez",
"A. Gianelle",
"S. Giani'",
"V. Gibson",
"L. Giubega",
"V. V. Gligorov",
"C. Göbel",
"D. Golubkov",
"A. Golutvin",
"A. Gomes",
"C. Gotti",
"M. Grabalosa Gándara",
"R. Graciani Diaz",
"L. A. Granado Cardoso",
"E. Graugés",
"G. Graziani",
"A. Grecu",
"E. Greening",
"S. Gregson",
"P. Griffith",
"L. Grillo",
"O. Grünberg",
"B. Gui",
"E. Gushchin",
"Yu. Guz",
"T. Gys",
"C. Hadjivasiliou",
"G. Haefeli",
"C. Haen",
"S. C. Haines",
"S. Hall",
"B. Hamilton",
"T. Hampson",
"X. Han",
"S. Hansmann-Menzemer",
"N. Harnew",
"S. T. Harnew",
"J. Harrison",
"J. He",
"T. Head",
"V. Heijne",
"K. Hennessy",
"P. Henrard",
"L. Henry",
"J. A. Hernando Morata",
"E. van Herwijnen",
"M. Heß",
"A. Hicheur",
"D. Hill",
"M. Hoballah",
"C. Hombach",
"W. Hulsbergen",
"P. Hunt",
"N. Hussain",
"D. Hutchcroft",
"D. Hynds",
"M. Idzik",
"P. Ilten",
"R. Jacobsson",
"A. Jaeger",
"J. Jalocha",
"E. Jans",
"P. Jaton",
"A. Jawahery",
"F. Jing",
"M. John",
"D. Johnson",
"C. R. Jones",
"C. Joram",
"B. Jost",
"N. Jurik",
"M. Kaballo",
"S. Kandybei",
"W. Kanso",
"M. Karacson",
"T. M. Karbach",
"S. Karodia",
"M. Kelsey",
"I. R. Kenyon",
"T. Ketel",
"B. Khanji",
"C. Khurewathanakul",
"S. Klaver",
"K. Klimaszewski",
"O. Kochebina",
"M. Kolpin",
"I. Komarov",
"R. F. Koopman",
"P. Koppenburg",
"M. Korolev",
"A. Kozlinskiy",
"L. Kravchuk",
"K. Kreplin",
"M. Kreps",
"G. Krocker",
"P. Krokovny",
"F. Kruse",
"W. Kucewicz",
"M. Kucharczyk",
"V. Kudryavtsev",
"K. Kurek",
"T. Kvaratskheliya",
"V. N. La Thi",
"D. Lacarrere",
"G. Lafferty",
"A. Lai",
"D. Lambert",
"R. W. Lambert",
"G. Lanfranchi",
"C. Langenbruch",
"B. Langhans",
"T. Latham",
"C. Lazzeroni",
"R. Le Gac",
"J. van Leerdam",
"J. -P. Lees",
"R. Lefèvre",
"A. Leflat",
"J. Lefrançois",
"S. Leo",
"O. Leroy",
"T. Lesiak",
"B. Leverington",
"Y. Li",
"T. Likhomanenko",
"M. Liles",
"R. Lindner",
"C. Linn",
"F. Lionetto",
"B. Liu",
"S. Lohn",
"I. Longstaff",
"J. H. Lopes",
"N. Lopez-March",
"P. Lowdon",
"H. Lu",
"D. Lucchesi",
"H. Luo",
"A. Lupato",
"E. Luppi",
"O. Lupton",
"F. Machefert",
"I. V. Machikhiliyan",
"F. Maciuc",
"O. Maev",
"S. Malde",
"A. Malinin",
"G. Manca",
"G. Mancinelli",
"A. Mapelli",
"J. Maratas",
"J. F. Marchand",
"U. Marconi",
"C. Marin Benito",
"P. Marino",
"R. Märki",
"J. Marks",
"G. Martellotti",
"A. Martens",
"A. Martín Sánchez",
"M. Martinelli",
"D. Martinez Santos",
"F. Martinez Vidal",
"D. Martins Tostes",
"A. Massafferri",
"R. Matev",
"Z. Mathe",
"C. Matteuzzi",
"A. Mazurov",
"M. McCann",
"J. McCarthy",
"A. McNab",
"R. McNulty",
"B. McSkelly",
"B. Meadows",
"F. Meier",
"M. Meissner",
"M. Merk",
"D. A. Milanes",
"M. -N. Minard",
"N. Moggi",
"J. Molina Rodriguez",
"S. Monteil",
"M. Morandin",
"P. Morawski",
"A. Mordà",
"M. J. Morello",
"J. Moron",
"A. -B. Morris",
"R. Mountain",
"F. Muheim",
"K. Müller",
"M. Mussini",
"B. Muster",
"P. Naik",
"T. Nakada",
"R. Nandakumar",
"I. Nasteva",
"M. Needham",
"N. Neri",
"S. Neubert",
"N. Neufeld",
"M. Neuner",
"A. D. Nguyen",
"T. D. Nguyen",
"C. Nguyen-Mau",
"M. Nicol",
"V. Niess",
"R. Niet",
"N. Nikitin",
"T. Nikodem",
"A. Novoselov",
"D. P. O'Hanlon",
"A. Oblakowska-Mucha",
"V. Obraztsov",
"S. Oggero",
"S. Ogilvy",
"O. Okhrimenko",
"R. Oldeman",
"G. Onderwater",
"M. Orlandea",
"J. M. Otalora Goicochea",
"P. Owen",
"A. Oyanguren",
"B. K. Pal",
"A. Palano",
"F. Palombo",
"M. Palutan",
"J. Panman",
"A. Papanestis",
"M. Pappagallo",
"L. L. Pappalardo",
"C. Parkes",
"C. J. Parkinson",
"G. Passaleva",
"G. D. Patel",
"M. Patel",
"C. Patrignani",
"A. Pazos Alvarez",
"A. Pearce",
"A. Pellegrino",
"M. Pepe Altarelli",
"S. Perazzini",
"E. Perez Trigo",
"P. Perret",
"M. Perrin-Terrin",
"L. Pescatore",
"E. Pesen",
"K. Petridis",
"A. Petrolini",
"E. Picatoste Olloqui",
"B. Pietrzyk",
"T. Pilař",
"D. Pinci",
"A. Pistone",
"S. Playfer",
"M. Plo Casasus",
"F. Polci",
"A. Poluektov",
"E. Polycarpo",
"A. Popov",
"D. Popov",
"B. Popovici",
"C. Potterat",
"E. Price",
"J. Prisciandaro",
"A. Pritchard",
"C. Prouve",
"V. Pugatch",
"A. Puig Navarro",
"G. Punzi",
"W. Qian",
"B. Rachwal",
"J. H. Rademacker",
"B. Rakotomiaramanana",
"M. Rama",
"M. S. Rangel",
"I. Raniuk",
"N. Rauschmayr",
"G. Raven",
"S. Reichert",
"M. M. Reid",
"A. C. dos Reis",
"S. Ricciardi",
"S. Richards",
"M. Rihl",
"K. Rinnert",
"V. Rives Molina",
"D. A. Roa Romero",
"P. Robbe",
"A. B. Rodrigues",
"E. Rodrigues",
"P. Rodriguez Perez",
"S. Roiser",
"V. Romanovsky",
"A. Romero Vidal",
"M. Rotondo",
"J. Rouvinet",
"T. Ruf",
"H. Ruiz",
"P. Ruiz Valls",
"J. J. Saborido Silva",
"N. Sagidova",
"P. Sail",
"B. Saitta",
"V. Salustino Guimaraes",
"C. Sanchez Mayordomo",
"B. Sanmartin Sedes",
"R. Santacesaria",
"C. Santamarina Rios",
"E. Santovetti",
"A. Sarti",
"C. Satriano",
"A. Satta",
"D. M. Saunders",
"M. Savrie",
"D. Savrina",
"M. Schiller",
"H. Schindler",
"M. Schlupp",
"M. Schmelling",
"B. Schmidt",
"O. Schneider",
"A. Schopper",
"M. -H. Schune",
"R. Schwemmer",
"B. Sciascia",
"A. Sciubba",
"M. Seco",
"A. Semennikov",
"I. Sepp",
"N. Serra",
"J. Serrano",
"L. Sestini",
"P. Seyfert",
"M. Shapkin",
"I. Shapoval",
"Y. Shcheglov",
"T. Shears",
"L. Shekhtman",
"V. Shevchenko",
"A. Shires",
"R. Silva Coutinho",
"G. Simi",
"M. Sirendi",
"N. Skidmore",
"T. Skwarnicki",
"N. A. Smith",
"E. Smith",
"E. Smith",
"J. Smith",
"M. Smith",
"H. Snoek",
"M. D. Sokoloff",
"F. J. P. Soler",
"F. Soomro",
"D. Souza",
"B. Souza De Paula",
"B. Spaan",
"A. Sparkes",
"P. Spradlin",
"S. Sridharan",
"F. Stagni",
"M. Stahl",
"S. Stahl",
"O. Steinkamp",
"O. Stenyakin",
"S. Stevenson",
"S. Stoica",
"S. Stone",
"B. Storaci",
"S. Stracka",
"M. Straticiuc",
"U. Straumann",
"R. Stroili",
"V. K. Subbiah",
"L. Sun",
"W. Sutcliffe",
"K. Swientek",
"S. Swientek",
"V. Syropoulos",
"M. Szczekowski",
"P. Szczypka",
"D. Szilard",
"T. Szumlak",
"S. T'Jampens",
"M. Teklishyn",
"G. Tellarini",
"F. Teubert",
"C. Thomas",
"E. Thomas",
"J. van Tilburg",
"V. Tisserand",
"M. Tobin",
"S. Tolk",
"L. Tomassetti",
"D. Tonelli",
"S. Topp-Joergensen",
"N. Torr",
"E. Tournefier",
"S. Tourneur",
"M. T. Tran",
"M. Tresch",
"A. Tsaregorodtsev",
"P. Tsopelas",
"N. Tuning",
"M. Ubeda Garcia",
"A. Ukleja",
"A. Ustyuzhanin",
"U. Uwer",
"V. Vagnoni",
"G. Valenti",
"A. Vallier",
"R. Vazquez Gomez",
"P. Vazquez Regueiro",
"C. Vázquez Sierra",
"S. Vecchi",
"J. J. Velthuis",
"M. Veltri",
"G. Veneziano",
"M. Vesterinen",
"B. Viaud",
"D. Vieira",
"M. Vieites Diaz",
"X. Vilasis-Cardona",
"A. Vollhardt",
"D. Volyanskyy",
"D. Voong",
"A. Vorobyev",
"V. Vorobyev",
"C. Voß",
"H. Voss",
"J. A. de Vries",
"R. Waldi",
"C. Wallace",
"R. Wallace",
"J. Walsh",
"S. Wandernoth",
"J. Wang",
"D. R. Ward",
"N. K. Watson",
"D. Websdale",
"M. Whitehead",
"J. Wicht",
"D. Wiedner",
"G. Wilkinson",
"M. P. Williams",
"M. Williams",
"F. F. Wilson",
"J. Wimberley",
"J. Wishahi",
"W. Wislicki",
"M. Witek",
"G. Wormser",
"S. A. Wotton",
"S. Wright",
"S. Wu",
"K. Wyllie",
"Y. Xie",
"Z. Xing",
"Z. Xu",
"Z. Yang",
"X. Yuan",
"O. Yushchenko",
"M. Zangoli",
"M. Zavertyaev",
"L. Zhang",
"W. C. Zhang",
"Y. Zhang",
"A. Zhelezov",
"A. Zhokhov",
"L. Zhong",
"A. Zvyagin"
] | 2014-08-01T19:50:06+00:00 | 2014-11-26T16:44:03+00:00 | [
"hep-ex"
] | Measurement of the B0bar-B0 and Bsbar-Bs production asymmetries in pp collisions at sqrt(s)=7 TeV | The B0bar-B0 and Bsbar-Bs production asymmetries, AP(B0) and AP(Bs), are
measured by means of a time-dependent analysis of B0 -> J/\psi K*0, B0 -> D-
pi+ and Bs -> Ds- pi+ decays, using a data sample corresponding to an
integrated luminosity of 1.0 fb^{-1}, collected by LHCb in pp collisions at a
centre-of-mass energy of 7 TeV. The measurements are performed as a function of
transverse momentum and pseudorapidity of the B0 and Bs mesons within the LHCb
acceptance. The production asymmetries, integrated over pT and eta in the range
4 < pT< 30 GeV/c and 2.5<eta<4.5, are determined to be AP(B0) = (-0.35 +/- 0.76
+/- 0.28)% and AP(Bs) = (1.09 +/- 2.61 +/- 0.66)%, where the first
uncertainties are statistical and the second systematic. |
1408.0276v1 | ## Asymptotic safety and the cosmological constant
## Kevin Falls
Instit¨t u f¨r u Theoretische Physik, Universit¨t a Heidelberg, Philosophenweg 16, 69120 Heidelberg, Germany (Dated: August 4, 2014)
We study the non-perturbative renormalisation of quantum gravity in four dimensions. Taking care to disentangle physical degrees of freedom, we observe the topological nature of conformal fluctuations arising from the functional measure. The resulting beta functions possess an asymptotically safe fixed point with a global phase structure leading to classical general relativity for positive, negative or vanishing cosmological constant. If only the conformal fluctuations are quantised we find an asymptotically safe fixed point predicting a vanishing cosmological constant on all scales. At this fixed point we reproduce the critical exponent, ν = 1 3, found in numerical lattice studies by Hamber. / This suggests the fixed point may be physical while solving the cosmological constant problem.
## Contents
| I. Introduction | 3 |
|--------------------------------------------------------------------------|-----|
| II. RG for gravity and asymptotic safety | 5 |
| III. Physical degrees of freedom | 8 |
| IV. Infra-red cutoff and the cosmological constant | 11 |
| V. Truncated heat kernel expansion | 13 |
| VI. Beta functions and UV fixed point | 14 |
| A. Flow equation and threshold constants | 14 |
| B. Regulator functions | 15 |
| C. Beta functions and fixed points | 16 |
| D. One-loop scheme independence | 18 |
| VII. Globally safe trajectories | 19 |
| VIII. Conformally reduced theory | 22 |
| A. Propagating conformal mode approximation | 23 |
| B. Physical conformal reduction | 23 |
| C. Critical exponents in d dimensions and the glyph[epsilon1] -expansion | 26 |
| D. Absence of essential divergences | 27 |
| IX. Summary and Conclusion | 29 |
| A. Summary | 29 |
| B. Conclusion | 30 |
| Acknowledgements | 31 |
| A. Heat kernels and flow equation | 31 |
| References | 32 |
## I. INTRODUCTION
Quantum gravity aims to combine the principles of quantum mechanics with the theory of gravity proposed by Einstein nearly a century ago. This classical theory, general relativity, is based on the equivalence principle for all observers. The theory is described by the Einstein field equations for the metric tensor g µν , which are generally covariant under arbitrary coordinate transformations. In the absence of matter these equations imply that the scalar curvature is given by R = 4Λ, where Λ is the cosmological constant, and that the theory describes a spin-two fluctuation corresponding to the graviton. However, quantum gravity runs into severe difficulties when standard perturbative methods are applied. In particular the theory is perturbatively nonrenormalisable already at one loop, in the presence of matter [1], and at two loops for pure gravity [2]. This leaves the possibility that gravity can be quantised non-perturbativley. Alternatively one must go beyond general relativity alone by adopting new degrees of freedom and/or symmetry principles.
Another conundrum of quantum gravity relates to the cosmological constant Λ. The standard folklore is that the cosmological constant is predicted to be of order the Planck scale M 2 Pl = G -1 N where G N is Newton's constant (here and throughout we use units ̵ h = 1 = c ). Such a prediction comes from naturalness arguments assuming that its value is set by Planck scale physics. On the other hand this reasoning is in contradiction with observation [3]. Indeed, assuming that Λ is responsible for the late time acceleration of the universe, the measured value of G N ⋅ Λ is some 122 orders of magnitude less than this prediction. Thus the standard ΛCDM-model of cosmology is called into question since it suffers from an apparent fine tuning problem for Λ.
One possibility is that Λ is exactly zero and that the acceleration of the universe comes from another source of dark energy or modified gravity. This would imply that flat Minkowski spacetime is the true vacuum of quantum gravity. That this is the case has been conjectured in [4] where a careful handling of conformal fluctuations g µν → e 2 σ g µν has been stressed. Furthermore in [5] it has been argued that Λ should not receive quantum corrections at all since it can always be set to unity by a conformal field redefinition of the metric tensor.
Conformal modes also cause a problem for the quantisation of gravity since they make the na¨ ıvely Wick rotated Euclidean action unbounded from below. On the other hand the conformal fluctuations are non-dynamical in general relativity. Therefore such apparently pathological fluctuations of σ are only influential off-shell or in the presence of matter. In [4] the correct treatment of the conformal mode has been derived at the semi-classical level. There it was observed that the proper Wick rotation of σ ensures that the action is bounded from below, while the dynamics of σ are cancelled by a Jacobian arising in the functional measure.
Ultimately to understand the stability of gravity with or without a cosmological constant we must appeal to the full quantum theory. After quantisation the classical action S ϕ [ ] of a theory is replaced by the effective action Γ [ φ ] , which results from a Legendre transform of the functional integral. This implies that the effective action is a convex functional of the mean field φ = ⟨ ϕ ⟩ such
that its second functional derivative is positive definite
$$\Gamma ^ { ( 2 ) } [ \phi ] > 0 \,.$$
This condition reflects the stability of the theory and allows for the determination of the vacuum state. If we wish to quantise gravity as a fundamental theory this necessitates that we compute Γ [ φ ] via non-perturbative methods. Making sure (1.1) continues to be satisfied when approximations are applied is therefore crucial for their consistency. At a technical level these considerations relate directly to the regulated functional measure of the path integral and therefore to how the gauge fixing and renormalisation schemes are implemented.
In this paper we shall investigate the non-perturbative quantisation of gravity at an ultra-violet (UV) fixed point of the renormalisation group (RG) [6], corresponding to a second order phase transition for quantum gravity. A theory defined at such a fixed point is said to be asymptotically safe provided the phase transition has finitely many relevant directions. In light of the above considerations we shall pay particular attention to the treatment of the cosmological constant, conformal fluctuations and ultimately the convexity condition (1.1). While we study a simple phase diagram, parameterised by only the Newtons coupling and the cosmological constant, we shall close the approximation scheme by a non-perturbative expansion ensuring that the effective action remains convex. In this way we aim to minimise unphysical contributions while capturing the physics of quantum general relativity namely the spin-two fluctuations of the graviton and the topological conformal modes.
Aside from asymptotic safety it has been suggested [7] that gravity could be quantised by first integrating out the conformal fluctuations and then obtaining a conformally invariant effective theory for the remaining degrees of freedom. Then, due to its conformal nature, one would expect the resulting theory to remain finite after further quantisation. These ideas came from observing that 'complementary' descriptions of evaporating black holes are related by conformal transformations [8]. The problem with this approach is that the conformal modes remain power counting non-renormalisable [7]. Therefore, the existence of an asymptotically safe UV fixed point for the conformal fluctuations would be desirable. Indeed an asymptotically safe fixed point implies that the theory becomes scale invariant at short distances and that small black hole horizons admit conformal scaling laws [9]. In addition to full quantum gravity, we shall therefore investigate the conformally reduced theory where only the conformal modes are quantised.
The rest of this paper is as follows. First we review the functional renormalisation group for gravity and the asymptotic safety scenario in section II. In section III we consider the physical and propagating degrees of freedom in quantum general relativity. We adopt a gauge fixing procedure which makes the nature of these degrees of freedom manifest while exactly cancelling the gauge variant fields with the Fadeev-Popov ghosts. In particular we are able to observe the topological stasis of the conformal mode. In section IV we consider the form of the IR regulator and revisit the convexity condition (1.1) for the regulated theory. Here we show how poles in the propagator can be avoided leading to a well behaved low energy limit provided the curvature satisfies R > 4Λ. In
light of this we employ an approximation scheme in section V whereby the early time heat kernel expansion is truncated rather than expanding in powers of the curvature. This allows us to close the Einstein-Hilbert approximation while not expanding around vanishing R . In the next three sections we present our results coming from these considerations while the explicit form of the flow equation is given in appendix A. The beta functions for G N and Λ are studied in section VI and the existence of a UV attractive fixed point is shown. Then in section VII we show how the renormalisation group flow possesses asymptotically safe trajectories with a classical limit for positive, negative and vanishing cosmological constant. We then turn to the conformally reduced theory in section VIII where only the conformal fluctuations are quantised and their topological nature is preserved. There we find a UV fixed point which predicts the vanishing of the cosmological constant Λ = 0 on all scales. We end in section IX with with a summary of our results and our conclusions.
## II. RG FOR GRAVITY AND ASYMPTOTIC SAFETY
Since perturbative methods fail to give a renormalisable theory of quantum gravity, or shed light on the cosmological constant problem, one can resort to non-perturbative methods. An indispensable tool for understanding non-perturbative physics is offered by the exact (or functional) renormalisation group [10, 11] (for reviews see [12-16]). Within this framework a perturbatively non-renormalisable field theory may still be renormalised at an asymptotically safe fixed point under RG transformations. At its root is the observation that couplings of the theory, such as G N and Λ, are not constants in the quantum theory but generally depend on the momentum scale at which they are evaluated. If at high energies they tend towards an asymptotically safe fixed point their low energy values can be determined by following their RG flow into the infra-red (IR). Given such a fixed point in gravity we can then follow the flow of G N ⋅ Λ to determine its observable value. To be a consistent theory of quantum gravity the low energy couplings must reproduce classical general relativity (plus corrections at high curvatures). Trajectories of the RG that fulfil asymptotic safety and give rise to a meaningful low energy limit can be said to be 'globally safe'.
There now exists are large amount of evidence for asymptotic safety in four dimensional gravity coming from functional RG calculations [17-24] (for reviews see [25-32]) and complimented by lattice [33-37] and perturbative calculations [38, 39]. Within the functional RG approach early work concentrated on simple approximations whereby only an action of the Einstein-Hilbert form was considered [17-20]. Later studies have gone beyond this by including higher curvature terms [21-24, 39], general actions of the f ( R ) type [24, 40, 41] and the effects of matter [42-46].
More recently more sophisticated calculations have been performed by including additional terms in the action which have a non-trivial background field dependence [47-50]. The nature of these non-covariant terms are in principle constrained by (modified) BRST invariance [51]. At leading order these take the form of the bare gauge fixing and ghost terms arising from the Faddeev-
Popov method. Beyond this approximation new terms should arise which depend on the explicit form of gauge fixing as well as the RG scheme. In [52] the background field dependence of such terms has been evaluated via the Nielsen identities for the geometric effective action. Although in other works the modified BRST invariance of such approximations has not been determined, the flow of covariance breaking couplings such as mass parameters [49], wave function renormalisation [48, 49] and purely background field couplings [47, 50] has been assessed, while in [49] the flow of the full momentum dependent graviton propagator was evaluated. Additionally, the scale dependence of the ghost sector has been studied in [53-55]. In each case a UV fixed point compatible with asymptotic safety has been found.
In addition to an asymptotically safe fixed point there is evidence of a non-trivial IR fixed point in quantum gravity [49, 52, 56-59]. While earlier work suggested that this fixed point led to a non-classical running of cosmological constant, in [49] it was found that this fixed point is for the unphysical mass parameter and that gravity behaves classically at this fixed point. Thus the existence of trajectories connecting the UV and IR fixed points imply that gravity is well defined on all length scales.
Here we will be studying the flow of the effective average action Γ k where k denotes the RG scale down to which quantum fluctuations have been integrated out in the path integral unsuppressed. This 'flowing' action obeys the exact functional renormalisation group equation [60]
$$\partial _ { t } \Gamma _ { k } [ \phi ; \bar { \phi } ] = \frac { 1 } { 2 } S \text{Tr} \frac { \partial _ { t } \mathcal { R } _ { k } [ \bar { \phi } ] } { \Gamma _ { k } ^ { ( 2 ) } [ \phi ; \bar { \phi } ] + \mathcal { R } _ { k } [ \bar { \phi } ] } \,, \quad \quad \quad$$
obtained by taking a derivative of the action with respect to the RG time t = log k k / 0 . In the context of quantum gravity [51] this equation has been the main tool of investigations into asymptotically safe gravity mentioned above. In general Γ k depends on both the dynamical fields φ = ⟨ ϕ ⟩ k , which are k dependent averages of the fundamental fields ϕ (in the presence of a source), and the non-dynamical background fields ¯ . φ The right hand side is a super-trace involving the second functional derivative Γ ( 2 ) k [ φ, φ ¯ ] of the action at fixed ¯ . φ The important ingredient entering (2.1) is regulator function or cutoff R [ k ¯ φ ] which vanishes for high momentum modes p 2 / k 2 →∞ while behaving as a momentum dependent mass term for low modes. Its presence in the denominator of the trace regulates the IR modes. Furthermore the appearance of ∂ t R [ k ¯ φ ] in the numerator means the trace is also regulated in the UV due to the vanishing of the regulator for high momentum. By construction the flowing action Γ k interpolates between the bare action S in the limit k →∞ and the full effective action Γ when the regulator is removed at k = 0. While the action Γ k need not be convex, the sum of the action and the regulator term is obtained from a Legendre transform of the regulated functional integral. This implies that the regulated inverse propagator be positive definite
$$\Gamma _ { k } ^ { ( 2 ) } [ \phi ; \bar { \phi } ] + \mathcal { R } _ { k } [ \bar { \phi } ] > 0 \,,$$
for all physical momentum modes included in the super-trace. Thus (2.2) generalises (1.1) in the presence of an IR regulator. In [61] it was shown how convexity of the effective action follows from
the flow equation (2.1) for scalar fields. Furthermore, in [62] it was shown that convexity arises as an IR fixed point in phases with spontaneous symmetry breaking.
In this paper we work in the Einstein-Hilbert approximation studying the flowing Euclidean action
$$\Gamma _ { k } [ g _ { \mu \nu }, \dots ; \bar { g } _ { \mu \nu } ] = \int d ^ { 4 } x \sqrt { \det g _ { \mu \nu } } \, \frac { 1 } { 1 6 \pi G _ { k } } ( 2 \Lambda _ { k } - R ( g _ { \mu \nu } ) + \dots,$$
corresponding to general relativity with k dependent couplings G k and Λ . The ellipses denote the k extra fields and action terms coming from the gauge fixing prescription which we specify in the next section. Here we assume the conformal mode σ has been Wick rotated from the Lorentzian action as derived from the functional measure [4] which ensures that the action is bounded from below. This action depends on two metrics, the dynamical metric g µν , and the non-dynamical background metric ¯ g µν . The background metric is needed both to regulate the theory and to implement the gauge fixing. Once we have inserted this action into the flow equation we shall identify ¯ g µν = g µν in order to determine the beta functions for the flowing couplings G k and Λ . k For a discussion of background field flows in the functional RG see [63]. For later convenience we also identify the wave function renormalisation of the metric g µν and the corresponding anomalous dimension
$$Z _ { k } \equiv \frac { G _ { N } } { G _ { k } } \,, \quad \eta \equiv \partial _ { t } \ln Z _ { k } \,,$$
where G N is a constant which can be identified with the the low energy Newton's constant G N = G 0 for trajectories with a classical limit. From the beta functions we will look for RG trajectories which emanate from a UV fixed point G k → k -2 g ∗ and Λ k → k λ 2 ∗ at high energies k → ∞ , while recovering classical k -independent couplings G 0 = G N and Λ 0 = Λ when the regulator is removed in the limit k → 0. Such globally safe trajectories suggest gravity is a well defined quantum field theory on all length scales.
At a non-gaussian fixed point where g ∗ and λ ∗ are finite the scaling is determined from the critical exponents θ n . These exponents appear in the linear expansion
$$\lambda ^ { i } - \lambda _ { * } ^ { i } = \sum _ { n } \mathcal { C } _ { n } V _ { n } ^ { i } e ^ { - t \theta _ { n } } \,,$$
where λ i is a basis of dimensionless couplings e.g λ i = { g, λ } = { k G ,k 2 k -2 Λ k } and the range of n is equal to the range of i . Here V i n are the eigen-directions and C n are constants. The exponents -θ n (note the minus sign) and the vectors V i n correspond to the eigenvalues and eigenvectors of the stability matrix
$$M ^ { i } \, _ { j } = \frac { \partial \beta ^ { i } } { \partial \lambda ^ { j } } \Big | _ { \lambda ^ { i } = \lambda ^ { i } _ { * } } \,,$$
where β i = ∂ λ t i are the beta functions which vanish for λ i = λ i ∗ . If θ n is positive it corresponds to a relevant (UV attractive) direction and supports renormalisable trajectories. For negative θ n the
direction is irrelevant and C n must be set to zero in order to renormalise the theory at the fixed point. Including more couplings in the approximation would introduce more directions in theory space. The criteria of asymptotic safety is that the number of relevant directions should be finite at such a UV fixed point [6]. The fewer number of relevant directions the more predictive the theory defined at the fixed point will be. High order polynomial expansions in R suggest there are just three relevant directions [24, 40, 41] while a general argument for f ( R ) theories imply that there is a finite number of relevant directions [64].
## III. PHYSICAL DEGREES OF FREEDOM
General relativity has just two massless propagating degrees corresponding to the two polarisations of the graviton. On the other hand conformal fluctuations, which are non-dynamical in the classical theory, are expected to play an important rˆle once the theory is quantised. o Our general philosophy in this paper will be to make the nature of these degrees of freedom as manifest as possible at the level of the flow equation (2.1). In this way we intended to optimise the Einstein-Hilbert approximation (2.3) to the physics which it contains.
In the covariant path integral quantisation, via the Faddeev-Popov prescription, the counting of propagating degrees of freedom comes from the ten components of the metric g µν minus the eight real degrees of freedom of the ghosts C µ and ¯ C µ , each of which counts once since the action is second order in derivatives (i.e. the propagator will have a single pole for each independent field variable). For d dimensions this gives d d ( + 1 )/ 2 -2 d = d d ( -3 )/ 2 propagating degrees of freedom. An alternative prescription [4] is to directly factor out of the path integral the four degrees of freedom of g µν corresponding to the volume of the diffeomorphism group
$$g _ { \mu \nu } \rightarrow g _ { \mu \nu } + \nabla _ { \mu } \epsilon _ { \nu } + \nabla _ { \nu } \epsilon _ { \mu } \,,$$
which removes four unphysical degrees of freedom. Following this procedure avoids the inclusion of ghosts in the semi-classical approximation. Instead the necessary field redefinitions leave behind a non-trivial Jacobian in the measure of the path integral corresponding to a further four negative degrees of freedom. Three of these (negative) degrees of freedom correspond to a transverse vector which remove the three additional degrees of freedom of the transverse-traceless fluctuations of the metric h ⊥ µν while an additional (negative) scalar degree of freedom cancels the conformal mode σ in the semi-classical approximation with R = 4Λ [4].
To make these cancelations visible in the flow equation (2.1) we will introduce the ghosts in such a way that they exactly cancel the gauge fixed degrees of freedom when evaluating the flow equation for ¯ g µν = g µν and C µ = 0 = ¯ C ν [65]. This then leaves just the auxiliary degrees of freedom coming from the Jacobian plus the gauge invariant physical degrees of freedom. For simplicity we will take the metric to be that of a four sphere which is sufficient to obtain the beta functions in the Einstein-Hilbert approximation.
To this end we employ the transverse-traceless (TT) decomposition of the metric fluctuation h µν ≡ δg µν given by [66]
$$h _ { \mu \nu } & = h _ { \mu \nu } ^ { \perp } + h \frac { 1 } { d } g _ { \mu \nu } + \nabla _ { \nu } \xi _ { \mu } + \nabla _ { \mu } \xi _ { \nu } + \nabla _ { \mu } \nabla _ { \nu } \psi - \frac { 1 } { d } g _ { \mu \nu } \nabla ^ { 2 } \psi \,,$$
$$h _ { \mu } ^ { \perp } \, ^ { \mu } = 0 \,, \quad \nabla _ { \mu } h _ { \nu } ^ { \perp } \, ^ { \mu } = 0 \,, \quad \nabla _ { \mu } \xi ^ { \mu } = 0 \,.$$
Here h ⊥ µν is the transverse-traceless fluctuation and ξ µ is a transverse vector. These differential constraints have the advantage of simplifying the differential operators entering the flow equation and facilitate its evaluation. Here the spacetime dimension is taken to be d = 4, however, there is an obvious generalisation to arbitrary dimension. In addition to the TT decomposition we re-define the trace h = h µ µ in terms of the (linear) conformal mode,
$$\sigma & = h - \nabla ^ { 2 } \psi \,,$$
which along with h ⊥ µν constitute the physical degrees of freedom.
Of course the parameterisation of the physical degrees of freedom depends on the gauge. Here we choose the gauge corresponding to S gf = 1 2 α ∫ d xF F d µ µ where F µ = ∇ λ h λ µ -1 d ∇ µ h λ λ and take Landau limit α → 0. In this gauge contributions to the flow equation from ξ and ψ will just come from the gauge fixing action S gf where the physical fields σ and h ⊥ are absent. The gauge variant fields { ξ, ψ } are fourth order in derivatives due to the field re-definitions ( ψ is momentarily sixth order but this shall be rectified shortly). In order that these contributions cancel exactly with the ghosts we also make the ghost sector fourth order by writing det M = ( det M 2 ) 1 2 before exponentiating the determinant of the Faddeev-Popov operator M [65]. This introduces a third real commuting ghost B µ as well as the anti-commuting ghosts C µ and ¯ C µ . We then perform the transverse decomposition of the ghosts and an additional field redefinitions of all the longitudinal modes ψ L ≡ { ψ,B,C,C ¯ }
$$C _ { \mu } = C _ { \mu } ^ { T } + \nabla _ { \mu } C \,, \quad \bar { C } _ { \mu } = \bar { C } _ { \mu } ^ { T } + \nabla _ { \mu } \bar { C } \,, \quad B _ { \mu } = B _ { \mu } ^ { T } + \nabla _ { \mu } B \,, \quad \psi _ { L } \to \frac { 1 } { \sqrt { - \nabla ^ { 2 } } } \psi _ { L } \,.$$
This procedure leads to the Jacobians
$$J _ { 0 } = ( \det ^ { \prime \prime } ( \Delta _ { 0 } ) ) ^ { \frac { 1 } { 2 } } \,, \quad J _ { 1 } = ( \det ^ { \prime } ( \Delta _ { 1 } ) ) ^ { \frac { 1 } { 2 } } \,,$$
arising from the functional measure of ψ and ξ . They are determinants of the differential operators ∆ 0 = -∇ -2 R d -1 and ∆ 1 = -∇ -2 R d acting on scalars and transverse vectors respectivly. The rescaling of the longitudinal modes (3.4) ensures that there is no Jacobian from the ghost sector and that J 0 is only second order in derivatives. The primes in (3.5) indicate that the lowest modes of ∆ i should be removed from the determinant corresponding to the negative mode and zero mode of ∆ and the zero mode of ∆ . 0 1 They are removed since the corresponding modes of ψ and ξ µ do not contribute to the physical metric fluctuations h µν . Exponentiating the determinants in terms
of auxiliary transverse fields j µ 1 = { c µ , c ¯ µ , φ µ } and scalars j 0 = { c, c, φ ¯ } (where { c µ , c ¯ µ , c, c ¯ } are anticommuting) will give the four negative degrees of freedom in addition to the six degrees of freedom h ⊥ µν and σ . The total bare action then reads
$$1 6 \pi G _ { N } \, S = S _ { \text{EH} } + S _ { \text{gf} } + S _ { \text{gh} } + \int d ^ { 4 } x \sqrt { \det g _ { \mu \nu } } ( j _ { 0 } \Delta _ { 0 } j _ { 0 } + j _ { 1 \mu } \Delta _ { 1 } j _ { 1 } ^ { \mu } ) \,.$$
In the semi-classical approximation to the functional integral the integration over ξ and ψ will be exactly cancelled by the ghosts. In turn the conformal mode integration σ will be cancelled by the Jacobian J 0 on-shell leaving only the negative mode σ -of ∆ . 0 To see these cancellations at the level of the flow equation (2.1) we define the differential operator
$$\Delta \equiv 1 6 \pi G _ { k } \, \Gamma _ { k } ^ { ( 2 ) } \,,$$
which takes the form ∆ = 16 πG N S ( 2 ) with the replacement Λ → Λ k where S ( 2 ) is the second variation of the bare action (3.6) after a Wick rotation of the conformal mode σ . Note that due to our field redefinitions ∆ is a matrix in field space. We will normalise the fields such that all components of ∆ have the form ∆ = -∇ 2 + ... ( or ∆ = (-∇ ) 2 2 + ... for the fourth order parts) in order to simplify formulas. Each transverse vectors ξ T ≡ { ξ µ , B T µ , C T µ , C ¯ T µ } and each longitudinal mode ψ L have the equal components of ∆ given by the fourth order differential operators
$$\Delta _ { T } = \Delta _ { 1 } ^ { 2 } \,, \quad \Delta _ { L } = \Delta _ { 0 } ^ { 2 } \,,$$
however under the super-trace the corresponding terms will exactly cancel in the background field approximation. This seen by observing that in both ξ T and ψ L there are an equal number of commuting and anti-commuting fields. The remaining components of ∆ are given by
$$\text{commuting neios. } 1 \, \text{ne remaining components or } \Delta \, \text{are given by} \\ \Delta _ { \perp } & = \Delta _ { 2 } + 2 \left ( \frac { R } { 4 } - \Lambda _ { k } \right ) \,, \\ \Delta _ { \sigma } & = \Delta _ { 0 } + \frac { 4 } { 3 } \left ( \frac { R } { 4 } - \Lambda _ { k } \right ) \,, \\ \Delta _ { 0 } & = - \nabla ^ { 2 } - \frac { R } { 3 } \,, \quad \Delta _ { 1 } = - \nabla ^ { 2 } - \frac { R } { 4 } \,, \\ \intertext { s the I.ichnerowicz I.anlacian and we have set $d=4$ Here the conformal mode}$$
where ∆ 2 = -∇ + 2 R 6 is the Lichnerowicz Laplacian and we have set d = 4. Here the conformal mode has been Wick rotated σ → iσ for all modes ∆ 0 ≥ 0 as derived from the functional measure [4]. On the other hand negative modes σ -of this operator should be wick rotated trivially [67]. On the sphere there is just one such mode corresponding to the constant mode which gives an eigenvalue of the operator -∆ σ of a -= + R 3 -4 3 ( R 4 -Λ k ) . Physically this mode corresponds to a rescaling of the radius of the four sphere [67]. Taking into account all contributions and the cancellation of the ghost and gauged fixed parts the flow equation reads
$$\dashrightarrow \sigma \dashrightarrow \sigma \dashrightarrow \sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashleftarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sappa \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrighteq \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma \dashrightarrow \ sigma
& \quad - \frac { 1 } { 2 } \text{Tr} ^ { \prime \prime } \left [ \frac { \partial _ { t } \mathcal { R } _ { 0, k } } { Z _ { k } \Delta _ { 0 } + \mathcal { R } _ { 0, k } } \right ] - \frac { 1 } { 2 } \text{Tr} ^ { \prime } \left [ \frac { \partial _ { t } \mathcal { R } _ { 1, k } } { Z _ { k } \Delta _ { 1 } + \mathcal { R } _ { 1, k } } \right ] \,,$$
where S i are the various traces i = {⊥ , σ, -, 0 1 , } and the prime indicates the excluded modes. We observe that by going on-shell Λ k = R / 4 we have ∆ σ = ∆ indicating that the conformal fluctuations 0 are removed by those of j 0 arising from the scalar Jacobian (3.5). The traverse vector fluctuations should then remove the three non-propagating degrees of freedom of h ⊥ µν .
Since the on-shell condition is not generally satisfied along the flow these cancellations do not occur exactly. However, the above reasoning implies a natural pairing of the contributions S grav ≡ S 2 +S 1 and S conf ≡ S σ +S 0 which carry two and zero propagating degrees of freedom respectively. These contributions are then identified with physical graviton and conformal fluctuations of spacetime. A standard approximation scheme to test asymptotic safety is to only quantise the conformal mode σ . At the level of (3.10) this could be achieved in two ways. On one hand we could make this approximation by only including S σ . On the other hand this would mean σ is a propagating degree of freedom since the Jacobian contribution is not there to cancel its on-shell dynamics 1 . This suggests that a more consistent approximation is achieved by keeping both contributions to S conf . We will come back to this point in section VIII where we consider these approximations.
## IV. INFRA-RED CUTOFF AND THE COSMOLOGICAL CONSTANT
We now turn to the form of the IR regulator R k which must be specified in order to evaluate the traces in (3.10). We will take particular care to regulate modes in such a way that the convexity condition (2.2) is satisfied. This point has been stressed [64] in the context of the f ( R ) approximation to asymptotic safety and was discussed in [68] for Yang-Mills coupled to gravity. We note that R k depends on the background field which translates to a dependence on the scalar curvature R . As we shall see this suggests a specific form of the regulator depending on R and the scale dependent cosmological constant Λ k . In general the form of the regulator will be
$$\mathcal { R } _ { k } = \frac { 1 } { 1 6 \pi G _ { k } } R _ { k } ( z ) \,,$$
where the cutoff function R k (not to be confused with the scalar curvature R ) should vanish in the limit k → 0 for all values of z > 0. Here z should be (the eigenvalue of) some differential operator of the form z = -∇ 2 + U where U is some potential. In the classifications of [21] a cutoff for which U = 0 is referred to as type I , whereas a curvature dependent potential U = U R ( ) with no k dependence is called a type II cutoff, finally a general k dependent potential U = U k ( R ) is termed type III .
In curvature expansions one expands the trace in powers of the curvature in order to extract the beta functions for the running couplings G k and Λ . This may lead to poles in the propagator which k
1 In f ( R ) gravity the conformal mode becomes fourth order and is a propagating degrees of freedom, however not including S 0 would then mean we have two propagating scalars.
can be seen by looking at the components of ∆ in (3.9) for the conformal and transverse traceless fluctuations. Setting R = 0 will create poles at -∇ 2 = 2Λ k and -∇ 2 = 4 3 Λ k in the unregulated propagator. These are clearly artefacts of expanding in the curvature and have no obvious physical meaning. On the other hand the graviton is a massless degree of freedom and should have a pole in its propagator at zero momentum. Indeed if we instead set the background metric to a solution of the equation of motion R = 4Λ k we have ∆ ⊥ = ∆ and ∆ 2 σ = ∆ . For the regulated propagators 0 of σ and h ⊥ µν we have potential poles at R = 0 for
$$P _ { \sigma } ( R = 0 ) \equiv - \nabla ^ { 2 } - \frac { 4 } { 3 } \Lambda _ { k } + R _ { \sigma, k } \stackrel {! } { = } 0 \,, \\.$$
$$P _ { \pm } ( R = 0 ) \equiv - \nabla ^ { 2 } - 2 \Lambda _ { k } + R _ { \pm, k } \stackrel {! } { = } 0 \,.$$
However taking R equal or greater to its on-shell value R ≥ 4Λ k ensures that ∆ ≥ 0 and that no unphysical pole can be present (note that ∆ 0 and ∆ 1 are positive definite since the negative and zero modes are not excluded). Now along the flow we only require Γ ( 2 ) +R > k 0 so the flowing Λ k need not satisfy Λ k ≤ R / 4 for all k . Instead we may regulate this potential pole by an appropriate choice of R k . On the other hand this must be done in such a way that the regulator function R k vanishes in the limit k → 0 such that all modes are integrated out unsuppressed.
Now say we choose a curvature independent type III cutoff z = -∇ -2 2Λ k in order that we remove the poles (4.2) then z can take negative values for eigenvalues p 2 of the Laplacian -∇ 2 for which p 2 < 2Λ . k For these eigenvalues the regulator would not vanish in the limit k → 0. For example if we take the optimised cutoff [69] R k ( z ) = ( k 2 -z ) θ ( k 2 -z ) at k = 0 we have R 0 ( z ) = -zθ (- ) z which only vanishes if z is positive and therefore not all modes will be integrated. If we instead take z = ∆, given by (3.9), we can ensure that z is positive at k = 0 provided the curvature satisfies R ≥ 4Λ . On the other hand modes for which 0 z < 0 for finite k can still be regulated. Here we will therefore use a type III regulator of the form
$$\mathcal { R } _ { k } = \frac { 1 } { 1 6 \pi G _ { k } } R _ { k } ( \Delta ) \,.$$
This choice has been studied in [21] where it was shown that asymptotically safe trajectories can reach a classical limit at k = 0 for positive Λ. Such a regulator is called a spectrally adjusted cutoff since it cuts off modes with respect to the full k dependent inverse propagator ∆. We observe that the vanishing of the regulator (4.4) at k = 0 for different values of the curvature R coincides with the convexity condition (1.1) provided G 0 > 0 . Here we will assume that R > 4Λ 0 such that R k indeed vanishes when we take the IR limit. In particular at classical infra-red fixed points for which G k and Λ k approach constants the condition on R in Planck units then depends on the value of the dimensionless product G 0 ⋅ Λ . 0 We will return to this in section VII where discuss renormalisable trajectories that reach a line of such fixed points.
## V. TRUNCATED HEAT KERNEL EXPANSION
To compute the beta functions of G k and Λ k we must evaluate the traces appearing on the right side of the flow equation. However in order close our equations an approximation scheme is needed since the traces will in general lead to curvature terms not present in our original action. We observe that each of the traces in (3.10) are functions f ( ∆ of the differential operator (3.9). ) As a first step we can express the trace in terms of the heat kernel via an anti-Laplace transform with respect to ∆ and expand in the early time s expansion. They then have the form
$$\mathcal { S } = \text{Tr} [ f ( \Delta ) ] = \int d s \, \text{Tr} [ e ^ { - \Delta s } ] \, \tilde { f } ( \tau ) \, \approx \, \frac { 1 } { ( 4 \pi ) ^ { \frac { d } { 2 } } } \, \sum _ { n = 0 } ^ { \infty } Q _ { \frac { d } { 2 } - n } [ f ] A _ { n } ( R, \Lambda _ { k } ) \,,$$
where we suppress the field index i . Here A n ( R, Λ k ) are the Seeley-DeWitt coefficients coming from the expansion of the heat kernel H ∆ ( s ) ≡ e -∆ s which obeys the heat equation ∆ H ∆ + ∂ H s = 0, subject to the initial condition H ( 0 ) = I where I is the identity operator. These coefficients depend on both the curvature and the scale dependent cosmological constant Λ k . The appearance of the cosmological constant inside the heat kernel coefficients is a direct consequence of the fact that the covariant momentum (i.e. eigenvalues of ∆) explicitly depends on Λ k . The functionals Q m [ f ] ≡ ∫ ∞ 0 dττ -n ˜ f ( τ ) depend on the argument f ( z ) of the traces given it (3.10). For m > 0 they are given by the following integrals over the covariant momentum z ,
$$Q _ { m } = \frac { 1 } { \Gamma ( m ) } \int _ { 0 } ^ { \infty } d z z ^ { m - 1 } f ( z ) \,.$$
Note that these integrals are over z ≥ 0 and therefore by adopting the heat kernel evaluation we automatically regulate modes z < 0 in a sharp way. This can be traced back to the anti-Laplace transform which only converges for ∆ ≥ 0.
Within the standard approach, where the momentum is independent of Λ , one would simply k expand to order R and neglect the higher order terms. Here we take a different approach and use the heat kernel expansion itself as the basis of our approximation scheme. That is we drop all heat kernel coefficients for n > n max where we take n max = 1. Additionally we drop the single negative conformal mode σ -whose contribution is proportion to d x 4 √ det g µν R 2 . To better the approximation we can increase n max systematically and assess the convergence properties [24]. Note that this differs from a curvature expansion since all higher order heat kernel coefficients will depend on terms linear in R Λ n -1 k and Λ n k (such terms have also been neglected in [70] in order to be able to go on-shell by assuming Λ k is of order R ). A truncation of the heat kernel expansion rather than the curvature expansion is therefore different approximation scheme which should have different convergence properties. Since it is not strictly a curvature expansion (around any point zero or otherwise) it does not necessitate that the curvature is 'small' however the early time heat kernel expansion should be expected to accurately evaluate the traces in the high momentum limit R z / ∼ R k / 2 → 0.
Our justification for this approximation is twofold. First this keeps the cosmological constant appearing to the combination R -4Λ k so as not to upset the on-shell limit. Another approach to this, put forward in [65], is to expand the trace around R = 4Λ k which involves evaluating the the trace via an approximation of the spectral sum. However, our second motivation is to get the approximation well suited to the power like divergence that renormalise Λ k and G k . These come from the large momentum limit of the trace. Since the early time heat kernel expansion correctly evaluates these terms in the asymptotic limit, embodied in the first two heat kernel coefficients, it is ideally suited to the Einstein-Hilbert approximation. What we neglect are the logarithmic divergences which renormalise the curvature squared terms at order n = 2 (and the IR divergent terms n > 2). Since these are also absent in the left hand side of the flow equation this approximation is self-consistent. These corrections are then naturally included in the n max = 2 approximation where curvature squared terms are included. This approach is then in line with the bootstrap approach to asymptotic safety [24] without having to specify R = 0 as an expansion point.
## VI. BETA FUNCTIONS AND UV FIXED POINT
We are now in the position to derive the beta functions β g = ∂ g t and β λ = ∂ λ t within the set-up outlined in the preceding sections. The vanishing of the beta functions for non-vanishing { g , λ ∗ ∗ } indicate a non-gaussian fixed point where the theory may be renormalised.
## A. Flow equation and threshold constants
The explicit form of the flow equation is given in the appendix A where we also give the heat kernel coefficients A n . Each component of ∆ in (3.9) has the form ∆ i = -∇ 2 + U i where the potentials U i = U i ( R, Λ k ) (given in (A1)) are linear in the scalar curvature R and the cosmological constant Λ . k The corresponding heat kernel coefficients A i,n = ∫ d x d √ det g µν a i,n which depend on these potentials are then given by (A2) in the appendix. We also need to evaluate the Q n functions (5.2) which depend on the regulator functions R k and the beta functions themselves since R k depends on both Λ k and G k . Here we will only need to evaluate Q m for m = 1 2 where, , in the sum (5.1), Q 2 appears at n = 0 and Q 1 appears at n = 1. For all m > 0 they have the form
$$Q _ { m, i } = ( - 1 ) ^ { [ i ] } \frac { k ^ { 2 m } } { 2 } \left ( \Phi _ { m } [ R _ { k } ] + \tilde { \Phi } _ { m } [ R _ { k } ] \eta + \hat { \Phi } _ { m } [ R _ { k } ] \dot { U } _ { i } \right ) \,.$$
Here the dot denotes a derivative with respect to the RG time t = ln ( k k / 0 ) . The anomalous dimension is given by η ≡ ˙ Z k / Z k = -η N ≡ -˙ G k / G k (see (2.4)) which we take to be the same for each field and takes the value η ∗ = 2 at a non-trivial fixed point. The [ ] i in the exponent of -1 takes values [ 2 ] = 0 = [ σ ] for the physical degrees of freedom and [ 0 ] = 1 = [ 1 ] for the 'anti'-degrees of
freedom as dictated by the super-trace. The 'threshold constants' Φ m , ˜ Φ m and Φ ˆ m are given by the following regulator R k dependent integrals evaluated for k = 1,
$$\Phi _ { m } = \int _ { 0 } ^ { \infty } d z z ^ { m - 1 } \frac { \dot { R } _ { 1 } ( z ) } { z + R _ { 1 } ( z ) }, \ \tilde { \Phi } _ { m } = \int _ { 0 } ^ { \infty } d z z ^ { m - 1 } \frac { R _ { 1 } ( z ) } { z + R _ { 1 } ( z ) }, \ \hat { \Phi } _ { m } = \int _ { 0 } ^ { \infty } d z z ^ { m - 1 } \frac { R _ { 1 } ^ { ^ { \prime } } ( z ) } { z + R _ { 1 } ( z ) } \,, \ ( 6. 2 )$$
where the prime denotes derivative with respect to the covariant momentum z . Since the threshold constants only depend on the shape function R k and are independent of the curvature and couplings they will just be numbers once the regulator is specified. We note that for any regulator function the threshold constants have a definite sign
$$\Phi _ { m } > 0 \,, \ \tilde { \Phi } _ { m } > 0 \,, \ \text{and} \quad \hat { \Phi } _ { m } < 0 \,.$$
This information allows us to determine physical fixed points and their properties without specifying the form of R k . The final form of the flow equation in terms of the dimensionless coupling g = k G 2 k , λ = k -2 Λ k and the constants (6.2) is given in (A3) with (A4),(A5) and (A6). We note that the 'one-loop' approximation where by Γ ( 2 ) k is replaced by S ( 2 ) in the right hand side of the flow equation translates to putting η = 0 = ˙ U i . This can also be achieved by setting Φ ˜ n = 0 = ˆ Φ . n We will consider this approximation in section VI D.
## B. Regulator functions
Here we consider the class of exponential functions of the form
$$R _ { k } ^ { \exp } ( z ) = k ^ { 2 } \frac { 1 } { 2 \exp \left [ c \frac { z ^ { b } } { k ^ { 2 b } } \right ] - 1 } \,,$$
where b is a free parameter which we study in the range 2 ≤ b ≤ 30 and we set c = ln 3 / 2. Increasing b sharpens the division between low and high modes. In addition to the exponential regulators we also consider the optimised regulator function [69]
$$R _ { k } ^ { \text{opt} } ( z ) = ( k ^ { 2 } - z ) \theta ( k ^ { 2 } - z ) \,,$$
where θ ( x ) is the Heaviside theta function. We use the notation opt = for quantitates evaluated with (6.5). Plugging these functions into the integrals (6.2) we obtain the numerical values for the threshold constants. For example with the optimised cutoff function we have Φ 1 opt = 2, Φ 2 opt = 1, ˜ Φ 1 opt = 1 2 , ˜ Φ 2 opt = 1 6 , ˆ Φ 1 opt = -1 and Φ ˆ 2 opt = -1 2 . The curvature dependence of the traces comes solely from heat kernel coefficients (A2).
## C. Beta functions and fixed points
Before specifying the regulator R k the beta functions β g = ∂ g t and β λ = ∂ λ t may be expressed explicitly in terms of the threshold constants (6.2) with (6.3),
$$\beta _ { g } = g \left ( 2 + \frac { g \left ( - 4 3 8 g \Phi _ { 2 } \hat { \Phi } _ { 1 } + \Phi _ { 1 } \left ( 7 5 2 g \lambda \hat { \Phi } _ { 1 } + 9 9 \left ( 6 \pi + 1 7 g \hat { \Phi } _ { 2 } \right ) \right ) \right ) } { 9 \left ( 6 \pi + 1 7 g \hat { \Phi } _ { 2 } \right ) \left ( - 4 \pi + 1 1 g \tilde { \Phi } _ { 1 } \right ) + 2 g \hat { \Phi } _ { 1 } \left ( 3 7 6 g \lambda \tilde { \Phi } _ { 1 } - 3 \left ( 5 0 \pi \lambda + 7 3 g \tilde { \Phi } _ { 2 } \right ) \right ) } \right ) \,,$$
$$\beta _ { \lambda } = - 2 \lambda - \frac { 9 g \left ( 2 \$$
These beta-functions are evidently non-perturbative. Solving for fixed points β g = 0 = β λ we find a gaussian fixed point { g = 0 , λ = 0 } and a pair of non-gaussian fixed points one of which is at positive g and λ for all cutoff functions. Due to the structure of the flow equation (A3) we always find exactly two non-gaussian fixed points in the complex plane for all regulators, one is at positive g ∗ and the other at negative g ∗ . To ensure the convexity condition (2.2) only the fixed point for positive g ∗ is physical. In terms of the threshold constants the physical fixed point couplings are given by
$$\text{given by} & & g _ { * } = \frac { 5 7 6 \pi }$$
which, due to (6.3), can be seen to be both manifestly real and positive. For the optimised cutoff we have
$$\begin{array} { l l$$
These quantities are not universal and may have a strong regulator dependence. On the other hand the dimensionless product G ∗ k ⋅ Λ ∗ k = g ∗ ⋅ λ ∗ is expected to be universal. For the optimised regulator function (6.5) the product is given by
$$g _ { * } \cdot \lambda _ { * } \approx 0. 0 7 8 8$$
In fig. 1 we plot the dependence of g ∗ ⋅ λ ∗ on the regulator parameter b for the regulator function (6.4). As b is increased we see a convergence.
Expressions for the critical exponents can also be obtained in terms of the threshold constants but they lengthy so we do not include them here. For all regulators considered they are each both real and relevant. Using the optimised cutoff (6.5) the critical exponents are given by
$$\theta _ { 0 } \stackrel { \text{opt} } { \approx } 3$$
Figure 1: We plot the regulator dependence of g ∗ ⋅ λ ∗ at the non-perturbative fixed point (6.8) as function of the parameter b appearing in the exponential cutoff (6.4). Increasing b sharpens the cutoff of IR modes and we observe a convergence of g ∗ ⋅ λ ∗
In fig. 2 we plots the dependence of the critical exponents on b for the exponential cutoff functions (6.4). We note that they are close to the values (6.11) and converge as b is increased. Here we use the convention that the more relevant critical exponent is denoted θ 0 .
Numerically the critical exponents calculated with the optimised cutoff function (6.5) are within ≈ 16% and ≈ 6% of the gaussian critical exponents θ G 0 , = 4 and θ G1 = 2 consistent with the bootstrap approach put forward in [24]. However it is also instructive to look at the corresponding eigenvectors. These, unlike the critical exponents, depend on the parameterisation of the fixed point coordinates. Since the (non-perturbative) power counting comes from the canonical dimension of the operators in the action (2.3) it therefore makes sense to consider the running vacuum energy ρ k = Λ k / G k and the running Planck mass (squared) M 2 k = G -1 k which appear as the coefficients of these operators. In this basis the eigenvectors are given by
$$\mathbf V _ { 0 } \equiv \{ V _ { 0 } ^ { \rho }, V _ { 0 } ^ { M ^ { 2 } } \} \approx \{ 0. 3 7 6 8 8, 0. 9 2 6 2 6 2 \} \,, \quad \mathbf V _ { 1 } \equiv \{ V _ { 1 } ^ { \rho }, V _ { 1 } ^ { M ^ { 2 } } \} \approx \{ 0. 9 8 7 8 9 8, 0. 1 5 5 1 0 6 \} \,. \quad \quad ( 6. 1 2 )$$
for the optimised cutoff. Interestingly we observe that the more relevant eigenvector V 0 points more strongly in the direction of M 2 k rather than the vacuum energy ρ k direction and vice versa for V 1 . This indicates that M 2 k becomes more relevant in the UV and ρ k less relevant. With the exponential cutoff (6.4), less relevant eigenvector also points more strongly in the ρ k direction.
It is intriguing to note that we obtain real critical exponents and not a complex conjugate pair found in previous Einstein-Hilbert approximations [17-20], including the on-shell approach [65]. However real exponents have been found in work that goes beyond this approximation by utilising
Figure 2: We plot the UV attractive critical exponents at the non-perturbative fixed point (6.8) evaluated using the exponential cutoff (6.4). As the sharpening parameter b is increased we see a convergence of both critical exponents. These values should also be compared to those obtained with the optimised cutoff (6.5) given by (6.11).
vertex expansions around flat space [48, 49, 59]. Also the critical exponents have been shown to be real provided a global f ( R ) -type fixed point solution exists [64]. This suggests that by not explicitly expanding in powers of the curvature we have a better approximation to such a solution.
## D. One-loop scheme independence
The semi-classical or 'one-loop' 2 approximation to the flow equation (2.1) is achieved by putting Γ ( 2 ) k = S ( 2 ) in the right hand side. This leads to the equation
$$\partial _ { t } \Gamma _ { k } ^ { \text{one-loop} } = \frac { 1 } { 2 } S \text{Tr} \left [ \frac { \partial _ { t } \mathcal { R } _ { k } } { S ^ { ( 2 ) } + \mathcal { R } _ { k } } \right ] \,,$$
where the regulator function R k should be modified accordingly. To obtain this approximation at the level of our beta-function we neglect the running of G k and Λ k on the right-hand side of the flow equation which is equivalent to putting Φ ˜ n = 0 = ˆ Φ . The beta-functions then simplify to the n
2 This is a slight abuse of language since the flow equation (2.1) is manifestly one-loop exact. By one loop we therefore mean the semi-classical approximation, keeping quantum effects up to order ̵ h .
form
$$\beta _ { g } = 2 g - \frac { 1 1 g ^ { 2 } \Phi _ { 1 } } { 4 \pi } \,,$$
$$\beta _ { \lambda } = \frac { - 2 4 \pi \lambda + g \lambda \Phi _ { 1 } + 6 g \Phi _ { 2 } } { 1 2 \pi } \,.$$
These beta-functions have a single non-trivial UV fixed point
$$g _ { * } = \frac { 8 \pi } { 1 1 \Phi _ { 1 } } \,, \quad \lambda _ { * } = \frac { 3 \Phi _ { 2 } } { 1 6 \Phi _ { 1 } } \,,$$
with regulator R k independent critical exponents
$$\theta _ { 0 } = 2 \,, \ \theta _ { 1 } = \frac { 6 4 } { 3 3 } \approx 1. 9 3 9 \,.$$
The more relevant exponent θ 0 = θ G, 1 is just the canonical mass dimension of the Planck mass squared M 2 Pl whereas θ 1 is a true quantum correction. In [49] a real critical exponent for g of θ = 2 has been found in agreement with the one-loop result found here. This scheme independence can be traced to our treatment of the cosmological constant and is directly linked to the use of the truncated heat kernel expansion suggesting that this approximation may better converge to the physical result. Ultimately this can be tested by increasing n max in a systematic way [24].
## VII. GLOBALLY SAFE TRAJECTORIES
We now turn to the renormalisable trajectories which leave the UV fixed point (6.8) and flow into the IR as k is decreased. To find infra-red fixed points other than the gaussian one λ = 0 = g we switch our parameterisation to { τ, g } where τ ≡ G k ⋅ Λ . k In nature we know that the product τ is very small, in particular if we Λ it to be the driving force of the late time expansion of the universe we get the numerical value G 0 ⋅ Λ 0 ≡ G N ⋅ Λ ≈ 10 -122 . It is therefore of interest to find RG trajectories consistent with this value.
In terms of τ and g the beta functions (6.6) read
$$\beta _ { \tau } = g \frac { 3 g \Phi _ { 2 } \left ( - 3 6 \pi - 1 4 6 \tau \hat { \Phi } _ { 1 } + 9 9 g \tilde { \Phi } _ { 1 } \right ) + \Phi _ { 1 } \left ( 5 7 6 \pi \tau + 7 5 2 \tau ^ { 2 } \hat { \Phi } _ { 1 } + 9 9 g \left ( 1 7 \tau \hat { \Phi } _ { 2 } - 3 g \tilde { \Phi } _ { 2 } \right ) \right ) } { 9 \left ( 6 \pi + 1 7 g \hat { \Phi } _ { 2 } \right ) \left ( - 4 \pi + 1 1 g \tilde { \Phi } _ { 1 } \right ) + \hat { \Phi } _ { 1 } \left ( 7 5 2 g \tau \tilde { \Phi } _ { 1 } - 6 \left ( 5 0 \pi \tau + 7 3 g ^ { 2 } \tilde { \Phi } _ { 2 } \right ) \right ) } \,, \quad ( 7. 1 )$$
$$\beta _ { g } = g \left ( 2 - \frac { g \left ( \Phi _ { 1 } \left ( 9 9 \left ( 1 7 g \hat { \Phi } _ { 2 } + 6 \pi \right ) + 7 5 2 \tau \hat { \Phi } _ { 1 } \right ) - 4 3 8 g \Phi _ { 2 } \hat { \Phi } _ { 1 } \right ) } { \hat { \Phi } _ { 1 } \left ( 4 3 8 g ^ { 2 } \tilde { \Phi } _ { 2 } - 7 5 2 g \tau \tilde { \Phi } _ { 1 } + 3 0 0 \pi \tau \right ) + 9 \left ( 1 7 g \hat { \Phi } _ { 2 } + 6 \pi \right ) \left ( 4 \pi - 1 1 g \tilde { \Phi } _ { 1 } \right ) } \right ) \,.$$
From which we recover the UV fixed point (6.8) as well as a line of classical IR at fixed points,
$$g _ { * } = 0 \,, \ \tau = \text{const} = G _ { N } \cdot \Lambda \,, \ \ G _ { k } = G _ { N } \,.$$
Figure 3: Phase diagram in the { λ, g } parameterisation. We plot the globally safe trajectories originating from the asymptotically safe fixed point (blue dot) for k →∞ and ending in classical general relativity for k = 0. Each trajectory corresponds to a different value of G N ⋅ Λ for k → 0 lying in the range -∞< G N ⋅ Λ ≤ τ max . The red lines are the trajectories G N ⋅ Λ = τ max = 18 π 25 , G N ⋅ Λ = 0 and G N ⋅ Λ = -∞ corresponding to the infinite fixed point λ = -∞ , g ∗ = 25 π 94 . In the regions where no trajectories are plotted no globally safe trajectories exist. Here we use the optimised cutoff (6.5).
This implies that G N ⋅ Λ can take any value for trajectories that reach this line as k → 0. Due to the considerations of section IV the regulator R k will vanish at such fixed points provided R ≥ 4Λ . 0 The interesting question is whether there exists renormalisable trajectories which reach the classical IR fixed point (7.3) and for which values of G N ⋅ Λ they correspond. These are the globally safe trajectories defined for all scales ∞≥ k ≥ 0 with the classical limit at k = 0. Since we have regulated the potential poles in λ arising in the type I and II regulators (i.e. Λ -independent k cutoff functions) there should be renormalisable trajectories for G N ⋅ Λ > 0 (as well as those for negative and vanishing G N ⋅ Λ). However, evaluating β τ / g at g = 0 there is a pole in the rescaled beta function at
$$\tau _ { \max } = - \frac { 1 8 \pi } { 2 5 \hat { \Phi } _ { 1 } } \stackrel { \text{opt} } { = } \frac { 1 8 \pi } { 2 5 } \,,$$
which is positive independent of the regulator due to (6.3). This value of τ = τ max then places a maximum value on τ in the IR for globally safe trajectories. That is we find that only trajectories with G N ⋅ Λ < τ max are globally safe. Note that in the one-loop approximation this pole is removed
Figure 4: Phase diagram in the { τ, g } parameterisation. Along g = 0 axis there is a line of classical IR fixed points in the range -∞ < G N ⋅ Λ < τ max . Here we plot several globally safe trajectories (thin lines) emanating from the asymptotically safe fixed point (blue dot) and ending along the line of classical fixed points. There corresponding values of G N ⋅ Λ can be read off the g = 0 axis.
since ˆ Φ 1 → 0. In addition to the IR and UV fixed points we find a non-gaussian solution β g = 0 = β 1 / τ given by
$$g _ { * } = \frac { 7 5 \pi } { 9 4 \left ( \Phi _ { 1 } + 2 \tilde { \Phi } _ { 1 } \right ) } \stackrel { \text{opt} } { = } \frac { 2 5 \pi } { 9 4 } \,, \quad 1 / \tau ^ { * } = 0 \,,$$
which corresponds to an infinite cosmological constant G N ⋅ Λ →±∞ . Due to the maximum (7.4) renormalisable trajectories will only reach this fixed point in the limit G N ⋅ Λ →-∞ . In Lorentzian signature this would correspond to universes of 'nothing' [71] i.e. anti-de-Sitter universes with vanishing radius. At the point (7.5) the critical exponents are given by
$$\theta = \left \{ - 2, 2 + \frac { 4 \tilde { \Phi } _ { 1 } } { \Phi _ { 1 } } \right \} \,.$$
The -2 corresponds to the IR attractive behaviour λ = Λ / k 2 →∞ whereas the IR repulsive direction indicates that this is a saddle point. The non-canonical scaling of the second critical exponent and the nontrivial fixed point for g ∗ show that this is not a classical fixed point and that no classical limit exists for 1 G N Λ →± 0. In fig. 3 we plot renormalisable trajectories in the standard parameterisation
{ λ, g } for the optimised cutoff. Additionally we plot the same trajectories in the parameterisation { τ, g } in fig. 4 . We observe that the saddle point (7.5) is approached for trajectories in the limit τ = -∞ which is a separatrix between globally safe trajectories and unphysical trajectories which are incomplete. For positive τ the pole at τ max provides the separatrix. These results suggest that gravity is asymptotically safe with a classical limit where the cosmological constant is a free parameter lying in the range -∞< G N ⋅ Λ < τ max .
We therefore find no evidence for non-classical behaviour in the IR within our approximation and in particular no non-trivial IR fixed point for positive λ . Instead flowing from UV fixed point into the IR, our choice of regulator has guaranteed that renormalisable trajectories exist which reach general relativity for k = 0 for all values of the cosmological constant in the range -∞ < Λ < τ max M 2 Pl where τ max ∼ 1. The exists of a non-trivial IR fixed point found in previous studies [49, 52, 56-59] can therefore be traced to expansions around flat space where the massless nature of gravity is obscured. In [49] zero graviton mass is nonetheless recovered at an IR fixed point which ensures convexity while Λ scales classically.
However, since we have used a truncation to only local operators there may still be non-trivial IR effects from non-local operators which are neglected due to our use of the early time heat kernel expansion. For discussions on IR effects in the functional RG approach to quantum gravity and the rˆle of non-local terms we refer to [41, 70] and to [72] where a screening of the cosmological o constant has been observed.
## VIII. CONFORMALLY REDUCED THEORY
In this section we consider the toy model where only the conformal mode σ is quantised. Asymptotic safety has also been studied in conformally reduced toy models [73-77]. In this case only the conformal fluctuations are quantised and the fluctuations of the other metric degrees of freedom are neglected. Such approximations depend on the whether the RG scheme breaks Weyl invariance [75]. Following the suggestion of [7] this route could also be understood as a first step towards a consistent theory of gravity.
As noted at the end of section III there are two conceptually different approaches to the conformal reduction at the level of the flow equation derived here (3.10). In one approach we only include the contribution S σ and neglect the other contributions 3 . However, this would mean that σ is a propagating degree of freedom since the contribution S 0 , coming from the Jacobian J 0 in (3.5), is not there to cancel it on-shell. In the second approach we quantise the conformal mode as a topological degree of freedom, as it is in full theory. This amounts to including both S σ and S 0 in the righthand side of the flow equation (3.10).
3 The contribution S -from the constant mode σ -should also be included but it is neglected in our approximation since it leads to R 2 terms.
## A. Propagating conformal mode approximation
First we consider the approach where we include just the conformal mode contribution S σ without the contribution S 0 . Here we find two non-gaussian fixed points at positive and negative g ∗ respectively, and both with negative λ ∗ . For the optimised regulator (6.5) the positive g ∗ fixed point is given by,
$$g _ { * } \overset { \cdot } { = } 1 8 \left ( 7 + \sqrt { 5 7 } \right ) \pi \approx 8 2 2 \,, \quad \lambda _ { * } \overset { \, \, \text{opt} } { = } - \frac { 1 } { 1 6 } \left ( 1 1 \sqrt { 5 7 } \right ) \approx 5. 1 9 \,.$$
Note that g ∗ is three orders of magnitude higher than the UV fixed point of the full approximation (6.8) indicating that this approximation is questionable. Evaluating the critical exponents for the optimised cutoff we find θ 0 opt ≈ 1 53784 . , θ 1 opt ≈ -19 6375 which suggests that there is just one . relevant operator at this fixed point. On the other hand using the exponential cutoff (6.4) we find that the critical exponents are both positive and that θ 0 depends strongly on the parameter b . For example with b = 2 we find θ 0 ≈ 367 403 . , θ 1 ≈ 1 48858 whereas for . b = 30 we have θ 0 ≈ 8 21878 . , θ 1 ≈ 1 44758. . We therefore see that the number of relevant directions is scheme dependant, implying that this is not a good approximation.
## B. Physical conformal reduction
We now turn to the physically well motivated approximation whereby we keep the scalar Jacobian contribution S 0 in addition to the conformal mode contribution S σ . This ensures the topological nature of the conformal mode. The beta functions then read
$$\beta _ { \lambda } = \lambda \left ( - 2 - \frac { 3 g \pi \Phi _ { 1 } } { - \left ( 3 \pi + g \hat { \Phi } _ { 2 } \right ) \left ( 6 \pi - g \tilde { \Phi } _ { 1 } \right ) + 2 g \lambda \hat { \Phi } _ { 1 } \left ( - 5 \pi + g \tilde { \Phi } _ { 1 } \right ) } \right ) \,,$$
$$\beta _ { g } = g \left ( 2 + \frac { g \Phi _ { 1 } \left ( 3 \pi + 2 g \lambda \hat { \Phi } _ { 1 } + g \hat { \Phi } _ { 2 } \right ) } { - \left ( 3 \pi + g \hat { \Phi } _ { 2 } \right ) \left ( 6 \pi - g \tilde { \Phi } _ { 1 } \right ) + 2 g \lambda \hat { \Phi } _ { 1 } \left ( - 5 \pi + g \tilde { \Phi } _ { 1 } \right ) } \right ) \,.$$
We observe that β λ is proportional to the cosmological constant and that therefore trajectories cannot cross the λ = 0 line. This is a direct consequence of the cancelations between the conformal mode and the Jacobian (3.5) and splits the phase diagram into three regions λ = 0, λ < 0 and λ > 0. The corresponding phase diagram for the conformally reduced toy model is plotted in fig 6.
Along the λ = 0 line there is a non-gaussian UV fixed point at
$$g _ { * } = \frac { 1 2 \pi } { \Phi _ { 1 } + 2 \tilde { \Phi } _ { 1 } } \,, \quad \lambda _ { * } = 0 \,,$$
with critical exponents
$$\theta _ { 0 } = 2 + \frac { 4 \tilde { \Phi } _ { 1 } } { \Phi _ { 1 } } \stackrel { \text{opt} } { = } 3 \,, \quad \theta _ { \text{irr} } = \frac { 8 \hat { \Phi } _ { 2 } } { \Phi _ { 1 } + 4 \hat { \Phi } _ { 2 } + 2 \tilde { \Phi } _ { 1 } } \stackrel { \text{opt} } { = } - 4 \,.$$
Figure 5: Here we plot the critical exponents of the λ ∗ = 0 fixed point (8.4) evaluated with the exponential cutoff (6.4). As b is increased the cutoff becomes increasingly sharp. We note that for increasing b the critical exponents tend towards the values given by the optimised cutoff (6.5) given by θ 0 = 3 and θ irr = -4. In particular ν = 1 / θ 0 tends towards ν = 1 3 obtained with the optimised cutoff and in agreement with / numerical lattice studies.
We note that the values obtained with the optimised cutoff (6.5) are integer. Setting λ = 0 and using the optimised cutoff the beta-function for g is given by
$$\beta _ { g } \overset { \text{opt} } { = } 2 g + \frac { 4 g ^ { 2 } } { g - 1 2 \pi } \,,$$
where the fixed point (8.4) is at g ∗ = 4 π and the critical exponent θ 0 = -∂β g ∂g ∣ g = g ∗ = 3 can be seen. The eigen-direction along the λ = 0 line corresponds to θ 0 and is relevant. The other direction corresponding to θ irr is irrelevant for all regulators considered. In fig. 5 we plot the dependence of the critical exponents on b for the exponential cutoff (6.4). Unlike the previous approximation of section VIII A the critical exponents show only a mild scheme dependence and appear to tend towards the optimised cutoff values { 3 , - } 4 as b is increased. Remarkably the value ν = 1 / θ 0 = 1 3 / obtained here is in agreement with lattice studies [33].
The fixed point (8.4) splits the phase space region λ = 0 into two regions. For g < g ∗ we recover flat space where as for g > g ∗ we recover the 'branched polymer' region [33] where g diverges and the renormalised metric,
$$\chi _ { \mu \nu } & \equiv Z _ { k } g _ { \mu \nu } \,,$$
Figure 6: Phase diagram in the conformally reduced approximation. For λ ∗ = 0 there exists a UV fixed point (blue dot) which has only a single globally safe trajectory running from { 0 , g ∗ } to { 0 0 , } (dark blue line) which ends in classical general relativity with a vanishing vacuum energy. For g > g ∗ and λ = 0 the theory has no classical limit and instead corresponds to a phase with a vanishing renormalised metric χ µν = 0. For negative λ there is a fully UV attractive fixed point (red dot). Trajectories run from this fixed point to negative values of G N ⋅ Λ in the infra-red limit. Trajectories for positive cosmological constant are not renormalisable and run into the singular line (thin blue curve) for finite k but still have a classical k → 0 limit.
tends to zero χ µν → 0 as k is decreased. This is observed by noting that wave function renormalisation Z k = G N / G k (see (2.4)) goes towards zero before hitting a pole for the renormalisable trajectory g > g ∗ . For g < g ∗ we instead recover classical scaling Z 0 = 1. At the fixed point χ µν scales as k 2 also running to zero for k = 0. The fixed point (8.4) therefore represents a second order phase transition for which χ µν is the order parameter. In figure 7 we plot the wave function renormalisation for the three renormalisable trajectories, g ∗ = 0, g ∗ < 0 and g ∗ > 0, as a function of the RG time t .
Note that the irrelevant critical exponent θ irr is proportional to Φ ˆ 2 which arises from the divergences of the vacuum energy and it is therefore the quantum fluctuations of the vacuum themselves that cause Λ k to be an irrelevant coupling. The renormalisable trajectory coming from this fixed point for g < g ∗ runs directly into the Gaussian fixed point at g = 0 = λ . This trajectory therefore provides a UV completion of gravity while also solving the cosmological constant problem; the UV theory predicts that the vacuum energy is exactly zero for all scales. That the critical exponent is
recovered in this approximation strongly suggests that the exists of the UV fixed point is due to topological degrees of freedom. This is in agreement to the observation of [78] that the fixed point is due to the dominance of paramagnetic interactions for which the Laplacian operator ∇ 2 plays no rˆle. o
For λ < 0 there is a further non-trivial fixed point with positive g ∗ given by
$$g _ { * } = \frac { 3 6 \pi } { 3 \Phi _ { 1 } - 2 \hat { \Phi } _ { 2 } + 6 \tilde { \Phi } _ { 1 } } \,, \quad \lambda _ { * } = - \frac { \hat { \Phi } _ { 2 } } { 2 \hat { \Phi } _ { 1 } } \,.$$
This fixed point has two relevant directions and trajectories emanating from it lead to a negative cosmological constant at low energies. A fundamental theory based on (8.8) is therefore less predictive than that of (8.4). It is inconsistent with the ΛCDM model of cosmology and would instead lead to anti-de-Sitter universes. For λ > 0 there is no UV completion since trajectories are not attracted to a fixed point in the UV. Instead trajectories run into a singularity for finite k . We conclude that asymptotic safety, based on this approximation, predicts either a vanishing cosmological constant when the theory is quantised at (8.4) or a negative Λ when the theory is quantised at (8.8). Note that the former case involves no fine tuning since we just set Λ = 0 in the bare action. From the λ > 0 region of the phase diagram (6) we conclude that a positive cosmological constant would be inconsistent with asymptotic safety.
## C. Critical exponents in d dimensions and the glyph[epsilon1] -expansion
Since the critical exponent (8.5) is a universal quantity we generalise to d dimensions where we have
$$\theta _ { 0 } \equiv 1 / \nu \stackrel { \text{opt} } { = } 2 d + \frac { 4 } { d } - 6 = \frac { 2 \epsilon ( 1 + \epsilon ) } { 2 + \epsilon } \,,$$
with glyph[epsilon1] = d -2. The large d limit θ 0 → 2 d is in agreement with previous studies [20]. We note that θ 0 = 0 in both d = 1 and d = 2 dimensions and that d = 4 lies on the radius of convergence of the small glyph[epsilon1] expansion. Expanding the critical exponent in glyph[epsilon1] we obtain
$$\theta _ { 0 } = \epsilon + \frac { \epsilon ^ { 2 } } { 2 } - \frac { \epsilon ^ { 3 } } { 4 } + \frac { \epsilon ^ { 4 } } { 8 } + \dots,$$
which for d = 4 gives the well known divergent series θ 0 = 2 + 2 -2 + 2 -2 + ... leading to θ 0 = 2 and θ 0 = 4 at alternating orders. This series indicates that the exact result in four dimensions could be obtained from a re-summation of the glyph[epsilon1] -expansion 4 . This expansion should be compared with the two loop result [79] that gives θ 0 = 2 and θ 0 ≈ 4 4 at the first two orders in . glyph[epsilon1] showing an error of ten percent between the two-loop calculation our result. This non-trivial agreement with perturbative
4 For example, defining x = θ 0 -2 2 we have x = 1 -1 + 1 -1 + ... and therefore ( x -1 ) + x = 0 which gives θ 0 = 3.
Figure 7: The wave function renormalisable trajectories for Λ = 0 as a function of RG time t . The green dotted line is the trivial trajectory that remains at the fixed point β g ( ∗ ) = 0 for all t . The red dashed line starts at g > g ∗ and Z k rapidly decreases before g runs into a pole. For the trajectory g > g ∗ the wave function renormalisation is given by the blue solid line and we observe that it reaches Z 0 = 1 for decreasing t corresponding to classical scaling as k → 0.
methods gives more evidence that the critical exponent (8.5) is physical and not an artefact of our approximation. Furthermore our analytical formula suggests the glyph[epsilon1] -expansion converges in d < 4 dimensions.
## D. Absence of essential divergences
To better understand our results we now make the one-loop approximation (6.13) while including both scalar contributions S 0 and S σ . Expressing the beta functions in terms of τ ≡ G k ⋅ Λ k and g we find that τ is scale independent
$$\beta _ { \tau } \equiv k \frac { \partial } { \partial k } ( G _ { k } \Lambda _ { k } ) = 0 \,.$$
This tells us that the cosmological constant (measured in Planck units) receives no quantum corrections from the conformal sector at one-loop. The beta function for g reads
$$\beta _ { g } = 2 g - \frac { g ^ { 2 } \Phi _ { 1 } } { 6 \pi } \,,$$
for which there is a fixed point for g ∗ = 12 π Φ 1 . The critical exponents are given by
$$\theta _ { 0 } = 2 \,, \quad \theta _ { \tau } = 0 \,,$$
independent of the regulator R k or the parameterisation of the couplings. Thus, at one-loop the conformally reduced model shows that the τ is exactly marginal and that G k is asymptotically safe. The former can be understood by noting that at one-loop S σ + S 0 is proportional to the equations of motion R -4Λ . k This follows from the on-shell cancelations between the conformal mode and the scalar Jacobian (3.5) and the fact that we neglect terms ∂ t Λ k in the right hand side of the flow equation. Therefore only the 'inessential' coupling G k runs. Here inessential refers to the fact that ∂ G t k appears as a coefficient of the equations of motion in the left hand side of the flow equation. This can be seen by writing the left hand side of (3.10) as
$$\partial _ { t } \Gamma _ { k } = \int d ^ { 4 } x \sqrt { \det g _ { \mu \nu } } \left ( \frac { \partial _ { t } \tau } { 8 \pi G _ { k } ^ { 2 } } + \frac { \partial _ { t } G _ { k } } { 1 6 \pi G _ { k } ^ { 2 } } ( R - 4 \Lambda _ { k } ) \right ) \,.$$
Normally inessential couplings do not require fixed points and can be removed via an appropriate field redefinition. However, G k can only be removed by a redefinition of the metric and since this would also rescale k [80] Newton's couplings also requires a fixed point for gravity to be asymptotically safe. Therefore, due to the double rˆle of the metric, as a force carrier and the o origin of scale, G k is promoted to an essential coupling.
In fact we can make a more general statement about the form of divergences coming from the conformal sector, S conf ≡ S 0 + S σ , beyond one loop and our truncation to the first two heat kernel coefficients. Let's first consider a type I or type II regulator, such that R k is independent of Λ , then k S σ is independent of ∂ t Λ . k It now follows that for R = 4Λ k we have S conf = 0 thus only inessential curvature terms, those proportional to ( R -4Λ k ) , can be generated. All essential divergences must cancel between the conformal fluctuations and the Jacobian J 0 (provided we choose the same regulator for both). That is
$$\mathcal { S } _ { \sigma } + \mathcal { S } _ { 0 } \in ( R - 4 \Lambda _ { k } ) \quad \text{for } \Lambda _ { k } - \text{independent.} \text{regulators} \,.$$
If we instead use a type III Λ k -dependent cutoff we will then gain additional essential divergences
$$\mathcal { S } _ { \sigma } + \mathcal { S } _ { 0 } \sim \partial _ { t } \Lambda _ { k } \,,$$
which are proportional to the scale derivative of the cosmological constant Λ k . This is the case we have encountered above (8.2) where there exists a fixed point at Λ k = 0. Along the renormalisable
trajectory ∂ t Λ k remains zero, therefore we would not generate any essential divergences in this case either. This leads us to conclude that these cancelations remain beyond the truncated heat kernel expansion and that therefore conformal fluctuations may generate no physical divergences. Furthermore if we are forced to use a type III cutoff to ensure stability (i.e. convexity of the effective action) this would only be possible for a vanishing cosmological constant. Whether or not we are forced to use a type III regulator we reach the conclusion that by setting Λ k →∞ = 0 at a UV fixed point (i.e. setting the cosmological constant to zero in the bare action) we would recover a vanishing renormalised cosmological constant Λ = Λ 0 in the classical limit without any fine tuning. This follows since the τ = G k ⋅ Λ k either receives no quantum corrections or the quantum corrections are proportional to ∂ t Λ . k
We note that the situation here is quite different from that encountered in f ( R ) gravity [81] where all operators where found to be inessential at a potential UV fixed point [82, 83]. In that case there existed no solutions to the equations of motion, and thus no essential operators were present. Here there are essential operators since the equation of motion has solutions, however no essential quantum corrections are generated and they can be consistently neglected.
## IX. SUMMARY AND CONCLUSION
## A. Summary
In this work we have revisited the renormalisation group flow of quantum gravity in the Einstein Hilbert approximation. In doing so we have made three novel steps:
i) In section III we have disentangled the gauge variant, topological and propagating degrees of freedom at the level of the renormalisation group equations by a careful treatment of the ghosts and auxiliary fields coming from the functional measure. While the gauge variant fields { ξ, ψ } have been made to cancel exactly with the ghosts [65], we have also identified the contributions from propagating graviton modes S grav = S ⊥ + S 1 and the topological conformal mode S conf = S σ + S 0 each of which have contributions from the Einstein-Hilbert action and the functional measure.
- ii) Further to this in section IV we have implemented the regularisation using a spectrally adjusted cutoff (4.4) depending on the full inverse propagator ∆ and determined the curvature constraint R > 4Λ 0 for which the regulator vanishes in the limit k → 0. This was done to obtain the correct IR limit of the flow equation while ensuring the convexity of the effective action (2.2).
iii) In section V we adopted a new non-perturbative approximation scheme whereby we truncate the early time heat kernel expansion at a finite order. In doing so we avoid an explicit curvature expansion to close our approximation while remaining sensitive to the UV divergences that renormalise G k and Λ . k
These modifications to the standard approach have had a direct effect on the physical results emerging from the resulting RG flow. First in the full theory we have the following results:
- a) In the UV there exists an asymptotically safe fixed point for positive g ∗ and λ ∗ in agreement
with all previous studies of the Einstein-Hilbert truncation in the background field approximation. However here we have found that the critical exponents are real and not a complex conjugate pair. This is in contrast to the standard background field approach but in agreement with vertex expansions which disentangle the background and dynamical metric and possible global fixed points in f ( R ) gravity.
- b) At one-loop the UV fixed point is still present and we find critical exponents are independent of the regulator function given by θ 0 = 2 , θ 1 = 64 33 .
- c) We have found globally safe RG flows which lead to classical general relativity at small distances compatible with a finite cosmological constant.
Only quantising the conformal mode σ as a topological (i.e. non-propagating) degree of freedom we find the following:
- d) For this theory we find two UV fixed points. One compatible with a negative cosmological constant and one at λ ∗ = 0 for which Λ k = 0 for all scales. For the λ ∗ = 0 fixed point we recover the critical exponent ν = 1 3 from non-perturbative lattice studies [33]. /
- e) At one-loop the essential parameter G k ⋅ Λ k has a vanishing beta function while G k reaches an asymptotically safe fixed point.
- f) We have argued that the integration over the topological conformal mode leads to no essential divergences at all loop orders, providing the first step towards a finite theory of quantum gravity along the lines suggested by 't Hooft [7].
## B. Conclusion
Since it seems highly unlikely that quantum gravity in four dimensions can be solved exactly we must always rely on approximations. Furthermore, since gravity becomes strongly coupled at high energies the approximation schemes used should be non-perturbative by construction. The question then arises on how to implement these schemes in a consistent manner. Here we have approached this question by concentrating on the convexity of the effective action and its relation to the physical degrees of freedom which are being quantised. Our attention has been focused on the UV behaviour of gravity assuming that the high energy theory is that of quantum general relativity.
Our results strongly suggest that gravity is asymptotically safe and that the low energy theory is consistent with Einstein's classical theory. In turn we have shed light on the cosmological constant problem finding a UV theory consistent with a vanishing cosmological constant on all scales. Although this fixed point is only found in the conformally reduced theory, the critical exponent ν = 1 3 is in agreement with lattice studies of full quantum gravity [33]. / This result is a clear vindication of our general philosophy to disentangle physical degrees of freedom at the level of the regulated functional integral. We therefore conclude that the methods developed here should be extend beyond the simple approximation studied here, and that the combination of lattice and continuum approaches to quantum gravity may prove fruitful in the near future.
## Acknowledgements
The author would like to thank Jan Pawlowski and Daniel Litim for discussions and helpful comments.
## Appendix A: Heat kernels and flow equation
To evaluate the traces in (3.10) we use the (truncated) early time heat kernel expansion (5.1) which depends on the heat kernel coefficients A n . Due to our field redefinitions we will obtain coefficients A i,n where i labels the field each of which takes the form A i,n = ∫ d 4 √ det g µν a i,n ( U i ) . Here U i are the potentials appearing in each component ∆ i = -∇ 2 + U i of the differential operator ∆ given in (3.9). These potentials are give by
$$U _ { \sigma } = - \frac { R } { 3 } - \frac { 4 } { 3 } \left ( \Lambda _ { k } - \frac { R } { 4 } \right ) \,, \ \ U _ { \perp } = \frac { R } { 6 } - 2 \left ( \Lambda _ { k } - \frac { R } { 4 } \right ) \,, \ \ U _ { 0 } = - \frac { R } { 3 } \,, \ \ U _ { 1 } = - \frac { R } { 4 } \,. \quad \quad ( A 1 )$$
which lead to the corresponding heat kernel coefficients
$$a _ { \sigma, 0 } & = a _ { 0, 0 } = 1 \,, \quad a _ { 0, 1 } = \frac { R } { 2 } \,, \quad a _ { \sigma, 1 } = \frac { R } { 2 } + \frac { 4 } { 3 } \left ( \Lambda _ { k } - \frac { R } { 4 } \right ) \,, \\ a _ { \perp, 0 } & = 5 \,, \quad a _ { \perp, 1 } = - \frac { 5 } { 3 } R + 1 0 \left ( \Lambda _ { k } - \frac { R } { 4 } \right ) \,, \quad a _ { 1, 0 } = 3 \,, \quad a _ { 1, 1 } = R \,.$$
To evaluate the right hand side of (3.10) we insert these coefficients along with the Q m,i functionals (6.1) into the trace formula (5.1) retaining terms up to n = n max = 1. The left hand side is then found by taking the scale derivative of the action (2.3). In terms of the threshold constants (6.2) this leads to the following flow equation
$$\tilde { V } \, \left [ \frac { \beta _ { \lambda } + 2 \lambda } { 8 \pi g } - \frac { \beta _ { g } - 2 g } { 1 6 \pi g ^ { 2 } } ( 2 \lambda - \tilde { R } ) \right ] = \sum _ { i } \mathcal { S } _ { i } \,,$$
where i = { 2 , σ, 1 0 , } sums over the various fields. Here we have introduced the dimensionless quantities g = k G 2 k , λ = k -2 Λ , k ˜ R = k -2 R and ˜ V = k 4 ∫ d x 4 √ det g µν and the beta functions β g = ˙ g = g ( 2 -η ) and β λ = ˙ λ . The terms on the right side are given by
$$\mathcal { S } _ { 0 } = - \frac { 2 \eta \tilde { \Phi } _ { 2 } + \eta \tilde { R } \tilde { \Phi } _ { 1 } + \tilde { R } \Phi _ { 1 } + 2 \Phi _ { 2 } } { 6 4 \pi ^ { 2 } } \tilde { V } \,, \quad \mathcal { S } _ { 1 } = - \frac { 3 \eta \tilde { \Phi } _ { 2 } + \eta \tilde { R } \tilde { \Phi } _ { 1 } + \tilde { R } \Phi _ { 1 } + 3 \Phi _ { 2 } } { 3 2 \pi ^ { 2 } } \tilde { V } \,, \quad \quad ( A 4 )$$
$$\mathcal { S } _ { \sigma } = \frac { \eta \tilde { \Phi } _ { 2 } - \frac { 1 } { 1 8 } ( 8 \lambda + \tilde { R } ) \left ( - 3 \eta \tilde { \Phi } _ { 1 } + 4 \hat { \Phi } _ { 1 } \left ( \beta _ { \lambda } + 2 \lambda \right ) - 3 \Phi _ { 1 } \right ) - \frac { 4 } { 3 } \hat { \Phi } _ { 2 } \left ( \beta _ { \lambda } + 2 \lambda \right ) + \Phi _ { 2 } } { 3 2 \pi ^ { 2 } } \tilde { V } \,, \quad ( A 5 )$$
$$\mathcal { S } _ { 2 } = \frac { 5 \left ( \eta \tilde { \Phi } _ { 2 } - 2 \hat { \Phi } _ { 2 } \left ( \beta _ { \lambda } + 2 \lambda \right ) + \Phi _ { 2 } \right ) + \frac { 5 } { 6 } \left ( 5 \tilde { R } - 1 2 \lambda \right ) \left ( - \eta \tilde { \Phi } _ { 1 } + 2 \hat { \Phi } _ { 1 } \left ( \beta _ { \lambda } + 2 \lambda \right ) - \Phi _ { 1 } \right ) } { 3 2 \pi ^ { 2 } } \tilde { V } \,. \quad ( A 6 )$$
- [1] G. 't Hooft and M. Veltman, Annales Poincare Phys.Theor. A20 , 69 (1974).
- [2] M. H. Goroff and A. Sagnotti, Nucl.Phys. B266 , 709 (1986).
- [3] S. Weinberg, Rev. Mod. Phys. 61 , 1 (1989), URL http://link.aps.org/doi/10.1103/RevModPhys. 61.1 .
- [4] P. O. Mazur and E. Mottola, Nucl.Phys. B341 , 187 (1990).
- [5] H. W. Hamber and R. Toriumi, Int.J.Mod.Phys. D22 , 1330023 (2013), 1301.6259.
- [6] S. Weinberg (1979), in General Relativity: An Einstein centenary survey, ed. S. W. Hawking and W. Israel, 790- 831.
- [7] G. 't Hooft (2010), gr-qc/1009.0669.
- [8] G. 't Hooft (2009), 0909.3426.
- [9] K. Falls and D. F. Litim, Phys.Rev. D89 , 084002 (2014), 1212.1821.
- [10] K. Wilson and J. B. Kogut, Phys.Rept. 12 , 75 (1974).
- [11] K. G. Wilson, Rev. Mod. Phys. 47 , 773 (1975), URL http://link.aps.org/doi/10.1103/ RevModPhys.47.773 .
- [12] J. Berges, N. Tetradis, and C. Wetterich, Phys.Rept. 363 , 223 (2002), hep-ph/0005122.
- [13] J. Polonyi, Central Eur.J.Phys. 1 , 1 (2003), hep-th/0110026.
- [14] J. M. Pawlowski, Annals Phys. 322 , 2831 (2007), hep-th/0512261.
- [15] H. Gies, Lect.Notes Phys. 852 , 287 (2012), hep-ph/0611146.
- [16] O. J. Rosten, Phys.Rept. 511 , 177 (2012), 1003.1366.
- [17] W. Souma, Prog. Theor. Phys. 102 , 181 (1999), hep-th/9907027.
- [18] M. Reuter and F. Saueressig, Phys. Rev. D65 , 065016 (2002), hep-th/0110054.
- [19] O. Lauscher and M. Reuter, Phys. Rev. D65 , 025013 (2002), hep-th/0108040.
- [20] D. F. Litim, Phys. Rev. Lett. 92 , 201301 (2004), hep-th/0312114.
- [21] A. Codello, R. Percacci, and C. Rahmede, Annals Phys. 324 , 414 (2009), 0805.2909.
- [22] D. Benedetti, P. F. Machado, and F. Saueressig, Mod.Phys.Lett. A24 , 2233 (2009), 4 pages, 0901.2984.
- [23] N. Ohta and R. Percacci, Class.Quant.Grav. 31 , 015024 (2014), 1308.3398.
- [24] K. Falls, D. Litim, K. Nikolakopoulos, and C. Rahmede (2013), hep-th/1301.4191.
- [25] D. F. Litim, AIP Conf. Proc. 841 , 322 (2006), hep-th/0606044.
- [26] M. Niedermaier and M. Reuter, Living Rev. Rel. 9 , 5 (2006).
- [27] M. Niedermaier, Class. Quant. Grav. 24 , R171 (2007), gr-qc/0610018.
- [28] Towards a New Understanding of Space,
- R. Percacci (2007), in 'Approaches to Quantum Gravity: Time and Matter' ed. D. Oriti, Cambridge University Press, 0709.3851.
- [29] D. F. Litim, PoS QG-Ph , 024 (2007), 0810.3675.
- [30] M. Reuter and F. Saueressig (2007), 0708.1317.
- [31] R. Percacci (2011), 1110.6389.
- [32] M. Reuter and F. Saueressig, New J.Phys. 14 , 055022 (2012), 1202.2274.
- [33] H. W. Hamber, Phys. Rev. D61 , 124008 (2000), hep-th/9912246.
- [34] H. W. Hamber and R. M. Williams, Phys. Rev. D70 , 124007 (2004), hep-th/0407039.
- [35] H. W. Hamber, Gen.Rel.Grav. 41 , 817 (2009), 0901.0964.
- [36] J. Ambjorn, A. Goerlich, J. Jurkiewicz, and R. Loll, Phys.Rept. 519 , 127 (2012), 1203.3591.
- [37] J. Ambjorn, A. Goerlich, J. Jurkiewicz, and R. Loll (2013), 1302.2173.
- [38] M. R. Niedermaier, Phys.Rev.Lett. 103 , 101303 (2009).
- [39] M. Niedermaier, Nucl.Phys. B833 , 226 (2010).
- [40] A. Codello, R. Percacci, and C. Rahmede, Int. J. Mod. Phys. A23 , 143 (2008), 0705.1769.
- [41] P. F. Machado and F. Saueressig, Phys. Rev. D77 , 124045 (2008), 0712.0445.
- [42] R. Percacci and D. Perini, Phys. Rev. D67 , 081503 (2003), hep-th/0207033.
- [43] R. Percacci and D. Perini, Phys. Rev. D68 , 044018 (2003), hep-th/0304222.
- [44] G. Vacca and O. Zanusso, Phys.Rev.Lett. 105 , 231601 (2010), 1009.1735.
- [45] A. Eichhorn and H. Gies, New J.Phys. 13 , 125012 (2011), 1104.5366.
- [46] P. Dona, A. Eichhorn, and R. Percacci (2013), 1311.2898.
- [47] E. Manrique, M. Reuter, and F. Saueressig, Annals Phys. 326 , 463 (2011), 1006.0099.
- [48] A. Codello, G. D'Odorico, and C. Pagani, Phys.Rev. D89 , 081701 (2014), 1304.4777.
- [49] N. Christiansen, B. Knorr, J. M. Pawlowski, and A. Rodigast (2014), 1403.1232.
- [50] D. Becker and M. Reuter (2014), 1404.4537.
- [51] M. Reuter, Phys. Rev. D57 , 971 (1998), hep-th/9605030.
- [52] I. Donkin and J. M. Pawlowski (2012), 1203.4207.
- [53] A. Eichhorn, H. Gies, and M. M. Scherer, Phys. Rev. D80 , 104003 (2009), 0907.1828.
- [54] A. Eichhorn and H. Gies, Phys. Rev. D81 , 104010 (2010), 1001.5033.
- [55] K. Groh and F. Saueressig, J. Phys. A43 , 365403 (2010), 1001.5032.
- [56] S. Nagy, J. Krizsan, and K. Sailer, JHEP 1207 , 102 (2012), 1203.6564.
- [57] D. Litim and A. Satz (2012), 1205.4218.
- [58] S. Rechenberger and F. Saueressig, Phys.Rev. D86 , 024018 (2012), 1206.0657.
- [59] N. Christiansen, D. F. Litim, J. M. Pawlowski, and A. Rodigast (2012), 1209.4038.
- [60] C. Wetterich, Phys. Lett. B301 , 90 (1993).
- [61] D. F. Litim, J. M. Pawlowski, and L. Vergara (2006), hep-th/0602140.
- [62] E. Marchais, Ph.D. thesis, University of Sussex (2013).
- [63] D. F. Litim and J. M. Pawlowski, Phys.Lett. B546 , 279 (2002), hep-th/0208216.
- [64] D. Benedetti, Europhys.Lett. 102 , 20007 (2013), 1301.4422.
- [65] D. Benedetti, New J.Phys. 14 , 015005 (2012), 1107.3110.
- [66] J. York, James W., J.Math.Phys. 14 , 456 (1973).
- [67] P. O. Mazur and E. Mottola, Absence of phase in the sum over spheres, Los Alamos preprint (1989).
- [68] S. Folkerts, D. F. Litim, and J. M. Pawlowski, Phys.Lett. B709 , 234 (2012), 1101.5552.
- [69] D. F. Litim, Phys.Rev. D64 , 105007 (2001), hep-th/0103195.
- [70] D. Dou and R. Percacci, Class. Quant. Grav. 15 , 3449 (1998), hep-th/9707239.
- [71] A. R. Brown and A. Dahlen, Phys.Rev. D85 , 104026 (2012), 1111.0301.
- [72] A. Kaya, Phys.Rev. D87 , 123501 (2013), 1303.5459.
- [73] M. Reuter and H. Weyer, Phys. Rev. D80 , 025001 (2009), 0804.1475.
- [74] M. Reuter and H. Weyer, Phys. Rev. D79 , 105005 (2009), 0801.3287.
- [75] P. F. Machado and R. Percacci, Phys. Rev. D80 , 024020 (2009), 0904.2510.
- [76] M. Demmel, F. Saueressig, and O. Zanusso, JHEP 1211 , 131 (2012), 1208.2038.
- [77] M. Demmel, F. Saueressig, and O. Zanusso, Journal of High Energy Physics 2014 , 26 (2014), URL http://dx.doi.org/10.1007/JHEP06%282014%29026 .
- [78] A. Nink and M. Reuter, JHEP 1301 , 062 (2013), 1208.0031.
- [79] T. Aida and Y. Kitazawa, Nucl.Phys. B491 , 427 (1997), hep-th/9609077.
- [80] R. Percacci and D. Perini, Class.Quant.Grav. 21 , 5035 (2004), hep-th/0401071.
- [81] J. A. Dietz and T. R. Morris, JHEP 1307 , 064 (2013), 1306.1223.
- [82] J. A. Dietz and T. R. Morris, JHEP 1301 , 108 (2013), 1211.0955.
- [83] D. Benedetti and F. Caravelli, JHEP 1206 , 017 (2012), 1204.3541. | 10.1007/JHEP01(2016)069 | [
"Kevin Falls"
] | 2014-08-01T19:53:55+00:00 | 2014-08-01T19:53:55+00:00 | [
"hep-th",
"gr-qc"
] | Asymptotic safety and the cosmological constant | We study the non-perturbative renormalisation of quantum gravity in four
dimensions. Taking care to disentangle physical degrees of freedom, we observe
the topological nature of conformal fluctuations arising from the functional
measure. The resulting beta functions possess an asymptotically safe fixed
point with a global phase structure leading to classical general relativity for
positive, negative or vanishing cosmological constant. If only the conformal
fluctuations are quantised we find an asymptotically safe fixed point
predicting a vanishing cosmological constant on all scales. At this fixed point
we reproduce the critical exponent, $\nu = 1/3$, found in numerical lattice
studies by Hamber. This suggests the fixed point may be physical while solving
the cosmological constant problem. |
1408.0277v1 | Cite this paper as J. Buongiorno, D. G. Cahill, C. H. Hidrovo, S. Moghaddam, A.J. Schmidt, L. Shi, 'Micro- and Nanoscale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces', Nanoscale and Microscale Thermophysical Engineering, vol. 18, p. 270-287 (2014). DOI:10.1080/15567265.2014.883454
Title: Micro- and Nano-scale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces
Authors: Jacopo Buongiorno, David G. Cahill, Carlos H. Hidrovo, Saeed Moghaddam, Aaron J. Schmidt, Li Shi
## Abstract
In this opinion piece, we discuss recent advances in experimental methods for characterizing phase change heat transfer. We begin with a survey of techniques for high resolution measurements of temperature and heat flux at the solid surface and in the working fluid. Next, we focus on diagnostic tools for boiling heat transfer and describe techniques for visualizing the temperature and velocity fields as well as measurements at the single bubble level. Finally, we discuss techniques to probe the kinetics of vapor formation within a few molecular layers of the interface. We conclude with our outlook for future progress in experimental methods for phase change heat transfer.
## 1. Introduction
There has been increasing interest in the potential for nanostructured surfaces to enhance phase change heat transfer, as discussed in several other papers in this issue. Heat transfer between a solid and a working fluid is fundamentally an interface phenomenon, and advancing both scientific understanding and engineering practice in this field requires experimental capabilities for high resolution characterization of the thermodynamics and kinetics of the underlying interfacial processes. The science and engineering discipline of phase-change heat transfer is not alone in facing this daunting challenge. Tribology, corrosion, water purification, electrochemical energy conversion, catalysis, photocatalysis, crystal growth, and adhesion of contacting surfaces are just some of the current problems in science and engineering that share these challenges. The traditional science of solid surfaces focuses on clean crystals in ultra-high vacuum, where powerful tools based on electron scattering and photoemission can be used to precisely probe structure and composition of a surface with high spatial and temporal resolutions. In the topical and more practical problems listed above, co-existence of multiple phases makes the ambient highly complex, the surfaces of interest are often inhomogeneous on multiple length scales, and elevated pressures and temperatures of the vapors and liquids involved greatly restrict the applicability of most experimental tools for the measurement of phase, composition, temperature, and heat flux.
Despite these challenges, there has been substantial work on experimental techniques to study different aspects of phase change heat transfer across many length and time scales, and the resulting data has shed light on many of the complex physical processes involved. In this opinion
Cite this paper as J. Buongiorno, D. G. Cahill, C. H. Hidrovo, S. Moghaddam, A.J. Schmidt, L. Shi, 'Micro- and Nanoscale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces', Nanoscale and Microscale Thermophysical Engineering, vol. 18, p. 270-287 (2014). DOI:10.1080/15567265.2014.883454
piece, we discuss recent advances in experimental methods for characterizing several aspects of phase change heat transfer from the nanoscale to the macroscale. We begin with a discussion on measurement challenges in phase change heat transfer in the context of the problem of nucleate boiling, and survey some techniques for high resolution measurements of temperature and heat flux at the solid surface and in the working fluid. We consider recent experiments based on microfabricated heater and sensor arrays, as well as a variety of optical techniques for visualizing the solid surface and the temperature and velocity fields in the adjacent fluid. Next we discuss recent microscale measurements at the single bubble level during boiling, and then go on to describe the use of photothermal and photoacoustic techniques to probe the kinetics of vapor formation within a few molecular layers of the interface. We conclude with our outlook for future progress and directions.
## 2. Measurement needs and existing approaches
The phenomenon of nucleate boiling highlights the experimental challenges associated with measuring phase change processes over multiple time and length scales. As a critical mechanism of heat transfer in numerous practical applications ranging from large-scale power plants to small-scale heat sinks, nucleate boiling has been studied for several decades. However, its physical nature is still plagued with large uncertainties, due in large part to the complex coupling of mass, momentum, and energy transport that occurs between the solid surface, the wetting liquid, and the vapor produced to generate the bubble. The complexity of the problem and the lack of physical understanding of the process have resulted in diverse hypotheses concerning the physics of the involved heat transfer processes [73-76]. For example, the microlayer [73,74] presence and its level of contribution to the bubble growth and the overall surface heat transfer is still a subject of debate and some studies (e.g. [75]) suggest that most heat transfer to the liquid takes place via a transient conduction process triggered by the bubble growth and departure.
Evaluation of analytical models requires new, highly-resolved data for the many individual (sub)phenomena that constitute what we call 'boiling'. In particular, a complete characterization of nucleate boiling would require knowledge of bubble size and shape throughout the ebullition cycle, bubble departure frequency (including growth and wait times), nucleation site density, wetted area fraction (i.e. fraction of the boiling surface that is in contact with the liquid phase), temperature field (2D on the boiling surface, and D within the fluid), velocity field, local heat flux, and evolution of the microlayer at the base of the bubble. To aid in the development and validation of models and simulations, the above quantities should be measured over a meaningful range of conditions; in dimensionless terms: P/Pcr (pressure), Bo (heat flux), Ja (subcooling) and Re (flow rate in flow boiling).
Cite this paper as J. Buongiorno, D. G. Cahill, C. H. Hidrovo, S. Moghaddam, A.J. Schmidt, L. Shi, 'Micro- and Nanoscale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces', Nanoscale and Microscale Thermophysical Engineering, vol. 18, p. 270-287 (2014). DOI:10.1080/15567265.2014.883454
Traditionally, boiling heat transfer has been studied with fairly simple apparatuses, consisting of a heater (typically a metallic plate or rod) instrumented with thermocouples, and high-speed cameras for visualization of the interactions among the fluid phases and the boiling surface. Thermocouples can only measure temperature at discrete locations on the boiling surface, thus little information on the temperature distribution can be obtained, while the usefulness of highspeed video is typically limited by poor optical access to the nucleation site and interference from adjacent bubbles, although the total reflection approach [35-38] allows for visualization of the boiling process from below the boiling surface through a transparent heater at any conditions.
Because no one diagnostic can generate all the desired data, a requirement key to the success of any modern study of boiling is combination/synchronization of multiple diagnostics. Secondgeneration two-phase flow diagnostics, such as multi-sensor conductivity and optical probes [39,40] and wire-mesh probes [41], can measure bubble diameter and velocity near the boiling surface. However, these approaches are intrusive, and also produce data only at discrete locations within the boiling fluid. On the other hand, X-ray and -ray tomography is nonintrusive, but rather costly/cumbersome as the radiation source has to be rotated at high speed around the test section, which also may limit the time and/or space resolution of the technique; ultra-fast approaches have been developed to increase the time and space resolution in recent years, including 2D non-tomographic approaches that can reach extraordinarily high spatial resolution (<1 m), but require massive X-rays sources [42]. Several modern and powerful experimental approaches are coming to full maturity for use in two-phase systems, and can be used to satisfy the data needs discussed above. For example, infrared-based visualization of thermal patterns on the boiling surface was pioneered by Theofanous et al. [43] in the early 2000s, and then adopted by many others [38, 44-48], while arrays of micro-heaters, individually controlled to achieve a constant temperature boundary condition, can be used to resolve the local instantaneous surface heat flux, as shown by Demiray and Kim [49]. In the remainder of this section we survey some of the approaches which could be combined in complementary experiments to study nucleate boiling as well as other phase change processes such as condensation and evaporation.
## 2.1 Contact techniques
The surface superheat temperature at the onset of nucleation on a superhydrophobic surface can be well within the measurement uncertainty of a thermocouple [1], and boiling sub-processes with drastically different physical nature could be only several micrometers apart [Moghaddam and Kiger, 2009]. Thus, techniques with high temperature measurement accuracy as well as high spatial and temporal resolutions are needed to quantify phase change processes on a boiling surface. Spatially resolved characterization of the temperature field at the solid surface can be accomplished with a number of techniques. Arrays of microfabricated electro-thermal
Cite this paper as J. Buongiorno, D. G. Cahill, C. H. Hidrovo, S. Moghaddam, A.J. Schmidt, L. Shi, 'Micro- and Nanoscale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces', Nanoscale and Microscale Thermophysical Engineering, vol. 18, p. 270-287 (2014). DOI:10.1080/15567265.2014.883454
transducers have been employed for surface temperature measurements in many phase change heat transfer experiments. For example, an array of ~100 m wide parallel long thin film Au lines were employed to measure the evolution of the surface temperature distribution when a water drop slowly evaporated from the surface kept at 60 K before the drop impinged [1]. Because each line only measured the average temperature along the line, the results were analyzed with a numerical simulation to obtain the radial distribution of the surface temperature under the drop. In addition, microcantilevers with an integrated resistive heater and thermometer have been employed for measuring the boiling curve of microjets [2]. However, there is still ample room to push to the limits in terms of both the spatial and temperature resolutions of these microfabricated temperature sensors.
Highly sensitive measurements of heat conduction in individual nanostructures have been achieved using a number of techniques [4]. With the use of microfabricated suspended thin film Pt serpentine line resistance thermometers and sophisticated differential measurement schemes, temperature and heat flow resolutions better than 1 x 10 -3 K and 10 -10 W/K have been demonstrated [5,6]. In addition, nanoscale thermocouple junctions have been fabricated for scanning thermal microscopy measurements of surface temperature distribution [7]. Continuous efforts along this direction during the past decade have allowed for quantitative surface temperature measurements with sub-100 nm spatial resolution and temperature resolution on the order of 1 K [8,9]. In a recent work that has pushed the limit of this nanoscale thermocouple based measurement technique, a temperature resolution approaching 1 x 10 -3 K was achieved with a unique modulation and averaging approach [10]. Nevertheless, these microfabricated electro-thermal transducers are contact sensors and often suffer from a long thermal time constant that is insufficient for observing the 10 -3 s scale dynamics of phase change processes. Furthermore, these are intrusive measurements that can greatly affect the dynamics of these processes. This has motivated the development non-contact techniques, some of which are described below.
## 2.2 Non-contact techniques for surface temperature measurements
A variety of optical techniques have been employed for thermal measurements of phase change heat transfer. For example, infrared thermometry techniques are commonly used for measuring the solid surface temperature during phase change [11]. Silicon is transparent in the infrared spectrum, making it convenient for infrared thermal imaging of phase change heat transfer processes on surfaces fabricated on a silicon wafer. Under proper calibration procedures, this technique can be used to clearly detect the liquid-vapor-solid triple contact line under even small temperature differences (~1°C) [12]. However, the spatial resolution of such measurements is typically limited by far field diffraction to be on the order of the infrared wavelength, about 10 microns, which is often insufficient for resolving the detailed temperature distribution on
Cite this paper as J. Buongiorno, D. G. Cahill, C. H. Hidrovo, S. Moghaddam, A.J. Schmidt, L. Shi, 'Micro- and Nanoscale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces', Nanoscale and Microscale Thermophysical Engineering, vol. 18, p. 270-287 (2014). DOI:10.1080/15567265.2014.883454
nanostructured surfaces being explored to enhance boiling or condensation. In addition, the relatively low temporal resolution of this measurement approach limits it ability to capture fast transient events that require several kHz temperature sampling rate, as described later in this study.
Spatial resolution can be improved with optical techniques in the visible spectrum, such as micro-Raman spectroscopy based thermometry techniques that have been increasingly used to study heat conduction in nanostructured materials and devices [13]. Silicon, III-V compounds, graphite and other solids possess strong and distinct Raman peaks, with the line width broadened and peak position shifted with the temperature because of scattering of Raman active optical phonons with intermediate and low frequency phonons. However, the peak shift can also be caused by strain, which can result in uncertainty in the temperature determined from the peak shift. In comparison, the optical phonon temperature can be measured unambiguously from the intensity ratio of the Stokes to anti Stokes peaks. Nevertheless, accurate measurements of the anti-Stokes peak intensity may require high sample temperature. These factors have often limited the temperature resolution of micro-Raman thermometry technique to be several Kelvins or even tens of Kelvins, whereas a spatial resolution of 0.5 micron is achievable. An advantage of the micro-Raman thermometry technique is that different materials yield distinctly different Raman peak frequencies, so that the temperature difference between two ultrathin layers in contact can be measured from the respective Raman peaks [14]. This capability may allow for highly localized measurement of surface heat flux during phase change heat transfer.
Similar to Raman spectroscopy, Brillouin light scattering (BLS) is an inelastic light scattering technique. While Raman peaks arise from inelastic scattering of photons and high-frequency optical phonons, BLS peaks origin from that between photons and low-frequency acoustic phonons or magnons that are energy quanta of spin waves in ferromagnetic materials [15]. Although BLS phonon signal is usually weak, it has been recently shown that the BLS phonon peak intensity of glass depends nearly linearly with temperature [16], because the Bose-Einstein distribution of the low-frequency acoustic phonon population is reduced to the classical limit at room temperature. A temperature resolution of a few Kelvins has been demonstrated based on the BLS phonon peak intensity of glass. Moreover, the BLS magnon peak intensity can be very strong on ferromagnetic materials, with its position shifted with temperature because of temperature dependence of the exchange constant and saturation magnetization. A temperature resolution on the order of 1K has been achieved based on the BLS magnon peak position measured on a sub-100 nm thick permalloy film, whereas the spatial resolution is 0.5 micrometer [16]. Hence, BLS combined with one or two thin ferromagnetic coatings near the surface can potentially allow for sensitive non-contact measurement of the surface temperature or heat flux distribution with high spatial resolution. Similarly, time-resolved ellipsometry has been used to assess the thermal accommodation coefficient between a solid-gas interface by measurement of the photoacoustics at the Brillouin frequencies [17]. Systems consisting of inert gases (Xe, Kr,
Cite this paper as J. Buongiorno, D. G. Cahill, C. H. Hidrovo, S. Moghaddam, A.J. Schmidt, L. Shi, 'Micro- and Nanoscale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces', Nanoscale and Microscale Thermophysical Engineering, vol. 18, p. 270-287 (2014). DOI:10.1080/15567265.2014.883454
Ar and Ne), tetrafluoroethane (R-134a), methanol and water vapor interacting with both a hydrophilic and hydrophobic Au substrate were characterized.
Sub-10 nm spatial resolution has been achieved with a thermal scanning electron microscopy (ThSEM) technique [18], where the electron diffraction pattern of Si and other crystals is used to obtain the surface temperature. Meanwhile, environmental scanning electron microscopy (ESEM) technique has been increasingly used for studying dropwise condensation [19,20]. Further development of multifunctional electron microscopy instruments may potentially allow high spatial resolution surface temperature mapping during phase change heat transfer process. Nevertheless, the temporal resolution of the ThSEM and the aforementioned inelastic light scattering techniques need to be improved by orders of magnitude in order for them to capture the fast dynamics of phase change processes.
## 2.3 Measurements of the liquid and vapor phases
The temperature distribution in the liquid and vapor phase are critical for understanding the phase change heat transfer process, yet options for temperature measurement in the liquid and vapor phase are rather limited. There have been a number of reports of using laser-induced fluorescence to measure the temperature of droplets. For example, laser induced fluorescence was used to measure the temperature of impinging droplets mixed with fluorescence dye [22]. Emission spectrum from Pyrene dye stabilized with a cationic surfactant, cetyldimethylbenzylammonium chloride (CDBAC), is used as the thermometer. The surfactant is used to prevent pyrene from grouping into micelles. Given the controversy on thermal properties of nanofluids that are liquid suspension of solid nanoparticles [23], it remains to be better understood whether the fluorescence dye can affect the phase change processes of the liquid and vapor phases. Similarly, thermochromic liquid crystal (TLC) thermography has been used to measure the temperature near the triple line in the evaporation front of a water filled microcapillary [24]. However, this technique has a very limited temperature range (~4°C) and the same concerns raised by the presence of long polymeric molecules in fluorescence techniques apply here with the introduction of the liquid crystals.
At MIT, researchers have expanded the infrared (IR) thermography approach by synchronizing it with high-speed video (HSV) from below and beside the boiling surface, and Particle Image Velocimetry (PIV) [50,51]. Also, the DEtection of Phase by Infrared Thermometry (DEPIcT) technique was developed, which uses the differences in emissivity of the liquid and vapor phases to measure the phase distribution (including the shape of the microlayer) on the boiling surface of IR-transparent heaters [52,53]. Examples of the data obtained at MIT from these two approaches are shown in Figures 1-4, and details are given in the figure captions.
Cite this paper as J. Buongiorno, D. G. Cahill, C. H. Hidrovo, S. Moghaddam, A.J. Schmidt, L. Shi, 'Micro- and Nanoscale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces', Nanoscale and Microscale Thermophysical Engineering, vol. 18, p. 270-287 (2014). DOI:10.1080/15567265.2014.883454
It is also desirable to develop other non-contact optical temperature measurement techniques without the need of seeding temperature sensitive materials in the heat transfer fluids. For example, a Global Rainbow Thermometry has been developed to measure the temperature of water droplets based on the temperature dependence of the refractive index and light scattering pattern of the water droplet [25]. Another possibility is the use of near infrared (NIR) absorption thermometry which has been used to measure water temperature in PDMS microchannels with a resolution of 0.2K [26]. These and other non-invasive techniques represent a research direction worth further investigations.
## 3 Measurement at the single bubble level.
Having discussed a variety of techniques for measuring the solid, fluid, and vapor phases, we now turn to an experiment designed to study the nucleation and departure of single bubbles. Measurement at the single bubble level is a key to deciphering the physics of the complex heat and mass transfer processes involved in boiling, and could pave the way for better understanding of the higher order effects resulting from multiple bubbles interactions. While it is common knowledge that bubble generation at the surface is responsible for the observed enhanced heat transfer in boiling, details of the heat transfer processes triggered by bubble formation and departure are not clearly understood. Different and often contradictory hypotheses have been proposed to describe the nature of the heat transfer processes [54-65], and although great efforts have been devoted to development and testing of these models, the necessary measurement tools have not been generally available to test the basic assumptions of the models at the single bubble level. This is because performance of the models has often been evaluated based on their ability to predict the overall surface heat transfer coefficient, which is the cumulative effect of all microscale boiling subprocesses. While this bulk measure is often the quantity of engineering interest, such a simple validation metric is insufficient to answer why a particular model fails.
Figure 5 shows the top view of a device that generates single bubbles and determines the surface temperature and heat flux at the surface-bubble interface using three layers of temperature sensors imbedded within a composite wall. The details of the device are discussed in Refs. [6669]. The device die is attached to a Pin Grid Array (PGA) that is installed on the bottom of a test liquid chamber and is connected from below to a custom-made Signal Conditioning Board (SCB) through a stack of sockets. The output of the SCB is directly connected to an A/D board installed in a PC. The temperature sensors are calibrated with an accuracy of ±0.1 °C. Data are collected during boiling of FC-72 liquid at saturation conditions under atmospheric pressure. Figure 6 shows images of a bubble at a surface temperature of 80.2 ºC, and Fig. 7 shows the surface temperature and heat flux results corresponding to the bubbling event.
Cite this paper as J. Buongiorno, D. G. Cahill, C. H. Hidrovo, S. Moghaddam, A.J. Schmidt, L. Shi, 'Micro- and Nanoscale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces', Nanoscale and Microscale Thermophysical Engineering, vol. 18, p. 270-287 (2014). DOI:10.1080/15567265.2014.883454
Comparison of the bubble images and the temperature data show that the initial formation of the bubble at t = 3.8 ms was associated with a sudden drop in surface temperature. The temperature drop started at the center of the array and progressed over the subsequent sensors. A second phase of surface temperature decline started after the bubble/surface contact area reached its maximum diameter and the apparent contact line started to advance over the contact area. The advancing liquid rewetted the dried out area contact area. This rewetting process began at about t = 5.8 ms. The process resulted in a continued decrease in temperature of sensor S-5 that had already significantly decreased due to the prior cooling event. The temperature decrease trend passed as a radially inward moving wave, corresponding to when the contact line successively passed over sensors S-4 to S-1. As can be seen in the lower panel of Fig. 9, the surface temperature outside the contact area remained unchanged during the entire bubble growth and departure process. Comparison of the bubble contact radius with the surface temperature history showed that the temperature drop at each sensor started after the apparent contact line passed over the sensor. This suggested that the observed temperature drop was due to surface cooling resulted from evaporation of a thin liquid layer (i.e., a 'microlayer') left over the surface, after the contact line rapidly receded. The surface temperature started to increase shortly after the microlayer was mostly evaporated. These data clearly reinforce the results of prior studies [7074] regarding the microlayer presence underneath a bubble, and suggest that the process is active primarily at the bubble/surface contact area, in contrast with the Mikic and Rohsenow [58] model that suggested an active area of twice the bubble diameter, and much closer to the more recent observations of Yaddanapudi and Kim [75]. Mikic and Rohsenow [58] cited Han and Griffith [34] as a basis for their assumption. However, Han and Griffith [76] did not provide a solid reason or any particular experimental evidence for their assumption. They simply assumed that following the departure of a bubble from the heating surface, a volume element of superheated liquid from an area twice the bubble diameter is brought into the bulk fluid.
The example results presented here suggest that in order to fully capture the heat transfer events at the bubble-surface interface, a temperature sampling rate on the order of 10 kHz is required. So, thermal imaging with 1-2 orders of magnitude lower sampling rate does not seem to be a suitable tool for capturing the details of the boiling sub-processes, although such tools can be used to capture the integral effect of the cooling events over a period of time. In addition, temperature sensors and the heated wall structure/properties should be carefully designed to isolate and study fast events (such as the microlayer evaporation process). For example, thermally thick walls (this is relative to the boiling characteristics of a particular fluid) do not allow proper isolation of different heat transfer events.
Cite this paper as J. Buongiorno, D. G. Cahill, C. H. Hidrovo, S. Moghaddam, A.J. Schmidt, L. Shi, 'Micro- and Nanoscale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces', Nanoscale and Microscale Thermophysical Engineering, vol. 18, p. 270-287 (2014). DOI:10.1080/15567265.2014.883454
## 4. Molecular- and nanoscale resolution of interfacial phenomena
There are few experimental methods that are capable of resolving the presence of a few molecular layers of liquid condensed on a surface. These adsorbed layers are critical, however, for determining the surface energies and the kinetics of evaporation and condensation. Sumfrequency vibrational spectroscopy can be used to probe the vibrational states at the interfaces but requires complex and expensive equipment and nearly perfect surfaces to suppress diffuse light scattering. X-ray or neutron reflectivity are also capable of directly resolving thin adsorbed layers but require large facilities to carry out the work and the sample requirements are highly restrictive. Photothermal and photoacoustic phenomena provide more generally applicable methods for probing the thermodynamics and kinetics of surfaces under conditions that are relevant for heat transfer applications. In the pump-probe ellipsometry approach, a short pump laser pulse creates a rapid temperature excursion of a surface and a time-delayed probe laser pulses measures the propagation of an acoustic wave in the vapor that is generated by heat transfer from the surface to the vapor [27] as well as the exchange of mass between the surface and vapor [17]. The time-resolved ellipsometry experiments revealed more extensive adsorption of R134a refrigerant on Au surfaces that were coated with OH-terminated self-assembled monolayers than on Au surfaces with CH4 termination [17]. The data show that while the macroscopic contact angle of refrigerants is nearly constant at very small values, the free energies of nanoscale adsorbed layers can nevertheless be affected by modifying surface chemistry.
The free energy of liquid-vapor interfaces creates a large nucleation barrier for the formation of a vapor bubble within a homogeneous liquid. At a solid surface, the nucleation barrier for the formation of vapor is reduced but homogeneous nucleation rates remain low. Therefore, rapid phase change of liquid-to-vapor in a heat-transfer application is usually thought to require preexisting vapor-liquid interfaces that circumvent the slow nucleation kinetics. At Illinois, researchers have investigated the evaporation rate of water droplets that adhere to hydrophilic surfaces [28] and the time-scale for the formation of a vapor layer during the droplet rebound from a hydrophobic surface [29]. In both cases, a small drop of liquid was brought rapidly into contact with a hot surface, and high frequency thermoreflectance measurements were used to determine when vapor forms at the solid-liquid interface with an ultimate time resolution on the order of 10 microseconds. A novel optical thermometry-based on the temperature dependence of two-photon absorption in an indirect gap semiconductor-enabled temperature measurements with similarly high time resolution. Measurements of temperature of Si using this approach have a spatial resolution on the order of 10 microns and a noise floor on the order of 1 K in a 1 kHz bandwidth [30].
Cite this paper as J. Buongiorno, D. G. Cahill, C. H. Hidrovo, S. Moghaddam, A.J. Schmidt, L. Shi, 'Micro- and Nanoscale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces', Nanoscale and Microscale Thermophysical Engineering, vol. 18, p. 270-287 (2014). DOI:10.1080/15567265.2014.883454
The apparatus used for the thermoreflectance measurements is shown in Figure 8, while Figure 9 shows illustrative transient temperature and effective thermal interface conductance data obtained during the impingement of dispensed water droplets at different surface temperatures. Because of the transient process of the impinging drops, the measurements of the transient absorption signal were synchronized to the trigger of drop dispenser. The time offset between the measurement and trigger of the dispenser was varied between 0 and 1500 ms. Results from a number of measurements made at the same time offset were averaged to obtain the evolution of the surface temperature and thermal conductance as a function of time after drop impingement on the substrate. The time evolution data were analyzed to obtain the heat transfer and residence time of the drop evaporation process, and it was found that the heat flux could exceed 500 W/cm 2 for a short duration on the order of 10 ms when the surface temperature was increased near 200 C. o
## 5. Conclusions
We have seen that there are many specialized measurement methods for studying phase change heat transfer, and that there is still much that can be learned from data generated today with stateof-the-art diagnostics. Indeed, transformative improvements in the understanding, prediction and control of phase change heat transfer may come from the development of new mechanistic models and direct numerical simulations, with inspiration and validation coming from a combination of existing experimental techniques that can probe temperature and heat flux with sufficient sensitivity over multiple length and time scales. However, there is great room for progress in experimental methods, and continuation of the following trends is advocated: i) more systematic use of combined (synchronized) diagnostics, to develop a complete physical picture of phase change phenomena; ii) development of temperature and velocity sensors with ever higher temporal resolution (>5 kHz) and spatial resolution (<5 m), particularly useful as we start to understand how phenomena at the microscale (e.g., microlayer evaporation) can affect macroscopic figures of merit such as the heat transfer coefficient and the Critical Heat Flux (CHF); and iii) more systematic use of engineered surfaces (engineered at the micro- and nanoscale), to control the composition and texture of the boiling surface, thus enabling investigation of the effects of roughness, wettability, porosity, presence of cavities, size and shape of cavities, and surface thermo-physical properties on boiling heat transfer.
We are confident that advances in experimental methods will significantly impact the discipline of phase change heat transfer, including the rapidly expanding research in phase change heat transfer on nanostructured surfaces. We emphasize, however, that further development of the field also requires clear accounting of what is known and what is unknown, i.e., that scientific progress requires the formulation of thoughtful questions that the field will ask and then subsequently answer as new capabilities for experiment come on-line. Theory and computational
Cite this paper as J. Buongiorno, D. G. Cahill, C. H. Hidrovo, S. Moghaddam, A.J. Schmidt, L. Shi, 'Micro- and Nanoscale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces', Nanoscale and Microscale Thermophysical Engineering, vol. 18, p. 270-287 (2014). DOI:10.1080/15567265.2014.883454
models typically require input parameters that are determined from experiment, and theory often includes simplifying assumptions that must be validated by experiment. This interplay between theory, modeling, and experiment will be critical for a better fundamental understanding of the key physical processes in phase change heat transfer, and for the translation of this understanding into improved performance in engineering applications.
## References
- 1. R. Betz, J. Jenkins, C.-J. C. Kim, and D. Attinger, "Boiling heat transfer on superhydrophilic, superhydrophobic, and superbiphilic surfaces," International Journal of Heat and Mass Transfer, vol. 57, pp. 733-741, 2013.
- 2. S. W. Paik, K. D. Kihm, S. P. Lee, and D. M. Pratt, "Spatially and Temporally Resolved Temperature Measurements for Slow Evaporating Sessile Drops Heated by a Microfabricated Heater Array," Journal of Heat Transfer, vol. 129, pp. 966-976, 2006.
- 3. J. Lee, H. Hunter, A. Glezer, and W. P. King, "Phase change and cooling characteristics of microjets measured using microcantilever heaters," Sensors and Actuators A: Physical, vol. 147, pp. 64-69, 2008.
- 4. Weathers and L. Shi, " Thermal Transport Measurement Techniques For Nanowires And Nanotubes," in Annual Review of Heat Transfer. vol. 16, G. Chen, Ed., ed, 2013, pp. 101-134.
- 5. M. C. Wingert, Z. C. Y. Chen, S. Kwon, J. Xiang, and R. Chen, "Ultra-sensitive thermal conductance measurement of one-dimensional nanostructures enhanced by differential bridge," Review of Scientific Instruments, vol. 83, p. 024901, 2012.
- 6. Weathers, K. Bi, M. T. Pettes, and L. Shi, "Reexamination of Thermal Transport Measurements of a Low-Thermal Conductance Nanowire with a Suspended MicroDevice," Review of Scientific Instruments, vol. In print, 2013.
- 7. L. Shi, O. Kwon, A. C. Miner, and A. Majumdar, "Design and batch fabrication of probes for sub-100 nm scanning thermal microscopy," Journal of Microelectromechanical Systems, vol. 10, pp. 370-378, Sep 2001.
- 8. K. Kim, J. Chung, J. Won, O. Kwon, J. S. Lee, S. H. Park, and Y. K. Choi, "Quantitative scanning thermal microscopy using double scan technique," Applied Physics Letters, vol. 93, p. 3, Nov 2008.
- 9. Jo, I. K. Hsu, Y. J. Lee, M. M. Sadeghi, S. Kim, S. Cronin, E. Tutuc, S. K. Banerjee, Z. Yao, and L. Shi, "Low-Frequency Acoustic Phonon Temperature Distribution in Electrically Biased Graphene," Nano Letters, vol. 11, pp. 85-90, Jan 2011.
- 10. W. Lee, K. Kim, W. Jeong, L. A. Zotti, F. Pauly, J. C. Cuevas, and P. Reddy, "Heat dissipation in atomic-scale junctions," Nature, vol. 498, pp. 209-212, 2013.
- 11. Gerardi, J. Buongiorno, L. W. Hu, and T. McKrell, "Study of bubble growth in water pool boiling through synchronized, infrared thermometry and high-speed video," International Journal of Heat and Mass Transfer, vol. 53, pp. 4185-4192, Sep 2010.
Cite this paper as J. Buongiorno, D. G. Cahill, C. H. Hidrovo, S. Moghaddam, A.J. Schmidt, L. Shi, 'Micro- and Nanoscale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces', Nanoscale and Microscale Thermophysical Engineering, vol. 18, p. 270-287 (2014). DOI:10.1080/15567265.2014.883454
- 12. H. Kim and J. Buongiorno, "Detection of liquid-vapor-solid triple contact line in twophase heat transfer phenomena using high-speed infrared thermometry," International Journal of Multiphase Flow, vol. 37, pp. 166-172, Mar 2011.
- 13. T. E. Beechem and J. R. Serrano, "Raman Thermometry of Microdevices: Choosing a
- Method to Minimize Error," Spectroscopy, vol. 26, pp. 36-44, Nov 2011.
- 14. Y. N. Yue, J. C. Zhang, and X. W. Wang, "Micro/Nanoscale Spatial Resolution Temperature Probing for the Interfacial Thermal Characterization of Epitaxial Graphene on 4H-SiC," Small, vol. 7, pp. 3324-3333, Dec 2011.
- 15. Berne and R. Pecora, Dynamics Light Scattering. New York: Dover, 2000.
- 16. R. Birt, K. An, A. Weathers, L. Shi, M. Tsoi, and X. Li, "Brillouin light scattering spectra as local temperature sensors for thermal magnons and acoustic phonons," Applied Physics Letters, vol. 102, pp. 082401-5, 2013.
- 17. C. K. Min, K. J. Chen, S. C. Bae, D. G. Cahill, and S. Granick, "Heat Transfer at SolidGas Interfaces by Photoacoustics at Brillouin Frequencies," Journal of Physical Chemistry C, vol. 116, pp. 10896-10903, May 2012.
- 18. X. Wu and R. Hull, "The material dependence of temperature measurement resolution in thermal scanning electron microscopy," Applied Physics Letters, vol. 102, p. 113107, 2013.
- 19. Y. C. Jung and B. Bhushan, "Wetting behaviour during evaporation and condensation of water microdroplets on superhydrophobic patterned surfaces," Journal of MicroscopyOxford, vol. 229, pp. 127-140, Jan 2008.
- 20. K. Rykaczewski, "Microdroplet Growth Mechanism during Water Condensation on Superhydrophobic Surfaces," Langmuir, vol. 28, pp. 7720-7729, May 2012.
- 21. P. Kukura, D. W. McCamant, and R. A. Mathies, "Femtosecond Stimulated Raman Spectroscopy," Annual Review of Physical Chemistry, vol. 58, pp. 461-488, 2007.
- 22. ala ar, o le , a ivera, easureme t of em eratures o Flight Water Droplets by Laser Induced Fluorescence Thermometry," Journal of Heat Transfer, vol. 126, pp. 279-285, 2004.
- 23. J. Buongiorno, D. C. Venerus, N. Prabhat, T. McKrell, J. Townsend, R. Christianson, Y. V. Tolmachev, P. Keblinski, L.-w. Hu, J. L. Alvarado, I. C. Bang, S. W. Bishnoi, M. Bonetti, F. Botz, A. Cecere, Y. Chang, G. Chen, H. Chen, S. J. Chung, M. K. Chyu, S. K. Das, R. D. Paola, Y. Ding, F. Dubois, G. Dzido, J. Eapen, W. Escher, D. Funfschilling, Q. Galand, J. Gao, P. E. Gharagozloo, K. E. Goodson, J. G. Gutierrez, H. Hong, M. Horton, K. S. Hwang, C. S. Iorio, S. P. Jang, A. B. Jarzebski, Y. Jiang, L. Jin, S. Kabelac, A. Kamath, M. A. Kedzierski, L. G. Kieng, C. Kim, J.-H. Kim, S. Kim, S. H. Lee, K. C. Leong, I. Manna, B. Michel, R. Ni, H. E. Patel, J. Philip, D. Poulikakos, C. Reynaud, R. Savino, P. K. Singh, P. Song, T. Sundararajan, E. Timofeeva, T. Tritcak, A. N. Turanov, S. V. Vaerenbergh, D. Wen, S. Witharana, C. Yang, W.-H. Yeh, X.-Z. Zhao, and S.-Q. Zhou, "A benchmark study on the thermal conductivity of nanofluids," Journal of Applied Physics, vol. 106, p. 094312, 2009.
- 24. C. Buffone and K. Sefiane, "Temperature measurement near the triple line during phase change using thermochromic liquid crystal thermography," Experiments in Fluids, vol. 39, pp. 99-110, Jul 2005.
- 25. J. P. A. J. van Beeck, D. Giannoulis, L. Zimmer, and M. L. Riethmuller, "Global Rainbow Thermometry For Average Temperature Measurement of Spray Droplets," in
Cite this paper as J. Buongiorno, D. G. Cahill, C. H. Hidrovo, S. Moghaddam, A.J. Schmidt, L. Shi, 'Micro- and Nanoscale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces', Nanoscale and Microscale Thermophysical Engineering, vol. 18, p. 270-287 (2014). DOI:10.1080/15567265.2014.883454
Proc. 10th International Symposium on Applications of Laser Techniques to Fluid Mechanics, Lisbon, 2001.
- 26. N. Kakuta, Y. Fukuhara, K. Kondo, H. Arimoto, and Y. Yamada, "Temperature imaging of water in a microchannel using thermal sensitivity of near-infrared absorption," Lab on a Chip, vol. 11, pp. 3479-3486, 2011.
- 27. C.K. Min, D. G. Cahill, and S. Granick, "Time-resolved ellipsometry for heat transfer studies at liquid/solid and gas/solid interfaces," Rev. Sci. Instrum. 81, 074902 (2010).
- 28. J.Y. Park, C.K. Min, S. Granick, and D.G. Cahill, "Heat transfer and residence time when water droplets hit a scalding surface," J. Heat Transfer 134, 101503 (2012).
- 29. J.Y. Park, A. Gardner, A. Ramesh, W.P. King, S. Granick, D.G. Cahill, "Droplet impingement and vapor layer formation on hot hydrophobic surfaces," submitted for publication.
- 30. C.K Min, J.Y. Park, D.G. Cahill, and S. Granick, "Fast, spatially-resolved thermometry of Si crystals by pump-probe two-photon absorption," J. Appl. Phys. 106, 013102 (2009).
- 31. N. Kurul, M.Z. Podowski, Multidimensional effects in forced convection subcooled boiling, Proc. 9th International Heat Transfer Conference, Jerusalem, Israel. pp. 21-25, 1990.
- 32. N. Kolev, How accurately can we predict nucleate boiling?, in Multiphase Flow Dynamics 2, Springer, 2002.
- 33. Sanna, C. Hutter, D.B.R. Kenning, T.G. Karayiannis, K. Sefiane, R.A. Nelson, Nucleate Pool Boiling Investigation on a Silicon Test Section with Micro-Fabricated Cavities, ECI International Conference on Boiling Heat Transfer, Florianópolis, Brazil, 3-7 May, 2009.
- 34. L. Gilman, E. Baglietto, Advancements on Wall Boiling Modeling in CFD: Leveraging New Understanding from MIT Flow Boiling Facility, Paper 633, The 15th International Topical Meeting on Nuclear Reactor Thermalhydraulics, NURETH-15, Pisa, Italy, May 12-15, 2013.
- 35. S. Nishio and H. Tanaka, Visualization of Boiling Structures in High Heat-Flux PoolBoiling, Int. J. Heat Mass Transfer, vol. 47, pp. 4559-4568, 2004.
- 36. H. J. Chung, H. C. No, Simultaneous visualization of dry spots and bubbles for pool boiling of R-113 on a horizontal heater, International Journal of Heat and Mass Transfer 46 2239-2251, 2003.
- 37. I.-C. Chu, H. C. NO, C.-H. Song, Observation of High Heat Flux Boiling Structures in a Horizontal Pool by a Total Refection Technique, Paper 401, 14th International Topical Meeting on Nuclear Reactor Thermal Hydraulics (NURETH-14), Toronto, Ontario, Canada, September 25-29, 2011.
- 38. S. Jung, H. Kim, Simultaneous Measurements of Liquid-Vapor Phase and Temperature Distributions on Boiling Surface with Synchronized Infrared Thermometry and Total Internal Reflection Techniques, Paper 387, The 15th International Topical Meeting on Nuclear Reactor Thermalhydraulics, NURETH-15, Pisa, Italy, May 12-15, 2013.
- 39. Barrau, N. Rivière, Ch. Poupot and A. Cartellier, Single and double optical probes in airwater two-phase flows: real time signal processing and sensor performance, Int. J. Multiphase Flow, vol. 25, no.2, pp. 229-256, 1999.
- 40. S. Kim, X. Y. Fu, X. Wang and M. Ishii, Development of the miniaturized four-sensor conductivity probe and the signal processing scheme, Int. J. Heat Mass Transfer, vol. 43, no. 22, 4101-4118, 2000.
Cite this paper as J. Buongiorno, D. G. Cahill, C. H. Hidrovo, S. Moghaddam, A.J. Schmidt, L. Shi, 'Micro- and Nanoscale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces', Nanoscale and Microscale Thermophysical Engineering, vol. 18, p. 270-287 (2014). DOI:10.1080/15567265.2014.883454
- 41. H. M. Prasser, A. Bottger, J. Zschau, A new electrode-mesh tomography for gas-liquid flows, Flow Measurement and Instrumentation, vol. 9, pp. 111-119, 1998.
- 42. C. Gerardi, K. Fezzaa, High-Speed X-Ray Visualization of Boiling Phenomena: Preliminary Demonstration, Paper 381, The 15th International Topical Meeting on Nuclear Reactor Thermalhydraulics, NURETH-15, Pisa, Italy, May 12-15, 2013.
- 43. T. G. Theofanous, J. P. Tu, A. T. Dinh and T. N. Dinh, The Boiling Crisis Phenomenon, J. Experimental Thermal Fluid Science, P.I: pp. 775-792, P.II: pp. 793-810, vol. 26, no. 6-7, 2002.
- 44. Wagner and P. Stephan, High-resolution measurements at nucleate boiling of pure FC-84 and FC-3284 and its binary mixtures, Journal of Heat Transfer, vol. 131, no. 12, p. 121008, 2009.
- 45. Golobic, J. Petkovsek, M. Baselj, A. Papez, D.B.R. Kenning, Experimental determination of transient wall temperature distributions close to growing vapour bubbles, Heat Mass Transfer, vol. 45, 857-866, 2009.
- 46. Golobic, J. Petkovsek, D.B.R. Kenning, Bubble growth and horizontal coalescence in saturated pool boiling on a titanium foil, investigated by high-speed IR thermography, Int. J. Heat Mass Transfer, vol. 55, 1385-1402, 2012.
- 47. T. H. Kim, E. Kommer, S. Dessiatoun, J. Kim, Measurement of two-phase flow and heat transfer parameters using infrared thermometry, Int. J. Multiphase Flow, vol. 40, 56-67, 2012.
- 48. C. Kunkelmann, K. Ibrahem, N. Schweizer, S. Herbert, P. Stephan, T. GambaryanRoisman, The effect of three-phase contact line speed on local evaporative heat transfer: Experimental and numerical investigations, Int. J. Heat Mass Transfer, vol. 55, 18961904, 2012.
- 49. Demiray, J. Kim, Microscale heat transfer measurements during pool boiling of FC-72: effect of subcooling, Int. J. Heat Mass Transfer, vol. 47, pp. 3257-3268, 2004.
- 50. C. Gerardi, J. Buongiorno, L. W. Hu, T. McKrell, Study of Bubble Growth in Water Pool Boiling through Synchronized, Infrared Thermometry and High-Speed Video, International Journal of Heat and Mass Transfer, vol. 53, 19-20 4185-4192, 2010.
- 51. X. Duan, B. Phillips, T. McKrell, J. Buongiorno, Synchronized High-Speed Video, Infrared Thermometry and PIV Data for Validation of Interface-Tracking Simulations of Nucleate Boiling Phenomena, Exp. Heat Transfer, 26:169-197, 2013.
- 52. Kim, J. Buongiorno, Detection of Liquid-Vapor-Solid Triple Contact Line in Two-Phase Heat Transfer Phenomena Using High-Speed Infra-Red Thermometry, Int. J. Multiphase Flow, vol. 37, 166-172, 2011.
- 53. Kim, Y. Park, J. Buongiorno, Measurement of Wetted Area Fraction in Subcooled Pool Boiling of Water Using Infrared Thermography, Nucl. Eng. Des., 2013, http://dx.doi.org/10.1016/j.nucengdes.2013.07.002.
- 54. M. Jakob, W. Linke, Boiling heat transfer, Phys. Z. 636 (1935) 267-273.
- 55. W.M. Rohsenow, A method of correlating heat transfer data for surface boiling of liquids, Journal of Heat Transfer. 74 (1951) 969-976.
- 56. C.L. Tien, Hydrodynamic model for nucleate pool boiling, International Journal of Heat and Mass Transfer. 5 (1962) 533-540.
- 57. N. Zuber, Nucleate boiling the region of isolated bubbles and the similarity with natural convection, International Journal of Heat and Mass Transfer. 6 (1963) 53-78.
Cite this paper as J. Buongiorno, D. G. Cahill, C. H. Hidrovo, S. Moghaddam, A.J. Schmidt, L. Shi, 'Micro- and Nanoscale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces', Nanoscale and Microscale Thermophysical Engineering, vol. 18, p. 270-287 (2014). DOI:10.1080/15567265.2014.883454
- 58. B.B. Mikic, W.M. Rohsenow, A new correlation of pool boiling data including the effect of heating surface characteristics, Journal of Heat Transfer. 91 (1969) 245-250.
- 59. H.K. Forster, R. Greif, Heat transfer to a boiling liquid-mechanism and correlations, Journal of Heat Transfer. 81 (1959) 43-54.
- 60. S.I. Haider, R.L. Webb, A transient micro-convection model of nucleate pool boiling, International Journal of Heat and Mass Transfer. 40 (1997) 3675-3688.
- 61. R.W. Graham, R.C. Hendricks, Assessment of convection, conduction, and evaporation in nucleate boiling, 1967.
- 62. R.L. Judd, K.S. Hwang, A comprehensive model for nucleate pool boiling heat transfer including microlayer evaporation, Journal of Heat Transfer. 98 (1976) 623-629.
- 63. R.J. Benjamin, A.R. Balakrishnan, Nucleate pool boiling heat transfer of pure liquids at low to moderate heat fluxes, International Journal of Heat and Mass Transfer. 39 (1996) 2495-2504.
- 64. P. Stephan, J. Hammer, A new model for nucleate boiling heat transfer, Heat Mass Transfer/Waerme-und Stoffuebertragung. 30 (1995) 119-125.
- 65. V.K. Dhir, Boiling heat transfer, Annual Review of Fluid Mechanics. 30 (1998) 365-401.
- 66. S. Moghaddam, K. Kiger, Physical mechanisms of heat transfer during single bubble nucleate boiling of FC-72 under saturation conditions-I. Experimental investigation, International Journal of Heat and Mass Transfer. 52 (2009) 1284-1294.
- 67. S. Moghaddam, K. Kiger, Physical mechanisms of heat transfer during single bubble nucleate boiling of FC-72 under saturation conditions- : heoretical a alysis, International Journal of Heat and Mass Transfer. 52 (2009) 1295-1303.
- 68. S. Moghaddam, Microscale study of nucleation process in boiling of low-surface-tension liquids, University of Maryland, College Park, MD, 2006.
- 69. S. Moghaddam, K.T. Kiger, A. Modafe, R. Ghodssi, A novel benzocyclobutene-based device for studying the dynamics of heat transfer during the nucleation process, Journal of Microelectromechanical Systems. 16 (2007) 1355-1366.
- 70. F.D. Moore, R.B. Mesler, The measurement of rapid surface temperature fluctuation during nucleate boiling of water, AIChE Journal. 7 (1961) 620-624.
- 71. R. Sharp, The nature of liquid film evaporation during nucleate boiling, NASA TN D1997, 1964.
- 72. R.C. Hendricks, R.R. Sharp, On the initiation of cooling due to bubble growth on a heating surface, NASA TN D-2290, 1961.
- 73. M.G. Cooper, The microlayer and bubble growth in nucleate pool boiling, International Journal of Heat and Mass Transfer. 12 (1969) 915-933.
- 74. L.D. Koffman, M.S. Plesset, Experimental observation of the microlayer in vapor bubble growth on a heated solid, International Journal of Heat and Mass Transfer. 105 (1983) 625- 632.
- 75. N. Yaddanapudi, J. Kim, Single bubble heat transfer in saturated pool boiling of FC-72, Multiphase Science and Technology. 12 (2001) 47-63.
- 76. C.Y. Han, P. Griffith, The mechanism of heat transfer in nucleate pool boiling-Part II: The heat flux-temperature difference relation, International Journal of Heat and Mass Transfer. 8 (1965) 905-914.
Cite this paper as J. Buongiorno, D. G. Cahill, C. H. Hidrovo, S. Moghaddam, A.J. Schmidt, L. Shi, 'Micro- and Nanoscale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces', Nanoscale and Microscale Thermophysical Engineering, vol. 18, p. 270-287 (2014). DOI:10.1080/15567265.2014.883454
## FIGURE CAPTIONS
Figure 1: (a) Schematic diagram of the experimental configuration, not to scale. The Al heater block is 1 cm thick and the diameter of the hole in the Al heater block is 1.5 cm. Droplet generator, sample, and objective lens are vertically aligned. (b) The sample is a 1 mm thick double-side polished Si wafer. The bottom film of TiO2 is an antireflection coating and the top film of TiO2 thermally isolated the Ti thin film from the Si wafer. The Pt film provides a chemically inert surface, stable against boiling water. (c) Optical layout of the pump-probe system used. Sample region is described in detail in Fig 1(a).
Figure 2: Illustrative data showing transient changes of temperature and effective thermal conductance created by the impingement of dispensed water volume of 0.19 mm . Panels (a) and 3 (c) are for a relatively low sample temperature of 130 ° C. Panels (b) and (d) are for a relatively high sample temperature of 210 C. Time zero is defined by the electronic trigger of the ° microdispenser. The series of water droplets arrives at the sample surface 50 ms after the trigger.
Figure 3: Representative synchronized PIV and high-speed video images of a growing and departing steam bubble at atmospheric pressure. (from [51])
Figure 4: Post-processing of the PIV data in Figure 3 yields the values of the z-component of the vorticity vector in the wake of a steam bubble. The abscissa represents time after bubble nucleation (bubble departure occurs at t=16 ms). Maximum uncertainty for vorticity values is ∼ 18%. (from [51])
Figure 5: Phase distribution on the boiling surface at various values of heat flux (normalized to CHF) and subcoolings, obtained with eth DEPIcT technique. Black represents dry regions, grey and white represent wet regions. The fluid is water at atmospheric pressure, boiling over hori o tal silico heaters Note that the subscri t 'CHF(-)' i icates the last heat flux ste before CHF occurs.
Figure 6: Post-processing of the DEPIcT data in Figure 5 yields the time-averaged wetted fraction (i.e. fraction of boiling surface that is contact with the liquid phase) as a function of normalized heat flux and subcooling.
Cite this paper as J. Buongiorno, D. G. Cahill, C. H. Hidrovo, S. Moghaddam, A.J. Schmidt, L. Shi, 'Micro- and Nanoscale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces', Nanoscale and Microscale Thermophysical Engineering, vol. 18, p. 270-287 (2014). DOI:10.1080/15567265.2014.883454
Figure 7: Close view of the MEM device showing the topmost sensor layer radially distributed. The bottom-right inset figure shows two sensors (i.e. H-1 and H-2) made at different levels beneath the sensor array. All temperature sensors are Resistance Temperature Detector (RTD). Material of the sensors is Ni and their thickness is approximately 10 nm. The H-1 and H-2 sensors are coil-shape with a diameter of 1 mm. Spacing between the H-1 and H-2 sensors is 7.5 μm he se sor array is 2 5 μm above the H-2 sensor. The sensor array is covered with a 0.2-μm thick polymer layer. The top-right image shows 0 7, 1 3, a 2 4 μm i iameter cavities
Figure 8: A bubbling event at surface temperature 80.2 °C. A waiting time of 2.9 ms exists between the bubbles. Time is in millisecond.
Figure 9: Surface temperature (upper panel) and heat flux (lower panel) variations during the bubbling event. Heat transfer during microlayer evaporation and transient conduction processes are marked. S-1 through S-8 indicate sensors 1-8 in Figure 7.
Cite this paper as J. Buongiorno, D. G. Cahill, C. H. Hidrovo, S. Moghaddam, A.J. Schmidt, L. Shi, 'Micro- and Nanoscale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces', Nanoscale and Microscale Thermophysical Engineering, vol. 18, p. 270-287 (2014). DOI:10.1080/15567265.2014.883454
Figure 1: Representative synchronized PIV and high-speed video images of a growing and departing steam bubble at atmospheric pressure. (from [51])
Cite this paper as J. Buongiorno, D. G. Cahill, C. H. Hidrovo, S. Moghaddam, A.J. Schmidt, L. Shi, 'Micro- and Nanoscale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces', Nanoscale and Microscale Thermophysical Engineering, vol. 18, p. 270-287 (2014). DOI:10.1080/15567265.2014.883454
Figure 2: Post-processing of the PIV data in Figure 3 yields the values of the z-component of the vorticity vector in the wake of a steam bubble. The abscissa represents time after bubble nucleation (bubble departure occurs at t=16 ms). Maximum uncertainty for vorticity values is ∼ 18%. (from [51])
Cite this paper as J. Buongiorno, D. G. Cahill, C. H. Hidrovo, S. Moghaddam, A.J. Schmidt, L. Shi, 'Micro- and Nanoscale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces', Nanoscale and Microscale Thermophysical Engineering, vol. 18, p. 270-287 (2014). DOI:10.1080/15567265.2014.883454
ATsub =30 %
Figure 3: Phase distribution on the boiling surface at various values of heat flux (normalized to CHF) and subcoolings, obtained with eth DEPIcT technique. Black represents dry regions, grey and white represent wet regions. The fluid is water at atmospheric pressure, boiling over hori o tal silico heaters Note that the subscri t 'CHF(-)' i icates the last heat flux ste before CHF occurs.
Cite this paper as J. Buongiorno, D. G. Cahill, C. H. Hidrovo, S. Moghaddam, A.J. Schmidt, L. Shi, 'Micro- and Nanoscale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces', Nanoscale and Microscale Thermophysical Engineering, vol. 18, p. 270-287 (2014). DOI:10.1080/15567265.2014.883454
Figure 4: Post-processing of the DEPIcT data in Figure 5 yields the time-averaged wetted fraction (i.e. fraction of boiling surface that is contact with the liquid phase) as a function of normalized heat flux and subcooling.
Cite this paper as J. Buongiorno, D. G. Cahill, C. H. Hidrovo, S. Moghaddam, A.J. Schmidt, L. Shi, 'Micro- and Nanoscale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces', Nanoscale and Microscale Thermophysical Engineering, vol. 18, p. 270-287 (2014). DOI:10.1080/15567265.2014.883454
Figure 5: Close view of the MEM device showing the topmost sensor layer radially distributed. The bottom-right inset figure shows two sensors (i.e. H-1 and H-2) made at different levels beneath the sensor array. All temperature sensors are Resistance Temperature Detector (RTD). Material of the sensors is Ni and their thickness is approximately 10 nm. The H-1 and H-2 sensors are coil-shape with a diameter of 1 mm. Spacing between the H-1 and H-2 sensors is 7.5 μm he se sor array is 2 5 μm above the H-2 sensor. The sensor array is covered with a 0.2-μm thick polymer layer. The top-right image shows 0 7, 1 3, a 2 4 μm i iameter cavities
Cite this paper as J. Buongiorno, D. G. Cahill, C. H. Hidrovo, S. Moghaddam, A.J. Schmidt, L. Shi, 'Micro- and Nanoscale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces', Nanoscale and Microscale Thermophysical Engineering, vol. 18, p. 270-287 (2014). DOI:10.1080/15567265.2014.883454
Figure 6: A bubbling event at surface temperature 80.2 °C. A waiting time of 2.9 ms exists between the bubbles. Time is in millisecond.
Cite this paper as J. Buongiorno, D. G. Cahill, C. H. Hidrovo, S. Moghaddam, A.J. Schmidt, L. Shi, 'Micro- and Nanoscale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces', Nanoscale and Microscale Thermophysical Engineering, vol. 18, p. 270-287 (2014). DOI:10.1080/15567265.2014.883454
Figure 7: Surface temperature (upper panel) and heat flux (lower panel) variations during the bubbling event. Heat transfer during microlayer evaporation and transient conduction processes are marked. S-1 through S-8 indicate sensors 1-8 in Figure 5.
Cite this paper as J. Buongiorno, D. G. Cahill, C. H. Hidrovo, S. Moghaddam, A.J. Schmidt, L. Shi, 'Micro- and Nanoscale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces', Nanoscale and Microscale Thermophysical Engineering, vol. 18, p. 270-287 (2014). DOI:10.1080/15567265.2014.883454
Figure 8: (a) Schematic diagram of the experimental configuration, not to scale. The Al heater block is 1 cm thick and the diameter of the hole in the Al heater block is 1.5 cm. Droplet generator, sample, and objective lens are vertically aligned. (b) The sample is a 1 mm thick double-side polished Si wafer. The bottom film of TiO2 is an antireflection coating and the top film of TiO2 thermally isolated the Ti thin film from the Si wafer. The Pt film provides a chemically inert surface, stable against boiling water. (c) Optical layout of the pump-probe system used. Sample region is described in detail in Fig 1(a).
Cite this paper as J. Buongiorno, D. G. Cahill, C. H. Hidrovo, S. Moghaddam, A.J. Schmidt, L. Shi, 'Micro- and Nanoscale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces', Nanoscale and Microscale Thermophysical Engineering, vol. 18, p. 270-287 (2014). DOI:10.1080/15567265.2014.883454
Figure 9: Illustrative data showing transient changes of temperature and effective thermal conductance created by the impingement of dispensed water volume of 0.19 mm . Panels (a) and 3 (c) are for a relatively low sample temperature of 130 ° C. Panels (b) and (d) are for a relatively high sample temperature of 210 C. Time zero is defined by the electronic trigger of the ° microdispenser. The series of water droplets arrives at the sample surface 50 ms after the trigger.
 | 10.1080/15567265.2014.883454 | [
"Jacopo Buongiorno",
"David G. Cahill",
"Carlos H. Hidrovo",
"Saeed Moghaddam",
"Aaron J. Schmidt",
"Li Shi"
] | 2014-08-01T19:55:38+00:00 | 2014-08-01T19:55:38+00:00 | [
"cond-mat.mes-hall",
"cond-mat.mtrl-sci"
] | Micro- and Nano-scale Measurement Methods for Phase Change Heat Transfer on Planar and Structured Surfaces | In this opinion piece, we discuss recent advances in experimental methods for
characterizing phase change heat transfer. We begin with a survey of techniques
for high-resolution measurements of temperature and heat flux at the solid
surface and in the working fluid. Next, we focus on diagnostic tools for
boiling heat transfer and describe techniques for visualizing the temperature
and velocity fields, as well as measurements at the single bubble level.
Finally, we discuss techniques to probe the kinetics of vapor formation within
a few molecular layers of the interface. We conclude with our outlook for
future progress in experimental methods for phase change heat transfer. |
1408.0278v2 | ## Hadroproduction of t anti-t pair with two isolated photons with PowHel
## A. Kardos and Z. Tr´cs´nyi o a
Institute of Physics and MTA-DE Particle Physics Research Group,
University of Debrecen, H-4010 Debrecen P.O.Box 105, Hungary
## Abstract
We simulate the hadroproduction of a t ¯ pair in association with two isolated t hard photons at 13 TeV LHC using the PowHel package. We use the generated events, stored according to the Les-Houches event format, to make predictions for differential distributions formally at the next-to-leading order (NLO) accuracy. We present predictions at the hadron level employing the cone-type isolation of the photons used by experiments. We also compare the kinematic distributions to the same distributions obtained in the t ¯ t H final state when the Higgs-boson decays into a photon pair, to which the process discussed here is an irreducible background.
## 1 Introduction
The Higgs-boson was discovered by the ATLAS [1] and CMS [2] experiments two years ago, and many of its properties have been measured since then. The results of these measurements are in agreement with the predictions of the Standard Model within the uncertainties of the measurements: (i) it is a J P = 0 + particle, (ii) the branching ratios are as predicted, and (iii) it couples to the masses of the heavy gauge bosons [3, 4]. Its mass is not predicted, but measured consistently by the two experiments: m / H GeV = 125 5 . ± 0 2 . stat ± 0 6 . syst by ATLAS [5] and 125 6 . ± 0 4 . stat ± 0 2 . syst by CMS [6].
The measurements in order to discover the properties of the Higgs-boson are not over. It is yet to measure its couplings to the fermions y f as well as to check if its triple and quartic self couplings are consistent with its mass. The smallness of these couplings require large integrated luminosity and such measurements become feasible only at the designed c.m. energy of the LHC. In this respect the t-quark is special among the fermions due to its large mass. As m H < m q , the Higgs-boson cannot decay into a t ¯ -pair, therefore, t the Yukawa-coupling y q can only be measured directly from t ¯ t H final states, which is an important goal at the LHC.
Measuring the t ¯ t H production cross section is very challenging due to the small production rates and in general large backgrounds. The experiments concentrate on studying many decay channels sorted into three main categories: (i) the hadronic, (ii) the leptonic and the (iii) di-photon channels. In the hadronic channel the Higgs-boson is assumed to decay into a b ¯ b or into τ + τ -pair, while one or both t-quarks decay leptonically (hadrons with single lepton or dilepton channels). In the leptonic channel the Higgs-boson decays into charged leptons and missing energy (through heavy vector bosons), while one or both t-quarks decay again leptonically. Finally, in the di-photon channel the Higgs-boson decays into a photon pair, while the t-quarks decay into jets (di-photon with hadrons), or into the semileptonic channel (di-photon with lepton). Common characteristics of all these channels is the large background from other SM processes. Thus a precise measurement needs to be aided by theory through precise predictions of distributions at the hadron level for the hadroproduction of a t ¯ -pair in association with one or two hard object(s) t X , with X = H [7], Z 0 [8, 9], W ± [10], an isolated photon [11], jet [12], b ¯ -pair [13], or a b pair of jets or isolated photons. The goal of the present article is to make high-precision predictions for the last of these.
Isolated hard photons are important experimental tools at the LHC. However, in perturbation theory these are rather cumbersome objects because the photons couple directly to quarks. If the quark that emits the photon is a light quark, treated massless in perturbative QCD, then the emission is enhanced at small angles and in fact, becomes singular for strictly collinear emission. Due to this divergence of the collinear emission, the usual experimental definition of an isolated photon, which allows for small hadronic activity even inside the isolation cone, cannot be implemented directly in a perturbative computation. One has to take into account the non-perturbative contribution of photon fragmentation, too [14].
Recently, a new method was presented to make predictions for processes with isolated hard photons in the final state [11]. This method builds on the POWHEG method [15, 16] as implemented in the POWHEG-Box [17]. The POWHEG-Box program requires input from the user - the Born phase space and various matrix elements: (i) the Born squared matrix element (SME), (ii) the spinand (iii) color-correlated Born SME, (iv) the SMEs for real and (v) the virtual corrections. In the PowHel framework [18] we obtain all these ingredients from the HELAC-NLO code [19]. The result of PowHel is pre-showered events (with kinematics up to Born plus first radiation) stored in Les Houches event (LHE) files [20]. These events can be further showered and hadronized using standard shower Monte Carlo (SMC) programs up to the hadronic stage, where any experimental cut can be used, so a realistic analysis becomes feasible.
The essence of the new method of predicting cross sections for processes with isolated photons is to introduce generation isolation [11] and generation cuts [12, 21] during event generation that are sufficiently small so that predictions at the hadron level obtained with physical cuts and isolation do not depend on them. If the generation isolation of the photon is smooth [22], that is infrared safe at all orders of perturbation theory, then the pre-showered events can effectively be considered as sufficiently inclusive event sample, which leads to predictions for distributions obtained with usual experimental isolation. While we neglect the fragmentation contribution of the photon, 'sufficiently inclusive' means that the physical predictions obtained with typical experimental isolation employed on these shoewerd LHEs are (i) independent of the generation isolation and (ii) the fragmentation contribution should be negligible compared to the scale dependence when the photons are harder than possible accompanying jets [11].
The new method of making predictions for processes including isolated photons in the final state at the next-to-leading order (NLO) accuracy matched with parton shower (PS) have been tested thoroughly for the case of t ¯ t γ hadroproduction in Ref. [11]. Here we use the same method for the case of t ¯ t γ γ hadropoduction, an irreducible background for t ¯ t H final states in the H → γγ decay channel. This process was first computed at leading-order accuracy in Refs. [23, 24]. Here we make predictions for this process at NLO accuracy and compare the predictions of PowHel to those of MADGRAPH5 using the Frixione isolation. Then we show that predictions at the hadronic stage are independent of the generation isolation and cuts and finally make predictions at the LHC at 13 TeV collision energy and compare the t ¯ t H H ( → γγ ) signal to the t ¯ t γ γ background.
## 2 Implementation
Details of generating LHEs with the POWHEG-Box have been discussed extensively in the literature. Here we only summarize those features of our implementation that are specific to the t ¯ t γ γ hadroproduction process.
The process p p p (¯) → t ¯ t γ γ involves several subprocesses. Not all of these are independent, but related by crossing. Hence matrix elements, both at tree-level and at
one-loop, are generated for a subset of subprocesses by HELAC-NLO and all others are obtained by crossing into relevant channels. In HELAC-NLO the matrix elements are stored in separate files using an internal notation, these are commonly dubbed as skeleton files. For this process, we generated skeleton files for the following subprocesses: g g → γ γ t ¯ , t u ¯ u → γ γ t ¯ and d ¯ t d → γ γ t ¯ for the Born and virtual, while t g g → γ γ t ¯ t g , u ¯ u → γ γ t ¯ t g and d ¯ d → γ γ t ¯ t g for the real-emission contribution. The ordering among final state particles is in compliance with that used by POWHEG-Box . We checked the correct evaluation of all the matrix elements by comparing them in several, randomly picked phase space points to the original ones in HELAC-NLO . The self-consistency between real emission contribution and subtraction terms is established by verifying that the limiting value of the ratio of the real emission part and the subtraction terms in all kinematically degenerate configurations approaches one.
The phase space for t ¯ t γ γ -production with two massless particles in the final state at the Born level is quite involved provided by possible singular photon-radiation from initial state (anti)quarks. Thus we decided to build the Born phase space in a recursive fashion using the method of [26, 27]. To construct an ( n +1)-particle phase space point from an n -particle one we use the inverse constructions of [16] with three random numbers of unity. A choice can be made between two possible inverse constructions, initial and final state ones, depending upon the underlying splitting. Due to the mass of the t-quark, singular photon radiation can only come from the initial state hence in our phase-space construction we considered only initial state as a possible source for photons, which we arranged symmetrically for the two photons.
## 3 Predictions at NLO
With the internal consistency of matrix elements having checked and a suitable underlying Born phase space set up, the code is validated to make predictions at NLO accuracy. With the publicly available independent programs MADGRAPH5 aMC@NLO / [25] we can compare the predictions of the two independent codes. For this comparison we used the following set of parameters: √ s = 8TeV at the LHC, CTEQ6L1 PDF with a 1-loop running α s for the LO and CTEQ6M PDF with a 2-loop running α s for the NLO comparison. We used equal factorization and renormalization scales set to the mass of the t-quark, m t = 172 5 GeV. . We set the fine-structure constant to α EM = 1 137 and kept fixed at that value. / We used five massless-quark flavores throughout the paper. We employed the following set of selection cuts: (i) both photons had to be hard, p ⊥ , γ i > 30 GeV, i ∈ { 1 2 , } and (ii) central, | y γ i | < 2 5, . i ∈ { 1 2 , } , (iii) a Frixione isolation [22] was applied to the photons with δ 0 < 0 4, . glyph[epsilon1] γ = n = 1. When this isolation is applied the partonic activity is limited around the photon if the separation is less than δ 0 . The partonic activity in a cone of δ around the photon is measured through the total transverse energy deposited by partons
| | σ LO [ab] | σ NLO [ab] |
|---------|-------------|--------------|
| aMC@NLO | 863 ± 1 | 1063 ± 2 |
| PowHel | 864 ± 1 | 1060 ± 7 |
Table 1: Cross sections at LO and NLO accuracy obtained with MADGRAPH5 aMC@NLO / and PowHel . We refer to the main text for the details of the computations.
⊥
⊥
Figure 1: Comparison of predictions at NLO to those of MADGRAPH5 aMC@NLO / [25] for transverse momentum of the top quark and the hardest photon. The lower panel shows the ratio of the predictions. For details on the selection cuts we refer to the text.
in this cone, this total transverse energy is limited,
$$E _ { \perp } = \sum _ { i \in \text{partons} } E _ { \perp, i } \Theta \left ( \delta - R ( p _ { \gamma }, \, p _ { i } ) \right ) \leq E _ { \perp, \gamma } \left ( \frac { 1 - \cos \delta } { 1 - \cos \delta _ { 0 } } \right ) \,, \quad \ \ ( 1 ) \\. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad.$$
where δ ≤ δ 0 . Our purpose with this comparison is only to show that the two codes lead to the same predictions.
The cross section corresponding to our selection cuts both at LO and NLO can be found in Table 1. For illustrative purposes four sample distributions are shown on Figs. 1 and 2. As it can be seen from comparing the numbers obtained for cross sections at LO and NLO and from the distributions the two calculations are in agreement with each other.
## 4 Technical issues
## 4.1 Generation cuts
In a recent work [11] we proposed a new method to generate sufficiently inclusive LHEs including photons, which can be used to make predictions for the hadroproduction of iso-
Figure 2: The same as Fig. 1 but for the top quark and hardest photon rapidities.
lated hard photons at the hadron level, formally correct at NLO accuracy in perturbative QCD. The essence of the method is to introduce generation isolation that is sufficiently loose so that the physical cross section with usual isolation parameters is independent of the parameters of the isolation needed for generating the events. For generation isolation we employ the smooth isolation of Frixione [22] with a small value for the parameter δ 0 . In addition to the isolation during event generation, we also need a generation cut on the transverse momentum of each photon, that we set to 5 GeV and varying it we checked that our physical predictions for hard photon production do not depend upon this cut.
## 4.2 Suppression factors
This process has two photons in the final state. As no photon-photon splitting is possible two technical cuts are sufficient to obtain a finite cross section at LO: lower limits are needed for the transverse momentum of the two photons. To enhance event generation in the physically relevant portion of phase space we use a suppression factor which could be used to suppress events in regions of phase space that are expected to be cut by physical selection cuts. Instead of this straightforward generalization of the suppression factor for the t ¯ t γ process [11] we suggest
$$\mathcal { F } _ { \text{supp} } = \frac { 1 } { 1 + \left ( \frac { p _ { \perp, \sup } ^ { 2 } } { p _ { \perp, \, \gamma _ { 1 } } ^ { 2 } } \right ) ^ { k } + \left ( \frac { p _ { \perp, \sup } ^ { 2 } } { p _ { \perp, \, \gamma _ { 2 } } ^ { 2 } } \right ) ^ { k } + \left ( \frac { m _ { \sup } ^ { 2 } } { m _ { \gamma \gamma } ^ { 2 } } \right ) ^ { k } } \,,$$
where p ⊥ , γ i is the transverse momentum for the i th photon, m γγ is the invariant mass of the two-photon system and k = 2 was chosen throughout. The suppression over the invariant mass of the two-photon system is introduced to allow for a much better population of the large and moderately large m γγ region. Our results were obtained with p ⊥ , supp = 100 GeV and m supp = 100 GeV.
In the POWHEG-Box the differential cross section correct up to NLO has the form of
$$\mathrm d \sigma ^ { N L O } = \mathrm d \Phi _ { B } B + \mathrm d \Phi _ { B } \mathcal { V } + \mathrm d \Phi _ { B } \int \mathrm d \Phi _ { \text{rad} } ( R - C ) + \mathrm d \Phi _ { B } \left ( G _ { \oplus } + G _ { \oplus } \right ) \,,$$
where B is the Born term, V is the regularized virtual contribution, R is the real-emission part, C is a short-hand for all the local subtraction terms regularizing the real-emission, G ⊕ and G glyph[circleminus] are remnants of collinear factorization, dΦ B is the underlying Born while dΦ rad is the one-particle phase space measure and the integral over momentum fractions is considered implicit. In the POWHEG-Box it is possible to decompose the real-emission part into two, disjoint contributions: the singular ( R s ) and remnant ( R r ) ones. Originally this bisection was introduced to deal with problems arising when the Born term becomes zero but not all the contributions [28]. This decomposition resulted in faster event generation and later on became standard [26]. With this decomposition the differential cross section takes the form of
$$d \sigma ^ { N L O } & = d \Phi _ { B } B + d \Phi _ { B } \mathcal { V } + d \Phi _ { B } \int d \Phi _ { \text{rad} } ( R ^ { s } - C ) + d \Phi _ { B } \left ( G _ { \oplus } + G _ { \ominus } \right ) + d \Phi _ { R } R ^ { r } = \\ & = d \Phi _ { B } \widetilde { B } + d \Phi _ { R } R ^ { r } \,,$$
by introducing ˜ B we can now speak about a ˜ B and a remnant contribution. To decide whether a real-emission contribution is the singular or the remnant one, the following criterion is used as default in the POWHEG-Box :
$$R ^ { s } = R - R ^ { r } \,, \quad R ^ { r } = R \, \Theta \left ( R - \xi \max ( C _ { \mathcal { C } }, C _ { \mathcal { S } } ) \right ) \,,$$
where C C is the sum of all the collinear while C S is the sum of all the soft subtraction terms and ξ is an arbitrary parameter set to 5 as default in the POWHEG-Box . When dealing with a process with photon(s) in the final state, hence having a generation isolation, large contributions can turn up when the separation between one of the photons in the final state and one final state massless quark becomes smaller than δ 0 . These contributions end up in the remnant provided by the large R/B ratio degrading the efficiency of remnanttype event generation. This is a consequence of the method of event generation which is based upon the hit-and-miss technique: an overall upper bound is determined which is used in a subsequent step to unweight on the phase space. Large contributions tend to push this upper bound very high hence large difference can build up between the upper bound and regular contributions characterizing the vast majority of phase space hampering the unweighting procedure.
When the underlying Born process is already singular close to the singularity the contributions are enhanced due to the apparent, unregularized singularity. These large contributions result in a large norm which forces the hit-and-miss based event generation mechanism to generate almost all the events close to the singular region. This problem can be successfully solved by introducing a suppression factor which, as its name suggests, suppresses these large contributions by multiplying them with a dynamics-dependent, though small number. In the generated event file the phase space coverage of the events is such that less events are generated in and near singular regions. Since the suppression
is a purely non-physical quantity the events can only bear physical meaning if the event weight incorporates the inverse of the suppression factor. Thus the suppression of the number of events in regions bearing no physical importance is compensated by scaling up the event weight accordingly.
The concept of the suppression factor can be used to improve the efficiency of remnanttype event generation. We modify the differential cross section by introducing a second suppression factor only affecting the remnant contribution:
$$d \sigma ^ { N L O } = d \Phi _ { B } \widetilde { B } + d \Phi _ { R } \mathcal { F } _ { \text{supp} } ^ { r } ( \Phi _ { R } ) R ^ { r } \,,$$
where F r supp (Φ R ) is the remnant suppression factor depending upon the real-emission phase space. To obtain physically meaningful events when a remnant event is being generated the event weight has to be multiplied by the inverse of the remnant suppression factor. Since large contributions to the remnant part of the cross section only come from those configurations where a final state (anti)quark is not well-separated from one of the photons the following form of remnant suppression was used during our calculations
$$\mathcal { F } _ { \text{supp} } ^ { r } = \prod _ { i = 1 } ^ { 2 } \frac { 1 } { 1 + \left ( \frac { p _ { \perp, \sup } ^ { r } } { p _ { \perp } ( \gamma _ { i }, q ) } \right ) ^ { k } } \,,$$
where p r ⊥ , supp is the remnant suppression parameter set to 20 GeV throughout our calculation, p ⊥ ( γ , q i ) is the relative p ⊥ of the i th photon and the final state (anti)quark and we use k = 4. The suppression is used only for those contributions where a massless (anti)quark is present in the final state. The presence and this form of remnant suppression significantly increased the efficiency of remnant event generation.
## 4.3 Choice of scale
The final state for t ¯ t γ γ production contains four particles at the Born level which leaves us with a rich set of kinematic variables usable for setting the renormalization and factorization scales for the process. In our previous work [13] for the hadroproduction of the t ¯ b ¯ t b final state, with massless b-quarks, we found that the half sum of the transverse masses of final state particles provides a good scale. It yields a moderate K-factor and small differences in the shapes of the distributions at LO and NLO accuracies. Thus we adopt the same choice for t ¯ t γ γ production for the central scale,
$$\mu _ { 0 } = \hat { H } _ { \perp } / 2 = \frac { 1 } { 2 } \left ( m _ { \perp, \mathfrak { t } } + m _ { \perp, \bar { \mathfrak { t } } } + p _ { \perp, \gamma _ { 1 } } + p _ { \perp, \gamma _ { 2 } } \right ) \,,$$
where the hat indicates that the underlying Born kinematics was used to construct the scale. We show the scale dependence of the cross section as the scale is varied around the central choice in Fig. 3. To make this prediction we used the following setup: 13 TeV LHC with CT10nlo as the chosen PDF set using a 2-loop α s , m t = 172 5 GeV and the . fine-structure constant kept fixed at α EM = 1 137. / The selection cuts coincide with the
Figure 3: LO and NLO cross sections as functions of the scale variation parameter. Details of the setup and cuts applied can be found in the text.
set used in Sect. 3 with the inclusion of a cut on the separation of the two hard photons measured in the rapidity azimuthal angle plane, ∆ R γ , γ ( 1 2 ) > 0 4. . The scales are varied by first making the renormalization and factorization scales coincide, µ R = µ F = µ , then µ is varied through µ = ξµ 0 , where ξ ∈ [1 / 16 8] and , µ 0 = ˆ H / ⊥ 2. Using this default scale, the NLO K-factor is K glyph[similarequal] 1 24. . If the scale is varied in the standard region of [ µ / , 0 2 2 µ 0 ] the scale uncertainty drops from +30 %-27 % to +14 %-13 % if NLO QCD corrections are included. It is interesting to note that choosing µ 0 = ˆ H / ⊥ 4 as default scale, the scale dependence remains the same, but the K-factor decreases to K glyph[similarequal] 1 08. .
## 4.4 The effect of the POWHEG Sudakov factor
When pre-showered events are generated, the POWHEG Sudakov factor may have an effect on measurable quantities through higher order terms in the perturbative expansion. Hence it is always informative to compare predictions at the NLO, that is at fixed order, and those predictions which are obtained from pre-showered events. In order to quantify the effect of the POWHEG Sudakov factor events were generated with the setup of Sect. 3 and the predictions drawn from these events were compared to those of the NLO calculation. Sample distributions can be found on Figs. 4-6. From these distributions, except
⊥
⊥
Figure 4: Comparison of the predictions at NLO accuracy to those obtained from preshowered events for a configuration listed in the text for the transverse momentum of the top quark and for the hardest photon. The lower panel depicts the ratio of the two predictions.
Figure 5: The same as Fig. 4 but for the rapidity of the top quark and the hardest photon.
for the transverse momentum of the extra parton, a good agreement can be seen between the fixed-order calculation and predictions drawn from pre-showered events. Taking a look at the prediction of the transverse momentum of the extra parton obtained using pre-showered events the usual Sudakov shape can be seen at smaller values while for large transverse momentum, log 10 ( p ⊥ / GeV) > 1 75, the LHE prediction merges into the fixed . order one. As we already pointed out in the case of t ¯ t γ production [11] the Sudakov shape gets distorted at around 0.75 and below due to the presence of the cut (Frixione isolation) acting on the real-emission part.
⊥
Figure 6: The same as Fig. 4 but for the invariant mass of the diphoton system and the transverse momentum of the extra parton.
| δ 0 | σ LHE [fb] | σ PS [fb] | σ SMC [fb] |
|--------|-----------------|-----------------|-----------------|
| 0 . 01 | 3 . 31 ± 0 . 03 | 3 . 03 ± 0 . 04 | 2 . 17 ± 0 . 02 |
| 0 . 05 | 3 . 30 ± 0 . 03 | 3 . 07 ± 0 . 05 | 2 . 13 ± 0 . 02 |
| 0 . 1 | 3 . 33 ± 0 . 03 | 3 . 05 ± 0 . 03 | 2 . 18 ± 0 . 03 |
Table 2: Cross sections obtained at different stages of event evolution for three different values of generation isolation using the cuts listed in the text.
## 5 Independence of the generation isolation
In our previous work [11] we carried out an extensive study to prove the independence of generation isolation at various stages of evolution of the pre-showered events. For the present process we have repeated all steps, but we restrict ourselves to present the case of full SMC. At other stages of event evolution the predictions taken with different generation isolations are in agreement with each other. The events were generated for the setup listed in Sect. 4.3. For concreteness the cross sections obtained at various levels for different generation isolations are listed in Table 2. These cross sections correspond to the following set of cuts:
- · Jets are reconstructed with the IR-safe antik ⊥ algorithm [29] with R = 0 4 and . p ⊥ ,j > 30 GeV using FastJet [30, 31].
- · Two hard photons, p ⊥ , γ > 30 GeV, are requested in the central region, | y γ | < 2 5. .
- · The photons should be isolated from the jets and each other such that ∆ R > 0 4. . The separation is measured in the rapidity-azimuthal angle plane.
- · The hadronic activity is limited around both photons in a cone of R γ = 0 4: . E max ⊥ , had = 3 GeV.
0
13 TeV,
50
100
µ
= ˆ
H /
150
⊥
2
200
250
SMC,
SMC,
SMC,
PY
PY
PY
300
350
δ
δ
0
0
= 0 01
.
= 0 05
.
δ
0
= 0 1
.
400
450
500
⊥
p
⊥
,γ
2
[GeV]
Figure 7: Comparison among predictions at the hadron level taken with three different generation isolation, δ 0 = 0 1 (green dashed), 0.05 (blue dotted) and 0.01 (red solid), for . the transverse momenta of the two hardest photons. The lower panels show the ratios with respect to the case of δ 0 = 0 01. .
The hadronic activity in a cone of R γ around the photon momentum, p γ , is measured through the total hadronic transverse energy deposited in this cone:
$$E _ { \perp, \text{had} } = \sum _ { i \in \text{tracks} } E _ { \perp, i } \Theta \left ( R _ { \gamma } - R ( p _ { \gamma }, p _ { i } ) \right ) < E _ { \perp, \text{had} } ^ { \max },$$
where the summation runs over all the hadronic tracks. To make our predictions at the hadron level we used PYTHIA-6.4.25 . At the hadron level we allowed tops to decay, but kept QED shower and multiparticle interactions inactivated. To see the independence of generation isolation we generated events with three different smooth isolation parameters ( δ 0 ∈ { 0 1 . , 0 05 . , 0 01 . } ) and compared these pairwise. Sample distributions at the hadron level are depicted on Figs. 7-9. Among these distributions t-quark related observables can be found too. These were obtained by reconstructing the t-quark using MCTRUTH . From these distributions we can see that the deviation of predictions from each other is only due to statistical fluctuations. All of the event sets taken with the three different generation isolations result in the same predictions at the hadron level which means that any of these can be safely used in analyses to provide theoretical predictions.
## 6 Phenomenology
The main consequence of the previous two sections is the ability to make physical predictions with experimental isolation criteria using pre-showered events prepared with a generation isolation chosen suitably small. The t ¯ t γ γ production plays an important role in Higgs physics when attention is turned to the properties of the Higgs boson. One
[fb/GeV]
2
,γ
10
10
10
⊥
d
10
d
-
-
-
-
1
2
3
4
σ/ p
10
-
5
ratio
1.0
ratio
1.1
0.9
1.1
0.9
1.0
,
,
,
⊥
Figure 8: The same as Fig. 7 but for the transverse momentum of the top quark and the two-photon invariant mass.
Figure 9: The same as Fig. 7 but for the rapidity of the hardest photon and the top quark.
such important property is the Yukawa coupling of the Higgs boson to various massive fermions. The Yukawa coupling of the Higgs boson to the top quark can be measured in t ¯ t H production. When the Higgs boson decays in the diphoton channel the irreducible background is t ¯ t γ γ production. As we provided predictions for t ¯ t H production in Ref. [7] at the hadron level, we decided to present predictions for both the signal and the background at the hadron level with the highest available precision. For the t ¯ t γ γ background we take the events generated with δ 0 = 0 05. For the signal we generated a new bunch of . pre-showered events sharing the same parameters with the t ¯ t γ γ sample, generated for the 13 TeV LHC with CT10nlo PDF and accordingly chosen 2-loop α s , m t = 172 5 GeV, .
m H = 125 GeV. We chose the renormalization and factorization scales equal to each other and set to µ R = µ F = m t + m / H 2. To make our predictions, we used PYTHIA-6.4.25 for simulating the evolution of the events to the hadron level, but with multiple interactions turned off.
The partonic final state contains a t-quark pair for both the signal and the background. There are two ways to detect t-quarks: either through their decay products, by looking for displaced secondary vertices as tagging for decaying b-quarks, or using t-tagging. We decided to use the latter as provided by the HEPTopTagger [32, 33]. In order to perform t-tagging we followed the following steps:
- · Jet reconstruction using all the hadronic tracks with the C/A algorithm with R = 1 5 . using FastJet [30, 31].
- · Only those jets were considered for which p ⊥ > 200 GeV and | y | < 5.
- · The t-quark candidate subjets were looked for in the jet mass range of 150 - 200 GeV.
- · We selected that particular subjet as a t-quark jet for which the jet mass was the closest to m t .
In respect to other parameters concerning the top tagging we kept the default values provided by the HEPTopTagger . The cuts applied to the hadronic events were the following:
- · Two hard photons were requested in the central region with p ⊥ , γ > 30 GeV and | y γ | < 2 5. .
- · The photons should be isolated from each other: ∆ R γ , γ ( 1 2 ) > 0 4. .
- · To isolate the photons from hadronic activity a jet clustering was performed on all the hadronic tracks with the antik ⊥ algorithm [29] as implemented in FastJet [30, 31] with R = 0 4 and . p ⊥ ,j > 30 GeV. The photons were requested to be isolated from these jets by ∆ R γ, j ( ) > 0 4 measured on the rapidity-azimuthal angle plane. .
- · Both of the hard photons had to be isolated from the top jet obtained by top tagging and from the three subjets of the top jet by ∆ R γ, j ( ) > 0 4 measured on . the rapidity-azimuthal angle plane.
- · One hard lepton was requested in the central region with p ⊥ , glyph[lscript] > 30 GeV and | y glyph[lscript] | < 2 5. . To select leptons we did not make any distinction between leptons and antileptons.
- · The lepton had to be isolated from both the jets and the photons with ∆ R glyph[lscript], ( i ) > 0 4 . , i ∈ { γ, j } measured on the rapidity-azimuthal angle plane.
- · Around both photons a minimal hadronic activity was allowed in a cone of R = 0 4 . such that E max ⊥ , had = 3 GeV.
Figure 10: The invariant mass and rapidity azimuthal separation is shown for the diphoton system at the hadron level for the signal (t ¯ t H , red) and for the background (t ¯ t γ γ , blue) such that the signal sits on the top of the background. Further details of the calculation can be found in the main text.
⊥
⊥
Figure 11: The same as Fig. 10 but for the transverse momentum of the two hard photons.
In order to reach reasonable statistics even for the signal we turned off all decay channels of the Higgs boson but the diphoton one and rescaled the event weights with the H → γ γ branching ratio. The most interesting distribution is the invariant mass of the two-photon system along with this a couple more distributions are depicted in Figs. 10 and 11. By taking a look at the distributions it is clear that the cross section drops into the attobarn range. Due to the narrow width of the Higgs boson the signal appears as a single, welldefined spike in the m γγ spectrum with excellent signal/background ratio. For all the other distributions the background overwhelms the signal.
## 7 Conclusions
We presented the first predictions for the hadroproduction of t ¯ t γ γ final state both at NLO accuracy and at NLO matched with parton shower. The predictions at NLO were
computed by two programs, PowHel and MADGRAPH5 and complete agreement was found. While the predicitions at NLO accuracy can be made only using smooth isolation prescription of Frixione for the hard photons, the matched predictions are based on LHEs that can be exposed to the usual cone-type isolation of experimenters. This is achieved by generating the events using generation cuts that are sufficiently loose, so that the predictions do not depend on those when usual physical isolations are used.
We found that using half of the sum of the transverse masses of particles in the final state the NLO corrections are about 25 %, and decrease the dependence on the renormalization and factorization scales significantly. The generated events are stored and can be downloaded from http://www.grid.kfki.hu/twiki/bin/view/DbTheory/TtaaProd. We generated similar events for the t ¯ t H final state, to which the t ¯ t γ γ process is an irreducible background in the H → γγ decay channel. We compared some kinematic distributions of both signal and background process and found that the invariant mass of the two hard photons shows a clearly visible narrow peak, but with small cross section, in the attobarn range for the LHC at 13 TeV. For other kinematic distributions the signal and background have rather similar shapes.
Acknowledgements This research was supported by the Hungarian Scientific Research Fund grant K-101482, the SNF-SCOPES-JRP-2014 grant 'Preparation for and exploitation of the CMS data taking at the next LHC run', the European Union and the European Social Fund through LHCPhenoNet network PITN-GA-2010-264564, and the Supercomputer, the national virtual lab TAMOP-4.2.2.C-11/1/KONV-2012-0010 project.
## References
- [1] ATLAS Collaboration Collaboration, G. Aad et. al. , Observation of a new particle in the search for the Standard Model Higgs boson with the ATLAS detector at the LHC , Phys.Lett. B716 (2012) 1-29, [ arXiv:1207.7214 ].
- [2] CMS Collaboration Collaboration, S. Chatrchyan et. al. , Observation of a new boson at a mass of 125 GeV with the CMS experiment at the LHC , Phys.Lett. B716 (2012) 30-61, [ arXiv:1207.7235 ].
- [3] ATLAS Collaboration Collaboration, G. Aad et. al. , Evidence for the spin-0 nature of the Higgs boson using ATLAS data , Phys.Lett. B726 (2013) 120-144, [ arXiv:1307.1432 ].
- [4] CMS Collaboration Collaboration, S. Chatrchyan et. al. , Measurement of Higgs boson production and properties in the WW decay channel with leptonic final states , JHEP 1401 (2014) 096, [ arXiv:1312.1129 ].
- [5] ATLAS Collaboration Collaboration, G. Aad et. al. , Measurements of Higgs boson production and couplings in diboson final states with the ATLAS detector at the LHC , Phys.Lett. B726 (2013) 88-119, [ arXiv:1307.1427 ].
| [6] | CMS Collaboration Collaboration, S. Chatrchyan et. al. , Measurement of the properties of a Higgs boson in the four-lepton final state , Phys.Rev. D89 (2014) 092007, [ arXiv:1312.5353 ]. |
|-------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [7] | M. Garzelli, A. Kardos, C. Papadopoulos, and Z. Tr´cs´nyi, o a Standard Model Higgs boson production in association with a top anti-top pair at NLO with parton showering , Europhys.Lett. 96 (2011) 11001, [ arXiv:1108.0387 ]. |
| [8] | A. Kardos, Z. Tr´cs´nyi, o a and C. Papadopoulos, Top quark pair production in association with a Z-boson at NLO accuracy , Phys.Rev. D85 (2012) 054015, [ arXiv:1111.0610 ]. |
| [9] | M. Garzelli, A. Kardos, C. Papadopoulos, and Z. Tr´cs´nyi, o a Z0 - boson production in association with a top anti-top pair at NLO accuracy with parton shower effects Phys.Rev. D85 (2012) 074022, [ arXiv:1111.1444 ]. |
| [10] | M. Garzelli, A. Kardos, C. Papadopoulos, and Z. Tr´cs´nyi, o a t ¯ t W + - and t ¯ t Z Hadroproduction at NLO accuracy in QCD with Parton Shower and Hadronization effects , JHEP 1211 (2012) 056, [ arXiv:1208.2665 ]. |
| [11] | A. Kardos and Z. Tr´cs´nyi, o a Hadroproduction of t anti-t pair in association with an isolated photon at NLO accuracy matched with parton shower , arXiv:1406.2324 . |
| [12] | A. Kardos, C. Papadopoulos, and Z. Tr´cs´nyi, o a Top quark pair production in association with a jet with NLO parton showering , Phys.Lett. B705 (2011) 76-81, [ arXiv:1101.2672 ]. |
| [13] | A. Kardos and Z. Trcsnyi, Hadroproduction of t anti-t pair with a b anti-b pair using PowHel , J.Phys. G41 (2014) 075005, [ arXiv:1303.6291 ]. |
| [14] | S. Catani, M. Fontannaz, and E. Pilon, Factorization and soft gluon divergences in isolated photon cross-sections , Phys.Rev. D58 (1998) 094025, [ hep-ph/9803475 ]. |
| [15] | P. Nason, A New method for combining NLO QCD with shower Monte Carlo algorithms , JHEP 0411 (2004) 040, [ hep-ph/0409146 ]. |
| [16] | S. Frixione, P. Nason, and C. Oleari, Matching NLO QCD computations with Parton Shower simulations: the POWHEG method , JHEP 0711 (2007) 070, [ arXiv:0709.2092 ]. |
| [17] | S. Alioli, P. Nason, C. Oleari, and E. Re, A general framework for implementing NLO calculations in shower Monte Carlo programs: the POWHEG BOX , JHEP 1006 (2010) 043, [ arXiv:1002.2581 ]. |
| [18] | M. Garzelli, A. Kardos, and Z. Tr´cs´nyi, o a NLO Event Samples for the LHC , PoS EPS-HEP2011 (2011) 282, [ arXiv:1111.1446 ]. |
| [19] | G. Bevilacqua, M. Czakon, M. Garzelli, A. van Hameren, A. Kardos, et. al. , HELAC-NLO , Comput.Phys.Commun. 184 (2013) 986-997, [ arXiv:1110.1499 ]. |
| [20] | J. Alwall, A. Ballestrero, P. Bartalini, S. Belov, E. Boos, et. al. , A Standard format for Les Houches event files , Comput.Phys.Commun. 176 (2007) 300-304, [ hep-ph/0609017 ]. |
|--------|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [21] | S. Alioli, P. Nason, C. Oleari, and E. Re, Vector boson plus one jet production in POWHEG , JHEP 1101 (2011) 095, [ arXiv:1009.5594 ]. |
| [22] | S. Frixione, Isolated photons in perturbative QCD , Phys.Lett. B429 (1998) 369-374, [ hep-ph/9801442 ]. |
| [23] | Z. Kunszt, Z. Tr´cs´nyi, o a and W. J. Stirling, Clear signal of intermediate mass Higgs boson production at LHC and SSC , Phys.Lett. B271 (1991) 247-255. |
| [24] | A. Ballestrero and E. Maina, t anti-b H production for an intermediate mass Higgs Phys.Lett. B299 (1993) 312-314. |
| [25] | J. Alwall, R. Frederix, S. Frixione, V. Hirschi, F. Maltoni, et. al. , The automated computation of tree-level and next-to-leading order differential cross sections, and their matching to parton shower simulations , JHEP 1407 (2014) 079, [ arXiv:1405.0301 ]. |
| [26] | J. M. Campbell, R. K. Ellis, R. Frederix, P. Nason, C. Oleari, et. al. , NLO Higgs Boson Production Plus One and Two Jets Using the POWHEG BOX, MadGraph4 and MCFM , JHEP 1207 (2012) 092, [ arXiv:1202.5475 ]. |
| [27] | A. Kardos, P. Nason, and C. Oleari, Three-jet production in POWHEG , JHEP 1404 (2014) 043, [ arXiv:1402.4001 ]. |
| [28] | S. Alioli, P. Nason, C. Oleari, and E. Re, NLO vector-boson production matched with shower in POWHEG , JHEP 0807 (2008) 060, [ arXiv:0805.4802 ]. |
| [29] | M. Cacciari, G. P. Salam, and G. Soyez, The Anti-k(t) jet clustering algorithm , JHEP 0804 (2008) 063, [ arXiv:0802.1189 ]. |
| [30] | M. Cacciari, G. P. Salam, and G. Soyez, FastJet User Manual , Eur.Phys.J. C72 (2012) 1896, [ arXiv:1111.6097 ]. |
| [31] | M. Cacciari and G. P. Salam, Dispelling the N 3 myth for the k t jet-finder , Phys.Lett. B641 (2006) 57-61, [ hep-ph/0512210 ]. |
| [32] | T. Plehn, M. Spannowsky, M. Takeuchi, and D. Zerwas, Stop Reconstruction with Tagged Tops , JHEP 1010 (2010) 078, [ arXiv:1006.2833 ]. |
| [33] | T. Plehn, M. Spannowsky, and M. Takeuchi, How to Improve Top Tagging , Phys.Rev. D85 (2012) 034029, [ arXiv:1111.5034 ]. | | 10.1016/j.nuclphysb.2015.05.032 | [
"Adam Kardos",
"Zoltán Trócsányi"
] | 2014-08-01T19:59:57+00:00 | 2015-05-18T21:26:32+00:00 | [
"hep-ph"
] | Hadroproduction of t anti-t pair with two isolated photons with PowHel | We simulate the hadroproduction of a t anti-t pair in association with two
isolated hard photons at 13 TeV LHC using the PowHel package. We use the
generated events, stored according to the Les-Houches event format, to make
predictions for differential distributions formally at the next-to-leading
order (NLO) accuracy. We present predictions at the hadron level employing the
cone-type isolation of the photons used by experiments. We also compare the
kinematic distributions to the same distributions obtained in the t anti-t H
final state when the Higgs-boson decays into a photon pair, to which the
process discussed here is an irreducible background. |
1408.0279v1 | ## Meteorological data for the astronomical site at Dome A, Antarctica
Yi Hu , Zhaohui Shang 1 2 1 , , Michael C. B. Ashley , Colin S. Bonner , 3 3
Keliang Hu , Qiang Liu , Yuansheng Li , Bin Ma , Lifan Wang 1 1 4 1 5 6 , , and Haikun Wen 7
## [email protected]
Received
;
accepted
1 National Astronomical Observatories, Chinese Academy of Sciences, Beijing 100012, P.
## R. China
2 Tianjin Astrophysics Center, Tianjin Normal University, Tianjin 300387, P. R. China
3 School of Physics, University of New South Wales, NSW 2052, Australia
4 Polar Research Institute of China, Pudong, Shanghai 200136, P. R. China
5 Purple Mountain Observatory, Nanjing 210008, P. R. China
6 Physics Department, Texas A&M University, College Station, TX 77843
7 Nanjing Institute of Astronomical Optics and Technology, Nanjing 210042, P. R. China
## ABSTRACT
We present an analysis of the meteorological data collected at Dome A, Antarctica by the Kunlun Automated Weather Station, including temperatures and wind speeds at eight elevations above the snow surface between 0 m and 14.5 m. The average temperatures at 2 m and 14.5 m are -54 C and ◦ -46 C, re-◦ spectively. We find that a strong temperature inversion existed at all heights for more than 70% of the time, and the temperature inversion typically lasts longer than 25 hours, indicating an extremely stable atmosphere. The temperature gradient is larger at lower elevations than higher elevations. The average wind speed was 1.5 m / s at 4 m elevation. We find that the temperature inversion is stronger when the wind speed is lower and the temperature gradient decreases sharply at a specific wind speed for each elevation. The strong temperature inversion and low wind speed results in a shallow and stable boundary layer with weak atmospheric turbulence above it, suggesting that Dome A should be an excellent site for astronomical observations. All the data from the weather station are available for download.
Subject headings: Site testing - Atmospheric effects - Methods: data analysis
## 1. Introduction
Dome A-the highest location on the Antarctic Plateau at an elevation of 4,089m-has long been considered to be an excellent site for astronomical observations over a broad range of the electromagnetic spectrum (e.g., Gillingham 1991, Marks 2002, Burton 2010, and references therein). Dome A's advantages include extremely low temperatures, an extremely dry atmosphere (Sims et al. 2012), low wind speed, and long periods of continuous cloud-free nighttime.
Other sites in Antarctica, such as the South Pole itself, Dome C, and Dome Fuji, are also promising, and have been extensively studied over many years with a variety of instruments. Astronomers are particularly interested in the layer within tens or hundreds of meters of the surface over the Antarctic plateau since a number of studies (e.g., Marks et al. 1999, Lawrence et al. 2004) have shown that the transition to the free atmosphere occurs within this height range, as opposed to 1-2 km from typical observatory sites at temperate latitudes.
At the South Pole, Marks et al. (1996) used microthermal sensors on a 27 m mast to measure the turbulence in the lowest part of the boundary layer at South Pole. Marks et al. (1999) extended the altitude coverage by using balloon-borne microthermals to estimate a free-atmosphere seeing component of 0.37 ′′ above a boundary layer with an average thickness of ∼ 220 m. Travouillon et al. (2003) confirmed these results using a sonic radar able to reach heights of 890 m above the surface, and derived a ground-based average seeing of 1.73 ′′ , dropping to 0.37 ′′ above 300 m.
At Dome C, Lawrence et al. (2004) used a MASS-DIMM instrument and a SODAR to measure a median free atmospheric seeing of 0.23 ′′ above a boundary layer that was less than 30 m high. Aristidi et al. (2005) analyzed two decades of temperature and wind speed data from an automated weather station, and four summers of measurements with
balloon-borne sondes. Trinquet et al. (2008) used balloon-borne measurements to find that the median value of the boundary layer height at Dome C is 33 m.
At Dome Fuji, astronomical seeing at times below 0.2 ′′ has been reported from a DIMM instrument 11m above the snow surface (Okita et al. 2013).
Dome A was first visited by a Chinese Antarctica Research Expedition (CHINARE) team in 2005, so opportunities for studying the site conditions have not existed for as long as for the other sites in Antarctica. Following the repeated visits in recent years by CHINARE teams, and the construction of the Kunlun Station at Dome A in 2009, several site-testing experiments have been installed (e.g., Yang et al. 2009) and encouraging results have been reported (e.g., Yang et al. 2010). For example, using data from a sonic radar, Bonner et al. (2010) found that the median boundary layer thickness at Dome A between February and August in 2009 was only 13.9 m, considerably lower than the median value of 33 m at Dome C (Trinquet et al. 2008) and 200-300 m at South Pole (Travouillon et al. 2003).
Missing up until now has been a detailed analysis of the meteorological conditions at Dome A from mast-based instruments. Such measurements can give direct information on the atmospheric turbulence and wind speed in the boundary layer, and are relevant to the operating environment of future telescopes, and the performance of adaptive optics systems (Aristidi et al. 2005). In this article, we present the first results from analyzing the weather data collected from the Kunlun Automated Weather Station (KL-AWS) at Dome A. Data acquisition and reduction are described in § 2. A statistical analysis of temperature and wind speed data are presented in § 3. Finally, a discussion is given in § 4.
## 2. Data acquisition and reduction
KL-AWS was installed in 2011 January by the 27th CHINARE team (see Fig. 1) at Dome A at the location 80 ◦ 25 ′ 03 ′′ S, 77 ◦ 05 ′ 32 ′′ E. KL-AWS consists of a 15 m mast equipped with nine Young 4-wire 41342 temperature sensors at heights of -1 m, 0 m, 1 m, 2 m, 4 m, 6 m, 9.5 m, 12 m and 14.5 m, four Young Wind Monitor-AQ model 05305V propeller anemometers at heights of 2 m, 4 m, 9 m and 14.5 m to measure both wind speed and direction, and one Young 61302V barometer at 1 m. The manufacturer's specified absolute measurement accuracy of the instruments is: temperature 0 32 C, wind speed . ◦ 0.2 m / s, and wind direction 0 5 . ◦ plus installation offset. Originally, the mast was planned to have four Young 85000 2D sonic anemometers, but an electrical fault prevented their operation. The anemometer at 2 m did not acquire data for unknown reasons. KL-AWS was powered by PLATO, an automated observatory (Lawrence et al. 2008), and it was connected to the supervisor computers in PLATO's instrument module. KL-AWS was placed 30 m south of PLATO to minimize any influence from nearby structures. KL-AWS acquired data from all sensors once every 36 seconds and stored the data on the hard disks of the supervisor computer. The data were sent back in real-time every 20 minutes through the Iridium satellite network in order to assist with monitoring and operating the other experiments on site.
KL-AWS began operation on 2011 January 17 and ran continuously for a year apart from a 20 day gap beginning on 2011 August 3 due to a computer script inadvertently stopping. The sampling rate was 1 / 36Hz up until 2011 August 3, and 1 / 60Hz from 2011 August 23. Since then, all the temperature sensors and the barometer were still working but the anemometer at 14.5 m was never recovered, and other two anemometers at 4 m and 9 m were recovered on October 13. On 2012 January 18, KL-AWS ceased operation, possibly as a result of damage to the power and data cables connecting it to PLATO.
Fig. 2 shows all the temperature data collected by KL-AWS.
When the CHINARE expedition revisited Dome A at the beginning of 2012 it was found that the top section of the weather tower had bent over and almost touched the ground. From looking at the data we determined that the tower likely failed on 2011 September 16 (see Fig. 3). Before that date there were often large temperature differences between the sensors, but after September 16 the higher sensors showed no differences. Unless otherwise stated, this paper restricts itself to the continuous period of data up until 2011 August 3.
To reduce the short-term fluctuations in the data, we have smoothed the temperature data by a boxcar of 10 adjacent data points. Fig. 4 shows an example of the results where the short-term fluctuations are removed while the temperature trends remains. All the following analyses are based on the smoothed temperature data.
## 3. Results
## 3.1. Temperature and Temperature Difference
Using data over the entire year, we show the temperature distributions at snow surface level and at -1 m (1 meter below surface) in Fig. 5. As expected, the snow temperature at -1 m has a much smaller range than the air temperature. Due to snow accumulation, we believe the sensor at the snow surface level was buried under snow later in the year and could not provide accurate air temperature measurements. This can be seen clearly in Fig. 3 when the surface temperature line (blue) becomes very smooth in the lower panel. We note the temperature distribution of -1 m shows multiple spikes in the right panel of Fig. 5 and they are real. By investigating the data, we find that the temperature at -1 m varied little around some values for considerable long periods, making such isolated spikes.
The temperature distributions at the other heights are shown in Fig. 6 with all data up to 2011 August 3. They all show a single peak distribution.
We plot the daily median temperature in Fig. 7 to show the long-term trend which follows the seasonal pattern. The daily median temperature at 2 m was ∼ -35 C in ◦ January, went down rapidly to ∼ -60 C in April, and was as low as ◦ ∼ -70 C during ◦ the dark winter before it slowly rose again. Compared with that of South Pole, we find that there is no significant difference in monthly median temperature between South Pole and Dome A (see Fig. 2 in Hudson and Brandt 2005). The trend of the daily median temperature is also similar to that at the South Pole.
Because of the radiatively-cooled snow surface, it is not surprising that there is strong temperature inversion just above the snow surface of Dome A. Fig 8 shows a good example of the temperature inversion at different heights. We define the temperature differences at 1 m, 2 m, 4 m, 6 m, 9.5 m, 12.5 m and 14.5 m as the temperature at the corresponding height minus the temperature at the adjacent lower height. For example, ∆ (2m) = T T (2m) -T (1m). We then calculate the temperature gradients (in ◦ C/m) at different heights and their distributions are are shown in Fig. 6. The distributions are relatively narrow except for the one involving ∆ T (1 m ), since the surface temperature that did not always measure the air temperature as explained in § 3.1. For all other heights, the distributions increase dramatically from negative values to a positive peak and go down relatively slowly with a tail. The temperature gradients are smaller at higher elevations. The existence of the positive temperature gradients indicates strong temperature inversion at all layers.
To compare the temperatures at two levels spanning more than 10 m in height, Hudson and Brandt (2005) found at South Pole that for 99.9% of the time there was a positive temperature difference between 13 m and 2 m. We found the same for our data at Dome
A between 12 m and 2 m. However, the median temperature difference at Dome A was 6 9 C, which is much larger than the difference of 1 1 C at South Pole. . ◦ . ◦ This shows that the strength of the temperature inversion at Dome A is much greater than that at South Pole.
We plot the temperature gradients at all levels between 2011 January and August in Fig. 9. This figure shows that most of the large negative temperature gradients happened during January and February while most of the large positive gradients happened in the wintertime. Higher layers tend to have smaller temperature gradients than lower layers as also seen in Fig. 6.
We also investigated whether the temperature inversion depends on the temperature or season. Figs. 10 and 11 show the temperature difference and temperature for different months during 2011. It is clearly seen that most of the largest positive temperature differences (e.g. > 5 C) occurred when the temperature was between ◦ -45 C and ◦ -65 C. ◦ The strong correlations between ∆ T and T at 1 m for later months are not real and are again caused by the snow buried sensor at 0 m measuring the snow temperature, which did not vary as much as the air temperature. The other plausible correlations for 2 m and 4 m after March could be real, suggesting that at lower layers a larger temperature inversion occurred at higher temperatures. There is a strong trend that after March, at 4 m, 6 m, and 9 m, the temperature differences are almost all positive, indicating long durations of temperature inversion.
Although the daily temperature obviously varies with the elevation of the Sun in the summer season as seen in the top panel of Fig. 3, we found that the temperature difference in a day has little relationship with the Sun. In Fig. 12 we fold the temperature difference at 6 m into one day for February, March and May, and also plot the temperature difference against the solar elevation. In February of 2011, the temperature difference at 6 m seems to be larger at night and seems to be related to the Sun, but in March when the Sun still rises
and sets at Dome A, we do not find a preferable time for larger temperature differences and there is no evidence of correlation between ∆ T and solar elevation.
As an indicator of the stability of the atmosphere, we are mostly concerned with how long a temperature inversion lasts. We show a clear example of two days' data in mid May, 2011 in Fig. 8 where a temperature inversion exists at all heights. We evaluated the temperature inversion (i.e., temperature difference) at each height with respect to the temperature sensor immediately below. To evaluate the duration of a temperature inversion at all heights, we need to define a real temperature difference quantitatively. While the temperature sensors have a specified absolute accuracy of 0 32 C, our data show that the . ◦ variation in consecutive measurements is much smaller than this (see Fig. 4). We have subtracted the smoothed data from the original data and calculated the standard deviation as σ = 0 033 C. Therefore, we set a 3 . ◦ σ temperature difference threshold of 0 14 C, above . ◦ which we regard it as a real temperature inversion rather than random measurement error. We choose all the temperature difference data points larger than 0 14 C, and counted how . ◦ long the inversion lasted. The results are shown in Fig. 13 and Table 1. A temperature inversion existed more than 70% of the time at all heights, and up to 95% at a height of 6 m. We note that when a temperature inversion occurred, it lasted at least 25 hours for more than ∼ 40% of the time at all heights and up to 78% at 6 m. For less than 10% of the time, the temperature inversion lasted under 1 hour.
We also note that the median duration of the temperature inversion increased from 2 m to 6 m and then decreased with height. There are two possible reasons for this trend. One is that the air turbulence is stronger at the lower elevation, the other is that sudden strong wind can destroy the temperature inversion at higher elevations (see the next subsection for wind speed statistics).
## 3.2. Wind speed and direction
The distributions and cumulative statistics for the wind speed at three heights are shown in Fig. 14. The peak at 0 m / s, accounting for about 10% of the time, simply means there was no wind, or the wind speed was too low to be detected by the sensors, or the sensors got stuck (see below). At 4 m, the wind speed was lower than 1.5 m / s for half of the time, and seldom reached 4 m / s. The wind speed increases with height above the snow. At 9 m and 14.5 m, the median wind speeds are 2.0 m / s and 2.4 m / s, respectively, and could often reach 4 m / s. This can also be clearly seen in the wind rose plots (Fig. 15). We notice that there is no dominant wind direction for all three heights. The obvious gaps in the wind roses are caused by the mast of KL-AWS which blocked wind in that direction.
We have shown the wind speed and direction between January and August in Fig. 16 and Fig. 17, and noticed that when there is virtually no wind, it was mostly during the polar night when the Sun never rose. However, for some extended periods during April to June, the wind speeds were flat in the plots and were 0 m / s or very low. This makes us to worry that it is possible that the propellers of the anemometers got stuck sometimes at low temperatures. Fig. 17 shows again that the wind directions are random and they are synchronized very well at the three heights.
We have compared the average wind speeds at Dome A, Dome C, and the South Pole in Table 2 and find that the horizontal wind speed at Dome A at 4 m was only 1.5 m / s, much smaller than the ground wind speed of 2.9 m / s at Dome C and 5.5 m / s at the South Pole.
In order to investigate whether the temperature gradient is related to wind speed, we plot the wind speeds at 4 m, 9 m and 14.5 m against temperature gradients at 4 m, 9.5 m and 14.5 m respectively in Fig. 18. Stronger temperature inversion always occurred when the wind speed was lower. When the wind speed is higher than 2.5 m / s at 4 m, the temperature
gradient was never larger than 0 7 C/m. . ◦ We also see the same trends at 9 m and 14.5 m, but with different specific wind speeds of 4.1 m / s and 5.3 m / s, respectively. It therefore seems that at the higher elevations, the positive temperature gradients are smaller, but more resistant to wind.
## 3.3. Air pressure
The barometer at the height of 1 m worked correctly for the entire year. Fig. 19 and 20 shows the data and the distribution. The average air pressure at Dome A was 586 hPa with a standard deviation of 8 hPa. We do not find any clear relationships between the air pressure and the temperature difference or wind speed at 4 m (Fig. 21).
## 4. Conclusion and discussion
We have analyzed the weather data collected from KL-AWS at Dome A during 2011. The average temperature between January and July was -54 C at 2 m and ◦ -46 C at 14.5 m. ◦ We have found that there is usually a strong temperature inversion at all the heights above the ground. The temperature inversion, defined here as a positive temperature difference larger than 0 14 C, existed more than 70% of the time at all heights and up to 95% at 6 m. . ◦ Such inversions lasted longer than 25 hours for more than about 40% of the time at all heights, and almost 80% of the time at 6 m. Most largest negative temperature differences (no inversion) occurred during January and February and most strong temperature inversion occurred when the temperature fell between -65 C and ◦ -45 C. ◦
The average wind speed at 4 m is 1.5 m / s, which may be the lowest reported from any site on earth. The wind speed at higher elevation is slightly stronger. We also notice that for ∼ 10% of the time there is almost no wind or the wind speed was too low to be
detected by the sensors, this usually happens during the dark polar night. However, there also seems to be evidence that the wind speed sensors might have been stuck sometimes at low temperatures. The wind on the Antarctic continent is dominated by katabatic winds, therefore the wind speed should be very small and the wind direction should be random above the highest point (Dome A). Hudson and Brandt (2005) found that the strongest temperature inversion often happened with a 3-5 m / s wind at the South Pole, but we find that strong temperature inversion happens when the wind speed is low, such as less than 2.5 m / s for the height of 4 m; when the wind speed is higher than 2.5 m / s the temperature gradient decreases sharply.
It has been known for a long time that the surface of the Antarctic plateau is prone to temperature inversions due to radiative cooling of ice, and hence a lower air temperature near the surface. As Basu and Port´ e-Agel (2006) pointed out, whenever a temperature inversion exists, turbulence is generated by shear and destroyed by negative buoyancy and viscosity. Competition between shear and buoyancy weakens turbulence and results in a shallower boundary layer. We list the monthly median boundary layer heights of Dome A in 2009 (Bonner et al. 2010) and the monthly median temperature difference at 4 m during 2011 in Table 3 and compared them in Fig. 22. If average conditions were similar between 2009 and 2011, this indicates, as expected, that stronger temperature inversions favor a thinner boundary layer, with the lowest in May and June.
The strong temperature inversion and low wind speed at Dome A both produce a very stable atmosphere and a stable boundary layer with little convection, especially in winter. When the temperature inversion happens as low as 2 m, we expect the boundary layer can be much lower too. Above the shallow boundary layer, we can reach the free atmosphere where the seeing will be good for astronomical observations. In other words, there will be periods when even telescopes that are very close to the ground will experience
free-atmosphere seeing approaching 0.25 ′′ .
We have demonstrated that the weather data collected by KL-AWS at different heights are very useful to characterize the stability of the near-ground atmosphere at Dome A and can provide insights for evaluating the site for astronomical observations. We plan to continue the experiment with a new weather tower and monitor the site for at least 2-3 years so that we can have better statistics and understanding of the site.
The data from KL-AWS are available in raw form at http://aag.bao.ac.cn/weather/ downloads/ and as plots at http://aag.bao.ac.cn/weather/plot\_en.php
The authors wish to thank all members of the 27th and 28th Chinese Antarctic Research Expedition teams for their effort in setting up the KL-AWS instruments and servicing the PLATO observatory. This research is supported by Chinese Polar Environment Comprehensive Investigation & Assessment Programmes under grant No. CHINARE201202-03, National Basic Research Program of China (973 Program 2013CB834900), and the National Natural Science Foundation of China under grant No. 11273019 and 11203039. This project is also supported by the Commonwealth of Australia under the Australia-China Science and Research Fund, and by the Australian Research Council and the Australian Antarctic Division.
## REFERENCES
Aristidi, E., Agabi, K., Fossat, E., Vernin, J., Travouillon, T., Lawrence, J.S., Meyer, C., Storey, J. W. V., Halter, B., Roth, W. L., & Walden, V. 2005, A&A, 430, 739.
Basu, S., & Port´-Agel, F. 2006, J. Atmos. Sci., 63, 2074. e
Bonner, C. S., Ashley, M. C. B., Cui, X., Feng, L., Gong, X., Lawrence, J. S., Luong-Van, D. M., Shang, Z., Storey, J. W. V., Wang, L., Yang, H., Yang, J., Zhou, X., & Zhu, Z., 2010, PASP, 122, 1122.
Burton, M. G., 2010, A&A Rev., 18, 417.
Gillingham, P.R., 1991, Proc. Astron. Soc. Australia, 9, 55.
Hudson, Stephen R., & Brandt, Richard E., 2005, J. Clim., 18, 1673.
Lawrence, J. S., Ashley, M. C. B., Tokovinin, A., & Travouillon, T., 2004, Nature, 431, 278.
Lawrence et al., 2008, Proc. SPIE, 7012, 701227.
Marks, R. D., Vernin, J., Azouit, M., Briggs, J. W., Burton, M. G., Ashley, M. C. B., and Manigault, J.-F., 1996, A&AS, 118, 385.
Marks, R., Vernin, J., Azouit, M., Manigault, J. F., & Clevelin, C., 1999, A&AS, 134, 161.
Marks, R. D., 2002, A&A, 385, 238.
Mefford, T., 2004, NOAA CMDL, http://www.cmdl.noaa.gov.
Okita, H., Ichikawa, T., Ashley, M. C. B., Takato, N., & Motoyama, H., 2013, A&A, 554, L5.
Sims, G., Ashley, M. C. B., Cui, X., Everett, J. R., Feng, L., Gong, X., Hengst, S., Hu, Z., Kulesa, C., Lawrence, J. S., Luong-van, D. M., Ricaud, P., Shang, Z., Storey, J. W. V., Wang, L., Yang, H., Yang, J., Zhou, X., & Zhu, Z., 2012, Publ. Astron. Soc. Pacific, 124, 74.
Travouillon, T., Ashley, M. C. B., Burton, M. G., Storey, J. W. V., & Loewenstein, R. F., 2003, A&A, 400, 1163.
Trinquet, H., Agabi, A., Vernin, J., Azouit, M., & Fossat, E., 2008, Publ. Astron. Soc. Pacific, 120, 203.
Yang et al., 2009, Publ. Astron. Soc. Pacific, 121, 174.
Yang, H., Kulesa, C. A., Walker, C. K., Tothill, N. F. H., Yang, J., Ashley, M. C. B., Cue, X., Feng, L., Lawrence, J. S., Luong-Van, D. M., Storey, J. W. V., Wang, L., Zhou, X., & Zhu, Z., 2010, Publ. Astron. Soc. Pacific, 122, 490.
This manuscript was prepared with the AAS LT E X macros v5.2. A
Table 1: Cumulative percentage of the duration of temperature inversion
| Height | > 25 hours | > 10 hours | Total |
|----------|--------------|--------------|---------|
| 1 m | 55 | 66 | 75 |
| 2 m | 39 | 51 | 71 |
| 4 m | 64 | 73 | 84 |
| 6 m | 78 | 89 | 95 |
| 9.5 m | 59 | 76 | 85 |
| 12 m | 57 | 73 | 86 |
| 14.5 m | 41 | 64 | 85 |
Note. - The temperature inversion is evaluated at each height with respect to the temperature sensor immediately below ( 3.1). §
Table 2: Average wind speed at Dome A, Dome C and the South Pole.
| Site | Wind speed ( m / | Reference |
|------------------|--------------------|------------------------|
| Dome A (4 m) | 1 . 5 | This paper |
| Dome C (3.3 m) | 2 . 9 | Aristidi et al. (2005) |
| South Pole (2 m) | 5 . 5 | Mefford (2004) |
Table 3: Dome A monthly median boundary layer height of 2009 (Bonner et al. 2010) and monthly median temperature difference of 2011 between 4 m and 2 m.
| Month | ∆ T (4 m - 2 m) ◦ C | Boundary layer height m |
|----------|-----------------------|---------------------------|
| January | 0 . 18 | · · · |
| February | 0 . 08 | 18 . 3 |
| March | 1 . 53 | 14 . 6 |
| April | 2 . 25 | 15 . 7 |
| May | 3 . 06 | 12 . 0 |
| June | 2 . 39 | 11 . 0 |
| July | 1 . 42 | 16 . 2 |
| August | · · · | 10 . 2 |
Fig. 1.- The Kunlun Automated Weather Station (KL-AWS) installed at Dome A in January 2011. The propeller anemometers at 2 m, 4 m, 9 m and 14.5 m are clearly visible.
Fig. 2.- Temperatures at nine heights, from 1 m under the snow surface to 14.5 m above the snow. The smoothed data (see § 2) are shown in this figure. The raw data are available for download (see § 4).
Fig. 3.- Temperatures recorded during the first 40 days of KL-AWS operation (top panel) in 2011 and 40 days following a restart of KL-AWS on 2011 August 23 (bottom panel). The changes in the temperature distribution from 2011 September 16 are believed to have been caused by the collapse of the upper part of the mast. This figure shows smoothed data (see § 2).
Fig. 4.- An example of temperature data smoothing for the 14.5 m and 12 m sensors for a one hour period on May 15, 2011. Note that the reproducibility of individual measurements on these timescales is much better than the absolute sensor accuracy of 0 32 . ◦ C .
Fig. 5.- Histograms and cumulative distributions (solid lines) of the temperature at the surface level and 1 m under the snow surface during 2011.
and 14.5 m during 2011, and the temperature gradients between every two adjacent levels.
Fig. 7.- Daily median temperatures during 2011. The bold black line is the temperature at 2 m.
Fig. 8.- Temperatures at different heights between 2011 May 15 and 2011 May 17. A positive temperature gradient existed as low as 1 m and lasted for more than two days at all levels. This is indicative of very low atmospheric turbulence.
Fig. 9.- The temperature gradient between 2011 January and August at all levels. The temperature gradients were calculated from the smoothed data (see § 2). And the data points were down-sampled by 100 for clarity.
Fig. 12.- Temperature difference between 6 m and 4 m folded into a day during February, March, and May (top) and the same temperature difference vs. solar elevation (bottom).
Fig. 13.- Cumulative distributions of the duration of the temperature inversion (defined as a positive temperature difference larger than 0 14 C) for all heights. . ◦
Fig. 14.- Wind speed histograms and cumulative distributions (solid lines) at 4 m, 9 m and 14.5 m during 2011.
Fig. 15.- The wind rose density at three heights during 2011. North is up. The gaps are caused by the weather mast which blocked the wind from the north direction. The grey scale is the logarithm of the number of data points per pixel. The obvious darker ring in the 4m plot corresponds to the dominant wind speed around about 1.7 m / s (see Fig. 14) of random directions. Similar patterns also exist in the other two plots, but not as obvious.
Fig. 16.- Wind speed during 2011 at three heights. Unsmoothed raw data are shown in this figure.
Fig. 17.- Wind direction during 2011 at three heights. Unsmoothed raw data are shown in this figure.
Fig. 18.- Wind speed against temperature gradient at the corresponding height. The grey scale is the logarithm of the number of data points per pixel. The vertical structures in the plots correspond to the real distribution of the data points in Fig. 14.
Fig. 19.- Air pressure during 2011. The raw data for the entire year are shown in this figure. Note that the barometric data should not be influenced by the collapse of the upper part of the mast in September.
Fig. 20.- Air pressure histogram and cumulative distribution (solid line) during 2011.
Fig. 21.- Temperature difference between the 2 m and 4 m heights (left panel) and wind speed at 4 m (right panel) versus air pressure during 2011.
Fig. 22.- Monthly median boundary layer height during 2009 (Bonner et al. 2010) versus the monthly median temperature difference between heights of 2 m and 4 m during 2011.
 | 10.1086/678327 | [
"Yi Hu",
"Zhaohui Shang",
"Michael C. B. Ashley",
"Collin S. Bonner",
"Keliang Hu",
"Qiang Liu",
"Yuansheng Li",
"Bin Ma",
"Lifan Wang",
"Haikun Wen"
] | 2014-08-01T00:50:40+00:00 | 2014-08-01T00:50:40+00:00 | [
"astro-ph.IM",
"physics.ao-ph"
] | Meteorological data for the astronomical site at Dome A, Antarctica | We present an analysis of the meteorological data collected at Dome A,
Antarctica by the Kunlun Automated Weather Station, including temperatures and
wind speeds at eight elevations above the snow surface between 0m and 14.5m.
The average temperatures at 2m and 14.5m are $-54^{\circ}$C and $-46^{\circ}$C,
respectively. We find that a strong temperature inversion existed at all
heights for more than 70% of the time, and the temperature inversion typically
lasts longer than 25 hours, indicating an extremely stable atmosphere. The
temperature gradient is larger at lower elevations than higher elevations. The
average wind speed was 1.5m/s at 4m elevation. We find that the temperature
inversion is stronger when the wind speed is lower and the temperature gradient
decreases sharply at a specific wind speed for each elevation. The strong
temperature inversion and low wind speed results in a shallow and stable
boundary layer with weak atmospheric turbulence above it, suggesting that Dome
A should be an excellent site for astronomical observations. All the data from
the weather station are available for download. |
1408.0280v1 | ## Development of Multi-gap Resistive Plate Chamber (MRPC) for Medical Imaging
A. Banerjee , A. Roy , S. Biswas a a a,b, ∗ , S. Chattopadhyay , G. Das , S. Pal a a a a Variable Energy Cyclotron Centre, 1 AF Bidhan Nagar, Kolkata-700 064, India / b GSI Helmholtzzentrum f¨ ur Schwerionenforschung GmbH, Dar mstadt, Germany-64291
## Abstract
The low cost and high resolution Multi-gap Resistive Plate Chamber (MRPC) opens up a new possibility to find an e ffi cient alternative detector for the Time of Flight (TOF) based Positron Emission Tomography, where the sensitivity of the system depends largely on the time resolution of the detector. In a layered structure, suitable converters can be used to increase the photon detection e ffi ciency. In this paper results of the cosmic ray test of a four-gap bakelite-based prototype MRPC operated in streamer mode and six-gap glass-based MRPC operated in avalanche mode are discussed.
Keywords: MRPC, Avalanche mode, Cosmic rays, Silicone, Time resolution PACS: 29.40.Cs
## 1. Introduction
Conventional PET scanners use expensive scintillators for detecting photon pairs with very high e ffi ciency and excellent energy resolution. However, the process of detecting scintillation light using PMT and associated electronics for charge integration circuits results in a large dead time. In addition to that, the scanners su ff er from the limitations of a short Field of View (FOV) [15-25 cm] and relatively poorer position resolution. In a Time-of-Flight (TOF) PET scanner, detectors with very good time resolution when positioned around the object can be used to record the time of arrival of two photons. The time di ff erence between two photons can be used to locate the point of annihilation. TOF information of the photons when coupled to the conventional PETs, improve the image quality significantly.
Apart from the limitations of the scintillator-based systems mentioned earlier, high quality scintillators are very expensive and e ff orts are being made to look for not-so-expensive alternatives. Extremely good time (50 ps) and position ( ∼ mm) resolution make the relatively less expensive MRPC-based TOFPETs to be good candidates [1, 2, 3, 4]. In MRPC-based system, the location of the origin of photons can be obtained in 3-dimensions. The detector cell provides the position in transverse plane (x-y plane) and the time of arrival on the detector gives the distance in z-direction.
An R&D e ff ort on the feasibility study of MRPC in PET imaging has been initiated at VECC. Small size prototype MRPCs have been built with both bakelite and glass and tested with cosmic rays in streamer and avalanche mode respectively. In this article, some experimental results of a four-gap (gas gap: 0.6 mm × 4) prototype bakelite MRPC, operated in streamer mode with argon, isobutane and tetrafluroethane (R-134a) in
∗ Corresponding author Email address: [email protected] (S. Biswas)
the volume ratio of 55 7.5 37.5 and a six-gap (gas gap: 0.2 mm / / × 6) prototype glass MRPC operated in avalanche mode with R-134a isobutane / / SF6 in 95 4.5 0.5 are reported. / /
## 2. Test setup and method of calculation
The MRPCs were tested in the same cosmic ray test bench described in Ref. [5]. Three scintillators, two placed above the MRPC and one below, were been used to construct the cosmic ray telescope. To measure the timing properties of the detectors the triple coincidence of the signals obtained from the three scintillators was taken as the START signal (master trigger) for the Time to Digital Converter (TDC). The STOP signal was taken from a single MRPC strip. The details of method of calculation are described in Ref. [6].
## 3. Results
The variation of e ffi ciency with that of the applied high voltage (HV) is shown in Fig. 1 for four-gap bakelite MRPC. The
Figure 1: The e ffi ciency as a function of high voltage for the four-gap bakelite MRPC.
e ffi ciency gradually increases and reaches the plateau at ∼ 95%.
Fig. 2 shows the distribution of the time di ff erence between the master trigger and the signal from one strip of the four gap bakelite MRPC. The intrinsic time resolution ( σ ) of the MRPC at 13.5 kV operating voltage appeared to be : 849 ± 17 ps. The
Figure 2: The distribution of the time di ff erence between the RPC and the master trigger for the four-gap bakelite MRPC.
Figure 3: The time resolution ( σ ) and the average signal arrival time with respect to the master trigger as a function of HV for the four-gap bakelite MRPC.
average signal arrival time, taken as the mean position of the fitted Gaussian peak in the timing spectrum (such as in Fig. 2) and the time resolution ( σ ) of the bakelite MRPC as a function of the applied HV are shown in Fig. 3. The time resolution improves and the average signal arrival time decreases with the increase of HV which is common to any gas filled detector.
The timing spectrum for the six-gap glass MRPC is shown in Fig. 4 where the intrinsic time resolution comes out to be ∼ 440 ps ( σ ).
## 4. Conclusions and outlook
MRPC-based PET system is potentially an attractive alternative to the scintillator-based system. As an e ff ort to obtain a least expensive PET imaging system capable of filtering the background, we have built a 4-gap Bakelite-based MRPC giving time resolution of ∼ 900 ps ( σ ) and a six-gap glass-based
Figure 4: The time spectrum for the six-gap glass MRPC .
Figure 5: The setup for glass-based MRPC system to be tested with gamma source
MRPC giving timing resolution of ∼ 440 ps ( σ ). Decreasing the individual gas gap and increasing the number of gaps the time resolution can be improved. Simulation to optimise the number of gaps and width of individual gap is in progress. For the proper application in PET imaging such a system with relatively lower time resolution can be useful. This also provides a platform to build a full glass-based high resolution MRPC system for PET imaging. The test set-up for one such glass-based system is shown in Fig. 5 where a Na 22 source, emitting two back-to-back 511 keV gamma ray, will be used to make the situation similar to PET imaging. The tests are in progress and the results of its performance will be communicated in a later stage.
## 5. Acknowledgement
This work is supported by the DAE-SRC award to S. Chattopadhyay under the scheme no.2008 21 07-BRNS 2738. / / /
## References
- [1] E. Cerron Zeballos, et al., Nucl. Instr. and Meth. A 374 (1996) 132.
- [2] A. Blanco, et al., Nucl. Instr. and Meth. A 508 (2003) 88.
- [3] C. Lippmann, et al., Nucl. Instr. and Meth. A 602 (2009) 735.
- [4] P. Fonte et al., Nucl. Instr. and Meth. A 449 (2000) 295.
- [5] S.Biswas, et al., Nucl. Instr. and Meth. A 602 (2009) 749.
- [6] S.Biswas, et al., Nucl. Instr. and Meth. A 617 (2010) 138. | 10.1016/j.nima.2012.08.064 | [
"A. Banerjee",
"A. Roy",
"S. Biswas",
"S. Chattopadhyay",
"G. Das",
"S. Pal"
] | 2014-08-01T05:09:19+00:00 | 2014-08-01T05:09:19+00:00 | [
"physics.ins-det",
"physics.med-ph"
] | Development of Multi-gap Resistive Plate Chamber (MRPC) for medical imaging | The low cost and high resolution Multi-gap Resistive Plate Chamber (MRPC)
opens up a new possibility to find an efficient alternative detector for the
Time of Flight (TOF) based Positron Emission Tomography, where the sensitivity
of the system depends largely on the time resolution of the detector. In a
layered structure, suitable converters can be used to increase the photon
detection efficiency. In this paper results of the cosmic ray test of a
four-gap bakelite-based prototype MRPC operated in streamer mode and six-gap
glass-based MRPC operated in avalanche mode are discussed. |
1408.0281v1 | ## Why the distance of PSR J0218+4232 does not challenge pulsar emission theories
## J. P. W. Verbiest 1 2 , /star and D. R. Lorimer 3 4 ,
1 2 3
- Fakult¨t f¨r Physik, Universit¨t Bielefeld, Postfach 100131, 33501 Bielefeld, Germany a u a
- Max-Planck-Institut f¨r Radioastronomie, Auf dem H¨gel 69, 53121 Bonn, Germany u u
- Department of Physics & Astronomy, West Virginia University, Morgantown, WV 26506, USA
- 4 National Radio Astronomy Observatory, Green Bank, WV 24944, USA
Accepted. Received; in original form 2014 March xx
## ABSTRACT
Recent VLBI measurements of the astrometric parameters of the millisecond pulsar J0218+4232 by Du et al. have suggested this pulsar is as distant as 6.3 kpc. At such a large distance, the large γ -ray flux observed from this pulsar would make it the most luminous γ -ray pulsar known. This luminosity would exceed what can be explained by the outer gap and slot-gap pulsar emission models, potentially placing important and otherwise elusive constraints on the pulsar emission mechanism. We show that the VLBI parallax measurement is dominated by the Lutz-Kelker bias. When this bias is corrected for, the most likely distance for this pulsar is 3 15 . +0 85 . -0 60 . kpc. This revised distance places the luminosity of PSR J0218+4232 into a range where it does not challenge any of the standard theories of the pulsar emission mechanism.
Key words: pulsars:general - pulsars:individual (PSR J0218+4232) - astrometry
## 1 INTRODUCTION
The millisecond pulsar (MSP) J0218+4232 was discovered by Navarro et al. (1995) and immediately stood out as a particularly bright MSP with a relatively high dispersion measure (DM). This DM placed it at a distance of 5.7 kpc based on the Galactic electron density model used at the time (Taylor & Cordes 1993). Subsequently, Verbunt et al. (1996) detected the pulsar in both X-rays and γ -rays and showed that the luminosities at these high frequencies were very high, though not impossibly so. Furthermore, these authors commented that the derived γ -ray efficiency was 10% and could only be predicted by the outer gap model for pulsar emission, not by the polar cap model. This situation changed, however, with the advent of a new electron density model for the Galaxy (Cordes & Lazio 2002), which dramatically changed the derived distance down to 2.7 kpc, in close agreement with the distance range of 2.5 to 4 kpc derived from optical observations of the pulsar's white dwarf companion (Bassa et al. 2003). At this distance, the pulsar's luminosity in γ -rays is still high, but no longer extreme. As noted by Kramer et al. (2003), this large variability in distances derived from electron density models, implies that care must be taken in deriving consequences from γ -ray luminosities based on such distances, in particular for PSR J0218+4232.
The debate on the distance (and thereby γ -ray luminosity) of PSR J0218+4232 changed again, when Du et al. (2014) recently presented the first VLBI measurements of this pulsar's astrometric parameters. They determined a total proper motion of 6 53 . ± 0 08 mas/yr and, more impor-. tantly, a parallax of 0 16 . ± 0 09 mas, translated into a dis-. tance of 6 3 . +8 0 . -2 3 . kpc. These results are significant because they make PSR J0218+4232 by far the most luminous γ -ray pulsar. Furthermore, the VLBI distance and proper motion imply a transverse speed of 195 +249 -71 km/s, which is also extreme for the class of MSPs. (Hobbs et al. 2005, derive an average transverse speed of 87 ± 13 km/s for this class of pulsar.)
- /star E-mail: [email protected]
In this paper, we demonstrate that the parallax value reported by Du et al. (2014) is dominated by the statistical Lutz-Kelker bias. We demonstrate that correction for this bias provides a distance that is more in line with the distance derived from the white-dwarf cooling models of Bassa et al. (2003) and with expected pulsar γ -ray fluxes and transverse velocities. Our analysis is given in Section 2, the implications for the pulsar velocity and γ -ray luminosity are detailed in Section 3 and our conclusions are summarised in Section 4.
Figure 1. The probability distributions for the parallax and distance of PSR J0218+4232. The combined probability distribution is indicated by the thick line; the probability distribution derived from the VLBI parallax measurement is given by the thin, continuous line; the dotted line shows the probability distribution derived from the Galactic distribution of pulsars (the so-called 'volumetric' probability); and the triple-dot-dashed line shows the probability distribution derived from the pulsar's published flux at 1.4 GHz. The most likely value and its 1σ uncertainties are indicated by the vertical lines.
## 2 CORRECTING BIASES IN PARALLAX MEASUREMENTS
The most important statistical bias that affects pulsar parallax measurements with limited measurement precision, is the Lutz-Kelker bias (Lutz & Kelker 1973). Because of the non-linear scaling of volume with parallax, the error volume on the lower-parallax end is larger than on the higher parallax end, resulting in anomalously large parallax measurements. This effect was investigated by Verbiest et al. (2010), who found that parallax measurements with significance below 2σ are likely dominated by this bias.
Applying the bias-correction code of Verbiest et al. (2012), we find that the 1.8σ measurement of Du et al. (2014) is indeed dominated by the Lutz-Kelker bias. The bias-corrected parallax value (see Figure 1) is 0 22 . +0 07 . -0 05 . mas, corresponding to a distance of 3 15 . +0 85 . -0 60 . kpc 1
With improved measurement precision, the impact of these biases will decrease, as shown in Figure 2. This indicates that, in order to reduce the effect of bias to within 0.5 σ , a parallax precision of better than 0.3 mas would be needed, requiring a factor three improvement in the published parallax value. This effort would be worthwhile, though our analysis suggests (as demonstrated in Verbiest et al. 2010), that it would be more likely that the measured parallax value would gradually increase as the measurement precision improved.
Figure 2 also demonstrates how this bias would worsen in case the Du et al. (2014) parallax uncertainty is underestimated. For example, if their measurement precision was underestimated even by only 33%, the bias-corrected distance would become as low as 2.7 kpc.
1 Note that due to the non-linear character of the parallax-todistance conversion, the most likely distance estimate is not the inverse of the most likely parallax value, unless in the absence of measurement uncertainties.
Figure 2. The strength of the bias as a function of the measurement uncertainty. For a parallax measurement of 0.16 mas, the thick line shows the bias-corrected distance (top) or parallax (bottom) as a function of the measurement uncertainty. The two horizontal lines are at the distance and parallax values proposed by Du et al. (2014) and as derived from our analysis. The vertical lines indicate 1, 2 and 3σ measurements (from right to left) and the recently published VLBI parallax measurement of Du et al. (2014) is shown by the black dot.
## 3 CONSEQUENCES OF THE BIAS-CORRECTED PARALLAX VALUE
Du et al. (2014) derive two significant results from their astrometric measurements. Firstly, they combine their proper motion and parallax measurements to derive a transverse speed for the pulsar of 195 +249 -71 km/s. This value is well above the mean MSP transverse speed of 87 ± 13 km/s (Hobbs et al. 2005). Redetermining the transverse speed of PSR J0218+4232 with our bias-corrected parallax value, we obtain 98 km/s, which lies well within the scatter inherent to the population.
More importantly, the VLBI distance was used to determine the γ -ray luminosity of PSR J0218+4232. This resulted in L γ = 2 2 . × 10 35 erg/s, which places it more than a factor of three above the next most luminous γ -ray MSP, implying a γ -ray efficiency in excess of 90%. Our bias-corrected distance of 3 15 kpc, however, results in a . γ -ray luminosity of 5 4 . × 10 34 erg/s, which lies well inside the range of γ -ray luminosities for MSPs and a γ -ray efficiency of 23%, also comparable to the population at large (Abdo et al. 2013).
## 4 CONCLUSIONS
We have quantified and corrected biases present in the recent VLBI parallax to PSR J0218+4232. Our results are collated in Table 1 and are as follows. For the parallax we obtain a most-likely value of 0 22 . +0 07 . -0 05 . mas and for the distance 3 15 . +0 85 . -0 60 . kpc. We have demonstrated that these bias-corrected values result in a transverse velocity of 98 km/s, which is comparable to values obtained for the rest of the MSP population; and in a γ -ray luminosity of 5 4 . × 10 34 erg/s, also in line with values obtained for other MSPs. The γ -ray emission of this pulsar, therefore, does not
Table 1. Summary of results. Shown are the parallax and distance from the Du et al. (2014) VLBI measurements (1), from our standard bias-correction code described in Verbiest et al. (2012) (2) and from the volumetric and luminosity information alone (i.e. the prior information excluding the parallax measurement; 3).
| Method | Parallax (mas) | Distance (kpc) |
|----------------------------------------------------|--------------------------------------------------------------|-------------------------------------------------------------------|
| (1) VLBI (2) Bias-corrected (3) vol. & lum. priors | 0 . 16 ± 0 . 09 0 . 22 +0 . 07 - 0 . 05 0 . 3 +0 . 3 - 0 . 1 | 6 . 3 +8 . 0 - 2 . 3 3 . 15 +0 . 85 - 0 . 60 0 . 5 +0 . 8 - 0 . 3 |
challenge the outer gap or slot-gap models for pulsar emission, as claimed by Du et al. (2014).
## ACKNOWLEDGMENTS
The authors wish to thank Ben W. Stappers and Adam Deller for discussion of the PSR J0218+4232 astrometry results, Michael Kramer for comments on the initial draft and the anonymous referee for a thorough and critical review.
## REFERENCES
Abdo A. A., Ajello M., Allafort A., Baldini L., Ballet J., Barbiellini G., Baring M. G., Bastieri D., Belfiore A., Bellazzini R., Bhattacharyya B., Bissaldi E., Bloom E. D., Bonamente E., Bottacini E., Brandt T. J., et al. 2013,
ApJS, 208, 17 Bassa C. G., van Kerkwijk M. H., Kulkarni S. R., 2003, A&A, 403, 1067 Cordes J. M., Lazio T. J. W., 2002, ArXiv:astroph/0207156 Du Y., Yang J., Campbell R. M., Janssen G., Stappers B., Chen D., 2014, ApJ, 782, L38 Hobbs G., Lorimer D. R., Lyne A. G., Kramer M., 2005, MNRAS, 360, 974 Kramer M., Bell J. F., Manchester R. N., Lyne A. G., Camilo F., Stairs I. H., D'Amico N., Kaspi V. M., Hobbs G., Morris D. J., Crawford F., Possenti A., Joshi B. C., McLaughlin M. A., Lorimer D. R., Faulkner A. J., 2003, MNRAS, 342, 1299 Lutz T. E., Kelker D. H., 1973, PASP, 85, 573 Navarro J., de Bruyn G., Frail D., Kulkarni S. R., Lyne A. G., 1995, ApJ, 455, L55 Taylor J. H., Cordes J. M., 1993, ApJ, 411, 674 Verbiest J. P. W., Lorimer D. R., McLaughlin M. A., 2010, MNRAS, 405, 564 Verbiest J. P. W., Weisberg J. M., Chael A. A., Lee K. J., Lorimer D. R., 2012, ApJ, 755, 39 Verbunt F., Kuiper L., Belloni T., Johnston H. M., de Bruyn A. G., Hermsen W., van der Klis M., 1996, A&A, 311, L9 | 10.1093/mnras/stu1560 | [
"J. P. W. Verbiest",
"D. R. Lorimer"
] | 2014-08-01T07:25:04+00:00 | 2014-08-01T07:25:04+00:00 | [
"astro-ph.HE",
"astro-ph.GA"
] | Why the distance of PSR J0218+4232 does not challenge pulsar emission theories | Recent VLBI measurements of the astrometric parameters of the millisecond
pulsar J0218+4232 by Du et al. have suggested this pulsar is as distant as 6.3
kpc. At such a large distance, the large {\gamma}-ray flux observed from this
pulsar would make it the most luminous {\gamma}-ray pulsar known. This
luminosity would exceed what can be explained by the outer gap and slot-gap
pulsar emission models, potentially placing important and otherwise elusive
constraints on the pulsar emission mechanism. We show that the VLBI parallax
measurement is dominated by the Lutz-Kelker bias. When this bias is corrected
for, the most likely distance for this pulsar is 3.15(+0.85/-0.60) kpc. This
revised distance places the luminosity of PSR J0218+4232 into a range where it
does not challenge any of the standard theories of the pulsar emission
mechanism. |
1408.0282v1 | ## A Polling Model with Multiple Priority Levels ∗
M.A.A. Boon †
[email protected]
I.J.B.F. Adan † [email protected]
O.J. Boxma †
[email protected]
August, 2009
## Abstract
In this paper we consider a single-server cyclic polling system. Between visits to successive queues, the server is delayed by a random switch-over time. The order in which customers are served in each queue is determined by a priority level that is assigned to each customer at his arrival. For this situation the following service disciplines are considered: gated, exhaustive, and globally gated. We study the cycle time distribution, the waiting times for each customer type, the joint queue length distribution of all priority classes at all queues at polling epochs, and the steady-state marginal queue length distributions for each customer type.
Keywords: Polling, priority levels, queue lengths, waiting times
## 1 Introduction
A polling model is a single-server system in which the server visits N queues Q , . . . , Q 1 N in cyclic order. Customers that arrive at Q i are referred to as type i customers. The special feature of the model considered in the present paper is that, within a customer type, we distinguish multiple priority levels. More specifically, we study a polling system which consists of N queues, Q , . . . , Q 1 N , and K i priority levels in Q i . The exhaustive, gated and globally gated service disciplines are studied.
Our motivation to study a polling model with priorities is that scheduling through the introduction of priorities in a polling system can improve the performance of the system significantly without having to purchase additional resources [27]. Priority polling systems can be used to study the Bluetooth and IEEE 802.11 wireless LAN protocols, or scheduling policies at routers and I/O subsystems in web servers. For example, Quality-of-Service (QoS) has become a very important issue in wireless LAN protocols because delays in responses or data transmissions do not have a major impact on, e.g., web browsing or email traffic, but video transmissions and Voice over Wireless LAN (VoWLAN) are very sensitive to delays or loss of data. The 802.11e amendment introduces different priority levels to differentiate between
∗ The research was done in the framework of the BSIK/BRICKS project, and of the European Network of Excellence Euro-FGI.
† Eurandom and Department of Mathematics and Computer Science, Eindhoven University of Technology, P.O. Box 513, 5600MB Eindhoven, The Netherlands
types of data in order to improve QoS for streaming traffic. In production environments the introduction of priorities can lead to smaller mean waiting times (see, e.g., [28]). It is shown in [1, 27] that assigning highest priority to jobs with a service requirement below a certain threshold level may reduce overall mean waiting times. Priority polling models also can be used to study traffic intersections where conflicting traffic flows face a green light simultaneously; e.g. traffic which takes a left turn may have to give right of way to conflicting traffic that moves straight on, even if the traffic light is green for both traffic flows. The last application area of polling models with priorities that we mention here, is health care (see, e.g., [9]).
Although there is an extensive amount of literature available on polling systems (see, e.g., the surveys of Takagi [24], Levy and Sidi [19], and Vishnevskii and Semenova [26]), very few papers treat priorities in polling models. Most of these papers only provide approximations or focus on pseudo-conservation laws [13, 22]. Wierman, Winands and Boxma [27] have obtained exact mean waiting time results using the Mean Value Analysis (MVA) framework for polling systems, developed in [29]. The MVA framework can only be used to find the first moment of the waiting time distribution for each customer type, and the mean residual cycle time. In a recent report [1] we have studied a polling model with two queues that are served according to the exhaustive, gated or globally gated service discipline. The first of these two queues contains customers of two priority classes. The main contribution of [1] is the derivation of Laplace-Stieltjes Transforms (LSTs) of the distributions of the marginal waiting times for each customer type; in particular it turned out to be possible to obtain exact expressions for the waiting time distributions of both high and low priority customers at a queue of a polling system. Probability Generating Functions (GFs) have been derived for the joint queue length distribution at polling epochs, and for the steady-state marginal queue length distribution of the number of customers at an arbitrary epoch. In the present paper these results are generalised to a polling model with N queues and K i priority levels in Q i ( i = 1 , . . . , N ). The actual probability distributions can be obtained by numerical inversion of the LSTs and GFs. A very efficient technique for numerical inversion of GFs and LSTs in polling models is discussed in [8].
The present paper is structured as follows: Section 2 gathers known results of nonpriority polling models which we shall need in the present study. Sections 3 (gated and exhaustive), and 4 (globally gated) give new results on the priority polling model. In each of the sections we successively discuss the joint queue length distribution at polling epochs, the cycle time distribution, the marginal queue length distributions and waiting time distributions. The mean waiting times are given at the end of each section.
## 2 The nonpriority polling model
The model that is considered in this section, is a polling model with N queues ( Q , . . . , Q 1 N ) without priorities. Each queue is served according to either the gated, or the exhaustive service discipline. The gated service discipline states that during a visit to Q i , only those type i customers are served who are present at the moment that the server arrives at Q i . All type i customers that arrive during the visit to Q i will be served during the next cycle. A cycle is the time between two successive visit beginnings (or completions) at a queue. The exhaustive service discipline states that when the server arrives at Q i , all type i customers that
are present at that polling epoch, and all type i customers that arrive during this particular visit to Q i , are served until no type i customer is present in the system. In Section 4 we also consider the globally gated service discipline, which is similar to the gated service discipline, except for the fact that the symbolic gate is being set at the beginning of a cycle for all queues. This means that during a cycle only those customers will be served that were present at the beginning of that cycle.
Customers of type i arrive at Q i according to a Poisson process with arrival rate λ i . Service times can follow any distribution. The LST of the distribution of the generic service time B i of type i customers is denoted by β i ( ). · The fraction of time that the server is serving customers of type i equals ρ i := λ E B i ( i ). Switches of the server from Q i to Q i +1 (all indices modulo N ), require a switch-over time S i . The LST of this switch-over time distribution is denoted by σ i ( ). · The fraction of time that the server is working (i.e., not switching) is ρ := ∑ N i =1 ρ i . We assume that ρ < 1, which is a necessary and sufficient condition for the steady state distributions of cycle times, queue lengths and waiting times to exist. We assume that all the usual independence assumptions apply to the model under consideration.
This model has been extensively investigated. Tak´ acs [23] studied this model, but with only two queues, without switch-over times and only with the exhaustive service discipline. Cooper and Murray [11] analysed this polling system for any number of queues, and for both gated and exhaustive service disciplines. Eisenberg [12] obtained results for a polling system with switch-over times (but only exhaustive service) by relating the GFs of the joint queue length distributions at visit beginnings, visit endings, service beginnings and service endings. Resing [21] and Fuhrmann [14] both pointed out the relation between polling systems and Multitype Branching Processes (MTBPs). Resing [21] considers the use of MTBPs with immigration in each state, which makes it possible to relate polling systems with and without switch-over times. His results can be applied to polling models in which each queue satisfies the following property:
Property 2.1 If the server arrives at Q i to find k i customers there, then during the course of the server's visit, each of these k i customers will effectively be replaced in an i.i.d. manner by a random population having probability generating function h i ( z , . . . , z 1 N ), which can be any N -dimensional probability generating function.
We use this property, and the relation to Multitype Branching Processes, to find results for our polling system with multiple priority levels, and gated, globally gated, and exhaustive service discipline. Notice that, unlike the gated and exhaustive service disciplines, the globally gated service discipline does not satisfy Property 2.1. But the results obtained by Resing also hold for a more general class of polling systems, namely those which satisfy the following (weaker) property that is formulated in [2]:
Property 2.2 If there are k i customers present at Q i at the beginning (or the end) of a visit to Q π i ( ) , with π i ( ) ∈ { 1 , . . . , N } , then during the course of the visit to Q i , each of these k i customers will effectively be replaced in an i.i.d. manner by a random population having probability generating function h i ( z , . . . , z 1 N ), which can be any N -dimensional probability generating function.
Globally gated and gated are special cases of the synchronised gated service discipline, which states that only customers in Q i will be served that were present at the moment that the
server reaches the 'parent queue' of Q i : Q π i ( ) . For gated service, π i ( ) = i , for globally gated service, π i ( ) = 1. The synchronised gated service discipline is discussed in [18], but no observation is made that this discipline is a member of the class of polling systems satisfying Property 2.2 which means that results as obtained in [21] can be extended to this model.
Borst and Boxma [3] combined the results of Resing [21] and Eisenberg [12] to find a relation between the GFs of the marginal queue length distribution for polling systems with and without switch-over times, expressed in the Fuhrmann-Cooper queue length decomposition form [15].
## 2.1 Joint queue length distribution at polling epochs
As discussed in [21], polling systems which satisfy Property 2.1 can be modelled as a MTBP. If the polling model has switch-over times, as is assumed in the present paper, we are dealing with a MTBP with immigration in each state. The immigration is formed by the customers that arrive during the switch-over times and all of their descendants. Polling models without switch-over times can be modelled as a MTBP with immigration in state zero (when the server is idle) only. Although the present paper studies a polling model with switch-over times, it is possible to relate results from a model with switch-over times to a model without switch-over times, as is shown in [3].
glyph[negationslash]
glyph[negationslash]
For now, we choose the beginning of a visit to Q 1 as start of a cycle. Property 2.1 states that each customer present at the beginning of a cycle will be 'replaced' by customers of type 1 , . . . , N during that cycle according to the probability generating function h i ( z , . . . , z 1 N ), which depends on the service discipline. If Q i ( i = 1 , . . . , N ) receives gated service, a customer of type i present at the beginning of the cycle is replaced by the type 1 , . . . , N customers that arrive during his service: h i ( z , . . . , z 1 N ) = β i ( ∑ N j =1 λ j (1 -z j )). For exhaustive service, a type i customer will not just be replaced by all type 1 , . . . , N customers that arrive during his service, but also all customers that arrive during the service of all type i customers that have arrived during the service of this tagged customer, etc., until no type i customer is present in the system. The consequence is that this type i customer has been replaced by the type j customers ( j = 1 , . . . , N ; j = ) that have arrived during the busy period (BP) i of type i customers that is initiated by the particular type i customer. This results in the expression h i ( z , . . . , z 1 N ) = π i ( ∑ j = i λ j (1 -z j )), where π i ( ) is · the LST of a BP distribution in an M/G/ 1 system with only type i customers, so it is the root in (0 , 1] of the equation π i ( ω ) = β i ( ω + λ i (1 -π i ( ω ))), ω ≥ 0 (cf. [10], p. 250).
In order to find the joint queue length distribution at the beginning of a cycle, we have to define the immigration GF and the offspring GF analogous to [21]. The first generation of offspring consists of the customers that have effectively replaced the customers present at the beginning of the cycle. The offspring GFs for queues N,N -1 , . . . , 1 are given below.
$$f ^ { ( N ) } ( z _ { 1 }, \dots, z _ { N } ) & = h _ { N } ( z _ { 1 }, \dots, z _ { N } ), \\ f ^ { ( i ) } ( z _ { 1 }, \dots, z _ { N } ) & = h _ { i } ( z _ { 1 }, \dots, z _ { i }, f ^ { ( i + 1 ) } ( z _ { 1 }, \dots, z _ { N } ), \dots, f ^ { ( N ) } ( z _ { 1 }, \dots, z _ { N } ) ), \quad i = 1, \dots, N - 1.$$
We define the GF for the n th generation of offspring recursively:
$$f _ { n } ( z _ { 1 }, \dots, z _ { N } ) & = ( f ^ { ( 1 ) } ( f _ { n - 1 } ( z _ { 1 }, \dots, z _ { N } ) ), \dots, f ^ { ( N ) } ( f _ { n - 1 } ( z _ { 1 }, \dots, z _ { N } ) ) ), \\ f _ { 0 } ( z _ { 1 }, \dots, z _ { N } ) & = ( z _ { 1 }, \dots, z _ { N } ).$$
The immigration GF is the GF of the joint distribution of the numbers of customers that are present at the beginning of a cycle due to the immigration during the switch-over times of the previous cycle. These customers can be divided into two groups: the customers that arrived during a switch after the service of their queue, and offspring of customers that arrived during a switch before the service of their queue. The immigration GFs are:
$$\hat { g } ^ { ( N ) } ( z _ { 1 }, \dots, z _ { N } ) & = \sigma _ { N } ( \sum _ { j = 1 } ^ { N } \lambda _ { j } ( 1 - z _ { j } ) ), \\ g ^ { ( i ) } ( z _ { 1 }, \dots, z _ { N } ) & = \sigma _ { i } ( \sum _ { j = 1 } ^ { i } \lambda _ { j } ( 1 - z _ { j } ) + \sum _ { j = i + 1 } ^ { N } \lambda _ { j } ( 1 - f ^ { ( j ) } ( z _ { 1 }, \dots, z _ { N } ) ) ), \quad i = 1, \dots, N - 1.$$
The total immigration GF is the product of these GFs:
$$g ( z _ { 1 }, \dots, z _ { N } ) = \prod _ { i = 1 } ^ { N } g ^ { ( i ) } ( z _ { 1 }, \dots, z _ { N } ).$$
This leads to the following recursive expression for the GF of the steady-state joint queue length at the beginning of a cycle (starting with a visit to Q 1 ) :
$$P _ { 1 } ( z _ { 1 }, \dots, z _ { N } ) & = P _ { 1 } ( f _ { 1 } ( z _ { 1 }, \dots, z _ { N } ) ) g ( z _ { 1 }, \dots, z _ { N } ). & ( 2. 1 )$$
Resing [21] shows how (2.1) can be used to obtain moments from the marginal queue length distribution at the beginning of a cycle. It is shown in [20] that iteration of (2.1) leads to
$$P _ { 1 } ( z _ { 1 }, \dots, z _ { N } ) = \prod _ { n = 0 } ^ { \infty } g ( f _ { n } ( z _ { 1 }, \dots, z _ { N } ) ).$$
Resing [21] proves that this infinite product converges if and only if ρ < 1.
We can relate the steady-state joint queue length distribution at other visit beginnings and endings to P 1 ( z , . . . , z 1 N ). We denote the GF of the joint queue length distribution at a visit beginning to Q i by V b i ( ), so · V b 1 ( ) = · P 1 ( ). · The queue length GF at a visit completion to Q i is denoted by V c i ( ). · The following relation holds:
$$V _ { b _ { i } } ( z _ { 1 }, \dots, z _ { N } ) = V _ { c _ { i - 1 } } ( z _ { 1 }, \dots, z _ { N } ) \sigma _ { i - 1 } ( \sum ^ { N } \lambda _ { j } ( 1 -$$
$$( 2. 3 )$$
$$V _ { b _ { i } } ( z _ { 1 }, \dots, z _ { N } ) & = V _ { c _ { i - 1 } } ( z _ { 1 }, \dots, z _ { N } ) \sigma _ { i - 1 } ( \sum _ { j = 1 } ^ { N } \lambda _ { j } ( 1 - z _ { j } ) ) \\ & = V _ { b _ { i - 1 } } ( z _ { 1 }, \dots, z _ { i - 2 }, h _ { i - 1 } ( z _ { 1 }, \dots, z _ { N } ), z _ { i }, \dots, z _ { N } ) \sigma _ { i - 1 } ( \sum _ { j = 1 } ^ { N } \lambda _ { j } ( 1 - z _ { j } ) ). \quad ( 2. 3 )$$
Since V b 1 ( ) is known, applying (2.3) · i -1 times expresses V b i ( ) into · V b 1 ( ) = · P 1 ( ). · Applying (2.3) N times gives expression (2.1).
## 2.2 Cycle time
The cycle time, starting at a visit beginning to Q 1 , is the sum of the visit times to Q , . . . , Q 1 N , and the switch-over times S , . . . , S 1 N which are independent of the visit times. Let θ i ( ) denote · the LST of the distribution of the time that the server spends at Q i due to the presence of one type i customer there. For gated service θ i ( ) = · β i ( ), for exhaustive service · θ i ( ) = · π i ( ). · Furthermore, we define
$$\psi _ { i } ( \omega ) & = \omega + \lambda _ { i } ( 1 - \theta _ { i } ( \omega ) ), & i = 1, \dots, N, \\ \psi _ { i, N } ( \omega ) & = \psi _ { i + 1 } ( \psi _ { i + 2 } ( \dots ( \psi _ { N } ( \omega ) ) ) ), & i = 1, \dots, N - 1, \\ \psi _ { N, N } ( \omega ) & = \omega.$$
It is shown in [7] that the LST of the distribution of the cycle time C 1 , γ 1 ( ), · is related to P 1 ( ) as follows: ·
$$\gamma _ { 1 } ( \omega ) = \prod _ { i = 1 } ^ { N } \sigma _ { i } ( \psi _ { i, N } ( \omega ) ) \, P _ { 1 } ( \theta _ { 1 } ( \psi _ { 1, N } ( \omega ) ), \dots, \theta _ { N } ( \psi _ { N, N } ( \omega ) ) ).$$
In the present paper we need expressions for the LST of the distribution of the cycle time starting at the beginning of a visit to an arbitrary queue. We denote the cycle time starting with a visit beginning to Q j by C j , and its LST by γ j ( ). · It is straightforward to see that
$$\gamma _ { j } ( \omega ) = \prod _ { i = 1 } ^ { N } \sigma _ { i } ( \psi _ { i, j - 1 } ( \omega ) ) \, V _ { b _ { j } } ( \theta _ { 1 } ( \psi _ { 1, j - 1 } ( \omega ) ), \dots, \theta _ { N } ( \psi _ { N, j - 1 } ( \omega ) ) ),$$
where we use the following notation:
$$\psi _ { i, j } ( \omega ) & = \psi _ { i + 1 } ( \dots ( \psi _ { j } ( \omega ) ) ), & j & = 1, \dots, N ; i < j, \\ \psi _ { i, j } ( \omega ) & = \psi _ { i + 1 } ( \dots ( \psi _ { N } ( \psi _ { 1 } ( \dots ( \psi _ { j } ( \omega ) ) ) ) ), & j & = 1, \dots, N ; i > j, \\ \psi _ { j, j } ( \omega ) & = \omega, & j & = 1, \dots, N.$$
When analysing a queue with exhaustive service, it is convenient to define C ∗ j to be the time between two successive visit completions to Q j . The LST of its distribution, denoted by γ ∗ j ( ), · is:
$$\gamma _ { j } ^ { * } ( \omega ) = \prod _ { i = 1 } ^ { N } \sigma _ { i } ( \psi _ { i, j } ^ { * } ( \omega ) ) \, V _ { c _ { j } } ( \theta _ { 1 } ( \psi _ { 1, j } ( \omega ) ), \dots, \theta _ { N } ( \psi _ { N, j } ( \omega ) ) ),$$
with V c j ( z , . . . , z 1 N ) = V b j ( z , . . . , z 1 j -1 , h j ( z , . . . , z 1 N ) , z j +1 , . . . , z N ), and glyph[negationslash]
$$\psi _ { i, j } ^ { * } ( \omega ) & = \psi _ { i, j } ( \omega ), & j & = 1, \dots, N ; i \neq j, \\ \psi _ { j, j } ^ { * } ( \omega ) & = \psi _ { j + 1 } ( \dots ( \psi _ { N } ( \psi _ { 1 } ( \dots ( \psi _ { j } ( \omega ) ) ) ) ), & j & = 1, \dots, N.$$
## 2.3 Marginal queue lengths and waiting times
We denote the GF of the steady-state marginal queue length distribution of Q i at the visit beginning by ˜ V b i ( z ) = V b i (1 , . . . , 1 , z, 1 , . . . , 1), with z as i th argument. Analogously we define
˜ V c i ( ). · It is shown in [3] that the steady-state marginal queue length of Q i can be decomposed into two parts: the queue length of the corresponding M/G/ 1 queue with only type i customers, and the queue length at an arbitrary epoch during the intervisit period of Q i , denoted by L i I | . An intervisit period of Q i is the time between a visit ending at Q i , and the next visit beginning at Q i . Borst [2] shows that by virtue of PASTA, L i I | has the same distribution as the number of type i customers seen by an arbitrary type i customer arriving during an intervisit period, which equals
$$E ( z ^ { L _ { i | I } } ) = \frac { E ( z ^ { L _ { i | I } } _ { b e g i n } ) - E ( z ^ { L _ { i | I } } _ { e n d } ) } { ( 1 - z ) ( E ( L _ { i | I } _ { e n d } ) - E ( L _ { i | I } _ { b e g i n } ) ) },$$
where L i I | begin is the number of type i customers at the beginning of an intervisit period I i , and L i I | end is the number of type i customers at the end of I i . Since the beginning of an intervisit period coincides with the completion of a visit to Q i , and the end of an intervisit period coincides with the beginning of a visit, we know the GFs for the distributions of these random variables: ˜ V c i ( ) and · ˜ V b i ( ). · This leads to the following expression for the GF of the steady-state queue length distribution of Q i at an arbitrary epoch, E z [ L i ]:
$$E [ z ^ { L _ { i } } ] = \frac { ( 1 - \rho _ { i } ) ( 1 - z ) \beta _ { i } ( \lambda _ { i } ( 1 - z ) ) } { \beta _ { i } ( \lambda _ { i } ( 1 - z ) ) - z } \cdot \frac { \widetilde { V } _ { c _ { i } } ( z ) - \widetilde { V } _ { b _ { i } } ( z ) } { ( 1 - z ) ( E ( L _ { i | I _ { e n d } } ) - E ( L _ { i | I _ { b e g i n } } ) ) }. \quad ( 2. 7 )$$
Keilson and Servi [16] show that the distributional form of Little's law can be used to find the LST of the distribution of the marginal waiting time W i : E z ( L i ) = E (e -λ i (1 -z )( W i + B i ) ), hence E (e -ωW i ) = E [(1 -ω λ i ) L i ] /β i ( ω ). This can be substituted into (2.7):
$$\cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \\ E [ e ^ { - \omega W _ { i } } ] & = \frac { ( 1 - \rho _ { i } ) \omega } { \omega - \lambda _ { i } ( 1 - \beta _ { i } ( \omega ) ) } \cdot \frac { \widetilde { V } _ { c _ { i } } \left ( 1 - \frac { \omega } { \lambda _ { i } } \right ) - \widetilde { V } _ { b _ { i } } \left ( 1 - \frac { \omega } { \lambda _ { i } } \right ) } { ( E ( L _ { i } | _ { I _ { e n d } } ) - E ( L _ { i } | _ { I _ { b e g i n } } ) ) \omega / \lambda _ { i } } \\ & = E [ e ^ { - \omega W _ { i } | M / G / 1 } ] E \left [ \left ( 1 - \frac { \omega } { \lambda _ { i } } \right ) ^ { L _ { i } | I } \right ]. \\ \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \end{cases}$$
The interpretation of this formula is that the waiting time of a type i customer in a polling model is the sum of two independent random variables: the waiting time of a customer in an M/G/ 1 queue with only type i customers, W i M/G/ | 1 , and the remaining intervisit time for a customer that arrives at an arbitrary epoch during the intervisit time of Q i .
For gated service, the number of type i customers at the beginning of a visit to Q i is exactly the number of type i customers that arrived during the previous cycle, starting at Q i . In terms of GFs: ˜ V b i ( z ) = γ i ( λ i (1 -z )). The type i customers at the end of a visit to Q i are exactly those type i customers that arrived during this visit. In terms of GFs: ˜ V c i ( z ) = γ i ( λ i (1 -β i ( λ i (1 -z )))). We can rewrite E L ( i I | end ) -E L ( i I | begin ) as λ E I i ( i ), because this is the mean number of type i customers that arrive during an intervisit time. In Section 2.4 we observe that λ E I i ( i ) = λ i (1 -ρ i ) E C ( ). Using these expressions we can rewrite Equation (2.8) for gated service to:
$$E [ \mathrm e ^ { - \omega W _ { i } } ] = \frac { ( 1 - \rho _ { i } ) \omega } { \omega - \lambda _ { i } ( 1 - \beta _ { i } ( \omega ) ) } \cdot \frac { \gamma _ { i } ( \lambda _ { i } ( 1 - \beta _ { i } ( \omega ) ) ) - \gamma _ { i } ( \omega ) } { ( 1 - \rho _ { i } ) \omega E ( C ) }.$$
For exhaustive service, ˜ V c i ( z ) = 1, because Q i is empty at the end of a visit to Q i . The number of type i customers at the beginning of a visit to Q i in an exhaustive polling system
is equal to the number of type i customers that arrived during the previous intervisit time of Q i . Hence, ˜ V b i ( z ) = ˜ I i ( λ i (1 -z )), where ˜ I i ( ) is the LST of the intervisit time distribution for · Q i . Substitution of ˜ I i ( ω ) = ˜ V b i (1 -ω λ i ) in (2.8) leads to the following expression for the LST of the steady-state waiting time distribution of a type i customer in an exhaustive polling system:
$$E [ e ^ { - \omega W _ { i } } ] = \frac { ( 1 - \rho _ { i } ) \omega } { \omega - \lambda _ { i } ( 1 - \beta _ { i } ( \omega ) ) } \cdot \frac { 1 - \widetilde { I } _ { i } ( \omega ) } { \omega E ( I _ { i } ) }.$$
In [1] another expression is found for E [e -ωW i ]. Let the cycle time C ∗ i be the time between two successive visit completions to Q i . The LST of the cycle time distribution, γ ∗ i ( ), is given · by (2.5). In [1] an equation is derived that expresses ˜ I i ( ) in · γ ∗ i ( ): ·
$$\widetilde { I } _ { i } ( \omega ) = \gamma _ { i } ^ { * } ( \omega - \lambda _ { i } ( 1 - \beta _ { i } ( \omega ) ) ).$$
Using this expression we can write E [e -ωW i ] in terms of E [e -ωC ∗ i ]:
$$E [ e ^ { - \omega W _ { i } } ] & = \frac { 1 - \gamma _ { i } ^ { * } ( \omega - \lambda _ { i } ( 1 - \beta _ { i } ( \omega ) ) ) } { ( \omega - \lambda _ { i } ( 1 - \beta _ { i } ( \omega ) ) ) E ( C ) } \\ & = E [ e ^ { - ( \omega - \lambda _ { i } ( 1 - \beta _ { i } ( \omega ) ) ) C _ { i, r e s } ^ { * } } ],$$
where C ∗ i, res is the residual length of C ∗ i .
Remark 2.3 Substitution of ω := s + λ i (1 -π i ( s )) in (2.12) leads to:
$$E [ e ^ { - ( s + \lambda _ { i } ( 1 - \pi _ { i } ( s ) ) ) W _ { i } } ] & = E [ e ^ { - ( s + \lambda _ { i } ( 1 - \pi _ { i } ( s ) ) - \lambda _ { i } ( 1 - \beta _ { i } ( s + \lambda _ { i } ( 1 - \pi _ { i } ( s ) ) ) ) C _ { i, r e s } ^ { * } } ] } \\ & = E [ e ^ { - ( s + \lambda _ { i } ( 1 - \pi _ { i } ( s ) ) - \lambda _ { i } ( 1 - \pi _ { i } ( s ) ) ) C _ { i, r e s } ^ { * } } ] \\ & = E [ e ^ { - s C _ { i, r e s } ^ { * } } ].$$
The interpretation, which can also be found in [4], is that the residual cycle time consists of two components; firstly the amount of work present at the beginning of the residual cycle, and secondly all work that arrives at Q i during the residual cycle. By virtue of PASTA, the amount of work at the beginning of the residual cycle is the waiting time of an arbitrary type i customer. The work that arrives during this waiting time, is equal to the length of the busy periods generated by the customers that have arrived during this waiting time.
## 2.4 Moments
The focus of this paper is on LST and GF of distribution functions, not on their moments. Moments can be obtained by differentiation or Taylor series expansion, and are also discussed in [27]. In this subsection we will only mention some results that will be used later.
First we derive the mean cycle time E C ( ). Unlike higher moments of the cycle time, the mean does not depend on where the cycle starts: E C ( ) = E S ( ) 1 -ρ , where S = ∑ N i =1 S i . This can easily be seen, because 1 -ρ is the fraction of time that the server is not working, but switching. The total mean switch-over time is ∑ N i =1 E S ( i ) = E S ( ).
The expected length of a visit to Q i is E V ( i ) = ρ E C i ( ), hence the mean length of an intervisit period for Q i is E I ( i ) = (1 -ρ i ) E C ( ). Notice that these expectations do not depend on the
service discipline used. The expected number of type i customers at polling moments does depend on the service discipline. For gated service the expected number of type i customers at the beginning of a visit to Q i is λ E C i ( ). For exhaustive service this is λ E I i ( i ).
Moments of the waiting time distribution for a type i customer at an arbitrary epoch can be derived from the LSTs given by (2.9), (2.10) and (2.12). We only present the first moment:
$$Q _ { i } \, \text{gated} \colon & & E ( W _ { i } ) = ( 1$$
$$E ( W _ { i } ) = ( 1 + \rho _ { i } ) \frac { E ( C _ { i } ^ { 2 } ) } { 2 E ( C ) },$$
$$Q _ { i } \text{ exhaustive} \colon$$
$$& + \rho _ { i } ) \frac { E ( C _ { i } ^ { 2 } ) } { 2 E ( C ) }, & & ( 2. 1 3 ) \\ & \frac { ( I _ { i } ^ { 2 } ) } { \L ( I _ { i } ) } + \frac { \rho _ { i } } { 1 - \rho _ { i } } \frac { E ( B _ { i } ^ { 2 } ) } { 2 E ( B _ { i } ) } \\ & - \rho _ { i } ) \frac { E ( C _ { i } ^ { * 2 } ) } { 2 E ( C ) }. & & ( 2. 1 4 )$$
$$E ( W _ { i } ) & = \frac { E ( I _ { i } ^ { 2 } ) } { 2 E ( I _ { i } ) } + \frac { \rho _ { i } } { 1 - \rho _ { i } } \frac { E ( B _ { i } ^ { 2 } ) } { 2 E ( B _ { i } ) } \\ & = ( 1 - \rho _ { i } ) \frac { E ( C _ { i } ^ { * } ^ { 2 } ) } { 2 E ( C ) }.$$
Notice that the start of C i is the beginning of a visit to Q i for gated service, whereas the start of C ∗ i is the end of a visit for exhaustive service. Equations (2.13) and (2.14) are in agreement with Equations (4.1) and (4.2) in [4], which also gives interpretations of these equations. Although at first sight these might seem nice, closed formulas, it should be noted that second moments of the cycle time and intervisit time are not easy to determine, requiring the solution of a set of equations. MVA is an efficient technique to compute mean waiting times, the mean residual cycle time, and also the mean residual intervisit time. We refer to [29] for an MVA framework for polling models.
## 3 The priority polling model with gated and/or exhaustive service
In this section we study a polling system with N queues. The service discipline of each queue is either gated, or exhaustive. Each queue contains customers that can be divided into one or more priority classes. Let K i be the total number of priority levels in Q i . Customers of level 1 receive highest priority, customers of level K i receive lowest priority. Customers in Q i of priority level k ( k = 1 , . . . , K i ) arrive according to a Poisson process with intensity λ ik , and have a service requirement B ik with LST β ik ( ). · If we consider the total number of customers in each queue at polling epochs, which means that we ignore the fact that they will be served according to their priority levels instead of ordinary First-Come-First-Served (FCFS), all the results from Subsections 2.1 and 2.2 still hold. In this situation the system should be regarded as a polling system with N queues where customers in Q i arrive according to a Poisson process with intensity λ i := ∑ K i k =1 λ ik and have service requirement B i with LST β i ( ) = · ∑ K i k =1 λ ik λ i β ik ( ). ·
We follow the same approach as in Section 2. First we study the joint queue length distribution of all customer types at polling epochs, then the cycle time distribution, followed by the marginal queue length distribution and waiting time distribution, which depend on the service discipline in the selected queue. Subsection 3.4 provides the first moment of these distributions. The last subsection gives a pseudo-conservation law for this system.
## 3.1 Joint queue length distribution at polling epochs
Equation (2.2) gives the GF of the joint queue length distribution at the beginning of a visit to Q 1 in a non-priority polling system. Equation (2.3) can be used to find the GF of the joint queue length distribution at the beginning of a visit to any other queue. The introduction of multiple priority levels to this polling system will only affect the queue length of a queue that is being served, but it will not affect the queue lengths at polling epochs. At the beginning of a visit to Q i , a waiting type i customer has priority level k with probability λ ik /λ i . We can express the GF of the joint queue length distribution in the polling system with priorities, V prio b i ( z 11 , . . . , z 1 K 1 , . . . , z N 1 , . . . , z NK N ), in terms of the GF of the joint queue length distribution in the polling system without priorities, V b i ( z , . . . , z 1 N ).
## Lemma 3.1
$$V _ { b _ { i } } ^ { p r i o } ( z _ { 1 1 }, \dots, z _ { 1 K _ { 1 } }, \dots, z _ { N 1 }, \dots, z _ { N K _ { N } } ) = V _ { b _ { i } } \left ( \frac { 1 } { \lambda _ { 1 } } \sum _ { k = 1 } ^ { K _ { 1 } } \lambda _ { 1 k } z _ { 1 k }, \dots, \frac { 1 } { \lambda _ { N } } \sum _ { k = 1 } ^ { K _ { N } } \lambda _ { N k } z _ { N k } \right ). \ \ ( 3. 1 )$$
## Proof:
Let X ijk be the number of customers with priority level k present in Q j at the beginning of a visit to Q i ( i, j = 1 , . . . , N ). Let X ij = ∑ K j k =1 X ijk . Since the type j customers with priority k are exactly those customers that arrived since the previous visit beginning (for gated service; for exhaustive: visit ending ) at Q j , we know that for i, j = 1 , . . . , N :
$$& P ( X _ { i j 1 } = n _ { 1 }, X _ { i j 2 } = n _ { 2 }, \dots, X _ { i j K _ { j } } = n _ { K _ { j } } | X _ { i j } = n ) \\ & = \frac { n! } { n _ { 1 }! \cdots n _ { K _ { j } }! } \left ( \frac { \lambda _ { j 1 } } { \lambda _ { j } } \right ) ^ { n _ { 1 } } \cdots \left ( \frac { \lambda _ { j K _ { j } } } { \lambda _ { j } } \right ) ^ { n _ { K _ { j } } },$$
under the condition ∑ K j k =1 n k = n Hence, for i, j = 1 , . . . , N :
.
$$\text{Hence, for } i, j = 1, \dots, N \colon \\ & E | z _ { j 1 } ^ { X _ { i j 1 } } \cdots z _ { j K _ { j } } ^ { X _ { i j K _ { j } } } | X _ { i j } = n ) \\ & = \sum _ { n _ { 1 } = 0 } ^ { \infty } \cdots \sum _ { n _ { K _ { j } } = 0 } ^ { \infty } z _ { j 1 } ^ { n _ { 1 } } \cdots z _ { j K _ { j } } ^ { n _ { K _ { j } } } P ( X _ { i j 1 } = n _ { 1 }, X _ { i j 2 } = n _ { 2 }, \dots, X _ { i j K _ { j } } = n _ { K _ { j } } | X _ { i j } = n ) \\ & = \sum _ { n _ { 1 }, \dots, n _ { K _ { j } } } \frac { n! } { n _ { 1 }! \cdots n _ { K _ { j } }! } \left ( \frac { \lambda _ { j 1 } } { \lambda _ { j } } z _ { j 1 } \right ) ^ { n _ { 1 } } \cdots \left ( \frac { \lambda _ { j K _ { j } } } { \lambda _ { j } } z _ { j K _ { j } } \right ) ^ { n _ { K _ { j } } } \\ & = \left ( \sum _ { k = 1 } ^ { K _ { j } } \frac { \lambda _ { j k } } { \lambda _ { j } } z _ { j k } \right ) ^ { n }.$$
Finally,
$$V _ { b _ { i } } ^ { p r i o } ($$
$$\text{Finally}, & & V _ { b _ { i } } ^ { p r i o } ( z _ { 1 1 }, \dots, z _ { 1 K _ { 1 } }, \dots, z _ { N 1 }, \dots, z _ { N K _ { N } } ) \\ & = \sum _ { n _ { 1 } = 0 } ^ { \infty } \dots \sum _ { n _ { N } = 0 } ^ { N } \prod _ { j = 1 } ^ { N } \left ( \sum _ { k = 1 } ^ { K _ { j } } \frac { \lambda _ { j k } } { \lambda _ { j } } z _ { j k } \right ) ^ { n _ { j } } P ( X _ { i 1 } = n _ { 1 }, \dots, X _ { i N } = n _ { N } ) \\ & = V _ { b _ { i } } \left ( \frac { 1 } { \lambda _ { 1 } } \sum _ { k = 1 } ^ { K _ { 1 } } \lambda _ { 1 k } z _ { 1 k }, \dots, \frac { 1 } { \lambda _ { N } } \sum _ { k = 1 } ^ { K _ { N } } \lambda _ { N k } z _ { N k } \right ), \quad \quad i = 1, \dots, N.$$
## 3.2 Cycle time
The LST of the cycle time distribution is still given by (2.4) (cycle starting at a visit beginning) or (2.5) (cycle starting at a visit completion) if we define λ i := ∑ K i k =1 λ ik and β i ( ) · := ∑ K i k =1 λ ik λ i β ik ( ), · for i = 1 , . . . , N , because the cycle time does not depend on the order of service within each queue.
## 3.3 Marginal queue lengths and waiting times
The LST of the marginal queue length distribution of Q i at an arbitrary epoch depends on the service discipline that is used to serve Q i , and so does the LST of the waiting time distribution. First we study a queue with gated service, afterwards a queue with exhaustive service.
## Gated service
We focus on a particular queue, say Q i , assuming it has gated service. We determine the LST of the waiting time distribution for a type i customer with priority level k , from now on called a 'type ik customer', using the fact that this customer will not be served until the next cycle (starting at a visit beginning to Q i ). The time from the start of the cycle until the arrival of this customer will be called 'past cycle time', denoted by C iP . The residual cycle time will be denoted by C iR . The waiting time of a type ik customer is composed of C iR , the service times of all higher priority customers that arrived during C iP + C iR , and the service times of all type ik customers that have arrived during C iP . Let L H ( T ) be the number of higher priority customers that have arrived during time interval T , and equivalently define L k ( T ) for the type ik customers. The arrival process of higher priority customers is a Poisson process with intensity λ H := ∑ k -1 j =1 λ ij . This intensity equals 0 if k = 1, which corresponds to the highest priority customers in Q i . The service time distribution of these higher priority customers is denoted by B H (suppressing i and k for notational reasons), with LST β H ( ) := · ∑ k -1 j =1 λ ij λ H β ij ( ). ·
Theorem 3.2 The LST of the waiting time distribution of a type ik customer in a polling system with Q i being served according to the gated discipline, is:
$$E \left [ \mathrm e ^ { - \omega W _ { i k } } \right ] = \frac { \gamma _ { i } \left ( \lambda _ { H } ( 1 - \beta _ { H } ( \omega ) ) + \lambda _ { i k } ( 1 - \beta _ { i k } ( \omega ) ) \right ) - \gamma _ { i } \left ( \omega + \lambda _ { H } ( 1 - \beta _ { H } ( \omega ) ) \right ) } { [ \omega - \lambda _ { i k } ( 1 - \beta _ { i k } ( \omega ) ) ] E ( C ) }.$$
## Proof:
$$E \left [ e ^ { - \omega W _ { i k } } \right ] = & E \left [ e ^ { - \omega ( C _ { i R } + \sum _ { j = 1 } ^ { L _ { H } ( C _ { i P } + C _ { i R } ) } B _ { H, j } + \sum _ { j = 1 } ^ { L _ { k } ( C _ { i P } ) } B _ { i k, j } ) } \right ] \\ = & \int _ { t = 0 } ^ { \infty } \int _ { u = 0 } ^ { \infty } \sum _ { m = 0 } ^ { \infty } \sum _ { n = 0 } ^ { \infty } E \left [ e ^ { - \omega ( u + \sum _ { j = 1 } ^ { m } B _ { H, j } + \sum _ { j = 1 } ^ { n } B _ { i k, j } ) } \right ] \\ & \cdot P ( L _ { H } ( C _ { i P } + C _ { i R } ) = m, L _ { k } ( C _ { i P } ) = n ) \, \mathrm d P ( C _ { i P } < t, C _ { i R } < u ) \\ = & \int _ { t = 0 } ^ { \infty } \int _ { u = 0 } ^ { \infty } e ^ { - \omega u } \sum _ { m = 0 } ^ { \infty } \sum _ { n = 0 } ^ { \infty } E \left [ e ^ { - \omega \sum _ { j = 1 } ^ { m } B _ { H, j } } \right ] E \left [ e ^ { - \omega \sum _ { j = 1 } ^ { n } B _ { i k, j } } \right ] \\ & \cdot \frac { ( \lambda _ { H } ( t + u ) ) ^ { m } } { m! } e ^ { - \lambda _ { H } ( t + u ) } \frac { ( \lambda _ { i k } t ) ^ { n } } { n! } e ^ { - \lambda _ { i k } t } \, \mathrm d P ( C _ { i P } < t, C _ { i R } < u )$$
$$= & \int _ { t = 0 } ^ { \infty } \int _ { u = 0 } ^ { \infty } \mathrm e ^ { - t \left ( \lambda _ { H } ( 1 - \beta _ { H } ( \omega ) ) + \lambda _ { i k } ( 1 - \beta _ { i k } ( \omega ) ) \right ) } \mathrm e ^ { - u \left ( \omega + \lambda _ { H } ( 1 - \beta _ { H } ( \omega ) ) \right ) } \mathrm d P ( C _ { i P } < t, C _ { i R } < u ) \\ = & \frac { \gamma _ { i } \left ( \lambda _ { H } ( 1 - \beta _ { H } ( \omega ) ) + \lambda _ { i k } ( 1 - \beta _ { i k } ( \omega ) ) \right ) - \gamma _ { i } \left ( \omega + \lambda _ { H } ( 1 - \beta _ { H } ( \omega ) ) \right ) } { [ \omega - \lambda _ { i k } ( 1 - \beta _ { i k } ( \omega ) ) ] E ( C ) }.$$
$$& J _ { t = 0 } \, J _ { u = 0 } \underset { m = 0 } { \underbrace { n = 0 } } \, \underbrace { \downarrow } & \quad \text{L} \\ & \cdot P ( L _ { H } ( C _ { i P } + C _ { i R } ) = m, L _ { k } ( C _ { i P } ) = n ) \, \mathrm d P ( C _ { i P } < t, C _ { i R } < u ) \\ = & \int _ { t = 0 } ^ { \infty } \int _ { u = 0 } ^ { \infty } e ^ { - \omega u } \sum _ { m = 0 } ^ { \infty } \sum _ { n = 0 } ^ { \infty } E \left [ e ^ { - \omega \sum _ { j = 1 } ^ { m } B _ { H, j } } \right ] E \left [ e ^ { - \omega \sum _ { j = 1 } ^ { n } B _ { i k, j } } \right ] \\ & \cdot \frac { ( \lambda _ { H } ( t + u ) ) ^ { m } } { m! } e ^ { - \lambda _ { H } ( t + u ) } \frac { ( \lambda _ { i k } t ) ^ { n } } { n! } e ^ { - \lambda _ { i k } t } \, \mathrm d P ( C _ { i P } < t, C _ { i R } < u ) \\ \\ = & \int _ { t = 0 } ^ { \infty } \int _ { u = 0 } ^ { \infty } e ^ { - t } \left ( \lambda _ { H } ( 1 - \beta _ { H } ( \omega ) ) + \lambda _ { i k } ( 1 - \beta _ { i k } ( \omega ) ) \right ) e ^ { - u } \left ( \omega + \lambda _ { H } ( 1 - \beta _ { H } ( \omega ) ) \right ) \, \mathrm d P ( C _ { i P } < t, C _ { i R } < u ) \\ & \overset { \chi } { \hat { \lambda } } ( \lambda u ( 1 - \beta u ( \hat { u } ) ) + \lambda _ { u } ( 1 - \beta u ( \hat { u } ) ) \right ) - \chi _ { i } ( \omega + \lambda u ( 1 - \beta u ( \hat { u } ) ) \right )$$
For the last step in the derivation of (3.2) we used
$$E [ \mathrm e ^ { - \omega _ { P } C _ { i P } - \omega _ { R } C _ { i R } } ] = \frac { E [ \mathrm e ^ { - \omega _ { P } C _ { i } } ] - E [ \mathrm e ^ { - \omega _ { R } C _ { i } } ] } { ( \omega _ { R } - \omega _ { P } ) E ( C ) },$$
see, e.g., [6]. The cycle time C i starts with a visit beginning to Q i , and its LST γ i ( ) is given · by (2.4). glyph[square]
We now present an alternative derivation of (3.2) by modelling the system in such a way that the Fuhrmann-Cooper decomposition [15] can be used. Fuhrmann and Cooper [15] showed that the waiting time of a customer in an M/G/ 1 queue with server vacations is the sum of two independent quantities: the waiting time of a customer in a corresponding M/G/ 1 queue without vacations, and the residual vacation time. Conditions that have to be satisfied in order to use this composition, are:
- 1. Poisson arrivals, service times are independent of each other, of the arrival process and of the sequence of vacation periods that precede a certain service time;
- 2. all customers are eventually served;
- 3. customers are served in an order that is independent of their service times;
- 4. service is nonpreemptive;
- 5. the rules that govern when the server begins and ends vacations do not anticipate future jumps of the Poisson arrival process.
In order to use the Fuhrmann-Cooper decomposition to determine E [ e -ωW ik ] , we first model the system as a polling system with N +2 queues ( Q , . . . , Q 1 i -1 , Q H , Q ik , Q L , Q i +1 , . . . , Q N ) and no switch-over times between Q ,Q H ik , and Q L . Q H contains all type i customers with
priority level higher than k (i.e. priority index less than k ), Q ik contains all type i customers of priority level k , and Q L contains all type i customers of priority level lower than k . Lower priority customers arrive according to a Poisson process with intensity λ L := ∑ K i j = +1 k λ ij (again suppressing index i for notational reasons). The service time distribution of these customers is denoted by B L , with LST β L ( ) · := ∑ K i j = +1 k λ ij λ L β ij ( ). · The service discipline of this equivalent system is synchronised gated, which is a more general version of gated. The gates for queues Q ,Q H ik , and Q L are set simultaneously when the server arrives at Q H , but the gates for the other queues are still set at the server's arrival at these queues. From the viewpoint of a type ik customer, and as far as waiting times are concerned, this system is an M/G/ 1 queue with generalised vacations (because service is not exhaustive and the length of a vacation is positively correlated with the length of a visit period). The vacation is the intervisit period of Q ik , which starts with a service to Q L and ends when service of Q H is completed. Fuhrmann and Cooper [15] state that the number of type ik customers in the system at an arbitrary epoch ( L ik ) is the sum of the queue length in an isolated M/G/ 1 queue with only type ik customers ( L ik M/G/ | 1 ), and the number of type ik customers at a random epoch during a vacation ( L ik I | ):
$$E \left [ z ^ { L _ { i k } } \right ] = \frac { ( 1 - \rho _ { i k } ) ( 1 - z ) \beta _ { i k } ( \lambda _ { i k } ( 1 - z ) ) } { \beta _ { i k } ( \lambda _ { i k } ( 1 - z ) ) - z } \cdot \frac { \widetilde { V } _ { c _ { i k } } ( z ) - \widetilde { V } _ { b _ { i k } } ( z ) } { ( 1 - z ) ( E ( L _ { i k | I _ { e n d } } ) - E ( L _ { i k | I _ { b e q i n } } ) ) }.$$
Equation (2.6) has been used to derive the second term at the right-hand side of (3.3), where ˜ V b ik ( z ) denotes the GF of the distribution of the number of type ik customers at the beginning of a visit to Q ik , and ˜ V c ik ( z ) denotes the GF at the completion of a visit to Q ik . The number of type ik customers at the beginning of a visit to Q ik is the number of type ik customers that arrived during the previous cycle (starting with a visit to Q i , which corresponds to a visit beginning to Q H in this model), plus the number of type ik customers that arrived during the service of the type H customers that arrived during the previous cycle. However, because of the synchronised gated service discipline only those type ik customers present at the visit beginning of Q H will be served:
$$\widetilde { V } _ { b _ { i k } } ( z ) & = \gamma _ { i } \Big ( \lambda _ { H } ( 1 - \beta _ { H } ( \lambda _ { i k } ( 1 - z ) ) ) + \lambda _ { i k } ( 1 - z ) \Big ), \\ \widetilde { V } _ { c _ { i k } } ( z ) & = \gamma _ { i } \Big ( \lambda _ { H } ( 1 - \beta _ { H } ( \lambda _ { i k } ( 1 - z ) ) ) + \lambda _ { i k } ( 1 - \beta _ { i k } ( \lambda _ { i k } ( 1 - z ) ) ) \Big ).$$
Furthermore, we have E L ( ik I | end ) -E L ( ik I | begin ) = λ ik E I ( ik ) = λ ik (1 -ρ ik ) E C ( ). Now, to derive the LST of the waiting time W ik , we can apply the distributional form of Little's law [16] to (3.3), because the required conditions are fulfilled for each customer class (including customer types H ik , , and L ): the customers enter the system in a Poisson stream, every customer enters the system and leaves the system one at a time in order of arrival, and for any time t the entry process into the system of customers after time t and the time spent in the system by any customer arriving before time t are independent. This leads to the following expression:
$$E \left [ e ^ { - \omega W _ { i k } } \right ] = & \, \frac { ( 1 - \rho _ { i k } ) \omega } { \omega - \lambda _ { i k } ( 1 - \beta _ { i k } ( \omega ) ) } \\ & \, \cdot \frac { \gamma _ { i } ( \lambda _ { H } ( 1 - \beta _ { H } ( \omega ) ) + \lambda _ { i k } ( 1 - \beta _ { i k } ( \omega ) ) ) - \gamma _ { i } ( \omega + \lambda _ { H } ( 1 - \beta _ { H } ( \omega ) ) ) } { ( 1 - \rho _ { i k } ) \omega E ( C ) }.$$
We recognise the first term on the right-hand side of (3.4) as the LST of the waiting time distribution of an M/G/ 1 queue with only type ik customers. Equation (3.4) can be rewritten to (3.2).
## Exhaustive service
Analysis of the model with exhaustive service requires a different approach. The key observation, made by Fuhrmann and Cooper [15], is that the polling system from the viewpoint of a type i customer is an M/G/ 1 queue with multiple server vacations. The M/G/ 1 queue with priorities and vacations has been extensively analysed by Kella and Yechiali [17]. We use their approach to find the waiting time LST for type ik customers. Kella and Yechiali [17] distinguish between systems with single and multiple vacations, and preemptive resume and nonpreemptive service. The polling system that we consider corresponds to the system with multiple vacations, because the server is never idling, but keeps on switching if the system is empty. In the present paper we do not focus on preemptive resume, but we will give some results on it later in this subsection. First we focus on the case labelled as NPMV (nonpreemptive, multiple vacations) in [17]. We consider the system from the viewpoint of a type ik customer to derive E [e -ωW ik ]. For the waiting time of a type ik customer, the order in which the higher priority customers are served is irrelevant. We will refer to all type i customers with higher priority than k as 'type H customers' arriving according to one Poisson stream with intensity λ H := ∑ k -1 j =1 λ ij . The service time distribution of these customers is denoted by B H , with LST β H ( ) := · ∑ k -1 j =1 λ ij λ H β ij ( ). · A busy period of type H customers is denoted by BP H with LST π H ( ) which is the root of the equation · π H ( ω ) = β H ( ω + λ H (1 -π H ( ω ))). In a similar way we define 'type L ' customers which arrive according to a Poisson stream with intensity λ L := ∑ K i j = +1 k λ ij . The service time of these customers is denoted by B L , with LST β L ( ) := · ∑ K i j = +1 k λ ij λ L β ij ( ). ·
From the viewpoint of a type ik customer and as far as waiting times are considered, a polling system is a nonpriority single server system with multiple vacations. The vacation can either be the intervisit period I i , or a busy period of type H customers, or the service of a type L customer. The LSTs of these three types of vacations are:
$$E [ \mathrm e ^ { - \omega I _ { i } } ] & = \widetilde { V } _ { b _ { i } } ( 1 - \omega / \lambda _ { i } ), \\ E [ \mathrm e ^ { - \omega B P _ { H } } ] & = \pi _ { H } ( \omega ), \\ E [ \mathrm e ^ { - \omega B _ { L } } ] & = \beta _ { L } ( \omega ).$$
The first equality in Equation (3.5) follows immediately from the fact that the number of type i customers at the beginning of a visit to Q i is the number of type i customers that have arrived during the previous intervisit period: ˜ V b i ( z ) = E [e -( λ i (1 -z )) I i ].
The key observation is that an arrival of a type ik customer will always take place within either an I H,ik cycle, or an L H,ik cycle. An I H,ik cycle is a cycle that starts with an intervisit period for Q i , followed by the service of all customers with priority level H or k , and ends at the moment that no type H or ik customers are left in the system. Notice that at the start of the intervisit period, no type H or ik customers were present in the system either. An L H,ik cycle is a similar cycle, but starts with the service of a type L customer. This cycle also ends at the moment that no type H or ik customers are left in the system.
The fraction of time that the system is in an L H,ik cycle is ρ L 1 -ρ H -ρ ik , because type L customers arrive with intensity λ L . Each of these customers will start an L H,ik cycle and the length of an L H,ik cycle equals E B ( L ) 1 -ρ H -ρ ik :
$$E ( L _ { H, i k } \text{ cycle} ) & = E ( B _ { L } ) + \lambda _ { H, i k } E ( B _ { L } ) E ( B P _ { H, i k } ) \\ & = E ( B _ { L } ) + \lambda _ { H, i k } E ( B _ { L } ) \frac { E ( B _ { H, i k } ) } { 1 - \rho _ { H, i k } } \\ & = \left ( 1 + \frac { \rho _ { H } + \rho _ { i k } } { 1 - \rho _ { H } - \rho _ { i k } } \right ) E ( B _ { L } ) = \frac { E ( B _ { L } ) } { 1 - \rho _ { H } - \rho _ { i k } }, \\. \quad \sim \quad \sim \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad.$$
where λ H,ik , B H,ik , BP H,ik , and ρ H,ik denote the intensity, service time, busy period and occupation rate corresponding to type i customers with priority level 1 , . . . , k .
The fraction of time that the system is in an I H,ik cycle, is 1 -ρ L 1 -ρ H -ρ ik = 1 -ρ i 1 -ρ H -ρ ik . This result can also be obtained by using the argument that the fraction of time that the system is in an intervisit period is the fraction of time that the server is not serving Q i , which is equal to 1 -ρ i . A cycle which starts with such an intervisit period and stops after the service of all customers with priority levels 1 , . . . , k that arrived during the intervisit period and their descendants of priority levels 1 , . . . , k , has mean length E I ( i ) + λ H,ik E I ( i ) E ( BP H,ik ) = E I ( i ) 1 -ρ H -ρ ik . This also leads to the conclusion that 1 -ρ i 1 -ρ H -ρ ik is the fraction of time that the system is in an I H,ik cycle.
A type ik customer arriving during an I H,ik cycle views the system as a nonpriority M/G/ 1 queue with multiple server vacations. The service requirement in this system is not the service requirement of a type ik customer, B ik , but the time that is required to serve a type ik customer and all higher priority customers that arrive during this service until no higher priority customers are present in the system. We will elaborate on this so-called completion time later. The vacation is the intervisit time I i , plus the service times of all type H customers that have arrived during that intervisit time and their type H descendants. We denote this extended intervisit time by I ∗ i with LST
$$\widetilde { I } _ { i } ^ { * } ( \omega ) = \widetilde { I } _ { i } ( \omega + \lambda _ { H } ( 1 - \pi _ { H } ( \omega ) ) ).$$
The mean of I ∗ i equals E I ( ∗ i ) = E I ( i ) 1 -ρ H .
A type ik customer arriving during an L H,ik cycle views the system as a nonpriority M/G/ 1 queue with multiple server vacations B L plus the busy periods generated by all type H customers that have arrived during the service time of the type L customer. We will denote this extended service time by B ∗ L with mean E B ( ∗ L ) = E B ( L ) 1 -ρ H , and LST
$$\beta _ { L } ^ { * } ( \omega ) = \beta _ { L } ( \omega + \lambda _ { H } ( 1 - \pi _ { H } ( \omega ) ) ).$$
We also have to take into account that a busy period of type ik customers might be interrupted by the arrival of type H customers. Therefore, in the alternative system that we are considering, the service time of a type ik customer is equal to B ik plus the service times of all type H customers that arrive during this service time, and all of their type H descendants. The LST of the distribution of this extended service time B ∗ ik is
$$\beta _ { i k } ^ { * } ( \omega ) = \beta _ { i k } ( \omega + \lambda _ { H } ( 1 - \pi _ { H } ( \omega ) ) ).$$
This extended service time is often called completion time in the literature, cf. [25]. In this alternative system, the mean service time of these customers equals E B ( ∗ ik ) = E B ( ik ) 1 -ρ H . The fraction of time that the system is serving these customers is ρ ∗ ik = ρ ik 1 -ρ H = 1 -1 -ρ H -ρ ik 1 -ρ H .
Theorem 3.3 The LST of the waiting time distribution of a type ik customer in polling system with Q i being served exhaustively, is:
$$E [ e ^ { - \omega W _ { i k } } ] = \frac { ( 1 - \rho _ { i k } ^ { * } ) \omega } { \omega - \lambda _ { i k } ( 1 - \beta _ { i k } ^ { * } ( \omega ) ) } \cdot \left [ \frac { 1 - \rho _ { i } } { 1 - \rho _ { H } - \rho _ { i k } } \cdot \frac { 1 - \widetilde { I } _ { i } ^ { * } ( \omega ) } { \omega E ( I _ { i } ^ { * } ) } + \frac { \rho _ { L } } { 1 - \rho _ { H } - \rho _ { i k } } \cdot \frac { 1 - \beta _ { L } ^ { * } ( \omega ) } { \omega E ( B _ { L } ^ { * } ) } \right ].$$
## Proof:
The system has been modelled as an M/G/ 1 queue with customers that have service requirement B ∗ ik , and multiple server vacations that have length I ∗ i with probability 1 -ρ i 1 -ρ H -ρ ik , and length B ∗ L with probability ρ L 1 -ρ H -ρ ik . Equation (3.6) follows immediately from application of the Fuhrmann-Cooper decomposition to this system. glyph[square]
Remark 3.4 Equation (3.6) can be rewritten in several forms, which lead to different interpretations. A compact form is
$$E [ e ^ { - \omega W _ { i k } } ] = \frac { \frac { 1 - \rho _ { i } } { E ( I _ { i } ) } [ 1 - \widetilde { I } _ { i } ( \omega + \lambda _ { H } - \lambda _ { H } \pi _ { H } ( \omega ) ) ] + \lambda _ { L } [ 1 - \beta _ { L } ( \omega + \lambda _ { H } - \lambda _ { H } \pi _ { H } ( \omega ) ) ] } { \lambda _ { i k } \beta _ { i k } ( \omega + \lambda _ { H } - \lambda _ { H } \pi _ { H } ( \omega ) ) - \lambda _ { i k } + \omega }, \quad ( 3. 7 )$$
which is the form used by Kella and Yechiali (see [17], Section 4.1).
Another form, which is more suitable for interpretation, is
$$\text{other form, which is more suitable for interpretation, is} \\ E [ e ^ { - \omega W _ { i k } } ] & = \frac { ( 1 - \rho _ { i k } ^ { * } ) \omega } { \omega - \lambda _ { i k } ( 1 - \beta _ { i k } ^ { * } ( \omega ) ) } \\ & \cdot \left [ \frac { 1 - \rho _ { i } } { 1 - \rho _ { H } - \rho _ { i k } } \cdot \frac { 1 - \widetilde { I } _ { i } ( \omega + \lambda _ { H } ( 1 - \pi _ { H } ( \omega ) ) ) } { ( \omega + \lambda _ { H } ( 1 - \pi _ { H } ( \omega ) ) ) E ( I _ { i } ) } \\ & \quad + \frac { \rho _ { L } } { 1 - \rho _ { H } - \rho _ { i k } } \cdot \frac { 1 - \beta _ { L } ( \omega + \lambda _ { H } ( 1 - \pi _ { H } ( \omega ) ) ) } { ( \omega + \lambda _ { H } ( 1 - \pi _ { H } ( \omega ) ) ) E ( B _ { L } ) } \right ] \\ & \cdot \left [ ( 1 - \rho _ { H } ) + \rho _ { H } \cdot \frac { 1 - \pi _ { H } ( \omega ) } { \omega E ( B P _ { H } ) } \right ]. \\ \\ \text{he first term on the right hand side of } ( 3 \, 8 \, \sl i \, the \, I \, T \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \t, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, \t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \. t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t \, t.$$
The first term on the right hand side of (3.8) is the LST of the waiting time distribution of a customer in a nonpriority M/G/ 1 queue with customers arriving with intensity λ ik and having service requirement LST β ∗ ik ( ), · i.e. the completion time of a type ik customer. The second term indicates that with probability 1 -ρ i 1 -ρ H -ρ ik a type ik customer has to wait a residual intervisit time, plus the busy periods of all higher priority customers that arrive during this residual intervisit time. With probability ρ L 1 -ρ H -ρ ik the customer has to wait a residual service time of a type L customer, plus the busy periods of all higher priority customers that arrive during this residual intervisit time. The last term in (3.8) indicates that with probability ρ H a customer also has to wait for the residual length of a busy period of higher priority customers.
Remark 3.5 Substitution of (2.11) in (3.7) leads to a different expression for E [e -ωW ik ]:
$$& \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \\ & E [ e ^ { - \omega W _ { i k } } ] \\ & = \frac { 1 - \gamma _ { i } ^ { * } ( \omega - \lambda _ { i k } ( 1 - \beta _ { i k } ( \omega + \lambda _ { H } ( 1 - \pi _ { H } ( \omega ) ) ) ) - \lambda _ { L } ( 1 - \beta _ { L } ( \omega + \lambda _ { H } ( 1 - \pi _ { H } ( \omega ) ) ) ) } { [ \omega - \lambda _ { i k } ( 1 - \beta _ { i k } ( \omega + \lambda _ { H } ( 1 - \pi _ { H } ( \omega ) ) ) ) ] E ( C ) } \\ & \quad + \frac { \lambda _ { L } [ 1 - \beta _ { L } ( \omega + \lambda _ { H } ( 1 - \pi _ { H } ( \omega ) ) ) ] } { \omega - \lambda _ { i k } ( 1 - \beta _ { i k } ( \omega + \lambda _ { H } ( 1 - \pi _ { H } ( \omega ) ) ) ) }. \\ & \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot$$
For the lowest priority customers this expression simplifies to
$$E [ \mathrm e ^ { - \omega W _ { i K _ { i } } } ] & = \frac { 1 - \gamma _ { i } ^ { * } \left ( \omega - \lambda _ { i K _ { i } } ( 1 - \beta _ { i K _ { i } } ( \omega + \lambda _ { H } ( 1 - \pi _ { H } ( \omega ) ) ) ) \right ) } { \left ( \omega - \lambda _ { i K _ { i } } ( 1 - \beta _ { i K _ { i } } ( \omega + \lambda _ { H } ( 1 - \pi _ { H } ( \omega ) ) ) ) \right ) E ( C ) } \\ & = E [ \mathrm e ^ { - ( \omega - \lambda _ { i K _ { i } } ( 1 - \beta _ { i K _ { i } } ( \omega + \lambda _ { H } ( 1 - \pi _ { H } ( \omega ) ) ) ) ) C _ { i, r e s } ^ { * } } ],$$
where type H customers in this case are all type i customers except for those with the lowest priority.
Remark 3.6 In the derivation of Equation (3.6) we followed the approach of Kella and Yechiali [17], but restricted ourselves to the nonpreemptive situation only. Using the decomposition (3.8), it only requires a small extra step to obtain the waiting time LST in a system with preemptive resume . The term 1 -β L ( ω + λ H (1 -π H ( ω ))) ( ω + λ H (1 -π H ( ω ))) E B ( L ) on the third line is the LST of a residual service time of a type L customer, plus the busy periods of all type H customers that arrive during this residual type L service time. If type ik customers are allowed to preempt the service of a type L customer, this term simply vanishes from the waiting time LST. This leads to the following expression for the waiting time LST of a type ik customer if the service policy is preemptive resume :
$$E [ e ^ { - \omega W _ { i k } } ] & = \frac { ( 1 - \rho _ { i k } ^ { * } ) \omega } { \omega - \lambda _ { i k } ( 1 - \beta _ { i k } ^ { * } ( \omega ) ) } \cdot \left [ ( 1 - \rho _ { H } ) + \rho _ { H } \cdot \frac { 1 - \pi _ { H } ( \omega ) } { \omega E ( B P _ { H } ) } \right ] \\ & \quad \cdot \left [ \frac { 1 - \rho _ { i } } { 1 - \rho _ { H } - \rho _ { i k } } \cdot \frac { 1 - \widetilde { I } _ { i } ( \omega + \lambda _ { H } ( 1 - \pi _ { H } ( \omega ) ) ) } { ( \omega + \lambda _ { H } ( 1 - \pi _ { H } ( \omega ) ) ) E ( I _ { i } ) } + \frac { \rho _ { L } } { 1 - \rho _ { H } - \rho _ { i k } } \right ].$$
Note that the sojourn time of a type ik customer equals the waiting time plus a completion time in this situation. Equation (3.10) is in agreement with the result in [17].
We will refrain from mentioning the GFs of the marginal queue length distributions here, because they can be obtained by applying the distributional form of Little's law as we have done before.
## 3.4 Moments
As mentioned in Section 2.4, we do not focus on moments in this paper, and we only mention the mean waiting times of type ik customers. The formulas in this subsection can also be obtained using MVA, as shown in [27], but we have obtained them by differentiating (3.2),
(3.6), and (3.10) respectively:
Q i gated:
Q i exhaustive (nonpreemptive):
Q
i
exhaustive
(preemptive resume):
$$E ( W _ { i k } ) = ( 1 + 2 \rho _ { H } + \rho _ { i k } ) E ( C _ { i, r e s } )$$
$$= ( 1 + 2 \sum _ { j = 1 } ^ { k - 1 } \rho _ { i j } + \rho _ { i k } ) E ( C _ { i, r e s } ),$$
$$E ( W _ { i k } ) = \frac { \sum _ { j = 1 } ^ { K _ { i } } \stackrel { \cdot } { \rho _ { i j } } E ( B _ { i j, r e s } ) + ( 1 - \rho _ { i } ) E ( I _ { i, r e s } ) } { ( 1 - \sum _ { j = 1 } ^ { k - 1 } \rho _ { i j } ) ( 1 - \sum _ { j = 1 } ^ { k } \rho _ { i j } ) },$$
$$E ( W _ { i k } ) = \frac { \sum _ { j = 1 } ^$$
Differentiation of (3.9) leads an to alternative expression for the mean waiting time of a type ik customer if Q i is served exhaustively:
$$\begin{array} {$$
Notice that the cycle time in this exhaustive case starts at a visit completion to Q i .
## 3.5 A pseudo-conservation law
Let V be the amount of work in the polling system at an arbitrary epoch. Boxma and Groenendijk [5] show that V can be written as the sum of two independent random variables:
$$V \stackrel { d } { = } V _ { M / G / 1 } + V _$$
with V M/G/ 1 the amount of work at an arbitrary epoch in the corresponding M/G/ 1 queue, and V I the amount of work in the system at an arbitrary epoch in an intervisit period when the server is idling. It is shown in [5] that relation (3.11) leads to a pseudo-conservation law for nonpriority polling systems which is not specified in more detail here. This law has been generalised for polling systems with multiple priority levels in each queue in [22] and [13]:
$$\circ \cdots \cdots \cdots \cdots \cdots \cdots \cd$$
where S = ∑ N i =1 S i and Z ii is the amount of work left behind by the server at Q i at the completion of a visit. Equation (3.12) is called a pseudo-conservation law because it indicates that the weighted sum of the mean waiting times ∑ N i =1 ∑ K i k =1 ρ ik E W ( ik ) is constant, except for the term Z ii which depends on the service discipline. For gated service, E Z ( ii ) = ρ E C 2 i ( ). For exhaustive service, E Z ( ii ) = 0 because Q i is empty at the completion of a visit to this queue. For globally gated service, which is discussed in the next section, E Z ( ii ) = ρ i ( E C ( ) ∑ i j =1 ρ j + ∑ i -1 j =1 E S ( j ) ) .
## 4 Globally gated service
In this section we discuss a polling model with N queues ( Q , . . . , Q 1 N ) and K i priority classes in Q i with globally gated service. For this service discipline, only customers that were present when the server started its visit to Q 1 are served. This feature makes the model exactly the same as a nonpriority polling model with ∑ N i =1 K i queues ( Q 11 , . . . , Q 1 K 1 , . . . , Q N 1 , . . . , Q NK N ).
It was observed in Section 2 that, although this system does not satisfy Property 2.1, it does satisfy Property 2.2 which implies that we can still follow the same approach as in the previous sections.
## 4.1 Joint queue length distribution at polling epochs
We define the beginning of a visit to Q 1 as the start of a cycle, since this is the moment that determines which customers will be served during the next visits to the queues. Arriving customers will always be served in the next cycle, so the offspring GFs are:
$$& \cdot \quad \cdot \quad \cdot \quad \cdot \quad \$$
The N immigration functions are:
$$g ^ { ( i ) } ( z _ { 1 1 }, \dots, z _ { 1 K _ { 1 } }, \dots, z _ { N 1 }, \dots, z _ { N K _ { N } } ) = \sigma _ { i } ( \sum _ { j = 1 } ^ { N } \sum _ { l = 1 } ^ { K _ { j } } \lambda _ { j l } ( 1 - z _ { j l } ) ), \quad \quad i = 1, \dots, N.$$
Using these definitions, the formula for the GF of the joint queue length distribution at the beginning of a cycle is similar to the one found in Section 2:
$$P _ { 1 } ( z _ { 1 1 }, \dots, z _ { 1 K _ { 1 } }, \dots, z _ { N 1 }, \dots, z _ { N K _ { N } } ) = \prod _ { n = 0 } ^ { \infty } g ( f _ { n } ( z _ { 1 1 }, \dots, z _ { 1 K _ { 1 } }, \dots, z _ { N 1 }, \dots, z _ { N K _ { N } } ) ). \quad ( 4. 1 )$$
Notice that in a system with globally gated service it is possible to express the joint queue length distribution at the beginning of a cycle in terms of the cycle time LST, since all customers that are present at the beginning of a cycle are exactly all of the customers that have arrived during the previous cycle:
$$P _ { 1 } ( z _ { 1 1 }, \dots, z _ { 1 K _ { 1 } }, \dots, z _ { N 1 }, \dots, z _ { N K _ { N } } ) = \gamma _ { 1 } ( \sum _ { i = 1 } ^ { N } \sum _ { k = 1 } ^ { K _ { i } } \lambda _ { i k } ( 1 - z _ { i k } ) ). \quad \quad ( 4. 2 )$$
## 4.2 Cycle time
Since only those customers that are present at the start of a cycle, starting at Q 1 , will be served during this cycle, the LST of the cycle time distribution is
$$\gamma _ { 1 } ( \omega ) = \prod _ { i = 1 } ^ { N } \sigma _ { i } ( \omega ) P _ { 1 } ( \beta _ { 1 1 } ( \omega ), \dots, \beta _ { 1 K _ { 1 } } ( \omega ), \dots, \beta _ { N 1 } ( \omega ), \dots, \beta _ { N K _ { N } } ( \omega ) ).$$
Substitution of (4.2) into this expression gives us the following relation:
$$\gamma _ { 1 } ( \omega ) = \prod _ { i = 1 } ^ { N } \sigma _ { i } ( \omega ) \, \gamma _ { 1 } ( \sum _ { i = 1 } ^ { N } \sum _ { k = 1 } ^ { K _ { i } } \lambda _ { i k } ( 1 - \beta _ { i k } ( \omega ) ) ).$$
Boxma, Levy and Yechiali [6] show (for a model without priorities) that this relation leads to the following expression for the cycle time LST:
$$\gamma _ { 1 } ( \omega ) = \prod _ { i = 0 } ^$$
where σ ( ) = · ∏ N i =1 σ i ( ), and · δ ( ) i ( ω ) is recursively defined as follows:
$$\cdot \, \_ \quad & \delta ^ { ( 0 ) } ( \omega$$
## 4.3 Marginal queue lengths and waiting times
The expression for E (e -ωW ik ) can be obtained by the method used in Section 3.3:
$$\begin{array} { l l$$
We can use the distributional form of Little's law to determine the LST of the marginal queue length distribution of Q ik . Substituting ω := λ ik (1 -z ) in (4.4) yields an expression for E z [ L ik ] , the GF of the distribution of the number of type ik customers at an arbitrary epoch.
## 4.4 Moments
We can obtain E W ( ik ) by differentiation of (4.4):
$$\cdot \cdot \cdot \cdot \cdot \cdot \cdot \cd$$
The interpretation of this expression is that the mean waiting time of a type ik customer in a globally gated polling system consists of the mean residual cycle time, the mean switch-over times S 1 + · · · + S i -1 , the mean service times of all type 1 , . . . , i -1 customers that arrive during the past cycle time plus the residual cycle time, the mean service times of all type i customers with higher priority level than k that arrive during the past cycle time plus the mean residual cycle time, and the mean service times of all type i customers of priority level k that have arrived during the past cycle time. Notice that the mean past cycle time is equal to the mean residual cycle time.
## 5 Numerical example
Consider a polling system with two queues, and assume exponential service times and switchover times. Suppose that λ 1 = 6 10 , λ 2 = 2 10 , E ( B 1 ) = E B ( 2 ) = 1 , E ( S 1 ) = E S ( 2 ) = 1. The workload of this polling system is ρ = 8 10 . This example is extensively discussed in [29] where MVA is used to compute mean waiting times and mean residual cycle times for the gated and exhaustive service disciplines, and in [1], where it is shown that the performance of this system can be improved by dividing jobs into two categories (small and large service times) and giving higher priority to jobs with smaller service times. In this example we illustrate how much further the performance of this system can be improved by increasing the number of priority levels in Q 1 . Let K 1 be the number of priority levels in Q 1 . We define thresholds t 1 , . . . , t K 1 -1 that divide the jobs arriving in Q 1 into K 1 classes: jobs with a service time less than t 1 receive highest priority, jobs with a service time between t 1 and t 2 receive second highest priority, and so on. The thresholds t 1 , . . . , t K 1 -1 are determined in such a way that the overall mean waiting time for customers in Q 1 is minimised. The values of these thresholds depend on the service discipline and are discussed in [27]. Figures 1, 2, and 3 show the overall mean waiting time for customers in Q 1 versus the number of priority levels K 1 for gated, exhaustive and globally gated service respectively. Notice that K 1 = 1 corresponds to a system without priorities in which customers are served FCFS. Both plots indicate that most of the improvement is obtained for values of K 1 up to 3 or 4. Creating more than 4 priority levels hardly improves this system anymore. The situation K 1 →∞ corresponds to a system with shortest-job-first (SJF) policy, which results in a mean waiting time of 10.38 for gated service, 3.53 for exhaustive service, and 9.75 for globally gated service. The dotted horizontal lines in Figures 1, 2 and 3 indicate these values.
From Figures 1 - 3 we can conclude that, as far as overall mean waiting time is concerned, the introduction of priority levels significantly improves the performance of this polling system of two queues. An interesting question is whether the introduction of multiple priority levels
still significantly improves the performance of a polling system consisting of more than two queues. In order to answer this question we compare polling systems with identical total occupation rate and identical total mean switch-over time. We only vary the number of queues N and, in order to keep the total occupation rate constant, the arrival intensities λ , i i = 1 , . . . , N . All systems are symmetrical, which means that in each system λ i = Λ /N,ρ i = ρ/N,E B ( i ) = β, E S ( i ) = E S /N,i ( ) = 1 , . . . , N . For this example we use Λ = 8 10 , B i exponentially distributed with mean β = 1, and deterministic switch-over times with E S ( ) = 2. Figures 4 (gated) and 5 (exhaustive) show the mean waiting time (which is identical for each queue, because of symmetry) versus the number of queues in the system. In each figure the mean waiting time is plotted for both the FCFS and the SJF policy. The mean waiting time of SJF provides a lower bound for what can be achieved by introducing priority levels. A conclusion that can be drawn from both figures is that the relative improvement by adding priority levels decreases quickly when the number of queues increases. This suggests that the introduction of priority levels has less influence in a system with more queues, because the server spends a greater part of each cycle serving other queues. Another typical feature that is illustrated in Figures 4 and 5 is that the mean waiting time of an arbitrary customer decreases when the number of queues increases in a gated polling system, whereas it increases in an exhaustive polling system. This can be verified using the pseudo-conservation law. For a polling system without priorities with FCFS service at each queue, Equation (3.12) reduces to
$$\sum _ { i = 1 } ^ { N } \rho _ { i } E ( W _$$
The Z ii ( i = 1 , . . . , N ) depend on the service discipline that is used and are discussed in Subsection 3.5. For an exhaustive polling system, E Z ( ii ) = 0, so if we substitute ρ i = ρ/N , we obtain
$$E ( W _ { i } ) = \frac { \rho } { 1 - \rho } \frac { E ( B _ { i } ^ { 2 } ) } { 2 E ( B _ { i } ) } + \frac { E ( S ^ { 2 } ) } { 2 E ( S ) } + \left ( 1 - \frac { 1 } { N } \right ) \frac { \rho E ( S ) } { 2 ( 1 - \rho ) },$$
which indeed gets larger as N increases. For gated service, we have E Z ( ii ) = ρ E S 2 i ( ) 1 -ρ , which results in
$$E ( W _ { i } ) = \frac { \rho } { 1 - \rho } \frac { E ( B _ { i } ^ { 2 } ) } { 2 E ( B _ { i } ) } + \frac { E ( S ^ { 2 } ) } { 2 E ( S ) } + \left ( 1 + \frac { 1 } { N } \right ) \frac { \rho E ( S ) } { 2 ( 1 - \rho ) }.$$
This expression indeed gets smaller as N increases. In practice this means that mean waiting times can be reduced when gated service is used, by creating more queues and assign arriving customers to a random queue. It is however questionable whether this can be realised without changing the total switch-over time distribution. Notice that for N → ∞ the mean waiting time in an exhaustive symmetric polling system is equal to the mean waiting time in a gated symmetric polling system.
Figure 1: Overall mean waiting time of customers in Q 1 in the gated polling system, versus number of priority levels K 1 .
Figure 2: Overall mean waiting time of customers in Q 1 in the exhaustive polling system, versus number of priority levels K 1 .
Figure 3: Overall mean waiting time of customers in Q 1 in the globally gated polling system, versus number of priority levels K 1 .
Figure 4: Mean waiting time E W ( i ) in the gated polling system, versus number of queues.
Figure 5: Mean waiting time E W ( i ) in the exhaustive polling system, versus number of queues.
## References
- [1] M. A. A. Boon, I. J. B. F. Adan, and O. J. Boxma. A two-queue polling model with two priority levels in the first queue. Eurandom report 2008-016, Eurandom , 2008. (to appear in Discrete Event Dynamic Systems, 2009)
- [2] S. C. Borst. Polling Systems , volume 115 of CWI Tracts . 1996.
- [3] S. C. Borst and O. J. Boxma. Polling models with and without switchover times. Operations Research , 45(4):536 - 543, 1997.
- [4] O. J. Boxma. Workloads and waiting times in single-server systems with multiple customer classes. Queueing Systems , 5:185-214, 1989.
- [5] O. J. Boxma and W. P. Groenendijk. Pseudo-conservation laws in cyclic-service systems. Journal of Applied Probability , 24(4):949-964, 1987.
- [6] O. J. Boxma, H. Levy, and U. Yechiali. Cyclic reservation schemes for efficient operation of multiple-queue single-server systems. Annals of Operations Research , 35(3):187-208, 1992.
- [7] O. J. Boxma, J. Bruin, and B. H. Fralix. Waiting times in polling systems with various service disciplines. Performance Evaluation , 66:621-639, 2009.
- [8] G. L. Choudhury and W. Whitt. Computing distributions and moments in polling models by numerical transform inversion. Performance Evaluation , 25(4):267-292, 1996.
- [9] M. Cicin-Sain, C. E. M. Pearce, and J. Sunde. On the application of a polling model with non-zero walk times and priority processing to a medical emergency-room environment.
Proceedings of the 23rd International Conference on Information Technology Interfaces, 2001 , 1:49-56, 2001.
- [10] J. W. Cohen. The Single Server Queue . North-Holland, Amsterdam, revised edition, 1982.
- [11] R. B. Cooper and G. Murray. Queues served in cyclic order. The Bell System Technical Journal , 48(3):675-689, 1969.
- [12] M. Eisenberg. Queues with periodic service and changeover time. Operations Research , 20(2):440-451, 1972.
- [13] L. Fournier and Z. Rosberg. Expected waiting times in polling systems under priority disciplines. Queueing Systems , 9(4):419-439, 1991.
- [14] S. W. Fuhrmann. Performance analysis of a class of cyclic schedules. Technical memorandum 81-59531-1, Bell Laboratories, March 1981.
- [15] S. W. Fuhrmann and R. B. Cooper. Stochastic decompositions in the M/G/ 1 queue with generalized vacations. Operations Research , 33(5):1117-1129, 1985.
- [16] J. Keilson and L. D. Servi. The distributional form of Little's Law and the FuhrmannCooper decomposition. Operations Research Letters , 9(4):239-247, 1990.
- [17] O. Kella and U. Yechiali. Priorities in M/G/ 1 queue with server vacations. Naval Research Logistics , 35:23-34, 1988.
- [18] A. Khamisy, E. Altman, and M. Sidi. Polling systems with synchronization constraints. Annals of Operations Research , 35:231 - 267, 1992.
- [19] H. Levy and M. Sidi. Polling systems: applications, modeling, and optimization. IEEE Transactions on Communications , 38:1750-1760, 1990.
- [20] M. P. Quine. The multi-type Galton-Watson process with immigration. Journal of Applied Probability , 7(2):411-422, 1970.
- [21] J. A. C. Resing. Polling systems and multitype branching processes. Queueing Systems , 13:409 - 426, 1993.
- [22] S. Shimogawa and Y. Takahashi. A pseudo-conservation law in a cyclic-service system with priority classes. IEICE Research Report , (IN88-86):13-18, 1988.
- [23] L. Tak´cs. Two queues attended by a single server. a Operations Research , 16(3):639-650, 1968.
- [24] H. Takagi. Queuing analysis of polling models. ACM Computing Surveys (CSUR) , 20: 5-28, 1988.
- [25] H. Takagi. Priority queues with setup times. Operations Research , 38(4):667-677, 1990.
- [26] V. M. Vishnevskii and O. V. Semenova. Mathematical methods to study the polling systems. Automation and Remote Control , 67(2):173-220, 2006.
- [27] A. Wierman, E. M. M. Winands, and O. J. Boxma. Scheduling in polling systems. Performance Evaluation , 64:1009-1028, 2007.
- [28] E. M. M. Winands. Polling, Production & Priorities . PhD thesis, Eindhoven University of Technology, 2007.
- [29] E. M. M. Winands, I. J. B. F. Adan, and G.-J. van Houtum. Mean value analysis for polling systems. Queueing Systems , 54:35-44, 2006. | 10.1016/j.peva.2010.01.002 | [
"Marko Boon",
"Ivo Adan",
"Onno Boxma"
] | 2014-08-01T09:43:15+00:00 | 2014-08-01T09:43:15+00:00 | [
"math.PR",
"cs.PF"
] | A Polling Model with Multiple Priority Levels | In this paper we consider a single-server cyclic polling system. Between
visits to successive queues, the server is delayed by a random switch-over
time. The order in which customers are served in each queue is determined by a
priority level that is assigned to each customer at his arrival. For this
situation the following service disciplines are considered: gated, exhaustive,
and globally gated. We study the cycle time distribution, the waiting times for
each customer type, the joint queue length distribution of all priority classes
at all queues at polling epochs, and the steady-state marginal queue length
distributions for each customer type. |
1408.0283v3 | ## Clocks around Sgr A*
## Raymond Ang´ elil 1 /star , Prasenjit Saha 2
1 Institut f¨r Rechnergest¨tztewissenschaften, Univers it¨t Z¨rich, Winterthurerstrasse 190, 8057 Z¨rich, Swi zerland u u a u u t 2 Physik-Institut, Universit¨t Z¨rich, Winterthurerstr asse 190, 8057 Z¨rich, Switzerland a u u
21 August 2014
## ABSTRACT
The S stars near the Galactic centre and any pulsars that may be on similar orbits, can be modelled in a unified way as clocks orbiting a black hole, and hence are potential probes of relativistic effects, including black hole spin. The high eccentricities of many S stars mean that relativistic effects peak strongly around pericentre; for example, orbit precession is not a smooth effect but almost a kick at pericentre. We argue that concentration around pericentre will be an advantage when analysing redshift or pulse-arrival data to measure relativistic effects, because cumulative precession will be drowned out by Newtonian perturbations from other mass in the Galactic-centre region. Wavelet decomposition may be a way to disentangle relativistic effects from Newton perturbations. Assuming a plausible model for Newtonian perturbations on S2, relativity appears to be strongest in a two-year interval around pericentre, in wavelet modes of timescale ≈ 6 months.
## 1 CLOCKS AS PROBES OF GRAVITY
An orbiting clock as a probe of general relativity is familiar from binary pulsars (Taylor 1994; Kramer et al. 2004) and from global navigation satellites (Ashby 2003). In the coming years, a new class of objects may join these.
The milliparsec region of the Galactic centre is home to a compact mass of ∼ 4 · 10 6 M /circledot at Sgr A*, presumably a black hole. This is known from a population of stars which orbit it at speeds up to a few percent of light, as shown by astrometric and spectroscopic observations (Sch¨del et al. 2002; Eisenhauer e t al. 2003; o Eckart et al. 2005; Martins et al. 2008; Gillessen et al. 2009a,b; Ghez et al. 2008; Meyer et al. 2012). Dozens of these 'S'-stars have been observed, and it is expected that many others with orbits even closer to Sgr A* await discovery by the next generation of telescopes such as the European Extremely Large Telescope (EELT). An even more exciting prospect is the possibility of pulsars near the black hole. A pulsar has recently been discovered in the region (Rea et al. 2013), and population models argue that there should be a few pulsars with periods < 1 yr and observable with the Square Kilometre Array (Cordes & Lazio 1997; Kramer et al. 2000; Pfahl & Loeb 2004; Macquart et al. 2010). Since the gravitational radius of the black hole is
$$\frac { G M } { c ^ { 2 } } \simeq 2 0 \, \text{light-sec}, \quad \ \ ( 1 ) \quad \ \ \ a \\ \quad \ \ a$$
and recalling that a parsec is 1 0 . × 10 8 light-seconds, we see that the S stars are at r ∼ 10 4 in relativistic units. This make their orbits the most relativistic of all known ballistic orbits, more than any known binary system, and far more than Mercury or artificial satellites (see, for example, Figure 2 in Ang´ elil, Saha & Merritt 2010) and provides incentive to search for relativistic effects.
Progress has been made to directly observe the event horizon silhouette(Doeleman 2010), and these two kinds of observations could potentially complement each other(Broderick et al. 2014).
An S star, or a pulsar on a similar orbit, can be considered as a clock in orbit around a black hole. The clock moves on a time-
Figure 1. Illustration of the basic scenario: an observer watches a clock in a nearly Keplerian orbit around black hole, but relativity changes both the clock's orbit, and the paths that the signals take to reach the observer. The orbits are initialised at zero proper time and then integrated forward and back, but observer time lags by about a day, because of the placement of the observer. Note how the tick rates are very similar until pericentre passage, but then the relativistic orbit appears to get ahead a little bit - that is pericentre precession. Spin effects are also included in the calculation, but too small to see at the resolution of this figure. The eccentricity of this orbit is e = 0 6 . , the semi-major axis is a = 0 06 . mpc, and the inclination I = 45 ◦ . For this and all other numerical results in this paper, we set the black hole mass to 20 light seconds, cf. equation (1) .
like geodesic and ticks at equal intervals of its proper time τ c = nν c . With each tick, the clock sends out photons in all directions on null geodesics. Some of these photons reach an observer, who records their arrival times as t a ( n ) . The observer can also choose to
calculate the frequency by taking the derivative of the arrival times with respect to the proper time of emission:
$$\nu _ { a } = \frac { 1 } { t _ { a } ( n + 1 ) - t _ { a } ( n ) } = \nu _ { c } \left ( \frac { d t _ { a } } { d \tau _ { e } } \right ) ^ { - 1 }. \quad \ ( 2 ) \quad \ \ p l a$$
Figure 1 shows an example of what might be measured. For pulsars, the ticks are simply the pulses. For an S star, there are no such discrete ticks, but the clock model still applies, because
$$c \ln ( \nu _ { c } / \nu _ { a } ) \quad \quad ( 3 ) \quad \colon$$
has the interpretation of redshift as usually measured from spectroscopy 1 . It is not essential for the observer to know the intrinsic frequency in advance, since ν c just introduces an additive constant into equation (3). The important thing is to be able to calculate t a ( n ) , for which one has to to compute time-like geodesics (orbits) and null geodesics (light paths), and solve the boundary-value problem for null geodesics from clock to observer. A moving observer can also be allowed for, if desired.
Many different relativistic effects are, in principle, measurable from a clock orbiting a black hole. First, the spacetime around the black hole dilates the clock time. Then, every term in the metric affects both the orbit of the clock and the photons from the clock, and imprints itself on the observables in its own distinctive way. The best-known examples are pericentre precession and the Shapiro time delay; the former concerns orbits while the latter influences light paths. Another difference is that precession is cumulative over many orbits, whereas the Shapiro delay is transient and does not get larger as one observes more orbits. In the solar system and for binary pulsars, both cumulative and transient effects are measurable. The circumstances of the Galactic-centre region, however, strongly favour the transients over the cumulatives for the following reasons:
(i) The orbital periods are long. Cumulative build-up needs multiple orbits which takes decades.
(ii) Orbits of S stars tend to be highly eccentric, e = 0 9 . being typical. Relativistic effects increase more steeply with small radius and high velocity than classical effects, and hence relativity is strongest around pericentre passage.
(iii) The extended stellar system will contribute significant noise, hampering in particular searches which rely on build-up over long time scales. Whereas it may be possible to disentangle transient effects from noise over short time scales because the time dependence of the former is well understood.
With a full 4-dimensional relativistic treatment, this paper performs numerical experiments - computing arrival times t a - to gain insight into transient relativistic behaviour on S Star redshifts. Section 2 discusses the more familiar tests of the Kerr metric and discusses a few examples of transient effects, section 3.1 formalises our redshift-calculating method, and discusses how the different effects scale with orbital period. Section 3 calculates these effects for mock S Star orbits. Finally, in section 4 we propose a novel strategy based on wavelet decomposition which may help separate relativistic behaviour from Newtonian noise.
1 Redshifts are conveniently stated in km/s, but in relativity no longer correspond to radial velocities.
## 2 FAMILIAR EFFECTS FROM KERR
There are multiple relativistic effects which grow over many orbits. This has been essential to observing them in artificial satellites, planets or pulsars. Here we list some of the well-known ones.
## 2.1 Cumulative
(i) The expected relativistic orbital precession has been discussed extensively in the context of S stars (e.g., Rubilar & Eckart 2001; Merritt et al. 2010; Sadeghian & Will 2011; Sabha et al. 2012) and pulsars (e.g., Liu et al. 2012). Relativity gives several contributions to the precession. The strongest cumulative relativistic effect comes from the first Schwarzschild contribution, resulting in a perihelion shift
$$\Delta \omega = \frac { 6 \pi } { a ( 1 - e ^ { 2 } ) } \, \quad \, \quad \, \quad \, ( 4 )$$
per orbit 2 .
(ii) There is another contribution to the precession if the black hole has internal angular momentum. This is characterised by a spin parameter; or angular momentum per unit mass s . Bodies near the black hole experience frame dragging in the spin direction. The precessional effect due to this is
$$\Delta \phi = - 8 \pi s \left [ a ( 1 - e ^ { 2 } ) \right ] ^ { - 3 / 2 } \text{ \quad \ \ } ( 5 )$$
per orbit. The phenomenon of frame dragging has been first observed only in recent years, by using laser ranging to accurately determine the orbit of the Lageos satellites and reveal the relativistic effect of the Earth's spin (Ciufolini & Pavlis 2004; Iorio 2010). The recently launched LARES satellite aims to measure the effect to an accuracy of 1%(Ciufolini et al. 2009).
(iii) Two further effects act on the spin of the star or pulsar. One is geodetic precession, wherein a vector attached to an orbiting body moves by (for circular orbits)
$$\Delta \phi = \frac { 3 \pi } { a }$$
per orbit (Fließbach 1990). Gravity Probe B has measured this effect in Earth's gravitational field (Everitt et al. 2011). The parallel transport of a vector along a geodesic is also influenced by frame-dragging. This is called the Lens-Thirring effect, and was also detected by Gravity Probe B (Everitt et al. 2011). It is possible that the spin axis of a close-in pulsar be parallel transported enough to change the pulse profile. Pulse-profile changes from geodetic precession have been observed in Binary Pulsar systems (Kramer 1998; Weisberg & Taylor 2002; Breton et al. 2008; Hotan, Bailes & Ord 2005), and could be observed in galactic centre pulsars.
Orbital decay due to gravitational radiation is another wellknown effect, but the time scales are too slow to be interesting for Galactic-centre stars.
## 2.2 Transients
Unlike the orbits of satellites, planets or pulsars, in the Galactic centre, orbital periods are much longer, so accumulating relativistic signals over many orbits is difficult even though the fields are
2 In this paper, all lengths are measured in units of the gravitational radius GM/c 2 .
Figure 2. Illustration of Schwarzschild precession. The orbit is like a 1/3-size version of the star S2 (Gillessen et al. 2009a), with semi-major axis a = 0 041 . ′′ and eccentricity e = 0 88 . , but viewed face-on. We have set the distance to the galactic centre at 8 31 . kpc. Along the trajectory we may associate a value for the argument of pericentre ω to the value it would take were the phase space position a solution to Kepler's equations. The right panel shows the instantaneous argument of pericentre against coordinate time for a distant observer. We see here that, for high eccentricities, precession is concentrated so strongly around pericentre that it looks nearly discrete. The dashed curve is the well-known formula (4) for the cumulative precession.
(mpc)
Figure 3. Illustration of frame-dragging precession. Since the effect is higher order than Schwarzschild precession, a smaller orbit is used than in Figure 2 in order to make the precession visible: the orbit is a 1/30-sized version of the star S2 (here with e = 0 88 . and a = 0 0041 . ′′ ), yet viewed face-on. The spin is maximal and perpendicular to the orbit; were the spin direction not perpendicular to the orbital plane, the orbital plane would also precess about the spin axis. Frame-dragging is even more strongly concentrated around pericentre than Schwarzschild precession. The dashed curve is the formula (5).
far stronger. Perhaps an even more serious problem is the Newtonian perturbations due to gas and other stars in the Galactic-centre region. So it is interesting to think about transient effects which occur over a single orbit. These may be measurable over a short time, and moreover a predictable time dependence could enable extracting the signal from the Newtonian background. In fact, there is a plethora of such effects, a few of which we describe here.
- (i) The strongest relativistic effect is gravitational time dilation,
one of the basic consequences of the Equivalence Principle. Time is dilated by a factor
$$g _ { t t } ^ { - 1 / 2 } = 1 - \frac { 2 } { r }$$
with no effect at this order on the orbit or the light paths. For a highly eccentric orbit, clearly there will be a peak at pericentre. GNSS satellites are sensitive to this shift. For navigation demands, it is enough for GNSS satellites to routinely step the on-board clock
time back, correcting for this effect. Gravitational time dilation has not yet been measured in the galactic centre, but is expected to be possible in the near future (Zucker et al. 2006). If observed, gravitational time dilation would provide a new test of the Einstein Equivalence Principle (Ang´ elil & Saha 2011).
(ii) Lensing effects of gravity on photons travelling to us are naturally also transient phenomena. Astrometric shifts due to gravitational lensing have been discussed in the Galactic-centre context (Bozza & Mancini 2009), as have time delays due to a curved space-time (Ang´ elil & Saha 2010), although none have yet been detected. With an impact parameter b , the deflection angle of a null ray is
$$\Delta \phi = \frac { 4 } { b }. \quad \quad \quad ( 8 ) \quad \colon$$
This is the leading-order Schwarzschild contribution. The ∼ b -2 effect is also relevant, and enters at the same order as the framedragging lensing contribution, which we discuss later. The extra delay induced in the arrival time of a packet of light compared to had it travelled in a straight line is the Shapiro delay (Shapiro 1964), and has been well-tested with the Mariner 9 and Viking spacecraft in the solar system(Shapiro et al. 1968, 1977; Reasenberg et al. 1979) and in binary pulsar systems(Stairs 2003; Demorest et al. 2010).
(iii) Underlying every type of orbit precession, there is a fleeting contribution which occurs around pericentre, the memory of which is not retained by the orbit's shape afterwards. Figure 2, which we shall return to later, shows our first example: the precession of the instantaneous pericentre of a highly eccentric orbit. Far from being the smooth effect suggested by equation (4), it consists almost of discrete kicks. In the derivation of (4), an oscillatory term crops up beside this term. Because this term imparts a momentary perturbation which time-averages to zero, it is dropped in textbook derivations.(Weinberg 1972; Misner, Thorne & Wheeler 1973; Schutz 2009; Carroll 2004). Analogously, Figure 3 shows pericentre precession due to frame dragging by the black hole spin, of which (5) is the average. These two examples are artificial and do not themselves correspond to observable quantities, for two reasons. First, for Figure 3, we have dropped lowerorder contributions from space curvature so as to isolate frame dragging. Second, the instantaneous pericentre of an orbit is defined as the pericentre of a Keplerian orbit with the same instantaneous position and momentum (the osculating elements, see e.g. Murray & Dermott 1999). In relativity, the instantaneous pericentre therefore becomes gauge-dependent and hence is not an observable quantity (cf. Preto & Saha 2009). Nonetheless Figures 2 and 3 do suggest that time-resolved observations could detect relativistic effects over a single orbit, especially around pericentre, where relativity is strongest and Newtonian perturbations are likely to be at their weakest.
In Section 3 below, we show how these and several other effects can be readily calculated numerically using a Hamiltonian formalism, and show various illustrative examples.
## 3 TIME DELAYS AND REDSHIFTS IN KERR
## 3.1 The Hamiltonian
The Hamilton equations for
$$H = \frac { 1 } { 2 } \, g ^ { \mu \nu } \, p _ { \mu } \, p _ { \nu } \,, \quad \quad \quad ( 9 ) \quad \quad 3$$
are simply the geodesic equations, with the affine parameter taking on the role of the independent variable. Since H does not depend explicitly on the affine parameter, H is constant along a geodesic. Proper time is √ | H | times the affine parameter, except for the case of H = 0 , corresponding to null geodesics.
In our case, g µν are the contravariant components of the Kerr metric. The Kerr metric is a vacuum solution to the Einstein Field Equations. This is the appropriate metric to use if we are interested in solving the forward problem for relativistic effects. This will allow us to investigate examples of transient relativistic effects in isolation. In section 4 we treat the more realistic case; we relax the vacuum assumption, and add other S stars as Newtonian perturbers to the system, and see whether we can uncover transient relativistic effects when the Newtonian noise is significantly large.
Assuming the orbits and light paths do not go close to the horizon, we can expand the Hamiltonian in powers of 1 /r . The result is available in Ang´ elil & Saha (2010). However because it is convenient to be able to set the black hole spin direction without having to rotate the observer and the orbit, we use a slightly different form. The Kerr geometry in Boyer-Lindquist coordinates necessarily aligns the axis of symmetry of the coordinate system with the axis of symmetry of the space-time geometry itself, and it is therefore not possible to disentangle the preferred direction of the coordinates with that of the spin in these coordinates. This means we need to first transform to pseudo-Cartesian coordinates. Table 1 contains the form of the Hamiltonian which we use here, in pseudoCartesian coordinates, and with the spin promoted to a 3-vector s = ( s x , s y , s z ) . (An equivalent table is given in Ang´ elil et al. 2014, but using Boyer-Lindquist variables.) For convenience, we use the short-form
$$s _ { \perp } \equiv \frac { s \times x } { r } \quad \text{and} \quad s _ { \| } \equiv \frac { s \cdot p } { r }. \quad \text{(10)}$$
Presenting the Hamiltonian in table form allows us to group the terms according to physical effects on orbits or light paths 3 . The Kepler/Rømer terms are classical. The leading relativisti c effect is time dilation, but as it is not associated with geodesics as such, it does not appear in the table. Relativistic terms not depending on the spin parameter s are labelled 'Schwarzschild'. Then there are various terms depending on spin. Of these, the term odd in s gives frame dragging.
We are now ready to use the Hamilton equations corresponding to the Hamiltonian in table 1 to explore the dynamics, and the consequences of the many terms in Table 1. While Ang´ elil, Saha & Merritt (2010) solves the inverse problem for relativity on S stars, here we attempt to give a more qualitative picture of exactly how relativity perturbs the orbit and redshifts/arrivaltimes, in particular for transient effects.
## 3.2 Numerical Experiments with S Stars and Pulsars
Wehave already referred to Figure 1, in passing in the Introduction. That figure compares the observable pulse rate from two cases: (i) a clock follows a relativistic orbit and the ticks are conveyed to the observer along null geodesics, and (ii) the classical case including Kepler and Rømer effects, and time dilation. The relativ istic case includes all terms in Table 1, other than the two highest-order 'not included' terms in the light path. The orbits are initialised at apocentre with e = 0 6 . and inclination I = 45 ◦ with respect to
3 While both orbits and light paths are geodesic in the same metric, the orders at which various terms affect the dynamics differ, due to the different behaviour of their momentum.
Table 1. Hamiltonian terms for orbits and light paths in a Kerr spacetime. The full Hamiltonian (9) is the sum of all the terms in the left column, plus higherorder terms that we have not considered. The middle and right terms group the terms by physical effect and scaling of the time delay ∆ t with period P , as explained in Section 3.1 Note that we are using geometrised units GM = c = 1 here. To put ∆ t and P in time units, simply multiply by a power of GM/c 3 so as to get the dimensions right.
| | Orbits | Light paths |
|---------------------------------------------------------------------------------------------------------------------------------|--------------------------------|-------------------------------------------------------|
| - p 2 t 2 | static | Rømer ∆ t ∼ P 2 / 3 |
| + p 2 2 | Kepler | Rømer ∆ t ∼ P 2 / 3 |
| - p 2 t r | | Shapiro ∆ t ∼ P 0 |
| - ( x · p ) 2 r 3 | Schwarzschild ∆ t ∼ P 0 | Shapiro ∆ t ∼ P 0 |
| - 2 p 2 t r 2 | | |
| - 2 p t p · ( s × x ) r 3 | frame-dragging ∆ t ∼ P - 1 / 3 | frame-dragging, spin-squared, Shapiro ∆ t ∼ P - 2 / 3 |
| + s 2 ⊥ 2 r 4 ( x · p ) 2 - 1 2 r 4 ( p · s × x ) 2 s 2 ⊥ - 1 2 r 4 1 - s 2 ⊥ 1 - s 2 ( ( p · s ) r - ( x · s ) ( x · p ) r ) 2 | | |
| ‖ | Spin (even), Schwarzschild | |
| + p 2 t r 3 s 2 ⊥ - 4 p 2 t r 3 | | not included |
the line of sight, and are integrated forward and back. The restframe tick rate of the clock is 1000 Hz and its orbital period is a week, while the assumed gravitational radius of the black hole is GM/c 2 = 20 light-sec (corresponding to Sgr A*) - these choices are only for the sake of putting axes on the figure and have no physical significance.
From Figure 1 we can infer that relativity makes the pericentre precess, but to see more detail we need to extract the difference between the relativistic and non-relativistic cases. It is especially interesting to see what different groups of terms from Table 1 do to the time-delay and redshift curves. To label different cases, let us introduce some shorthand, as follows.
- (i) H sch means that Schwarzschild terms but not spin terms have been included in the orbits, while no relativistic terms have been included for the light paths. These are the forth and fifth rows in Table 1. H sch means that Shapiro terms have been included for the
light paths, while the orbits are classical. These are the third and fourth rows in Table 1.
(ii) H s ( H s ) means that only the spin p t ( x × p ) /r 3 term has been added to the classical terms, and only for orbits (light paths). This term is found in the sixth row in Table 1.
(iii) Similarly, H s 2 ( H s 2 ) means classical plus spin-squared terms in the orbits (light paths). These terms are the remaining rows of Table 1.
Figures 2 and 3, mentioned in the previous section, show the orbit effects H sch and H s as a perturbation to pure Newtonian motion. Cumulative precession is a discrete phenomenon which occurs at pericentre. However, the effect is not completely step-like, with transient behaviour before and after the pericentre kicks. The complicated H s 2 orbit evolution effects are shown in Figure 4. The evolution depends on the relative orientation of the orbital angular momentum with the black hole spin. We do not have any inter-
Figure 4. Orbital effects of H s 2 (even-spin terms). The orbit is like a 1/300-size version of S2 ( a = 0 02 . mpc and e = 0 88 . ), initially in the x z -plane, while the spin is along x . Schwarzschild and frame-dragging terms have been omitted, so this is a completely artificial example. It is nonetheless interesting, as it illustrates the complexity of the spin-derived effects, from which none of the Keplerian orbital elements are exempt from change.
pretation that helps understand the dynamics generated by these higher-order terms, and merely show this orbit as an example.
That leaves only the p t x × p term to deal with. To do that, we consider the geometric mean of p 2 and p /r 2 t 4 to get
Moving now to light-path effects, Figure 5 shows the contribution of H sch , and Figure 6 shows the contributions of H s and H s 2 . The even-spin signals on timing and redshift are capable of a wide variety of signal shapes, which depend on the orbit geometry relative to the observer and the spin-direction. Timing delays due to spin effects influencing photon paths have been calculated for binary pulsars (see for example Fig. 5 in Wex & Kopeikin 1999).
## 3.3 Scaling
Table 1 also gives the scaling of the time delay, which depends on some power of the orbital period P . For the classical Kepler or Rømer effect ∆ t ∼ P 2 / 3 . With respect to the relevant terms in the Hamiltonian, we thus have
$$\mathbf p ^ { 2 } \, \Rightarrow \, \Delta t \sim P ^ { 2 / 3 } \,. \quad \quad \ ( 1 1 ) \quad \ \stackrel { \mu } { s }$$
The p /r 2 t behaves differently for orbits and light paths. For orbits, it is of course part of Keplerian dynamics. For light paths it is part of the Shapiro delay, which depends only logarithmically on r . Accordingly, we write
$$\frac { p _ { t } ^ { 2 } } { r } \, \Rightarrow \, \Delta t \sim \begin{cases} P ^ { 2 / 3 } & \text{orbits} \\ P ^ { 0 } & \text{light paths.} \end{cases} \quad \text{(12)} \quad \text{$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$$$
Table 1 also has many terms which look like increasingly elaborate versions of the classical ones. The scaling of ∆ t for such terms is simple: provided we are not close to the horizon, a factor of 1 /r in a Hamiltonian term introduces a factor P -2 / 3 in the time delay.
$$\frac { p _ { t } x \times { \mathbf p } } { r ^ { 3 } } \, \Rightarrow \, \Delta t \sim \begin{cases} P ^ { - 1 / 3 } & \text{orbits} \\ P ^ { - 2 / 3 } & \text{light paths.} \end{cases} \quad ( 1 3 )$$
Table 1 includes all terms with ∆ t contributions up to P -2 / 3 . Redshifts scale like
$$\frac { \Delta t } { P }$$
as will be evident from Equations (2) and (3). This assumes, as before, that orbits and light paths are not too close to the horizon. This suggests that prospects for testing relativity as period sizes decrease improve quicker for stellar orbits than pulsar orbits.
Wecan test these scalings with numerical experiments. In Figure 7, we show how the transient-relativistic contribution of H sch , H s and H s 2 scale with the orbital period P . We isolate the transient signal by calculating the most-positive plus most-negative difference in the observables, upon initialising two orbits at pericentre integrating over one period with and without the Hamiltonian terms in question. All orbits have e = 0 6 . , I = 45 ◦ , as in Figure 1, but the period was varied. As we can see in the figure, the predicted scalings from Table 1 are borne out. Note, in particular, that the leading-order Schwarzschild effects on the orbit makes for timing signals which remain constant as the orbital size deceases. The redshift contribution of Schwarzschild however, scales as 1 /P . Figure 8 then shows how the light-path contributions scale with period. For the latter figure, the spin is maximal and perpendicular to the orbit, but this detail is unimportant for the scaling.
For both orbit and light-path effects, our simulations show that
Figure 5. Light-path contributions to time delays and redshifts of Schwarzschild terms. The left panel is simply the well-known Shapiro delay. The redshift is the derivative of the time-delay, stated in eq. (2). The orbital period is 3 yrs - a scaled version of the orbits used for Fig. 1: e = 0 6 . , a = 0 041 . ′′ , I = 45 ◦ , and R 0 = 8 31 . kpc. The time delay depends only logarithmically on the period P , but the redshift signal scales as 1 /P . Because this is a light-propagation effect, there is no cumulative component to this effect.
- (i) the relativistic contributions are concentrated around pericentre, and
- (ii) vary along the orbit in a complicated way, (especially when spin is included), yet
- (iii) nonetheless agree with the orbital period scalings in Table 1.
As we shall see in the next section, (i) and (iii) will prove useful for extracting relativistic signals from extended mass noise.
Figure 6. Light-path contributions of spin terms to time delays and redshifts. The orbit in each case is 1/50-sized version of the one in Figure 5 ( a = 0 0008 . ′′ , e = 0 6 . ), while the spin is maximal. Only the spin direction changes: in the top row, the spin is perpendicular to the orbit; in the middle row, the spin is along the line of sight, and hence the frame-dragging signal is null, leaving only even-spin contributions; in the bottom row, the spin is perpendicular to the line of sight.
Figure 7. Transient orbital contributions to time delays and redshifts of different relativistic terms (Schwarzschild, frame-dragging and spin-squared), as a function of orbital period. The cumulative components of these relativistic effects have significantly smaller amplitude. The orbit size used in Figure 1 corresponds to the short-period end of these panels. The orbital geometry we maintain over this calculation has I = 45 ◦ , and e = 0 6 . .
Figure 8. Light-path contributions to time delays and redshifts of different relativistic terms. This figure complements Figure 7. The main difference is in how the odd-spin (or frame-dragging) term scales.
## 4 FILTERING NEWTONIAN PERTURBATIONS
Orbit fitting in the pure Kerr case poses no fundamental problems (Ang´ elil, Saha & Merritt 2010), however, critical to being able to resolve relativistic effects on galactic centre stars will be the handling of other perturbations. The most significant are expected to be those from the extended mass distribution, mainly from other stars, but also perhaps from a significant dark matter component. Merritt et al. (2010), Antonini & Merritt (2013) and Iorio (2011) compare the cumulative effects of extended mass and relativity.
In this section we are interested in transient relativistic signals over a single orbit. A star whose redshift/time-delay is expected to be influenced by relativity is the target star. The redshift/time-delay of this star is also affected by the Newtonian attraction of other black hole-orbiting stars in the neighbourhood, which we call the perturbers.
While the relativistic time dilation signal is likely to be stronger than extended Newtonian signals, the next-strongest effects (Schwarzschild and Shapiro) may be partially obscured. In this section we first discuss how to calculate the Newtonian perturbations on the target star, before introducing a wavelet decomposition method as a tool which could be used to help distinguish them from relativistic perturbations.
## 4.1 Newtonian perturbers
The classical leading-order perturbation due to other stars orbiting the black hole is given by a Hamiltonian contribution
$$H _ { \text{stellar} } = \sum _ { j } \frac { m _ { j } } { M } \left ( \frac { \vec { x } \cdot \vec { x } _ { j } } { | \vec { x } _ { j } | ^ { 3 } } - \frac { 1 } { | \vec { x } - \vec { x } _ { j } | } \right ), \quad ( 1 5 ) \\ \text{sources} \, \sqrt { \, } \colon \, \text{the store bins above and} \, \text{and} \, \text{...} \, \sqrt { \, } \text{-- for to not\,\,\text{throws} }$$
where /vector x is the star being observed and m ,/vector j x j refer to perturbing stars. For a derivation, see Wisdom & Holman (1991), especially their equation (17), and disregard the mutual perturbations of the /vector x j stars. Note however, that the back-reaction on the observed star due to the perturbed position of the black hole must be included. We model the perturbations by adding the classical perturbation (15) to the relativistic Hamiltonian from Table 1. Will (2014) shows that new relativistic terms appear in general N -body problems, if there is a tidal force or a quadrupole of the same order as the dominant monopole. If the star being observed were in a binary, such terms would arise, but for the simpler problem we are considering, the approximation of simply adding the classical perturbers appears to be valid.
As an example of the effect of Newtonian perturbers, we consider a target star at /vector x on an S2-like orbit (Gillessen et al. 2009b) with semi-major axis a = 30000 in geometric units, and eccentricity e = 0 9 . . The perturbers at /vector x j are 100 stars, all of equal mass, together making up 1% of the black hole's mass or /similarequal 4 × 10 4 M /circledot . These are distributed according to a power-law profile ρ ( r ) ∝ exp( -γr ) with γ = 0 5 . . Their eccentricity distribution is uniform. Figure 9 contrasts the Newtonian and relativistic perturbations on the target star's semi-major axis a and periapsis argument ω . As we see, the relativistic perturbations are completely submerged under the Newtonian stellar perturbations. We may recall that for Mercury, Newtonian perturbations from other masses are an order of magnitude larger than the relativistic effects, yet the accumulation of ∆ ω /similarequal 0 1 . ′′ per orbit is measurable. What makes such a measurement possible is that in the solar system, planetary masses are known accurately and hence the Newtonian perturbations can be subtracted off. Near the Galactic Centre, there is no prospect of measuring all the perturbing masses accurately. Hence,
]
a [r
Figure 9. The evolution of the instantaneous Keplerian elements a and ω . The Newtonian perturbations from other stars are distributed throughout the orbit, whereas relativistic perturbations are concentrated near pericentre. (Note that the instantaneous a and ω are not directly observable in relativity, because they are gauge-dependent and don't take signal propagation into account. The observable quantities are arrival times and redshifts.)
if the model perturbers in Figure 9 are at all representative, relativistic effects would be drowned under Newtonian perturbations.
## 4.2 Wavelets
However, the situation is not hopeless. Because the transient relativistic effects have a very specific time dependence that is known in advance, it may be possible to extract them from under the Newtonian background. Matched-filter techniques, well known from gravitational-wave searches (see for example Sathyaprakash & Schutz 2009), will not work because the observables are non-linear in the perturbing effects. But progress may be possible using wavelets.
A wavelet decomposition (Daubechies 1988, 1990), allows one to identify features by breaking down a signal according not just to the frequency at which they occur, but also according to the time they occur. In contrast to a Fourier decomposition, where each basis function carries frequency information only, a wavelet
Table 2. The coefficient structure of a wavelet-transformation. The first index gives the time scale, the second index give the localisation.
Figure 10. In this demonstration, a sum of two Gaussians (top panel) is decomposed into the Daubechies 20 wavelet basis (other panels). Each of the wavelet panels corresponds to (17) for the stated value of n . (The constant 0 -th mode is not shown.) The sum of the lower panel curves yields the parent shape.
basis function includes both frequency and localisation information. Relativistic perturbations and perturbations due to the extended mass affect the dynamics in different ways, as Figure 9 illustrates, at different frequencies and different localisations. We are interested in designing a procedure which helps identify relativistic signals when shrouded by significant extended-mass noise. Because redshift curves over a single orbit have no periodicity, and because relativistic perturbations are most prevalent around pericentre, wavelets are a natural choice for designing filters. As a result of relativistic effects being most pronounced around pericentre and non-lingering due to their oft transient nature - we can expect high-frequency coefficients, localized around pericentre passage, to be of greatest value in retaining information from relativistic effects. We would expect the extended mass perturbations to also impart transient, high-frequency effects, such as close encounters, but those would not be concentrated around pericentre.
In a typical wavelet decomposition, such as the Daubechies 4
Figure 11. The upper panel shows the pure signals z GR -z Kepl and z Pert -z Kepl . In our chosen example, the perturbation from the extended mass distribution is ∼ 30 times larger than the relativistic signal. The lower panel shows the signals after shifting epochs to minimise the area under the curves. It is the latter we plug into the wavelet procedure detailed in section 4.
and Daubechies 20 wavelet types, a signal is expressed as
$$z ( t ) = \sum _ { n } \sum _ { i = 1 } ^ { 2 ^ { n } } C _ { n i } \psi _ { n i } ( t ). \quad \quad ( 1 6 )$$
The wavelet basis functions ψ ni ( ) t are the scaled and translated versions of a single function, called the mother wavelet, while the C ni are the expansion coefficients. Table 2 schematically outlines the wavelet coefficient structure. Each row of this table corresponds to a particular time scale, which is twice as fast as in the row above it. The n -th row has 2 n coefficients, each of which correspond to different time windows (or localisations). Let us write
$$\mathcal { W } _ { n } \, z ( t ) \equiv \sum _ { i = 1 } ^ { 2 ^ { n } } C _ { n i } \psi _ { n i } ( t ). \quad \quad ( 1 7 )$$
The operator W n isolates a particular time scale in the signal. 4
Figure 10 shows the result of applying the W n operator to an example curve, consisting of two superposed Gaussians with different means and widths. We see that n = 2 the first (wider) Gaussian dominates, at n = 3 the second Gaussian starts to take over, and from n = 5 the first Gaussian has been completely filtered out and only the narrower Gaussian contributes.
## 4.3 Filtering relativistic signals with wavelets
Wenow consider an S star (or S pulsar) whose redshift (or pulse arrival times) are contaminated by significant noise from an extended mass system, and investigate how the wavelet coefficients are influenced by relativistic versus extended-mass perturbations. Starting with an unperturbed Keplerian orbit, we proceed as follows.
First, we generate three redshift curves for this orbit: z Kepl has no perturbations, z Pert includes the Newtonian perturbation by including the effects of (15), and z GR includes only the relativistic Schwarzschild and Shapiro perturbations. The differences z GR -z Kepl and z Pert -z Kepl are plotted in the upper panel of Figure 11. Here we use the same extended Newtonian mass system example as earlier in this section, corresponding to Figure 9. In this mock data example, the relativistic redshift signal is ∼ 30 times weaker than that due to the extended mass perturbations.
Before taking wavelet transforms, another step is necessary: we need to choose the reference orbit z Kepl anew, because of course the 'original' unperturbed orbit will not be provided by data. It would be natural to choose a reference orbit that best fits the data, but any consistent convention can be used. For simplicity, we shift the epoch of z Kepl so as to minimize the integrated difference from z GR and z Pert respectively. We denote the shifted Keplerian curves as ˜ z Kepl and ¯ z Kepl . The differences z GR -˜ z Kepl and z Pert -¯ z Kepl are plotted in Figure 11's second panel. In this example, they have approximately the same amplitude.
Wethen decompose the signals into different modes according to frequencies and plot the differences
$$\mathcal { W } _ { n } \left ( z _ { G R } - \tilde { z } _ { K e p l } \right ) \quad \ \ ( 1 8 ) \quad \ \ t _ { t }$$
and
$$\mathcal { W } _ { n } \left ( z _ { \text{Per} } - \bar { z } _ { \text{Kepl} } \right ) \quad \quad ( 1 9 ) \quad t |$$
for n = 1 . . . 9 . These are plotted in Figure (12), which shows the results using two different wavelet basis functions.
In our example, the wavelet-reconstructed perturber signal is stronger than the relativistic ones over all wavelet scales, except at n = 5 (with twice the amplitude) and n = 6 (with almost the same amplitude). This decomposition procedure indicates that given the geometry of our chosen orbit, Schwarzschild effects, although obscured by extended mass perturbations with a signal-to-noise S/N ∼ 1 30 / , will impart a significant contribution on the n = 6 frequency modes. Alternatively, one can say that Schwarzschild effects, though about 30-fold weaker overall than extended-mass perturbations (in this model), nonetheless stand out over Newtonian perturbations over a two-year interval around pericentre, in wavelet modes of timescale ≈ 6 months.
The above suggests that subtracting off a Keplerian orbit, applying a wavelet transform to the residual, and then considering a
4 We will speak of wavelet frequencies in this section, even though we really mean time-scalings of the wavelets.
specific subset of the wavelet coefficients may succeed in filtering out Newtonian perturbations.
## 5 CONCLUSIONS AND OUTLOOK
Galactic centre stars travel upon the most relativistic orbits known. However, the accessible relativistic effects are not simply extensions of similar experiments in the solar system and in binary pulsars. S stars, and S pulsars if they exist, live in stronger fields than binary pulsars, but their orbital periods are much longer. The combination of the strong fields, long orbital time-scales, and the typically high eccentricities push otherwise negligible aspects of dynamics near a black hole into the observables. For example, precession is not a steady process, as the well-known orbit-averaged formulas (4) and (5) may suggest, but nearly a shock that happens at pericentre. The concentration of dynamical effects around pericentre passage applies even more to effects which depend on the spin of the black hole. These pericentre shocks will be important when separating relativistic signals from noise sources.
If the observed dynamics is found to be in agreement with a Kerr space-time plus perturbations from the surrounding astrophysical environment, Einstein gravity will be tested to a new level. A further benefit is that because we test gravity by tracking freelyfalling bodies, as well as photon paths, by inferring the components of the metric by looking at their effects on the behaviour of geodesics, we probe not just the field equations, but implicitly test the notion requisite to describing gravity with geometry - the principle of equivalence.
The spectrometers of the Keck and VLT telescopes have independently observed the spectra of the S stars, managing to achieve spectral resolution to ∼ 10 kms -1 in the best cases. The next generation of instruments, such as the High Resolution Near-infrared Spectrograph (SIMPLE) on the E-ELT is expected to achieve ∼ 2 kms -1 (Origlia, Olivia & Maiolino 2010). If an S star with period ∼ 1 year is discovered, observations clustered around pericentre passage at this level of accuracy could provide a measurement for frame-dragging. Were S pulsars with stable periods to be detected with orbits similar to the already-known S stars, pulsar timing even at the msec level would be, in principle, enough for all the effects summarised in Table 1. The challenge would be removing the Newtonian 'foreground' due to the extended mass distribution around Sgr A*. Separating cumulative effects into Newtonian versus relativistic is a challenging task, yet with transient effects that vary along an orbit in different ways, one can be more optimistic.
Daubechies 4 wavelet
Daubechies 20 wavelet
Figure 12. Reconstructed signals from wavelet frequency modes. Due to the linearity of the wavelet transformation, the sum of the signals in each column yield the lower panel curves of figure 11. The wavelets down the left column are the Daubechies 4 variety, and those down the right the Daubechies 20. Despite the raw relativistic signal being 30 times less than that from Newtonian effects due to the perturbing stars, the relativistic signal manages to significantly dominate at n = 6 . This happens for both wavelet types used here, Daubechies 4 and Daubechies 20. This procedure highlights the specific time and frequency localisation properties of the two effect types, and tools like this could aid future searches in decorrelating them.
## ACKNOWLEDGEMENTS
We thank S. Gillessen and S. Tremaine for many helpful insights and B.P. Schmidt for some suggestions. We thank the referees, whose comments have helped to improve the paper. R.A. acknowledges support from the Swiss National Science Foundation.
## REFERENCES
Ang´ elil R., Saha P., 2010, The Astrophysical Journal, 711, 157 Ang´ elil R., Saha P., 2011, The Astrophysical Journal Letters, 734,
L19
Ang´ elil R., Saha P., Bondarescu R., Jetzer P., Sch¨ arer A., Lundgren A., 2014, Physical review D
Ang´ elil R., Saha P., Merritt D., 2010, The Astrophysical Journal, 720, 1303
Antonini F., Merritt D., 2013, Astrophysical Journal Letters, 763, L10
Ashby N., 2003, Living Reviews in Relativity, 6, 1
Bozza V., Mancini L., 2009, Astrophysical Journal, 696, 701
Breton R. P. et al., 2008, Science, 321, 104
Broderick A. E., Johannsen T., Loeb A., Psaltis D., 2014, ApJ, 784, 7
Carroll S. M., 2004, Spacetime and geometry. An introduction to general relativity
Ciufolini I., Paolozzi A., Pavlis E. C., Ries J. C., Koenig R., Matzner R. A., Sindoni G., Neumayer H., 2009, Space Science Reviews, 148, 71
Ciufolini I., Pavlis E. C., 2004, Nature, 431, 958
Cordes J. M., Lazio T. J. W., 1997, The Astrophysical Journal, 475, 557
Daubechies I., 1988, Information Theory, IEEE Transactions on, 34, 605
Daubechies I., 1990, Information Theory, IEEE Transactions on, 36, 961
Demorest P. B., Pennucci T., Ransom S. M., Roberts M. S. E., Hessels J. W. T., 2010, Nature, 467, 1081
Doeleman S., 2010, in 10th European VLBI Network Symposium and EVN Users Meeting: VLBI and the New Generation of Radio Arrays
Eckart A., Sch¨ odel R., Moultaka J., Straubmeier C., Viehma nn T., Pfalzner S., Pott J.-U., 2005, in American Institute of Physics Conference Series, Vol. 783, The Evolution of Starbursts, S. H¨ttmeister, E. Manthey, D. Bomans, & K. Weis, ed ., u pp. 17-25
Eisenhauer F., Sch¨ odel R., Genzel R., Ott T., Tecza M., Abut er R., Eckart A., Alexander T., 2003, Astrophysical Journal Letters, 597, L121
Everitt C. W. F. et al., 2011, Physical Review Letters, 106, 221101 Fließbach T., 1990, Allgemeine Relativit¨ atstheorie., Fl ießbach, T., ed.
Ghez A. M. et al., 2008, The Astrophysical Journal, 689, 1044 Gillessen S., Eisenhauer F., Fritz T. K., Bartko H., Dodds-Eden K., Pfuhl O., Ott T., Genzel R., 2009a, Astrophysical Journal Letters, 707, L114
Gillessen S., Eisenhauer F., Trippe S., Alexander T., Genzel R., Martins F., Ott T., 2009b, The Astrophysical Journal, 692, 1075 Hotan A. W., Bailes M., Ord S. M., 2005, The Astrophysical Journal, 624, 906
Iorio L., 2010, Central European Journal of Physics, 8, 25, 10.2478/s11534-009-0060-6
Iorio L., 2011, Phys. Rev. D, 84, 124001
Kramer M., 1998, The Astrohpyscial Journal, 509, 856
Kramer M., Backer D., Cordes J., Lazio T., Stappers B., Johnston S., 2004, New Astronomy Reviews, 48, 993 , ¡ce:title¿Science with the Square Kilometre Array¡/ce:title¿ ¡xocs:fullname¿Science with the Square Kilometre Array¡/xocs:fullname¿
Kramer M., Klein B., Lorimer D., M¨ uller P., Jessner A., Wiel ebinski R., 2000, in Astronomical Society of the Pacific Conference Series, Vol. 202, IAU Colloq. 177: Pulsar Astronomy - 2000 and Beyond, M. Kramer, N. Wex, & R. Wielebinski, ed., p. 37
Liu K., Wex N., Kramer M., Cordes J. M., Lazio T. J. W., 2012, The Astrophysical Journal, 747, 1
Macquart J.-P., Kanekar N., Frail D. A., Ransom S. M., 2010, The Astrophysical Journal, 715, 939
- Martins F., Gillessen S., Eisenhauer F., Genzel R., Ott T., Trippe S., 2008, Astrophysical Journal Letters, 672, L119
Merritt D., Alexander T., Mikkola S., Will C. M., 2010, Phys. Rev. D, 81, 062002
Meyer L. et al., 2012, Science, 338, 84
Misner C. W., Thorne K. S., Wheeler J. A., 1973, Gravitation,
Misner, C. W., Thorne, K. S., & Wheeler, J. A., ed.
Murray C. D., Dermott S. F., 1999, Solar system dynamics
Origlia L., Olivia E., Maiolino R., 2010, Simple: A high resolution near-infrared spectrograph for the e-elt
Pfahl E., Loeb A., 2004, The Astrophysical Journal, 615, 253
Preto M., Saha P., 2009, Astrophysical Journal, 703, 1743
Rea N. et al., 2013, APJL, 775, L34
Reasenberg R. D. et al., 1979, ApJL, 234, L219
Rubilar G. F., Eckart A., 2001, Astronomy and Astrophysics, 374, 95
Sabha N. et al., 2012, ArXiv e-prints
Sadeghian L., Will C. M., 2011, Classical and Quantum Gravity, 28, 225029
Sathyaprakash B. S., Schutz B. F., 2009, Living Reviews in Relativity, 12, 2
Sch¨ odel R. et al., 2002, Nature, 419, 694
Schutz B., 2009, A First Course in General Relativity
Shapiro I. I., 1964, Phys. Rev. Lett., 13, 789
Shapiro I. I., Pettengill G. H., Ash M. E., Stone M. L., Smith W. B., Ingalls R. P., Brockelman R. A., 1968, Physical Review Letters, 20, 1265
Shapiro I. I. et al., 1977, 82, 4329
Stairs I. H., 2003, Living Reviews in Relativity, 6
Taylor, Jr. J. H., 1994, Reviews of Modern Physics, 66, 711 Weinberg S., 1972, Gravitation and Cosmology: Principles and Applications of the General Theory of Relativity, Weinberg, S., ed.
Weisberg J. M., Taylor J. H., 2002, The Astrophysical Journal, 576, 942
Wex N., Kopeikin S. M., 1999, The Astrophysical Journal, 514, 388
Will C. M., 2014, Phys. Rev. D, 89, 044043
Wisdom J., Holman M., 1991, Astronomical Journal, 102, 1528 Zucker S., Alexander T., Gillessen S., Eisenhauer F., Genzel R., 2006, Astrophysical Journal Letters, 639, L21 | 10.1093/mnras/stu1686 | [
"Raymond Angélil",
"Prasenjit Saha"
] | 2014-08-01T20:00:00+00:00 | 2014-08-20T14:45:33+00:00 | [
"astro-ph.GA",
"astro-ph.HE",
"gr-qc"
] | Clocks around Sgr A* | The S stars near the Galactic centre and any pulsars that may be on similar
orbits, can be modelled in a unified way as clocks orbiting a black hole, and
hence are potential probes of relativistic effects, including black hole spin.
The high eccentricities of many S stars mean that relativistic effects peak
strongly around pericentre; for example, orbit precession is not a smooth
effect but almost a kick at pericentre. We argue that concentration around
pericentre will be an advantage when analysing redshift or pulse-arrival data
to measure relativistic effects, because cumulative precession will be drowned
out by Newtonian perturbations from other mass in the Galactic-centre region.
Wavelet decomposition may be a way to disentangle relativistic effects from
Newton perturbations. Assuming a plausible model for Newtonian perturbations on
S2, relativity appears to be strongest in a two-year interval around
pericentre, in wavelet modes of timescale approximately 6 months. |
1408.0284v1 | ## Discovery of a new Y dwarf: WISE J030449.03-270508.3
- D. J. Pinfield 1 /star , M. Gromadzki 2 3 , , S. K. Leggett , J. Gomes , N. Lodieu 4 1 5 6 , ,
- R. Kurtev 3 2 , , A. C. Day-Jones , M. T. Ruiz , N. J. Cook , C. V. Morley , 1 7 1 8
- M. S. Marley , F. Marocco , R. L. Smart 9 1 10 , H. R. A. Jones , P. W. Lucas , 1 1
- Y. Beletsky 11 , V. D. Ivanov 12 , B. Burningham , J. S. Jenkins , C.Cardoso 1 7 10 ,
- J. Frith , J. R. A. Clarke , M. C. G´lvez-Ortiz 1 2 a 13 , Z. Zhang 5 6 ,
- 1 Centre for Astrophysics Research, Science and Technology Research Institute, University of Hertfordshire, Hatfield AL10 9AB, UK 2 The Millennium Institute of Astrophysics (MAS), Santiago, Chile
- 3 Instituto de F´ ısica y Astronom´ ıa, Universidad de Valpara´ ıso, Av. Gran Breta˜a 1111, Playa Ancha, Casilla 5030, Chile n
- 4 Gemini Observatory, Northern Operations Center, 670 N. A'ohoku Place, Hilo, HI 96720, USA
- 5 Instituto de Astrof´ ısica de Canarias (IAC), Calle V´ ıa L´ctea s/n, E-38200 La Laguna, Tenerife, Spain a
- 6 Departamento de Astrof´ ısica, Universidad de La Laguna (ULL), E-38206 La Laguna, Tenerife, Spain
- 7 Universidad de Chile, Santiago, Casilla 36-D, Chile
- 8 Department of Astronomy and Astrophysics, University of California, Santa Cruz
- 9 NASA Ames Research Center
10 Istituto Nazionale di Astrofisica, Osservatorio Astronomico di Torino, Strada Osservatrio 20, 10025 Pino Torinese, Italy 11 Las Campanas Observatory, Carnegie Institution of Washington, Colina el Pino, 601 Casilla, La Serena, Chile 12 ESO, Av. Alonso de Cordova 3107, 19001 Casilla, Santiago 19, Chile
- 13 Centro de Astrobiolog´ ıa (CSIC-INTA), Ctra. Ajalvir km 4, E-28850 Torrej´n de Ardoz, Madrid, Spain o
5 August 2014
## ABSTRACT
We present a new Y dwarf, WISE J030449.03-270508.3, confirmed from a candidate sample designed to pick out low temperature objects from the WISE database. The new object is typed Y0pec following a visual comparison with spectral standards, and lies at a likely distance of 10-17 pc. Its tangential velocity suggests thin disk membership, but it shows some spectral characteristics that suggest it may be metalpoor and/or older than previously identified Y0 dwarfs. Based on trends seen for warmer late type T dwarfs, the Y -band flux peak morphology is indicative of subsolar metallicity, and the enhanced red wing of the J -band flux peak offers evidence for high gravity and/or low metallicity (with associated model trends suggesting an age closer to ∼ 10 Gyr and mass in the range 0.02-0.03 M /circledot ). This object may thus be extending the population parameter-space of the known Y0 dwarfs.
Key words: surveys - stars: low-mass, brown dwarfs
## 1 INTRODUCTION
The discovery of Y dwarfs (Cushing et al. 2011) has extended the T eff parameter-space of brown dwarfs (Chabrier & Baraffe 2000) below 500 K, with Y0 dwarfs generally 400-450 K (Dupuy & Kraus 2013), Y1 dwarfs 350-400 K (e.g. Kirkpatrick et al. 2013), and the most extreme example potentially as cool as ∼ 250 K (Luhman 2014). WISE J1828+2650 was first identified as the archetypal Y dwarf ( /greaterorequalslant Y2; Cushing et al. 2011) due to its extremely red nearmid-infrared colours and its red J -H colour (suggestive of a Wien tail flux collapse), though parallax measurements have subsequently shown that this object is warmer than the
Y0 dwarfs, with its extreme nature presumably influenced by other properties (Dupuy & Kraus 2013). A significant decrease in the width of the J -band flux peak (compared with the T9 spectral standard UGPS 0722-05; Lucas et al. 2010; Cushing et al. 2011), has become the practical means for separating late-T and Y dwarfs. Though some Y dwarf spectra (with sufficient signal-to-noise) also show possible evidence of NH 3 in the H -band, and some show enhanced Y -band flux (but not all; see fig 22 of Mace et al. 2013a), believed to be due to reduced absorption in the KI wings as potassium starts to form KCl (a trend that starts for the latest T dwarfs; Leggett et al. 2010; Lodieu et al. 2013).
At time of writing there are seventeen spectroscopically confirmed Y dwarfs (Cushing et al. 2011; Kirkpatrick et al. 2011, 2012; Liu et al. 2012; Tinney et al.
2012; Kirkpatrick et al. 2013; Cushing et al. 2014b), and three additional objects with similar T eff (or lower) for which it has not been possible to measure spectra due to their faintness in the near-infrared (WD 0806-661B, CFBDSIR J1458+1013B, and WISE J0855-0714; Luhman et al. 2011; Liu et al. 2011; Luhman 2014).
This new spectroscopic class was revealed with the advent of the Wide-field Infrared Survey Explorer (WISE; Wright et al. 2010) observatory, whose sensitivity in the mid-infrared opened up this detection-space. Previous large-scale surveys had searched for brown dwarfs in the near-infrared and red-optical, uncovering field L and T dwarfs (Kirkpatrick 2005) with the Two Micron All Sky Survey (2MASS; Skrutskie et al. 2006) and the Sloan Digital Sky Survey (SDSS; York et al. 2000), and then probing to the coolest T dwarfs (e.g. Warren et al. 2007; Burningham et al. 2008; Delorme et al. 2008a; Burningham et al. 2010; Lucas et al. 2010) with the UKIRT Infrared Deep Sky Survey (UKIDSS; Lawrence et al. 2007), the Canada France Brown Dwarf Survey (CFBDS; Delorme et al. 2008b), and more recently the Visible and Infrared Survey Telescope for Astronomy (VISTA; McMahon et al. 2013; Lodieu et al. 2012).
WISE surveyed the whole sky in four filters centred at 3.4 µm , 4.6 µm , 12 µm , and 22 µm (W1, W2, W3 and W4 respectively). In particular W1 covers a deep CH 4 absorption band in the spectra of cool brown dwarfs with W2 covering a spectral region of relatively low opacity. This leads to very red W 1 -W 2 colours for cool brown dwarfs (Mainzer et al. 2011a) and facilitates colour-based brown dwarf searches (e.g. Kirkpatrick et al. 2012). WISE also has a time-domain, with baselines ranging from ∼ 36 hours (identifying e.g. nearearth objects or NEOs as part of the NEOWISE programme; Mainzer et al. 2011b) up to 6-12 months when the main cryogenic mission data is combined with data from the postcryogenic survey phase (the AllWISE data release). This facilitates proper motion searches such as Luhman (2013) and Kirkpatrick et al. (2014). It also offers greater scope to assess the nature of WISE sources via an analysis of their multiple measurements and comparison with known populations. Pinfield et al. (2014) followed this approach, using a control sample (from SDSS) of isolated point-like nonvariable non-moving sources to guide the identification of brown dwarf candidates down to low signal-to-noise.
Detailed studies of the full spectral diversity of Y dwarfs are important for our understanding of ultra-cool atmosphere physics (e.g. Burrows et al. 2003; Baraffe et al. 2003; Burrows et al. 2007; Saumon & Marley 2008; Allard et al. 2011; Burrows et al. 2011; Morley et al. 2012, 2014), with implications for giant exoplanet research as well as brown dwarfs (the brown dwarf exoplanet connection; e.g. Pinfield et al. 2013; Beichman et al. 2014; Faherty et al. 2013; Marocco et al. 2014; Liu et al. 2013; Canty et al. 2013). Given the widely varying Y dwarf spectral energy distributions (evident not only for WISE J1828+2650, but amongst other Y dwarfs sharing very similar T eff ; Dupuy & Kraus 2013), there is great incentive to calibrate this diversity against physical properties.
This requires the identification of Y dwarfs across their full parameter-space, as well as benchmarks systems with independently constrained properties (following on from e.g. Pinfield et al. 2006; Burningham et al. 2009;
Day-Jones et al. 2011; Pinfield et al. 2012; Gomes et al. 2013; Zhang et al. 2013; Deacon et al. 2014). Such samples will include the most unusual (and informative) observational outliers, as well as offering a means to calibrate the spectral sensitivities directly. Growth and increased diversity in the known Y dwarf population is thus important not only to test formation models (e.g. Stamatellos et al. 2011; Bate 2012) through population studies (following e.g. Pinfield et al. 2008; Kirkpatrick et al. 2012; Burningham et al. 2013; Day-Jones et al. 2013), but also to reveal the rare examples that will offer the strongest tests for atmosphere models.
In this paper we present a new Y dwarf, WISE J030449.03-270508.3 (hereafter WISE J0304-2705), confirmed from amongst the candidate sample identified by Pinfield et al. (2014). Section 2 summarises the WISE candidate selection method. Sections 3 and 4 describes our nearinfrared photometric and spectroscopic follow-up, with section 5 detailing our proper motion measurement. Section 6 discusses the spectral classification of WISE J0304-2705, and estimates for distance and kinematics. In addition we discuss our interpretation of the J -band spectral morphology of WISE J0304-2705, and the resulting theoretical implications for its surface gravity, age and mass. Conclusions are presented in Section 7.
## 2 IDENTIFICATION
WISE J0304-2705 was first identified as a candidate latetype object in Pinfield et al. (2014). The search method identified WISE All-Sky sources detected in the W 2band only, and probed down to low signal-to-noise levels ( w2snr /greaterorequalslant 8), targeting objects with faint W 2 magnitudes and red W 1 -W 2 colour. Spurious sources were removed using database selection criteria defined through analysis of a control sample comprising isolated point-like non-variable nonmoving sources from the SDSS. A basic summary of the selection and rejection criteria is given below (for a more detailed description see Pinfield et al. 2014);
- · Select sources only detected in the W 2 band with signal-to-noise w2snr /greaterorequalslant 8.
- · Require at least 8 individual exposures (in all bands; w /star m /greaterorequalslant 8 ) covering the sky position.
- · The line of sight extinction must be Av < 0.8 to remove reddened contamination.
- · Reject non-point-like sources for which the reduced χ 2 of the W 2 profile fit photometry ( w2rchi2 ) was > 1.2.
- · Reject sources for which the scatter in the multiple measurement photometry was higher than expected from the integrated flux uncertainty, if log ( w2sigp1 -w2sigmpro ) > 1 3 . -1 38 log ( . w2snr ).
- · Reject sources for which the number of detections in the individual W 2 frames was less than expected, if ( w2nm w2m / ) < . 1 8 log ( w2snr ) -1 7. .
- · Reject faint sources ( W 2 signal-to-noise from 8-10) within extended bright star halo regions, by comparing with 2MASS point-source-catalogue positions (see figure 2 of Pinfield et al. 2014).
- · Select final candidates by visual inspection, rejecting artefacts, resolved extended structures (e.g. nebulosity and
Figure 1. Finder images of WISE J0304-2705 (indicated with a red arrow). The top image is from Four Star in the J 3-band, and is 2 × 1 arcmin. The lower two are WISE images in the W 2- and W 1-bands, and are 5 × 5 arcmins on the side. In the top image north and east are indicated in blue, and a 20 arcsecond scale bar is shown. In the WISE images the size of J -band image is indicated with a dashed line box.
galaxies), badly blended sources, and sources with visual W 1, W 3 or W 4 detections.
Figure 1 shows WISE and follow-up images of WISE J0304-2705 (which is indicated with a red arrow). A J -band image is at the top (2 × 1 arcmins), with the WISE W 2 and W 1 images below (each 5 × 5 arcmins).
## 3 NEAR-INFRARED FOLLOW-UP
WISE J0304-2705 was initially followed up using the SofI (Son of ISAAC) instrument mounted on the European Southern Observatory's New Technology Telescope on La Silla. Observations were obtained on 2013 January 1 under photometric conditions, seeing 0.8 arcsec, and dark conditions. SofI is equipped with a 1024 × 1024 Hawaii HgCdTe offering a field-of-view of 4.9 × 4.9 arcmin and a pixel size of 0.292 arcsec in its Large-field configuration. J -band images were measured using integrations of 10 seconds repeated 6 times with a 10-point dither pattern yielding a total exposure of 10 min. Dome flats were obtained during the afternoon preceding the observations. The images were reduced with the ESO SofI pipeline (via the gasgano GUI). A faint low signal-to-noise source was seen in the resulting J -band image near the WISE position, so further follow-up was implemented.
Magellan / Four Star observations were obtained on 2013 August 15. Images were taken using the J 3 and J 2 filters. The J 2 filter is centred at 1.144 microns (FWHM=0.136 microns), where methane absorption produces strong suppression for late-T and Y brown dwarfs. The J 3 filter is centred at 1.288 microns (FWHM=0.140 mi- crons), an opacity hole for cool brown dwarf photospheres (see figure 1 of Tinney et al. 2012). Observations used individual integrations of 20 seconds, with 6 co-adds and a random 11 point jitter pattern, leading to total exposure times of 22 mins. The target was centered on detector 2 as it has the most uniform pixel sensitivity, providing a 5.4 arcmin FOV. The data were reduced using standard IRAF routines. Calibration was achieved in two ways. For the J 3 photometry 2MASS stars in the images were used (with 2MASS to MKO conversions from Carpenter 2001) to define the J 3 zero points, following Tinney et al. (2012). Saturation limited the choice of calibrators, and we thus had to make use of a limited number of red calibrators (foregoing the 0 4 . < J ( -K ) MKO < 0 8 requirement of Tinney et al. . 2012). Our J 3 zero point thus has sizable uncertainties at the level of ∼ 0.2 magnitudes. However, we were able to calibrate ( J 3 -J 2) colour directly by measuring the instrumental colours of many sources in the J 3 and J 2 images, and calculating an offset that led to an average colour of ( J 3 -J 2)=0. This offset was then applied to the instrumental ( J 3 -J 2) colour of WISE J0304-2705, which in turn was used to calibrate J 2. We determined ( J 3 -J 2) = -1 55 . ± 0 1, . providing strong evidence for significant methane absorption (which is known to produce colours of J 3 -J 2 < -0 5; . Tinney et al. 2012).
In addition, J and H photometry was obtained with FLAMINGOS-2 on Gemini South (Eikenberry et al. 2004) on 2013 December 22 (via program GS-2013B-Q-16), in the 'bright mode' read mode. Nine 60-second images were obtained at J , and thirty-six 10 second images at H , for total on-source exposures of 9 minutes and 6 minutes at J and H respectively. Routine photometric calibration is still being established for the instrument. We determined photometric zero-points using 15th to 16th magnitude 2MASS stars in the image, as well as observations made of photometric standards and of known late-type T dwarfs via a different Gemini program (Leggett personal communication 2014). These measurements show that the photometric system is equivalent to the Mauna Kea Observatories system (Tokunaga & Vacca 2005), and that non-linearity is not significant over the range of data counts in our images. We adopted zero-points of 24.91 ± 0.05 at J and 25.15 ± 0.05 at H , and used apertures of diameter 2 . ′′ 9 and 2 . ′′ 2 at J and H respectively, with annular sky regions and aperture corrections determined from stars in the image.
The photometric measurements of WISE J0304-2705, along with a summary of other measured and inferred properties, are presented in Table 1. The table gives WISE AllSky and AllWISE magnitudes, with colours (here and hereafter) derived using AllWISE measurements.
## 4 SPECTROSCOPY
Long-slit spectroscopy was obtained with FLAMINGOS-2 on Gemini South on 2013 September 29 (via program GS2013B-Q-16), in the 'faint mode' read mode. Twenty-four 300 second images were obtained, for a total on-source time of two hours. The JH grism and filter were used with the 4-pixel slit, producing a spectrum from 1.0 to 1.8 µ m (see Figure 2). Due to a problem with the delivered image quality across the field of view, the resolution varied from 4 nm in
Table 1. Coordinates, photometry and proper motion for WISE J0304-2705.
| Measured and inferred properties of WISE J0304-2705 | Measured and inferred properties of WISE J0304-2705 |
|-------------------------------------------------------|-------------------------------------------------------|
| R.A.(WISE) J2000 | 03h04m49.03s |
| Dec(WISE) J2000 | -27 ◦ 05'08.3' |
| Spectral type | Y0pec |
| Magnitudes | Magnitudes |
| W 2(All-Sky) | 15.60 ± 0.10 |
| W 1(All-Sky) | > 18.95 |
| W 2(AllWISE) | 15.59 ± 0.09 |
| W 1(AllWISE) | > 19.14 |
| W 3(AllWISE) | > 12.78 |
| W 4(AllWISE) | > 8.71 |
| J MKO | 20.79 ± 0.09 |
| H MKO | 21.02 ± 0.16 |
| J 3 | 21.26 ± 0.21 |
| J 2 | 22.81 ± 0.21 |
| Colours (using AllWISE magnitudes) | Colours (using AllWISE magnitudes) |
| W 1 - W 2 | > 3.55 |
| J MKO - W 2 | 5.20 ± 0.13 |
| H MKO - W 2 | 5.43 ± 0.18 |
| ( J - H ) MKO | -0.23 ± 0.18 |
| J 3 - J 2 | -1.55 ± 0.10 |
| Proper motion (arcsec/yr) | Proper motion (arcsec/yr) |
| µ α cos δ | -0.03 ± 0.10 |
| µ δ | 0.65 ± 0.10 |
| µ tot | 0.65 ± 0.14 |
| H W 2 | 19.7 ± 0.5 |
| Distance and kinematics | Distance and kinematics |
| D | 10-17 pc |
| V tan | 24-64 km s - 1 |
the center of the spectral range, to 7 nm at the extremes of the range, for a resolving power of 140 to 300. This resolution is sufficient for the purposes of classifying the object and broadly assessing its spectral morphology. The F3V star HD 27975 was used as a calibrator to remove telluric features. Flat fields and spectral arcs were obtained using lamps in the calibration unit mounted on the telescope. The data were reduced in the standard way using IRAF routines, with flux calibration achieved through division by the IRTF Spex library spectrum of HD 26015 (Rayner et al. 2009). Although HD 27975 and HD 26015 have B-V of 0.47 and 0.36 respectively (suggesting some extinction for HD 27975), we did not correct for this as it would not be significant above the level of the spectral signal-to-noise. The signal-to-noise in the peak of the Y -, J -and H -bands is 4, 5 and 3 respectively.
## 5 PROPER MOTION MEASUREMENT
We measured the proper motion of WISE J0304-2705 using our SofI J -band image as a first epoch (2013 January 1), and our FLAMINGOS-2 J image as a second epoch (2013 December 22), leading to a baseline of 0.97 years. We selected 21 reference stars to define an inter-epoch positional transform, which had associated root-mean-square residuals (when compared to the reference stars) of 0.05 arcsecs. J -
Figure 2. The near-infrared Flamingos-2 spectrum of WISE J0304-2705 (plotted in red). The top plot shows the J -band region, and the lower plot the full YJH region. Several spectra are over-plotted for spectral typing and comparison; UGPS 0722-05 (T9 spectral standard in blue), WISE J1738+2732 (Y0 spectral standard in green), WISE J0350-5658 (Y1 spectral standard in orange), and red-optical WISE J2056+1459 (Y0 in purple). All spectra are normalised to an average of unity from 1.265-1.270 microns.
band counterpart centroiding uncertainty was estimated to be 0.08 arcsecs by assessing the scatter in centroid measurements of a simulated population of sources with Gaussian point spread functions and added Poisson noise. Fit residuals and centroiding uncertainties were combined in quadrature to give full proper motion uncertainties. WISE J0304-2705 was found to be a high proper motion source, moving at 0.65 ± 0.14 arcsec/yr. The proper motion measurements are listed in Table 1.
## 6 DISCUSSION
## 6.1 Spectral classification
The top plot in Figure 2 shows a zoom-in of the J -band flux peak region, with the full Y JH spectrum in the lower plot. The spectrum of WISE J0304-2705 is shown in red. For visual spectral typing and comparison we over-plot the spectral standards UGPS 0722-05 (T9; blue), WISE J1738+2732 (Y0; green), and WISE J0350-5658 (Y1; orange), as well as the red-optical spectrum of the Y dwarf WISE J2056+1459 (Cushing et al. 2011; Leggett et al. 2013). The spectra have been normalised to an average of unity in the 1.265-1.270 micron range (including WISE J2056+1459, though nearinfrared not shown).
The highest signal-to-noise is in the central region of the J -band flux peak, and from 1.22-1.29 microns the spectrum of WISE J0304-2705 is well represented by the Y0 spectral standard. It is a poor match to the T9 standard in this range, and at shorter J -band wavelengths the noise precludes a useful comparison. However, over the 1.29-1.33 micron range the average flux level of WISE J0304-2705 is brighter than the Y dwarf standards, being closer to the
T9 standard. In addition, in the Y -band (over at least the 0.974-1.033 micron range covered by the spectra) it can be seen that WISE J0304-2705 is relatively bright compared to the comparison Y0 dwarf WISE J2056+1459. Aside from these two main differences (which will be discussed further in Section 6.3) the full spectrum of WISE J0304-2705 is well represented by the Y0 spectral standard. To assess the J -band flux enhancement we define a new spectral ratio J-wing as follows.
$$J - \text{wing} & = \frac { \tilde { F } _ { 1. 2 9 0 - 1. 3 3 0 \mu m } } { \tilde { F } _ { 1. 2 6 5 - 1. 2 7 0 \mu m } } & ( 1 ) & \frac { \frac { \mu } { S _ { f } } } { R _ { f } }$$
The denominator wavelength range is the same as for the J-narrow ratio, but the numerator fully samples the enhanced flux region for WISE J0304-2705. We note that the CH4-J ratio (Burgasser et al. 2006b) also samples this region at some level, but only > 1.315 microns. Table 2 presents spectral ratio measurements for WISE J0304-2705 as well as for the T9, Y0 and Y1 spectral standards (median ratios were derived following Mace et al. 2013a). Spectral type average ratios are also listed aside the T9 and Y0 spectral standard values (from Mace et al. 2013a). We measured J -band ratios (including previously defined ratios and the new J-wing ratio), as well as flux peak ratios involving the Y , J and H flux peaks. We do not present H -band ratios since the signal-to-noise of the WISE J0304-2705 spectrum is too low for these to be useful.
In the J -band peak, the J-narrow ratio value is midway between the Y0 and Y1 spectral standard values. The Wj ratio value is similar to T9 expectations, but the uncertainties are relatively large due to the low signal-to-noise in the numerator region, and this ratio does not strongly differentiate between T9 and Y. The value of J-wing for WISE J0304-2705 is comparable to expectations for T9, being a factor ∼ 2 greater than for the Y0 spectral standard. Also, the value of CH4-J is similar to T9, being rather greater than the range expected for Y dwarfs (with a previously observed maximum CH4-J for Y0 dwarfs of 0.096 ± 0.010 for WISE J2056+1459; see Mace et al. 2013a). The higher values for J-wing and CH4-J result from the relatively enhanced flux in the red wing of the J -band flux peak. The H2O-J ratio is saturated, and the measured value does not differentiate between late-T and Y types. Considering the full Y JH spectrum it can be seen that the Y/J ratio is significantly greater than expectations for either T9 or Y0 spectral types. A Y -band enhacement is also clearly seen by direct comparison with the spectrum of WISE J2056+1459. The H/J ratio is very similar to the Y0 spectral standard and Y0 average value, being rather lower than the T9 and Y1 values.
We base our spectral classification of WISE J0304-2705 primarilly on the visual comparison with the spectral standards in the J -band wavelength range 1.22-1.29 microns. This region offers the best signal-to-noise, and avoids the 1.29-1.33 micron range where we note relative enhancement (as will be discussed in Section 6.3, models suggest this region is sensitive to gravity and metallicity, and as such is not ideal for spectral classification). Our typing also acknowledges the unusual J -band morphology that makes the object and outlier in e.g. J-narrow J-wing space. We thus classify WISE J0304-2705 as Y0pec. Only one other Y0 dwarf has been classified as peculiar (WISE J1405+5334), which was as a result of its H -band flux peaking at a wavelength
| WISE J0304-2705 | WISE J0304-2705 | |
|---------------------------|---------------------------|---------------------------|
| Ratio | Median | Integrated |
| H2O-J | -0.009 ± 0.037 | -0.016 ± 0.026 |
| CH4-J | 0.150 ± 0.031 | 0.167 ± 0.021 |
| W J | 0.181 ± 0.025 | 0.200 ± 0.018 |
| J-narrow | 0.759 ± 0.054 | 0.776 ± 0.040 |
| Y/J | 0.733 ± 0.052 | 0.702 ± 0.035 |
| H/J | 0.538 ± 0.034 | 0.532 ± 0.023 |
| J-wing | 0.389 ± 0.039 | 0.409 ± 0.026 |
| Spectral type comparisons | Spectral type comparisons | |
| Ratio | Spectral standard 1 | Average (per half type) 2 |
| T9 | T9 | T9 |
| H2O-J | 0.041 ± 0.027 2 | 0.032 ± 0.037 |
| CH4-J | 0.156 ± 0.026 2 | 0.117 ± 0.030 |
| W J | 0.204 ± 0.052 2 | 0.203 ± 0.038 |
| J-narrow | 0.862 ± 0.040 2 | 0.879 ± 0.053 |
| Y/J | 0.420 ± 0.050 2 | 0.448 ± 0.095 |
| H/J | 0.530 ± 0.020 2 | 0.555 ± 0.030 |
| J-wing | 0.320 ± 0.005 | - |
| Y0 | Y0 | Y0 |
| H2O-J | 0.039 ± 0.009 2 | -0.022 ± 0.050 |
| CH4-J | 0.052 ± 0.010 2 | 0.045 ± 0.031 |
| W J | 0.122 ± 0.028 2 | 0.117 ± 0.043 |
| J-narrow | 0.839 ± 0.027 2 | 0.778 ± 0.050 |
| Y/J | - | 0.423 ± 0.131 |
| H/J | 0.460 ± 0.020 2 | 0.467 ± 0.061 |
| J-wing | 0.164 ± 0.012 | - |
| Y1 | Y1 | Y1 |
| H2O-J | -0.055 ± 0.071 2 | - |
| CH4-J | 0.068 ± 0.049 2 | - |
| W J | 0.116 ± 0.096 2 | - |
| J-narrow | 0.676 ± 0.095 2 | - |
| Y/J | - | - |
| H/J | 0.751 ± 0.081 | - |
| J-wing | 0.136 ± 0.038 | - |
- 1 T9=UGPS 0722-05, Y0=WISE J1738+2732
- Y1=WISE J0350-5658
- 2 Taken from Mace et al. (2013a).
Table 2. Spectral ratios for WISE J0304-2705, and the T9, Y0 and Y1 comparison objects. Median flux ratios were calculated using multiple realizations of a Monte Carlo spectrum model with flux density values (at each wavelength) randomly drawn from a normal distribution with mean and standard deviation set by the measured values and associated uncertainties.
60 ˚ redder compared to other Y dwarfs and the T9 spectral A standard.
## 6.2 Distance, kinematics and reduced proper motion
In Table 3 we present spectrophotometric distances estimates for WISE J0304-2705, derived using the available J -, H -and W 2-band photometry. We did the calculations assuming a spectral type of Y0, and defined absolute magnitudes using Dupuy & Kraus (2013), converting from M [4 5] . into M W 2 using an average colour of W 2 -[4 5]=0.08 . ± 0.06 (for the 3 Y0 dwarfs used in their calculation).
The three bands yield distance constraints that over-
Table 3. Spectrophotometric distance estimates for WISE J03042705. Estimates are made using the J , H and W 2 apparent magnitudes, and assuming a Y0 spectral type.
| | Distance (pc) |
|-----------------|-----------------|
| J MKO H MKO W 2 | 12.2-15.6 |
| | 10.4-14.1 |
| | 12.9-16.9 |
lap, covering a range 10.4-16.9 pc. This is not surprising when one considers Figure 3. These plots place WISE J0304-2705 in the colour-spectral-type space of the lateT and Y dwarf population (data from Leggett et al. 2013; Mace et al. 2013b; Pinfield et al. 2014; Kirkpatrick et al. 2014; Burningham et al. 2014), and it can be seen that WISE J0304-2705 occupies essentially the same colour space as the previously known Y0 dwarfs, though at the bluer end in J -W 2 and H -W 2. Our best estimate for the distance of WISE J0304-2705 is thus 10-17 pc, appropriate for a single object. However, if WISE J0304-2705 was found to be an unresolved binary then it could be as distant as 24 pc.
By combining our distance constraint with the measured proper motion of WISE J0304-2705, we estimate a tangential velocity of V tan =24-64 km s -1 which is consistent with thin disk membership (e.g. Nissen 2004; Faherty et al. 2009). In addition Figure 4 plots WISE J0304-2705 in reduced proper motion diagrams (as a filled red symbol). Reduced proper motion ( H W 2 = W 2 + 5 log µ + 5) is shown against both W 1 -W 2 colour and spectral type, with disk L and T dwarfs shown as filled blue circles, and other Y dwarfs in green. Sub dwarfs from the thick-disk/halo are blue open circles (L dwarfs) or orange symbols (T dwarfs; Mace et al. 2013b; Pinfield et al. 2014; Burningham et al. 2014). Black lines delineate V tan ranges within the diagrams, and WISE J0304-2705 can be seen to lie in the middle of the Y dwarf population. Kinematically, WISE J0304-2705 thus appears to be a typical thin disk Y dwarf.
## 6.3 Unusual spectral features
The full spectrum of WISE J0304-2705 (see Figure 2) shows that the Y -band flux peak is unusual. It is significantly broadened to the blue, with a peak between ∼ 1.02-1.04 microns. By contrast the Y -band spectrum of the comparison Y0 dwarf WISE J2056+1459 (in purple) is fainter and rising steeply over this range, with its Y -band flux peaking closer to the typical value for most late-T dwarfs ( ∼ 1.08 microns). These unusual Y -band characteristics have been considered likely indicators of sub-solar metallicity for some time (Burrows et al. 2002; Burgasser et al. 2006a), and have recently been observed in the spectrum of the benchmark T8 Wolf 1130B, a companion to an M subdwarf with metallicity [Fe/H]=-0.64 ± 0.17 (Mace et al. 2013b). By contrast the mildly metal-poor ([Fe/H]=-0.38 ± 0.06) T8 benchmark BD+01 2920 has a Y -band flux peak that is very similar to the T8 spectral standard 2MASS J0415-09 (see fig 2 of Pinfield et al. 2012). Assuming this metallicity sensitivity extends over the T/Y transition, then WISE J03042705 may have lower metallicity than the Y0 dwarf WISE J2056+1459, and could have sub-solar composition.
WISE J0304-2705 also shows a relative flux enhancement in the J -band (over the 1.29-1.33 micron range), when compared to the Y spectral standards. To assess if unresolved binarity might lead to such spectral morphology we considered a T9+Y0 composite spectrum. Two such binaries were studied by Liu et al. (2012), with the components of WISE J1217+1626AB having a J -band brightness difference of 2.10 magnitudes and the components of CFBDSIR J1458+1013AB a difference of 2.02 magnitudes. The average J -band brightness difference between single T9 and Y0 dwarfs is ∼ 1.7 magnitudes (Dupuy & Kraus 2013). We compared the spectrum of WISE J0304-2705 to composite T9+Y0 spectra (constructed using the spectral standard spectra) covering a range of possible J -band magnitude differences from 1.5-2.0 magnitudes. None of the composite spectra came close to replicating the J -band morphology of WISE J0304-2705.
For a single object explanation of the J -band flux enhancement comparison with T8 benchmarks is once again instructive, as the red wing of the J -band flux peak of Wolf 1130B also shows an unusual morphology. Figure 5 of Mace et al. (2013b) shows that this object has greatly enhanced flux in the red wing of the J -band flux peak compared to BD+01 2920 (which has a similar J -band flux peak to the T8 standard 2MASS J0415-09). This type of spectral morphology is thus characteristic of very late objects that are old (with relatively high surface gravity) and/or have low metallicity, although the enhancement seen for WISE J0304-2705 is at a significantly lower level than for Wolf 1130B.
We have analysed this spectroscopic feature in the context of log g sensitivity, through comparison with the cloudy models of Morley et al. (2012). In Figure 5 model spectra are shown for T eff values of 500K, 450K and 400K (covering the Y0 T eff range in e.g. Dupuy & Kraus 2013), and log g of 4.0, 4.5 and 5.0 (red, blue and green respectively) appropriate for the T eff range. For each T eff and log g combination the models are available with up to four different sedimentation efficiencies ( fsed =2, 3, 4, 5). The 1.29-1.33 micron range is enclosed by dashed lines in Figure 5, and it can be seen that the models have significant sensitivity to log g in this range.
We can use the calculated trends with gravity for the J-wing spectral ratio to estimate gravity and hence age for WISE J0304-2705. Table 2 shows that the value of this spectral ratio is a factor of at least two larger than the equivalent value for the Y0 standard (Table 4). Using the model suite described in the previous paragraph, we calculated J-wing ratios as a function of gravity for each set of [Teff,fsed]. Table 4 gives the resulting enhancement as gravity is increased (where an enhancement factor of 1 means no enhancement).
For the theoretical log g changes shown in Table 4 the only one that yields enhancement consistent with WISE J0304-2705 is log g =4.0 → 5.0, which would suggest an age close to 10 Gyr and a mass of 0.02-0.03 M /circledot (Saumon & Marley 2008). However, a more complete theoretical analysis should also assess the effects of sub-solar metallicity as well as high surface gravity, which requires the generation of new model spectra.
We have also made a direct comparison with the model spectra. Figure 6 shows the Y JH spectrum of WISE J0304-2705 (grey) with the indicated models over-plotted ( T eff =400 and 450 K, log g =4.5 and 5.0, and f sed =4 and
Figure 3. Colour versus spectral type plots. WISE J0304-2705 is plotted in red. L and T dwarfs from Leggett et al. (2013) are plotted in blue (ignoring known unresolved binaries). Late-T subdwarfs with thick-disk/halo kinematics are shown in orange. Available Y dwarf photometry is plotted in green.
Figure 4. Reduced proper motion (H W 2) plotted against W 1 -W 2 colour (left hand plot) and spectral type (right hand plot). WISE J0304-2705 is plotted in red. L and T dwarfs (from the WISE team's census and other WISE cross-matches from the literature) are plotted as blue circles (subdwarfs are open symbols). Y dwarfs with measured proper motion are plotted in green. Four late-T sub-dwarfs with thick-disk/halo kinematics are plotted in orange (WISE J0013+0634; WISE J0833+0052; Wolf 1130B; ULAS J1316+0755). Rightand down-facing triangles indicate lower limits. Tracks of constant tangential velocity are shown in black.
5). These models offer a reasonable matched with the observed relative brightness of the J -and H -band flux peaks. The observed Y -band flux peak shows significantly different morphology than the models, as expected from our previous discussion. In the J -band the 400 K models give the best match to the blue wing of the flux peak, with the 450 K models being relatively brighter. In the red wing of the flux peak the gravity sensitive enhancement is apparent between the log g =4.5 and 5.0 models, though the high gravity 450 K model gives the closest match.
of an unresolved WISE multiple system. A comparison between the J -band spectrum of WISE J0304-2705 and previously known Y dwarfs shows that this object has some unusual spectral features, including relatively enhanced flux in the J - (1.29-1.33 microns) and Y - (inclusive of 0.974-1.033 microns) bands. Based on observed trends for the warmer late-type T dwarfs these enhancements are both characteristic of sub-solar metallicity objects that may be old and high gravity.
## 7 CONCLUSIONS AND FUTURE WORK
The discovery of WISE J0304-2705 identifies the 18th spectroscopically confirmed Y dwarf, which includes 17 that were identified as WISE sources and one that was a component
In the future, additional higher signal-to-noise spectroscopy of WISE J0304-2705 can assess the stability of the relative flux enhancements, constraining any variability in these features (i.e. if Y dwarf weather is important; Cushing et al. 2014a). Adaptive optics imaging will be useful to assess if WISE J0304-2705 is an unresolved multiple or a single object, and thus if its spectrum may be a composite of more than one spectral type. Radial velocity mea-
Figure 5. Plots of the J -band spectral region showing the Morley et al. (2012) model spectra. The three plots show model T eff =400K, 450K and 500K respectively. Different surface gravities (log g =4.0, 4.5 and 5.0) are plotted in red, blue and green respectively. For each value of log g there are up to four spectra representing different sedimentation efficiencies ( f sed =2, 3, 4, 5). Vertical dotted lines enclose the spectral range 1.29-1.33 microns.
Figure 6. The spectrum of WISE J0304-2705 with models over-plotted, as indicated. All spectra have been normalised to an average of unity from 1.265-1.270 microns.
surements, as a further test of multiplicity and to determine space motion, are currently beyond the capabilities of the largest telescopes. But a parallax measurement combined with Spitzer photometry will be crucial for determining a robust T eff constraint, and K -band photometry of WISE J0304-2705 (and other Y dwarfs) will show if this object is K -band suppressed, a known signature of high surface gravity and/or low metallicity for T dwarfs (e.g. Pinfield et al. 2012), caused by increased collisionally induced H 2 absorption.
We did not find any common proper motion companions to WISE J0304-2705 out to a separation of 10 arcmins, using the Hipparcos catalogue (van Leeuwen 2007) and the SuperCOSMOS Science Archive (Hambly et al. 2001). Also, we found no kinematic membership of the moving groups listed in Clarke et al. (2010) and Torres et al. (2008) after assess- ing proper motion and distance constraints for a possible range of radial velocity (-200 km s -1 to +200 km s -1 ). However, as the Y dwarf population grows, benchmark systems should become more numerous. We note that WISE J06476232 (Y1: Kirkpatrick et al. 2014) could be a benchmark Y dwarf in the Columba moving group, though radial velocity confirmation is not currently possible. And WD 0806-661B is a benchmark object but does not have a measured spectrum. As more Y dwarfs are uncovered, the confluence of observed spectral variations with predictions from theory and calibration from benchmark systems, should lead to an improved understanding of brown dwarfs and giant exoplanets with T eff < 500 K.
Table 4. Enhancement levels in the 1.29-1.33 micron range, as represented by the J-wing ratio. In the top of the table the relative enhancement for WISE J0304-2705 is given relative to the Y0 spectral standard. The lower portion of the table presents theoretical predictions for T eff =400-500K, f sed =2,3,4,5.
| Ratio / Ratio | Enhancement factor |
|-------------------------------|----------------------|
| J-wing (0304) / J-wing (Y0) 1 | 1.99-2.82 |
| ∆log g | Enhancement factor |
| 4.0 → 4.5 | 1.18-1.41 |
| 4.5 → 5.0 | 1.30-1.46 |
| 4.0 → 5.0 | 1.56-2.06 |
| 1 Y0=WISE J1738+2732 | |
## ACKNOWLEDGMENTS
This publication makes use of data products from the Widefield Infrared Survey Explorer, which is a joint project of the University of California, Los Angeles, and the Jet Propulsion Laboratory/California Institute of Technology, funded by the National Aeronautics and Space Administration. This paper includes data gathered with the 6.5 meter Magellan Telescopes located at Las Campanas Observatory, Chile. Based on observations obtained at the Gemini Observatory, which is operated by the Association of Universities for Research in Astronomy, Inc., under a cooperative agreement with the NSF on behalf of the Gemini partnership: the National Science Foundation (United States), the National Research Council (Canada), CONICYT (Chile), the Australian Research Council (Australia), Ministrio da Ciˆncia, Tecnoloe gia e Inova¸˜o (Brazil) and Ministerio de Ciencia, Tecnolog´ ıa ca e Innovaci´n Productiva (Argentina). Based on observations o made with ESO Telescopes at the La Silla Paranal Observatory under programme ID ppp.c-nnnn. DP, NL and ADJ have received support from RoPACS during this research and JG was supported by RoPACS, a Marie Curie Initial Training Network funded by the European Commissions Seventh Framework Programme. BB is supported by a Science and Technology Research Council (STFC) Consolidated Grant. MG is financed by the GEMINI-CONICYT Fund, allocated to the project 32110014. SKL is supported by the Gemini Observatory, which is operated by AURA, on behalf of the international Gemini partnership of Argentina, Australia, Brazil, Canada, Chile, the United Kingdom, and the United States of America. NL is a Ramon y Cajal at the IAC in Tenerife (fellowship number 08-303-0102) and is funded by the national programme AYA201019136 of the Spanish ministry of economy and competitiveness (MINECO). RK acknowledges partial support from FONDECYTthrough grant 1130140. ADJ was supported by a Fondecyt Postdoctorado under project number 3100098. MTR received support from PB06 (CONICYT). This research has made use of the SIMBAD database, operated at CDS, Strasbourg, France.
## REFERENCES
Allard F., Homeier D., Freytag B., 2011, in Astronomical Society of the Pacific Conference Series, Vol. 448, 16th Cambridge Workshop on Cool Stars, Stellar Systems, and the Sun, Johns-Krull C., Browning M. K., West A. A., eds., p. 91
- Baraffe I., Chabrier G., Barman T. S., Allard F., Hauschildt P. H., 2003, A&A, 402, 701
Bate M. R., 2012, MNRAS, 419, 3115
- Beichman C., Gelino C. R., Kirkpatrick J. D., Cushing M. C., Dodson-Robinson S., Marley M. S., Morley C. V., Wright E. L., 2014, ApJ, 783, 68
- Burgasser A. J., Burrows A., Kirkpatrick J. D., 2006a, ApJ, 639, 1095
- Burgasser A. J., Geballe T. R., Leggett S. K., Kirkpatrick J. D., Golimowski D. A., 2006b, ApJ, 637, 1067
- Burningham B. et al., 2013, MNRAS, 433, 457
- Burningham B. et al., 2008, MNRAS, 391, 320
- Burningham B. et al., 2009, MNRAS, 395, 1237
- Burningham B. et al., 2010, MNRAS, 406, 1885
- Burningham B., Smith L., Cardoso C. V., Lucas P. W., Burgasser A. J., Jones H. R. A., Smart R. L., 2014, MNRAS, 440, 359
- Burrows A., Burgasser A. J., Kirkpatrick J. D., Liebert J., Milsom J. A., Sudarsky D., Hubeny I., 2002, ApJ, 573, 394
- Burrows A., Heng K., Nampaisarn T., 2011, ApJ, 736, 47 Burrows A., Hubeny I., Budaj J., Knutson H. A., Charbonneau D., 2007, ApJ, 668, L171
- Burrows A., Sudarsky D., Lunine J. I., 2003, ApJ, 596, 587 Canty J. I., Lucas P. W., Roche P. F., Pinfield D. J., 2013, MNRAS, 435, 2650
- Carpenter J. M., 2001, AJ, 121, 2851
- Chabrier G., Baraffe I., 2000, ARA&A, 38, 337
- Clarke J. R. A. et al., 2010, MNRAS, 402, 575
- Cushing M., Hardegree-Ullman K., Trucks J., 2014a, in American Astronomical Society Meeting Abstracts, Vol. 223, American Astronomical Society Meeting Abstracts, p. 425.08
- Cushing M. C. et al., 2011, ApJ, 743, 50
- Cushing M. C., Kirkpatrick J. D., Gelino C. R., Mace G. N., Skrutskie M. F., Gould A., 2014b, AJ, 147, 113
Day-Jones A. C. et al., 2013, MNRAS, 430, 1171
Day-Jones A. C. et al., 2011, MNRAS, 410, 705
- Deacon N. et al., 2014, in American Astronomical Society Meeting Abstracts, Vol. 223, American Astronomical Society Meeting Abstracts, p. 116.03
Delorme P. et al., 2008a, A&A, 482, 961
Delorme P. et al., 2008b, A&A, 484, 469
Dupuy T. J., Kraus A. L., 2013, Science, 341, 1492
- Eikenberry S. S. et al., 2004, in Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series, Vol. 5492, Ground-based Instrumentation for Astronomy, Moorwood A. F. M., Iye M., eds., pp. 1196-1207
- Faherty J. K., Burgasser A. J., Cruz K. L., Shara M. M., Walter F. M., Gelino C. R., 2009, AJ, 137, 1
- Faherty J. K., Rice E. L., Cruz K. L., Mamajek E. E., N´ u˜ez A., 2013, AJ, 145, 2 n
- Gomes J. I. et al., 2013, MNRAS, 431, 2745
Hambly N. C. et al., 2001, MNRAS, 326, 1279
Kirkpatrick J. D., 2005, ARA&A, 43, 195
| Kirkpatrick J. D., Cushing M. C., Gelino C. R., Beichman C. A., Tinney C. G., Faherty J. K., Schneider A., Mace G. N., 2013, ApJ, 776, 128 J. D. et al., 2011, ApJS, 197, 19 J. D. et al., 2012, ApJ, 753, 156 J. D. et al., 2014, ApJ, 783, 122 |
|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| Kirkpatrick Kirkpatrick |
| Kirkpatrick |
| Lawrence A. et al., 2007, MNRAS, 379, 1599 Leggett S. K. et al., 2010, ApJ, 710, 1627 |
| S. K., Morley C. V., Marley M. S., Saumon Fortney J. J., Visscher C., 2013, ApJ, 763, 130 |
| Leggett D., Liu M. C. et al., 2011, ApJ, 740, 108 |
| Liu M. C., Dupuy T. J., Bowler B. P., Leggett S. K., W. M. J., 2012, ApJ, 758, 57 |
| Best Liu M. C. et al., 2013, ApJ, 777, L20 |
| Lodieu N., B´jar e V. J. S., Rebolo R., 2013, A&A, 550, L2 Lodieu N. et al., 2012, A&A, 548, A53 |
| P. W. et al., 2010, MNRAS, 408, L56 K. L., 2013, ApJ, 767, L1 |
| Lucas Luhman Luhman K. L., 2014, ApJ, 786, L18 |
| Luhman K. L., Burgasser A. J., Bochanski J. J., |
| 2011, ApJ, 730, L9 |
| G. N. et al., 2013a, ApJS, 205, 6 |
| Mace G. N. et al., 2013b, ApJ, 777, 36 |
| Mace Mainzer A. et al., 2011a, ApJ, 726, 30 |
| Mainzer A. et al., 2011b, ApJ, 743, 156 |
| Marocco F. et al., 2014, MNRAS, 439, 372 |
| McMahon R. G., Banerji M., Gonzalez E., |
| Bejar V. J., Lodieu N., Rebolo R., the |
| Koposov S. E., 2013, The Messenger, 154, 35 C. V., Fortney J. J., Marley M. S., Visscher |
| VHSCollaboration, Morley C., Saumon D., Leggett S. K., 2012, ApJ, 756, 172 |
| Morley C. V., Marley M. S., Fortney J. J., Lupu R., Saumon D., Greene T., Lodders K., 2014, ArXiv e-prints Nissen P. E., 2004, Origin and Evolution of the 154 |
| Elements, Pinfield D. J., Beaulieu J.-P., Burgasser A. J., Delorme P., Gizis J., Konopacky Q., 2013, Mem. Soc. Astron. Italiana, 84, 1154 |
| D. J. et al., 2012, MNRAS, 422, 1922 |
| D. J. et al., 2008, MNRAS, 390, 304 |
| D. J. et al., 2014, MNRAS, 437, 1009 |
| Pinfield D. J., Jones H. R. A., Lucas P. W., Kendall T. |
| R., Folkes S. L., Day-Jones A. C., Chappelle R. J., I. A., 2006, MNRAS, 368, 1281 |
| Steele J. T., Cushing M. C., Vacca W. D., 2009, ApJS, |
| 185, 289 |
| D., Marley M. S., 2008, ApJ, 689, 1327 |
| M. F. et al., 2006, AJ, 131, 1163 D., Maury A., Whitworth A., Andr´ e P., |
| MNRAS, 413, 1787 C. G., Faherty J. K., Kirkpatrick J. D., E. L., Gelino C. R., Cushing M. C., Griffith R. L., |
| G., 2012, ApJ, 759, 60 A. T., Vacca W. D., 2005, PASP, 117, 421 |
| C. A. O., Quast G. R., Melo C. H. F., Sterzik M. F., 2008, Young Nearby Loose Associations, p. 757 |
| Leeuwen F., 2007, A&A, 474, 653 |
| S. J. et al., 2007, MNRAS, 381, 1400 |
| E. L. et al., 2010, AJ, 140, 1868 |
| York D. G. et al., 2000, AJ, 120, 1579 |
| Wright |
| van Warren |
| Torres |
| Wright Salter |
| 2011, |
| Rayner Saumon Skrutskie |
| Pinfield |
| Pinfield |
| Pinfield |
| Stamatellos |
| Tinney |
| Tokunaga | | 10.1093/mnras/stu1540 | [
"D. J. Pinfield",
"M. Gromadzki",
"S. K. Leggett",
"J. Gomes",
"N. Lodieu",
"R. Kurtev",
"A. C. Day-Jones",
"M. T. Ruiz",
"N. J. Cook",
"C. V. Morley",
"M. S. Marley",
"F. Marocco",
"R. L. Smart",
"H. R. A. Jones",
"P. W. Lucas",
"Y. Beletsky",
"V. D. Ivanov",
"B. Burningham",
"J. S. Jenkins",
"C. Cardoso",
"J. Frith",
"J. R. A. Clarke",
"M. C. Galvez-Ortiz",
"Z. Zhang"
] | 2014-08-01T20:00:08+00:00 | 2014-08-01T20:00:08+00:00 | [
"astro-ph.SR"
] | Discovery of a new Y dwarf: WISE J030449.03-270508.3 | We present a new Y dwarf, WISE J030449.03-270508.3, confirmed from a
candidate sample designed to pick out low temperature objects from the WISE
database. The new object is typed Y0pec following a visual comparison with
spectral standards, and lies at a likely distance of 10-17 pc. Its tangential
velocity suggests thin disk membership, but it shows some spectral
characteristics that suggest it may be metal-poor and/or older than previously
identified Y0 dwarfs. Based on trends seen for warmer late type T dwarfs, the
Y-band flux peak morphology is indicative of sub-solar metallicity, and the
enhanced red wing of the J-band flux peak offers evidence for high gravity
and/or low metallicity (with associated model trends suggesting an age closer
to ~10 Gyr and mass in the range 0.02-0.03 Mo). This object may thus be
extending the population parameter-space of the known Y0 dwarfs. |
1408.0285v1 | ## Observations of nitrogen isotope fractionation in deeply embedded protostars
S. F. Wampfler 1 2 , , J. K. Jørgensen 2 1 , , M. Bizzarro 1 , and S. E. Bisschop 1 2 ,
- 1 Centre for Star and Planet Formation, Natural History Museum of Denmark, University of Copenhagen, Øster Voldgade 5-7, DK-1350 København K, Denmark
- 2 Niels Bohr Institute, University of Copenhagen, Juliane Maries Vej 30, DK-2100 København Ø, Denmark
Received 7 March 2014 / Accepted 26 July 2014
## ABSTRACT
Context. The terrestrial planets, comets, and meteorites are significantly enriched in 15 N compared to the Sun and Jupiter. While the solar and jovian nitrogen isotope ratio is believed to represent the composition of the protosolar nebula, a still unidentified process has caused 15 N-enrichment in the solids. Several mechanisms have been proposed to explain the variations, including chemical fractionation. However, observational results that constrain the fractionation models are scarce. While there is evidence of 15 Nenrichment in prestellar cores, it is unclear how the signature evolves into the protostellar phases.
Aims.
The aim of this study is to measure the
14
N 15
/
N ratio around three nearby, embedded low- to intermediate-mass protostars.
Methods. Isotopologues of HCN and HNC were used to probe the 14 N 15 / N ratio. A selection of J = 3-2 and 4-3 transitions of H 13 CN, HC 15 N, HN 13 C, and H 15 NC was observed with the Atacama Pathfinder EXperiment telescope (APEX). The 14 N 15 / N ratios were derived from the integrated intensities assuming a standard 12 C 13 / C ratio. The assumption of optically thin emission was verified using radiative transfer modeling and hyperfine structure fitting.
Results. Two sources, IRAS 16293A and R CrA IRS7B, show 15 N-enrichment by a factor of ∼ 1 5 . -2 5 in both HCN and HNC . with respect to the solar composition. IRAS 16293A falls in the range of typical prestellar core values. Solar composition cannot be excluded for the third source, OMC-3 MMS6. Furthermore, there are indications of a trend toward increasing 14 N 15 / N ratios with increasing outer envelope temperature.
Conclusions. The enhanced 15 Nabundances in HCN and HNC found in two Class 0 sources ( 14 N 15 N / ∼ 160 -290) and the tentative trend toward a temperature-dependent 14 N 15 / N ratio are consistent with the chemical fractionation scenario, but 14 N 15 / N ratios from additional tracers are indispensable for testing the models. Spatially resolved observations are needed to distinguish between chemical fractionation and isotope-selective photochemistry.
Key words. Astrochemistry - ISM: abundances - ISM: molecules - Stars: formation - Stars: individual: IRAS 16293-2422, R CrA IRS7B, OMC-3 MMS6
## 1. Introduction
The largest isotopic anomalies in the solar system are found in highly volatile elements like hydrogen, carbon, oxygen, and nitrogen (e.g., Birck 2004; Anders & Grevesse 1989). Thereof, the most extreme variations among solar system objects exist in the deuterium-to-hydrogen ratio, D H, and the ratio of the two / stable nitrogen isotopes, 14 N 15 N / /star . Recent results from the Genesis mission (Marty et al. 2011), which sampled the solar wind, have demonstrated that there is a large discrepancy in the nitrogen isotopic composition between the solar system solids and the gas reservoir: The 14 N 15 N ratios of the terrestrial planets, / comets (e.g., Hutsemékers et al. 2005; McKeegan et al. 2006; Bockelée-Morvan et al. 2008; Manfroid et al. 2009; Jehin et a l. 2009; Mumma & Charnley 2011; Shinnaka et al. 2014), meteorites (e.g., Kerridge 1993; Briani et al. 2009; Bonal et al. 2010), and interplanetary dust particles (IDPs, Messenger 2000; Floss et al. 2006) are significantly enhanced compared to the Sun ( 14 N 15 / N = 441 ± 6, Marty et al. 2011) and Jupiter (435 ± 57,
Owen et al. 2001; Fouchet et al. 2004; Abbas et al. 2004). The derived ratios for the Earth's atmosphere (272, e.g., Junk & Svec 1958; Mariotti 1983), Venus (Ho ff man et al. 1979), and the interior of Mars (Mathew & Marti 2001) are very similar and also lie within the range spanned by most primitive meteorites (e.g., Fig. 2 from Meibom et al. 2007). The nitrogen isotopic composition of the Sun and Jupiter, on the other hand, is substantially 15 N-poorer (Marty et al. 2011; Hashizume et al. 2000) and believed to be representative of the presolar nebula. This idea is supported by the results from an isotopic measurement of a high-temperature condensate - titanium nitride (TiN), named osbornite - in a calcium-aluminum-rich inclusion (CAI) of the 'Isheyevo' meteorite. CAIs are believed to be the first solids that condensated from the gas of the presolar nebula. The 14 N 15 N / ratio measured from this osbornite (424 ± 3, Meibom et al. 2007) is therefore interpreted as reflecting the nitrogen isotope composition of the solar nebula at the time of CAI formation 4567 3 . ± 0 16 Myr ago (Connelly et al. 2012). .
/star In cosmochemistry, nitrogen isotope ratios are generally reported in per mil deviations from the terrestrial atmospheric standard, denoted δ 15 N = [ ( 15 N 14 N / ) sample / ( 15 N 14 N / ) air -1 ] × 10 3 where 15 N 14 Nair / = 3 676 . × 10 -3 .
The aim of this paper is to study the 14 N 15 N ratio around / low-mass protostars to investigate whether fractionation of nitrogen isotopes occurs in the early phases of star formation and to better understand what mechanisms could be responsible for the solar system heterogeneity.
## 1.1. 15 N-enrichments in primitive solar system matter
Meteorites, interstellar dust particles, and comets represent the most pristine solar system material. For meteorites, the nitrogen isotope composition is highly variable among the different classification groups, among meteorites belonging to the same group, and even within individual meteorites (e.g., Alexander et al. 2007; Bonal et al. 2010). The heaviest nitrogen isotope signature in any solar system material was measured to date in the 'Isheyevo' chondrite ( 14 N 15 N / ∼ 50, Briani et al. 2009; Bonal et al. 2010). However, the nitrogen isotope composition in Isheyevo is also extremely variable, ranging from an almost solar-like ( 14 N 15 N / = 394 ± 12) to a very 15 N-heavy isotopic composition. The 15 N-enrichment consists of two components, a di ff use background enrichment ( 14 N 15 N / ∼ 160) and localized micron-sized zones with extremely high 15 N-anomalies, so-called hotspots. Generally, the highest 15 N-enrichments seem to be found in the most pristine meteoritic components (Busemann et al. 2006; Alexander et al. 2007; Briani et al. 2009; Bonal et al. 2010). The carrier of the signature is most likely carbonaceous (Bonal et al. 2010), but the exact functional groups carrying the 15 N-enrichment have not yet been identified.
Interplanetary dust particles (IDPs) are collected in the stratosphere of the Earth and believed to be of cometary or potentially asteroidal origin. They show similar 15 Ncharacteristics as chondrites, with a modest di ff use enrichment and hotspots, but with generally more moderate 15 Nenhancements (Busemann et al. 2006; Floss et al. 2006). The 'Stardust' samples from comet 81P Wild-2 have bulk / nitrogen isotope compositions that cluster around the terrestrial composition, but moderate enhancements are also measured (McKeegan et al. 2006). Hotspot values on submicrometer scale are similar to IDPs. An isotopically light component, consistent with the solar value, is also found in the Wild-2 particles, which is not the case for IDPs.
The average 14 N 15 N ratio from CN spectroscopy in a sam-/ ple of comets is 147 8 . ± 5 6 (Manfroid et al. 2009) with slight . di ff erences for Jupiter family ( 14 N 15 N / = 156 8 . ± 12 2, probable . Kuiper belt origin) and Oort cloud ( 14 N 15 N / = 144 ± 6 5) comets. . These results are consistent with the 14 N 15 N ratio from HCN / measurements by Bockelée-Morvan et al. (2008) in the comets 'Hale-Bopp' and 'Holmes' (see also the review by Jehin et al. 2009) as well as measurements from NH2 in comet C 2012 S1 / (ISON, 14 N / 15 N = 139 ± 38, Shinnaka et al. 2014) and the average spectrum of 12 comets ( ∼ 130, Rousselot et al. 2014).
An overview of 14 N 15 / N ratios in solar system bodies from the literature is provided in Table B.1 in the appendix.
## 1.2. Proposed scenarios: chemical fractionation vs. isotope selective photodissociation
A variety of processes have been proposed to explain the observed nitrogen isotope heterogeneity, including Galactic chemical evolution (GCE, e.g., Adande & Ziurys 2012, and references therein), isotope-selective N2 photodissociation (Clayton 2002; Lyons et al. 2009; Heays et al. 2014), and chemical fractionation (Terzieva & Herbst 2000; Rodgers & Charnley 2008a,b; Wirström et al. 2012; Hily-Blant et al. 2013a).
The term 'Galactic chemical evolution' refers to a scenario in which the observed 14 N 15 N ratio variations are the result of / an enrichment of the interstellar medium (ISM) with the products of stellar nucleosynthesis over time. GCE models predict the 14 N 15 N ratio to vary with distance from the Galactic center / as a result of di ff erent production mechanisms for 14 N and 15 N, the details of which are not yet fully understood (see discussion and references in Adande & Ziurys 2012). Observational support comes from surveys of the nitrogen isotopic composition of the Galaxy (by, e.g., Linke et al. 1977; Wannier et al. 1981; Guesten & Ungerechts 1985; Dahmen et al. 1995; Wilson 1999; Adande & Ziurys 2012). Most of these studies found a radial gradient in the 14 N 15 N ratio of the Galaxy, arguing for Galac-/ tic evolution as the main cause for deviations in the isotopic composition of di ff erent star-forming regions. However, significant source-to-source variations within individual clouds (e.g., Ikeda et al. 2002) as well as evidence for spatial variations in prestellar cores (Hily-Blant et al. 2013a) hint at local processes being more important for the nitrogen isotope composition of the protosolar disk than Galactic inheritance.
Liang et al. (2007) demonstrated that isotopically selective photodissociation of N2 plays an important role in the explanation of the nitrogen isotope anomaly inferred in the upper atmosphere of Titan from HCN observations (Niemann et al. 2005). The radiation field component corresponding to the dissociation energy of 14 N2 cannot penetrate as deep into the atmosphere as the one for 14 N 15 N, because of 14 N2 being more abundant than 14 N 15 N, an e ff ect which is referred to as photochemical selfshielding. To what extent N2 self-shielding could have a ff ected the nitrogen isotope composition of the protosolar nebula is not yet clear, but is currently being investigated (e.g., Lyons et al. 2009; Bertin et al. 2013). Li et al. (2013) and Heays et al. (2014) have recently presented photodissociation rates for 14 N2 and 14 N 15 N. In addition, Heays et al. (2014) have incorporated these rates into chemical models with conditions simulating an interstellar cloud and a protoplanetary disk with grain growth. Their results show 15 N-enhancements by up to a factor of 10 in HCN in the protoplanetary disk model. Based on these results they argue that the observed nitrogen fractionation in protostellar environments could be caused by a combination of chemical and photolytic reactions.
As for chemical fractionation, the astrochemical gas-grain models by Rodgers & Charnley (2008a) predict a considerable 15 N-enhancement for di ff erent nitrogen-bearing molecules in the gas phase of dense cloud cores through isotope exchange reactions in low-temperature chemistry. Their results indicate that the fractionation mechanisms for nitriles, i.e., molecules bearing the nitrile functional group (-CN), and nitrogen hydrides (-NH) are di ff erent. The slow route ( ∼ 10 6 yrs) through the dissociative ionization of N2, leading to N + , is responsible for the fractionation in nitrogen hydrides, including ammonia (NH3) and diazenylium (N2H + ). The more rapid route ( ∼ 10 5 yrs) starts from atomic nitrogen and drives the 15 N-enrichment of (iso-)nitriles and associated species such as cyanide (CN), hydrogen cyanide (HCN), and hydrogen isocyanide (HNC). The largest 15 N-enhancement in this chemical fractionation model is predicted for nitrile species. With substantial 15 N-enrichments predicted by both chemical fractionation and photodissociation models, species like HCN should therefore be particularly good tracers of a potential isotope fractionation in nitrogen. Refined chemical fractionation models by Wirström et al. (2012) include the dependence of the ion-molecule reactions on the H2 spinstate and show that a large 15 N-enhancement is not necessarily expected to correlate with an extreme D-enrichment, depending on the carriers of the two signatures. The lack of a correlation between the 15 N and D anomalies measured in meteorites had previously often been invoked as an argument against an inheritance of the isotopic signature measured in primitive solar system matter from local cloud or disk processes in early phases of solar system formation (e.g., Aléon 2010).
Table 1. Literature values of the 14 N 15 / N ratio from low-mass prestellar cores and protostars (top). For comparison, the most nearby sources from the galactic survey by Adande & Ziurys (2012) are listed as well (bottom).
| Source | HCN | HNC | CN | NH 3 | NH 2 D | N 2 H + | Remarks |
|------------------------------------------------------------------------|--------------------------------------------------------------|---------------------------|---------------------------|-----------------------------|--------------------------------------------------------------------------------|----------------------------|---------------------------------------------------------------------------------|
| L1544 L1498 L183 L1521E L1521F | 140 - 360 a 69 - 154 e > 813 f 140 - 250 a 151 ± 16 f > 51 e | > 27 e > 90 e 24 - 31 e | 510 ± 70 b 476 ± 70 b | | > 700 c | 1000 ± 200 d | prestellar core prestellar core prestellar core prestellar core prestellar core |
| Barnard 1 NGC1333 IRAS4 NGC1333 DCO + NGC1333 L134N(S) L1689N Cha-MMS1 | 330 + 60 - 50 g | 225 + 75 - 45 g 135 i | 290 + 160 - 80 g | 300 + 55 - 40 g 344 ± 173 h | 230 + 105 - 55 g > 270 c 360 + 260 - 110 c 530 + 570 - 180 c 810 + 600 - 250 c | 400 + 100 - 65 g , > 600 g | protostar |
| IRAS 17233-3606 DR 21 Orion KL Orion Bar | 276 ± 62 | 159 ± 31 159 ± 40 338 311 | 243 ± 66 234 ± 47 361 141 | | | | HII region HII region PDR |
±
±
References. a Hily-Blant et al. (2013a) - the range reflects spatial variations within the core, b Hily-Blant et al. (2013b), c Gerin et al. (2009) based on 15 NH2D observations at the IRAM 30m, d Bizzocchi et al. (2013), revised from the previously published value of 446 ± 71 (Bizzocchi et al. 2010), e Milam & Charnley (2012), f Ikeda et al. (2002), g Daniel et al. (2013), h Lis et al. (2010) based on observations of 15 NH3 inversion lines with the Green Bank Telescope, i reported in Milam & Charnley (2012) based on Tennekes et al. (2006), j Adande & Ziurys (2012)
Table 2. Coordinates and properties of the observed sources.
| Source | RA [h ms] | Dec [ ◦ ′ ′′ ] | v lsr [km s - 1 ] | Class | d [pc] | L bol [L /circledot ] | M env [M /circledot ] |
|-------------|------------------|------------------|---------------------|---------------------------|----------|-------------------------|-------------------------|
| OMC-3 | MMS605:35:23.5 a | - 05:01:32.0 a | 10 | intermed.-mass (Class 0) | 450 | 60 | 36 |
| Elias 29 | 16:27:09.4 b | - 24:37:18.6 b | 5 | Class I | 125 | 14.1 | 0.04 |
| IRAS 16293A | 16:32:22.87 c | - 24:28:36.4 c | 3 | Class 0 | 120 | 27 | 5.4 |
| R CrA IRS7B | 19:01:56.4 d | 36:57:28.3 d | 5.6 | transitionary Class 0 / I | 130 | 4.7 | 2.2 |
-
References. a Johnstone et al. (2003), b Jørgensen et al. (2008), c Jørgensen et al. (2011), d Lindberg & Jørgensen (2012)
Observations of the 14 N 15 N ratio in individual low-mass pre-/ and protostellar sources in the literature are scarce and have so far led to inconclusive results about whether or not fractionation of nitrogen isotopes occurs in pre- and protostellar environments. Moreover, it still an open question whether chemical fractionation or isotope-selective photodissociation could explain the observed solar system variations. Results from HCN observations of three dense cores by Ikeda et al. (2002) show large variations between the sources and the result for L1521E ( 14 N 15 N / = 151 ± 16) is consistent with the chemical fractionation scenario. Gerin et al. (2009) on the other hand derive 14 N 15 N ratios of / ∼ 360 -810 from 15 NH2D in several interstellar clouds that are consistent with the solar composition and thus argue against nitrogen isotope fractionation. Follow-up observations by Lis et al. (2010) using measurements of 15 NH3 in Barnard 1 and NGC 1333 yield ratios that fall in between the solar and terrestrial values. Bizzocchi et al. (2010) reported a 14 N 15 N ratio of 446 / ± 71 from N 15 NH + observations toward the prestellar core L1544, which was recently revised to be as high as 1000 ± 200 based on new observations of 15 NNH + (Bizzocchi et al. 2013), implying a depletion of 15 N in N2H . + Results from mapping of HCN isotopologues in the two prestel- lar cores L183 and L1544 by Hily-Blant et al. (2013a) show an overall enhancement of 15 N as well as some spatial variations within each source, strongly favoring a fractionation scenario. For L1544, Hily-Blant et al. (2013a) find an enhancement of 15 N in HCN (140 < 14 N 15 N / < 360), but there are no indications for isotope fractionation in CN ( 14 N 15 N / = 500 ± 75, Hily-Blant et al. 2013b). These di ff erences likely arise from the variety of molecular species and isotopologues used to probe the nitrogen isotope ratio in di ff erent sources. For example, the low 14 N 15 N ratios inferred from nitriles like HCN seem to sup-/ port the chemical fractionation scenario, while 14 N 15 N ratios / derived from nitrogen hydrides, like NH3 and N2H , show no + evidence for nitrogen isotope fractionation or even hint at a 15 Ndepletion. Wirström et al. (2012) and Hily-Blant et al. (2013b) proposed refined chemical fractionation scenarios with di ff erent fractionation paths that may be able to explain the discrepancies between the nitrogen isotope composition seen in individual molecular tracers. To constrain the fractionation models, it is thus important to measure isotope ratios from an entire set of species. However, the prestellar core L1544 and the very young Class 0 binary B1b are so far the only sources for which comprehensive observational data sets of 14 N 15 N ratios from both /
nitriles and nitrogen hydrides exist (cf. Table 1). While evidence for a 15 N-enrichment in some molecules is found for L1544 (Hily-Blant et al. 2013a), the results on B1b by Daniel et al. (2013) show no evidence for nitrogen isotope fractionation in any class of molecules. Whether this di ff erence is caused by the more evolved status of B1b, a source-to-source variation, or any other e ff ect, is unclear and needs to be investigated on a larger source sample. A similar data set is not yet available for other sources, in particular not for more evolved evolutionary stages.
In contrast to previous studies, which mainly focused on the prestellar core phase, we present in this paper measurements of the nitrogen isotope composition in the gas around three Class 0 YSOs from HCN and HNC isotopologues. The aim of this work is to investigate whether signatures of nitrogen isotope fractionation also occur in the embedded protostellar stages. The paper is organized as follows: Section 2 describes the observations and the data reduction procedure. The results are presented in Sect. 3. They are compared to previous results and their implications are discussed in Sect. 4. Conclusions are presented in Sect. 5.
## 2. Observations and data reduction
## 2.1. Observations
Observations of HCN and HNC isotopologues were carried out using the Atacama Pathfinder EXperiment telescope 1 (APEX, Güsten et al. 2006) in service mode. Wobbler switching with an amplitude of 150 ′′ was used throughout. A detailed overview on the observations is provided in Table A.1 in the appendix.
1 APEX is a collaboration between the Max-Planck-Institut für Radioastronomie, the European Southern Observatory, and the Onsala Space Observatory.
Table 3. Molecular data of the HCN and HNC isotopologues observed with APEX. HCN isotopologue data are from the CDMS catalog (Müller et al. 2001, 2005) and HNC isotopologues are from the JPL catalogue (Pickett et al. 1998), all obtained through www.splatalogue.net.
| Species | Transition | Frequency [GHz] | E up [K] | log( A ul ) [s - 1 ] |
|-------------|--------------|-------------------|------------|------------------------|
| | | APEX-1 | | |
| D 13 CN | J = 3-2 | 213.51987 | 20.49 | -3.36216 |
| DCN | J = 3-2 | 217.23854 | 20.85 | -3.33964 |
| H 13 C 15 N | J = 3-2 | 251.17523 | 24.11 | -3.15217 |
| HC 15 N | J = 3-2 | 258.15700 | 24.78 | -3.11641 |
| H 13 CN | J = 3-2 | 259.01180 /star | 24.86 | -3.11216 |
| HN 13 C | J = 3-2 | 261.26331 | 25.08 | -3.18837 |
| HCN | J = 3-2 | 265.88643 | 25.52 | -3.07662 |
| H 15 NC | J = 3-2 | 266.58780 | 25.59 | -3.16196 |
| HNC | J = 3-2 | 271.98114 | 26.11 | -2.94523 |
| DCN | J = 4-3 | 289.64492 | 34.75 | -2.94907 |
| | | APEX-2 | | |
| HC 15 N | J = 4-3 | 344.20011 | 41.30 | -2.72583 |
| H 13 CN | J = 4-3 | 345.33977 /star | 41.43 | -2.72161 |
| HN 13 C | J = 4-3 | 348.34027 | 41.80 | -2.79774 |
| DCN | J = 5-4 | 362.04575 | 52.13 | -2.64860 |
| HNC | J = 4-3 | 362.63030 | 43.51 | -2.55463 |
Notes. ( /star ) Hyperfine structure only partially resolved.
The source sample consists of three Class 0 YSOs, IRAS 16293A, R CrA IRS7B, and OMC-3 MMS6, and the Class I protostar Elias 29. The coordinates and properties of the sources can be found in Table 2. The targets were selected to cover di ff erent source properties: IRAS 16293-2422 is an extremely well-studied (e.g., Blake et al. 1994; van Dishoeck et al. 1995; Schöier et al. 2002; Takakuwa et al. 2007; Caux et al. 2011; Jørgensen et al. 2011; Pineda et al. 2012), isolated pr otobinary in ρ Oph with a hot corino and known deuterium fractionation (e.g., Loinard et al. 2000; Persson et al. 2013). Our observations were pointed at the A component, IRAS 16293A (Jørgensen et al. 2011). R CrA IRS7B is located in the vicinit y of the Herbig star R CrA and therefore exposed to a strong external radiation field (Lindberg & Jørgensen 2012). OMC-3 MMS6 is an intermediate-mass protostar (Johnstone et al. 2003) in the vigorous environment of the Orion star-forming complex, also exposed to enhanced irradiation (Jørgensen et al. 2006). El ias 29 (e.g., Boogert et al. 2002) is one of the most luminous protostars in the ρ Oph cloud and the most evolved source in our sample.
IRAS 16293A was observed in H 13 CN ( J = 3-2 and 4-3), HC 15 N (3-2 and 4-3), H 13 C 15 N(3-2), DCN(3-2, 4-3, and 5-4), and D 13 CN(3-2) on 2012-04-01 and between 2012-06-04 and 2012-06-13 (project 089.F-9316A, PI J. Jørgensen). The observations of HN 13 C(3-2) and H 15 NC(3-2) were carried out on 2013-08-19 20 (project 092.F-9324A, PI S. Wampfler). / OMC3 MMS6, Elias 29, and R CrA IRS7B were observed in HCN(32), H 13 CN (3-2 and 4-3), HC 15 N(3-2), HNC(3-2), HN 13 C (3-2 and 4-3), and H 15 NC(3-2) with APEX between 2012-08-16 and 2012-10-01 (project 090.F-9317A, PI S. Wampfler). The frequencies of the transitions and the molecular data are provided in Table 3.
The Swedish Heterodyne Facility Instrument (SHeFI, Vassilev et al. 2008) single-sideband SIS receivers APEX-1 (211 -275 GHz) and APEX-2 (267 -370 GHz) were used in combination with the eXtended bandwidth Fast Fourier Transform Spectrometer (XFFTS). Two XFFTS units with an instantaneous bandwidth of 2.5 GHz each are available, which partially overlap and o ff er full coverage of the 4 GHz IF bandwidth of the SHeFI receivers. The large bandwidth of this backend type permits several isotopologues to be observed simultaneously, thereby reducing the calibration uncertainty from the absolute value down to the relative calibration uncertainty across the band. For IRAS 16293A, H 13 CN (3-2) and HC 15 N (3-2) were observed in a combined setting, as were H 13 CN (4-3) and HC 15 N (4-3). In the case of OMC-3 MMS6, Elias 29, and R CrA-IRS7B, the observations of H 13 CN (3-2), HC 15 N (3-2), and HN 13 C (3-2) were carried out simultaneously. Moreover, H 15 NC (3-2) and HCN(3-2) were observed together. H 13 CN (43) and HN 13 C (4-3) were also jointly measured.
## 2.2. Data reduction
The data reduction was carried out using the GILDAS-CLASS software package 2 . The data were calibrated to T mb scale using the default forward e ffi ciency of 0.95 (Dumke & Mac-Auli ff e 2010) for APEX, and beam e ffi ciencies 3 of 0.75 for APEX-1 and 0.73 for APEX-2, respectively. The half-power beam width of APEX is 27 ′′ at 230 GHz and 18 ′′ at 345 GHz.
Several observations were split between di ff erent observing dates (cf. Table A.1). Before averaging those data sets, we checked for deviations in the intensity and line profile on strong
2 http://www.iram.fr/IRAMFR/GILDAS/
3 values from http://www.apex-telescope.org/telescope/efficiency
Fig. 1. Spectra of the HCN and HNC isotopologues observed with APEX toward IRAS 16293A. Where needed, the spectra have been scaled by the factors indicated in the individual panels.
lines present in the spectrum, and, if the signal-to-noise ratio for the individual observations permits, also on our target lines. The H 15 NC(3-2) observations from R CrA IRS7B carried out on the 09-30-2012 show substantially larger deviations than the calibration uncertainty ( ∼ 30%) when compared to other days and this dataset was consequently discarded. Deviations in the line profiles between di ff erent days also occur for the detected main isotopologues lines HCN and HNC in Elias 29 (rarer isotopologues are not detected). Moreover, the line profiles are a mixture of absorption and emission, which could be an indication for emission in one of the o ff positions (wobbler switching mode). Therefore, we do not report integrated intensities for Elias 29.
Apotential contamination from the o ff positions for the other sources was assessed by re-pipelining some data sets using only one or the other o ff position for H 13 CN and H 13 CO + , a stronger line in the same frequency setup. The resulting spectra, which can be found in appendix A, were then examined for di ff erences. There is no evidence for a substantial emission from the o ff positions.
Table 4. Peak temperatures Tpeak, integrated intensities ∫ T mb dv , line widths dv , and central peak position v of the HCN and HNC isotopologues. The integrated intensites for IRAS 16293A were obtained from integration over the line, while Gaussian fitting was used for the two other sources. Also listed are the obtained noise level (Trms), error on the integrated intensity σ rms , and the total error σ tot (combined error of the integrated intensity and the calibration).
| Transition | T peak [K] | ∫ T mb dv [K km s - 1 ] | dv [km s - 1 ] | v [km s - 1 ] | T rms [K] | σ rms [K km s - 1 ] | σ tot [K km s - 1 ] |
|-------------------|--------------|---------------------------|-------------------|----------------------|-------------|-----------------------|-----------------------|
| IRAS 16239A | IRAS 16239A | IRAS 16239A | IRAS 16239A | IRAS 16239A | IRAS 16239A | IRAS 16239A | IRAS 16239A |
| HCN(3-2) | 9.66 | 27.52 | 6 . 81 ± 0 . 02 | 3 . 72 ± 0 . 01 | 0.04 | 0.03 | 2.3 |
| H 13 CN(3-2) | 0.68 | 3.24 | 5 . 70 ± 0 . 08 | 3 . 75 ± 0 . 03 | 0.03 | 0.02 | 0.3 |
| HC 15 N(3-2) | 0.27 | 1.37 | 6 . 2 ± 0 . 2 | 3 . 71 ± 0 . 09 | 0.04 | 0.03 | 0.1 |
| H 13 CN(4-3) | 1.08 | 5.56 | 6 . 11 ± 0 . 09 | 4 . 28 ± 0 . 04 | 0.07 | 0.04 | 0.8 |
| HC 15 N(4-3) | 0.44 | 2.02 | 6 . 4 0 . 2 | 3 . 11 ± 0 . 09 | 0.06 | 0.04 | 0.3 |
| DCN(3-2) | 0.87 | 2.77 | ± 3 . 6 ± 0 . 1 | 3 . 87 ± 0 . 04 | 0.05 | 0.03 | 0.2 |
| DCN(4-3) | 0.76 | 2.45 | 3 . 74 ± 0 . 07 | 3 . 92 ± 0 . 03 | 0.05 | 0.03 | 0.4 |
| DCN(5-4) | 0.80 | 3.04 | 4 . 47 ± 0 . 14 | 3 . 86 ± 0 . 06 | 0.09 | 0.05 | 0.4 |
| D 13 CN(3-2) | - | - | - | - | 0.02 | - | - |
| HNC(3-2) | 6.97 | 12.93 | 2 . 934 ± 0 . 002 | 3 . 959 ± 0 . 001 | 0.02 | 0.009 | 1.1 |
| HNC(4-3) | 5.57 | 11.00 | 3 . 14 ± 0 . 02 | 3 . 671 ± 0 . 001 | 0.11 | 0.05 | 0.9 |
| HN 13 C(3-2) | 0.40 | 0.77 | 3 . 0 ± 0 . 3 | 3 . 77 ± 0 . 09 | 0.05 | 0.03 | 0.1 |
| H 15 NC(3-2) 0.09 | | 0.22 | 2 . 5 ± 0 . 4 | 4 . 1 ± 0 . 2 | 0.03 | 0.01 | 0.03 |
| R CrA IRS7B | R CrA IRS7B | R CrA IRS7B | R CrA IRS7B | R CrA IRS7B | R CrA IRS7B | R CrA IRS7B | R CrA IRS7B |
| HCN(3-2) 6.59 | | 29 . 48 ± 0 . 03 | 4 . 202 ± 0 . 005 | 5 . 987 ± 0 . 002 | 0.02 | 0.01 | 2.4 |
| H 13 CN(3-2) | 0.90 | 1 . 87 ± 0 . 03 | 1 . 96 ± 0 . 03 | 5 . 92 ± 0 . 01 | 0.04 | 0.02 | 0.2 |
| HC 15 N(3-2) 0.24 | | 0 . 45 ± 0 . 02 | 1 . 79 ± 0 . 11 | 5 . 92 ± 0 . 05 | 0.04 | 0.02 | 0.04 |
| H 13 CN(4-3) | 0.46 | 0 . 92 ± 0 . 02 | 1 . 87 ± 0 . 05 | 6 . 05 ± 0 . 02 | 0.04 | 0.01 | 0.1 |
| HNC(3-2) | 6.35 | 14 . 3 ± 0 . 1 | 2 . 12 ± 0 . 02 | 5 . 925 ± 0 . 008 | 0.2 | 0.07 | 1.2 |
| HN 13 C(3-2) | 0.45 | 0 . 75 ± 0 . 02 | 1 . 56 ± 0 . 05 | 5 . 6 ± 0 . 02 | 0.04 | 0.02 | 0.06 |
| H 15 NC(3-2) | 0.14 | 0 . 20 ± 0 . 01 | 1 . 40 ± 0 . 08 | 5 . 82 ± 0 . 03 | 0.02 | 0.006 | 0.02 |
| HN 13 C(4-3) | 0.23 | 0 . 32 ± 0 . 02 | 1 . 32 ± 0 . 08 | 5 . 19 ± 0 . 04 | 0.04 | 0.01 | 0.05 |
| OMC-3 MMS6 | OMC-3 MMS6 | OMC-3 MMS6 | OMC-3 MMS6 | OMC-3 MMS6 | OMC-3 MMS6 | OMC-3 MMS6 | OMC-3 MMS6 |
| HCN(3-2) 3.25 | | 13 . 10 ± 0 . 03 | 3 . 786 ± 0 . 009 | 11 . 119 ± 0 . 002 | 0.01 | 0.01 | 1.09 |
| H 13 CN(3-2) | 0.51 | 0 . 53 ± 0 . 02 | 0 . 98 ± 0 . 04 | 11 . 12 ± 0 . 01 | 0.03 | 0.01 | 0.05 |
| HC 15 N(3-2) 0.10 | | 0 . 10 ± 0 . 02 | 1 . 00 (fixed) | 11 . 14 ± 0 . 07 | 0.03 | 0.01 | 0.01 |
| H 13 CN(4-3) | 0.18 | 0 . 19 0 . 04 | 1 . 00 (fixed) | 11 . 09 0 . 10 | 0.06 | 0.02 | 0.03 |
| HNC(3-2) | 5.55 | ± 6 . 86 ± 0 . 09 | 1 . 16 ± 0 . 02 | ± 11 . 177 ± 0 . 007 | 0.19 | 0.06 | 0.57 |
| HN 13 C(3-2) | 0.46 | 0 . 36 ± 0 . 01 | 0 . 74 ± 0 . 03 | 10 . 85 ± 0 . 01 | 0.03 | 0.008 | 0.03 |
| H 15 NC(3-2) 0.09 | | 0 . 054 ± 0 . 004 | 0 . 54 ± 0 . 04 | 11 . 07 ± 0 . 02 | 0.01 | 0.003 | 0.005 |
| HN 13 C(4-3) | - | - | - | - | 0.01 | - | - |
First order baselines were applied to the individual scans before averaging them. In particular, the DCN(5-4) spectrum su ff ers from standing waves, which cannot easily be fitted with a sinusoidal baseline, as amplitude and period change. Therefore, the first order polynomial baseline fits were applied locally whenever required. For lines that were covered by both XFFTS units (overlapping spectral range), the processed spectra from the individual parts were finally averaged after visual inspection.
## 3. Results
## 3.1. Line intensities
The line intensities were obtained by Gaussian fitting except for IRAS 16293A, because the line shapes of this source differ substantially from a simple Gaussian profile (cf. Fig. 1). For IRAS 16293A, the intensities were derived from integrating over a velocity range around the source velocity ( -10 km s -1 < /v1 < 16 km s -1 for broad lines and -4 km s -1 < /v1 < 12 km s -1 for narrow lines). Table 4 lists the resulting integrated line intensi- ties, widths, and the obtained noise level (rms). The spectra from R CrA IRS7B and OMC-3 MMS6 are shown in Figs. 2 and 3, respectively.
While the HCN(3-2) line has prominent outflow wings in IRAS 16293A and OMC-3 MMS6, the line width of HNC(3-2) is narrower and does not feature the broad wings at the obtained signal-to-noise ratio.
For IRAS 16293A, H 13 C 15 N(3-2) and D 13 CN(3-2) were targeted as well, but not detected at the achieved noise level. The H 13 C 15 N(3-2) was only observed for a fraction of the time intended, and the sideband was changed during the observation because of a better receiver performance in the upper sideband. Unfortunately, the upper sideband tuning su ff ers from a tuning instability, which went unnoticed during the observations. A reflection from the tuning instability shows up in one channel at the tuning frequency, and a reliable determination of the noise level is therefore not possible. Hence, we do not give an upper limit on the H 13 C 15 N(3-2) intensity.
In the Class I source Elias 29, the only lines detected are HCN(3-2) and HNC(3-2), which are factors of 50-100 times
Fig. 2. Spectra of the HCN and HNC isotopologues observed with APEX toward R CrA IRS7B. Where needed, the spectra have been scaled by the factors indicated in the individual panels
weaker than the same lines from the Class 0 targets. Therefore, none of the rarer isotopologues are detected in Elias 29 at the achieved rms, maybe because of the much smaller envelope mass compared to the Class 0 sources in the sample. Both HCN(3-2) and HNC(3-2) have a line profile that shows emission at velocities higher than the source velocity, and absorption at the lower velocities, which could either be a real source feature or indicate that there was some molecular emission at the o ff position.
## 3.2. 14 N/ 15 N isotopic ratios
Isotope ratios can generally be measured from molecular lines at sub-millimeter wavelengths using three di ff erent approaches, each with its advantages and caveats:
- 1. Singly substituted molecules: The 14 N 15 / N ratio can be measured from, e.g., CN isotopologues (C 15 N Adande & Ziurys 2012; Hily-Blant et al. 2013b), N2H + isotopologues ( 15 NNH , + 15 NNH , + Womack et al. 1992; Bizzocchi et al. 2010, 2013) or ammonia isotopologues ( 15 NH3, Guesten & Ungerechts 1985; Lis et al. 2010). The advantage of this method is that the main isotopologue is usually abundant and thus its lines are strong. On the other hand, the high abundance often implies a high optical depth, which must be accounted for using radiative transfer modeling. If the molecular transition has a resolved hyperfine structure pattern, the intensity ratios of the hyperfine transitions can also be used to infer and correct for the optical depth.
- 2. Doubly substituted molecules: An alternative to the often optically thick main isotopologues are molecules with two atoms replaced by a less abundant isotope, like the ammonia isotopologue 15 NH2D(Gerin et al. 2009). However, the lines
Fig. 3. Spectra of the HCN and HNC isotopologues observed with APEX toward OMC-3 MMS6. Where needed, the spectra have been scaled by the factors indicated in the individual panels
of such doubly substituted molecules are often very weak because of their low abundance.
- 3. Double isotope method: The nitrogen isotope ratio can also be measured from two singly substituted molecules, e.g., H 13 CN and HC 15 N or HN 13 Cand H 15 NC, avoiding the main isotopologue. However, in this example the 12 C 13 / C ratio needs to be known from independent measurement or an assumption has to be made.
Using the double isotope method and assuming that all lines are optically thin, which will be discussed in Sect. 3.3, the 14 N 15 N ratio can be obtained from the H / 13 CN and HC 15 N integrated intensities using the relation
$$\begin{smallmatrix} \text{itio can} & \frac { \frac { 1 4 } { 1 5 } } { 2 5 N } & \frac { 1 4 } { 1 5 N } & = & \frac { \int T d v \, ( H ^ { 1 3 } C N ) } { \int T d v \, ( H C ^ { 1 5 } N ) } \times \frac { 1 2 } { 1 3 C } & & & ( 1 ) \\ \int \frac { 1 } { 2 } \infty. & \dots & \dots &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. &. & \end{smallmatrix}$$
and from HNC isotopologues accordingly. For IRAS 16293A, the non-detection of D 13 CN hampers a direct determination of the 12 C 13 / Cfrom D 13 CN and DCN, but only provides a 3 σ upper limit of 12 C 13 C / /lessorsimilar 120. Because no 12 C 13 / C ratio measurements for our sources are available from the literature, we assume the local standard ISM value of 69 (Wilson 1999; Milam et al. 2005). The resulting ratios are illustrated in Fig. 4, together with solar system measurements from the literature, and listed in Table 5. The table also includes the ratio from the integrated intensities of H 13 CN(4-3) and HC 15 N(4-3) reported in Watanabe et al. (2012). The H 13 CN(4-3) line may have a contribution of SO2(132 12 , -121 11), but given that the isotopic ratios , derived from H 13 CN(3-2) and (4-3) in IRAS 16293A agree well, the contamination is expected to be minor in this source. For the other sources, H 13 CN(4-3) was not used in the determination of
Fig. 4. Left panel: Illustration of the nitrogen isotope heterogeneity of the solar system, adapted from Meibom et al. (2007); Mumma & Charnley (2011). Comets, meteorites, and interstellar dust particles (IDPs) are located over a range of distances from the Sun and the plotted d Sun values should be regarded as an average value only. Right panel: 14 N 15 / N ratios obtained from APEX measurements (black, this work) and ASTE (grey, Watanabe et al. 2012). Diamonds and triangles indicate values from HCN and HNC isotopologues, respectively. Error bars are 1 σ for combined uncertainties on the calibration and the intensity. The gray bars ('cores') indicate the prestellar core values from HCN by Hily-Blant et al. (2013a).
Table 5. 14 N 15 / N ratios derived from APEX observations of HCN and HNC isotopologues using a standard 12 C 13 / C ratio of 69, except for the H 13 CN HC 15 N(4-3) ratio, where the data are taken from the survey by / Watanabe et al. (2012).
all errors. In the cases where the 14 N 15 N ratios are derived from / HCN isotopologues, the errors listed should therefore be an upper limit on the uncertainty.
| Source | H 13 CN(3 - 2) HC 15 N(3 - 2) | HN 13 C(3 - 2) H 15 NC(3 - 2) | H 13 CN(4 - 3) HC 15 N(4 - 3) |
|-------------|---------------------------------|---------------------------------|---------------------------------|
| IRAS 16293A | 163 ± 20 | 242 ± 32 | 190 ± 38 |
| R CrA IRS7B | 287 ± 36 | 259 ± 34 | (285 ± 139) |
| OMC3 MMS6 | 366 86 | 460 65 | - |
±
±
the 14 N 15 N ratio, just as a constraint for the radiative transfer / models.
The smallest error is expected on the ratio inferred from H 13 CN(3-2) and HC 15 N(3-2) for all sources as well as the H 13 CN(4-3) and HC 15 N(4-3) for IRAS 16293A, because the lines were measured simultaneously. The 14 N 15 / N ratios from the HNC isotopologues are more uncertain, as the involved lines were observed in two individual spectral setups. For the errors, we use a combined error from the intensity measurement and the calibration uncertainty. The error on the integrated intensity is taken from the uncertainty in the fit of the area under the Gaussian profile, provided by the CLASS software, for R CrA IRS7B and OMC-3 MMS6, and for IRAS 16293A calculated as σ rms = √ ∆ /v1 × δ v × T rms where ∆ /v1 is the full width at half maximum of the line (assumed to be the same as that of a Gaussian fit to the line), δ v is the velocity resolution of the spectrum, and T rms is the noise level. For the calibration error σ calib , we used the calibration uncertainties on the integrated intensity provided in the APEX calibration report (Dumke & Mac-Auli ff e 2010) of 8.3% and 14.1% for APEX-1 and APEX-2, respectively. The overall error on a line intensity is then the combination of the two errors, σ tot = √ σ 2 rms + σ 2 calib , and propagated on to the intensity ratio using regular error propagation. Even though some lines were measured in a joint spectral setup, the full calibration uncertainty is included in the determination of
## 3.3. Optical depth considerations
## 3.3.1. Hyperfine structure fitting for IRAS 16293A
The assumption that the lines, in particular H 13 CN(3-2), are optically thin and hence their intensity is a direct measure of the number of molecules, needs to be verified.
For IRAS 16293-2422, we have a H 13 CN(1-0) spectrum (Jørgensen et al. 2004) from the SEST telescope 4 with clearly resolved hyperfine structure (cf. Fig. 5, using a main beam e ffi ciency of 0.75). The pointing is slightly di ff erent ( α = 16:32:22.80, δ = -24:28:33.0) from our recent observations, but given the large beam the spectrum should still provide an estimate of the opacity. The three hyperfine transitions were fitted using the 'HFS method' of the CLASS software to derive the opacity. The HFS routine fits Gaussians to all hyperfine components simultaneously under the assumption of a common excitation temperature, with relative positions and intensities provided by the user (see the CLASS manual or appendix A of Hily-Blant et al. 2013a, for more details). The best fit, illustrated in Fig. 5, yields an opacity for the main component of τ main = 0 274 . ± 0 405 and hence the line is optically thin. .
## 3.3.2. Radiative transfer modeling
The optical depth on higher excited lines can be larger than the opacity of H 13 CN(1-0). Therefore, line radiative transfer models can be used as an alternative method to estimate the optical depth in the isotopologues and check the validity of Eq. 1. The spherically symmetric source model from Schöier et al. (2002) for IRAS 16293-2422, the R CrA IRS7B model with external irradi-
4 The Swedish-ESO 15m Submillimeter Telescope (SEST) was decommissioned in 2003.
Fig. 5. H 13 CN(1-0) spectrum of IRAS 16293-2422 from the SEST telescope. The solid line shows the best fit from the CLASS software.
ation from Lindberg & Jørgensen (2012), and the best fit model of OMC-3 MMS6 from Jørgensen et al. (2006) are used as an input for the 1D version of the 'RATRAN' radiative transfer code (Hogerheijde & van der Tak 2000). A comparison of the source models, i.e., density and temperature profiles, is shown in Fig. 6. We use molecular data for the HCN isotopologues from the LAMDAdatabase 5 (Schöier et al. 2005). The transition frequencies, energy levels, and Einstein A coe ffi cients in the LAMDA files are based on data from the CDMS catalog (Müller et al. 2001, 2005), and the collisional rate coe ffi cients are scaled and extrapolated HCN-He collision rates from Green & Thaddeus (1974). Individual models were checked against newer collision rates from Dumouchel et al. (2010) and the di ff erences were found to be at the 5-10% level in these cases, i.e., well within the calibration uncertainty.
The abundance profile is chosen according to the results by Jørgensen et al. (2005a), i.e., a drop abundance profile accounting for freeze-out of the molecule at temperatures below a certain threshold and su ffi cient densities. Following Jørgensen et al. (2005a), the freeze-out region for HCN in NGC 1333 IRAS 2A is defined as the part of the envelope with densities above n > 7 × 10 -4 cm -3 and temperatures below T < 90 K, which corresponds to the region where CO and HCO + are also depleted. However, the density at which HCN is frozen out is not a fixed value but time and source dependent (Jørgensen et al. 2005b).
Therefore, models with three di ff erent freezeout evaporation / temperatures (30 K, 60 K, and 90 K) and various abundances were tested for IRAS 16293-2422. The density is higher than n > 7 × 10 -4 cm 3 throughout the envelope -in this model and a depletion zone was therefore not employed. A Doppler b-parameter of db = 2 3 . km s -1 was assumed, corresponding to a line width (FWHM) of ∼ 3 8 km s . -1 . The model FWHM was chosen a bit lower than the observed line
5 http://home.strw.leidenuniv.nl/~moldata/
Fig. 6. Temperature and density as function of radius for IRAS 162932422 (solid line), OMC-3 MMS6 (dotted), and R CrA IRS7B (dashed).
widths, because no additional velocity profile (e.g. infall) was used in the models.
The model with the abundance jump at 90 K and x out ≈ 6 × 10 -12 (for T < 90 K) and x in ≈ 1 × 10 -9 (where T ≥ 90 K) can reproduce the observed H 13 CN(3-2 and 4-3) integrated intensities from IRAS 16293-2422 within a few percent. However, this model underproduces the HC 15 Nlines, in particular the J = 4 -3 transition, by at least 20% for H 13 CN HC 15 N abundance ratios / in the range of 2 6 . -6 3, corresponding to a . 14 N 15 N ratio of / 180 -440 assuming 12 C 13 C / = 69 (Wilson 1999). In addition, an abundance jump of more than two orders of magnitude is almost a factor of five higher than what was found in earlier work (Jørgensen et al. 2005a; Brinch et al. 2009). Therefore, we a lso tested models where the jump occurs at 60 K and 30 K instead of 90 K. For 60 K, the H 13 CN intensities are well modelled using x in ≈ 2 × 10 -10 and x out ≈ 8 × 10 -12 . Using an HC 15 N abundances of x in ≈ 7 7 . × 10 -11 and x out ≈ 3 1 . × 10 -12 , corresponding to 14 N 15 N / = 180 as obtained using Eq. 1, yields a very good fit to the observed integrated intensities of HC 15 N. In contrast, using a solar ratio of 14 N 15 N / = 441 ( x in ≈ 3 1 . × 10 -11 and x out ≈ 1 3 . × 10 -12 ) underproduces the line intensities by ∼ 50%. For 30 K, a good fit to the H 13 CNlines is obtained using x in ≈ 4 × 10 -11 and x out ≈ 5 × 10 -12 . Again, 14 N 15 N / = 180 ( x in ≈ 1 5 . × 10 -11 and x out ≈ 1 9 . × 10 -12 ) gives an excellent fit to the HC 15 N data, whereas a solar abundance ratio, x in ≈ 6 3 . × 10 -12 and x out ≈ 7 8 . × 10 -13 , underestimates the observed integrated intensites by ∼ 50%.
So while the 90 K jump model, which becomes moderately optically thick for H 13 CN ( τ 3 -2 ≈ 1 5 towards line and source . center), can reproduce the H 13 CN lines, it cannot fit HC 15 N for 14 N 15 N ratios ranging from the most extreme values found from / the simple analysis (Eq. 1) to solar abundances. The optically
thin models with the jump occuring at 60 K and 30 K on the other hand are also able to reproduce the HC 15 N integrated intensities for 14 N 15 N / = 180, but not for solar abundances. Therefore, the radiative transfer results support the assumption that the line emission is optically thin, and that Eq. 1 is valid in this case.
For OMC-3 MMS6, a fairly well constrained model is also available. An abundance profile with x ≈ 8 × 10 -13 for T < 90 K and n > 1 × 10 6 cm 3 -and x ≈ 7 × 10 -11 elsewhere yields integrated intensities that are consistent with the observed values within the uncertainties ( ∼ 20%). The corresponding opacities are in the range of τ ∼ 0 5 for . db = 0 6 (FWHM . ∼ 1 km s -1 ) and indicate that the H 13 CN emission from this source is optically thin. For HC 15 N(3-2), the integrated intensity can be reproduced within 20% using models with 14 N 15 N / = 370 (as derived from Eq. 1) as well as 14 N 15 N / = 441 (solar ratio). As already concluded in Sec. 3.2, it is therefore not possible to distinguish between a slightly fractionated and a solar 14 N 15 N ra-/ tio for OMC-3 MMS6. The higher freeze-out density compared to IRAS 16293-2422 could indicate that this source is in a less evolved state.
Because of the complex physical structure of R CrA IRS7B, including a disk (Lindberg et al. 2014), the spherically symmetric model is less reliable for this source than for IRAS 162932422 and OMC-3 MMS6. An agreement within 20% of the modeled and the observed intensities can be obtained with an H 13 CN abundance of x ≈ 2 × 10 -10 for T < 90 K and n > 7 × 10 4 cm -3 and x ≈ 2 × 10 -8 elsewhere. The optical depth in the central line channel and for a ray directly through the source center are τ ∼ 1 -3. The observed HC 15 N emission can be reproduced within the uncertainties with a range of 14 N 15 N ratios / ( ∼ 270 -441) because the model is not well constrained and only one HC 15 N line is available. However, the required high abundances are an indication that the physical model underestimates the observed column density, which results in a higher optical depth. Compared to the IRAS 16293-2422 model, which has a similar outer radius, but higher densities, R CrA IRS7B is not expected to be significantly optically thick. Given the model uncertainties and the degeneracies in the parameter space, the simple analysis (Eq. 1) is likely to provides the most reliable results for this source.
## 4. Discussion
At least two measurements of the 14 N 15 N ratio have been ob-/ tained in all three sources. For IRAS 16239A, the two measurements from the J = 3-2 and 4-3 lines of H 13 CN and HC 15 N, 14 N 15 / N = 163 ± 20 and 14 N 15 / N = 190 ± 38, agree well. The 14 N 15 N / = 242 ± 32 value from the HNC(3-2) isotopologues is di ff erent from the HCN values at the 1 σ level. In the case of R CrA IRS7B, the HCN and HNC values are consistent within the error bars, including also the ratio derived from the data by Watanabe et al. (2012). The 14 N 15 N ratios from HCN and / HNC in OMC-3 MMS6 agree within errors, but the uncertainties are larger in this source than in the two others. Overall, the agreement between di ff erent measurements of the same source is fairly good. The comparison to other measurements and literature values is discussed in the subsections below.
Comparing Table 5 with the temperatures in the outer regions of the spherically symmetric envelope models of the three sources in the sample, it is apparent that the lowest 14 N 15 N ratios are found in the source with the lowest temper-/ ature in the outermost regions, IRAS 16293A. It is followed by R CrA IRS7B, which has a somewhat elevated temperature because it is being irradiated by a nearby Herbig Be star,
Fig. 7. Average 14 N 15 / Nratios of the three young stellar objects plotted versus their outer envelope temperature at the projected beam radius. Error bars reflect the spread in measured 14 N 15 / N ratios.
R CrA (Lindberg & Jørgensen 2012). The highest, almost solar , 14 N 15 N ratios are from OMC-3 MMS6, located in the active re-/ gion of Orion and thus subject to a strong radiation field, which has the highest temperatures in the outer envelope. More quantitatively, this behavior is illustrated in Fig. 7, showing the average of the 14 N 15 N ratio for each source versus the temperature / at the projected radius of the beam. The APEX measurements listed in Table 5 were averaged for each source and the errors bars reflect the spread of the measured ratios. The temperature was taken from the spherically symmetric source models, evaluated at the radius of the APEX beam ( ∼ 24 ′′ ) at the corresponding source distance ( T = 16 5 K for IRAS 16239A, 25 5 K for . . R CrA IRS7B, and 37 K for OMC-3 MMS6). A larger sample would be needed to draw any conclusions about whether or not there is a direct correlation between temperature and 15 Nenhancement, but our result provides at least an indication for a temperature dependence of the 14 N 15 N ratio in these embedded / protostars.
## 4.1. Comparison with previous HCN and HNC measurements in the target sources
Isotopologues of HCN and HNC in IRAS 16293-2422 were previously measured in the spectral surveys by Caux et al. (2011, IRAM and JCMT) and Blake et al. (1994); van Dishoeck et al. (1995, JCMT and CSO). The integrated intensities from the Caux et al. (2011) and van Dishoeck et al. (1995) surveys are listed in Table B.2 in the appendix. Note that the observations by Caux et al. (2011) were pointed at the B component of the protobinary, whereas our observations are pointed directly at the A component. The two components are separated by ∼ 4 ′′ and thus lie both within the beam. However, in the 230 GHz band observations at the IRAM 30m, the flux of the A component is considerably attenuated (see discussion in Caux et al. 2011). The resulting 14 N 15 N ratios are, assuming a calibration uncer-/ tainty of 15%, 276 ± 74 from the H 13 CN(1-0) and HC 15 N(1-0) intensities, 219 ± 71 for the 3-2 transitions, and 320 ± 106 for the 4-3 lines and thus comparable to our values. No H 15 NC intensities are reported by Caux et al. (2011).
From the integrated intensities of HC 15 N(3-2) and H 13 CN(32) measured by van Dishoeck et al. (1995) we find 14 N 15 N / = 131. From the fact that HC 15 N is detected, they argue - pre-
sumably based on the assumption of a local interstellar medium value for the nitrogen isotope ratio (Wilson & Rood 1994) - that H 13 CN is still slightly optically thick. However, from our model results and the fit to the hyperfine structure of H 13 CN(1-0) we conclude that the lines are optically thin and that the assumption of a standard nitrogen isotope ratio likely does not hold because nitrogen is fractionated.
R CrA IRS7B was surveyed in the 345 GHz spectral window by Watanabe et al. (2012) using the ASTE telescope. The integrated intensities are 0 23 . ± 0 09 K km s . -1 for HC 15 N(4-3), 0 95 . ± 0 10 K km s . -1 for H 13 CN(4-3), 0 32 . ± 0 06 K km s . -1 for HN 13 C(4-3), and 0 05 . ± 0 04 K km s . -1 for H 15 NC(4-3). Based on exactly the same procedure as for our own measurements but with a calibration uncertainty of 19% for ASTE (Watanabe et al. 2012), the inferred 14 N 15 N / ratios are 285 ± 139 from HCN and 442 ± 382 from HNC. The ratios derived from the work by Watanabe et al. (2012) are consistent with our measurements, but the error bars are significantly larger than for our APEX observations.
Schöier et al. (2006) observed HCN(4-3) and HNC(4-3) with APEX in R CrA IRS7B, resulting in integrated intensities of 30 4 K km s . -1 and 11 3 K km s . -1 , respectively. Their intensities are very similar to the values found from our observations of HCN(3-2) and HNC(3-2), which are 30 2 K km s . -1 and 14 2 K km s . -1 . For optically thin HCN lines, similar line strengths for the 4-3 and 3-2 transitions are predicted by radiative transfer models if a constant abundance is assumed (see also Jørgensen et al. 2005a, for a discussion of the HCN abundance structure). Therefore, the similar intensities of the 4-3 and 32 transitions could be an indication for little variations in the HCN abundance with radius in this source, and thus no significant freeze-out of HCN.
No HCN and HNC isotopologue measurements were found in the literature for OMC-3 MMS6.
## 4.2. Comparison with 14 N/ 15 N ratios from the literature
14 N 15 N ratios were previously measured in a selection of low-/ mass pre- and protostellar environments from a variety of molecular tracers. The literature results are summarized in Table 1. The table clearly illustrates the discrepancy of 14 N 15 N ratios / derived from HCN and NH3, which may be caused by different fractionation routes for nitriles and nitrogen hydrides (Wirström et al. 2012; Hily-Blant et al. 2013a,b). It also highlights the lack of a comprehensive set of 14 N 15 N ratios from / di ff erent tracers for a source, which makes the interpretation and comparison of individual measurements very challenging. For instance, Adande & Ziurys (2012) found a reasonable agreement between the 14 N 15 / N ratios inferred from HNC and CN. We find that the ratios derived from HCN and HNC are relatively similar. On the other hand, the prestellar core nitrogen isotope composition varies between HCN and CN (Hily-Blant et al. 2013a,b), so it is not clear if there is any correlation between the 14 N 15 N ratios from HCN, HNC, and CN despite the / fact that they all belong to to the 'nitrile' fractionation route (Wirström et al. 2012). Moreover, the HNC values reported in Milam & Charnley (2012) are substantially lower than any other measurement for reasons which are unclear.
For IRAS 16293A, the inferred ratios from HCN (two transitions) and HNC (one measurement) fall into the range of the typical prestellar core HCN values (Hily-Blant et al. 2013a). The similarity of the nitrogen isotope ratio in IRAS 16293A to the prestellar cores is an indication that either fractionation (not necessarily chemical) continues in the envelope of this YSO or that an enhanced 15 N signature from the prestellar core phase has survived into the Class 0 phase. An enhanced radiation field seems to lower or even suppress nitrogen fractionation, as can be seen for the two sources with an enhanced external radiation field, R CrA IRS7B and OMC-3 MMS6. The 14 N 15 N ratio in / R CrA IRS7B lies at the upper end of the prestellar core values and the 14 N 15 N ratio from OMC-3 MMS6 is even higher, / outside the range spanned by the prestellar cores.
## 4.3. Comparison to the local interstellar medium
To decide whether a source shows an enhancement of 15 N in a molecule, one would like to compare the 14 N 15 N / ratio from the source to a reference value of the interstellar medium (ISM). Di ff erent 14 N 15 N measurements of the local interstel-/ lar medium are available from the literature. Wilson (1999) list 14 N 15 / N = 388 ± 32 for the local ISM (updated from the earlier value of 14 N 15 N / = 450 ± 22, Wilson & Rood 1994) based on the HCN surveys data from Wannier et al. (1981, for the northern sources) and Dahmen et al. (1995, for the southern sky) as well as the NH3 survey by Guesten & Ungerechts (1985). However, Wielen & Wilson (1997) argue that the ISM value should be 14 N 15 N / = 414 ± 32, taking into account that the Sun formed at a smaller galactocentric radius that its current location. A local ISM value of 14 N 15 N / = 290 ± 40 is reported by Adande & Ziurys (2012) from measurements of HNC and CN in warm clouds and the 12 C 13 / C ratios from Milam et al. (2005). While the local value from the work by Adande & Ziurys (2012) is inferred from a linear fit to the data throughout the Galaxy, the local ratio reported in Wilson (1999) is the average of 8 sources close to the solar galactic distance. The Wilson (1999) and Adande & Ziurys (2012) results are not consistent within error bars, and the di ff erence presumably arises from the di ff erent carbon isotope ratios ( 12 C 13 / C) assumed for the double isotope method in the case of HCN and HNC. The 12 C 13 / C ratio from CN and CO isotopologues, used in Adande & Ziurys (2012), are generally lower than those obtained from H2CO isotopologues, included in Wilson (1999). However, the local 12 C 13 / C ratios from Wilson (69 ± 6 from CO and H2CO, 1999) and Milam et al. (2005, 68 ± 15 from CO, CN, and H2CO) agree well, but the values for individual sources close to the solar neighborhood from Milam et al. (2005) are partly lower, which is then transferred onto the 14 N 15 N ratios from HNC (Adande & Ziurys 2012). The / Adande & Ziurys (2012) result agrees with the average value of 14 N 15 N / = 237 + 27 -21 measured in two di ff use clouds along the line of sight toward a strong extragalactic source from absorption spectroscopy of HCN (Lucas & Liszt 1998). It was however noticed that there is a significant scatter of the data points from the Galactic isotope surveys, and thus it is unclear how meaningful the comparison of the obtained 14 N 15 N ratios to an average / local ISM value is to decide whether or not the nitrogen isotopes are fractionated in a specific source. Moreover, the local ISM value is also subject to change, as galactic evolution models by Romano & Matteucci (2003) for instance predict that the 14 N 15 N ratio of the ISM increases with galactocentric distance / and decreases with time.
Our IRAS 16293A results from HCN lie outside the uncertainty range of both the Adande & Ziurys (2012) and Wilson (1999) local ISM values, while the ratio inferred from HNC lies within the uncertainties of Adande & Ziurys (2012), but clearly outside Wilson (1999), as do the R CrA HCN and HNC ratios. The HCN measurement from OMC-3 MMS6 is agreement with both ISM values, while the ratio from HNC, which is the most uncertain of all data points, is consistent with the Wilson (1999)
ISMresult but falls above the one from Adande & Ziurys (2012). Hence, from the comparison with ISM measurements, the nitrogen isotopes in IRAS 16293A show indications for fractionation, and fractionation may possibly also occur in R CrA IRS7B, while OMC-3 MMS6 shows no evidence for fractionation if the comparison with a local ISM value is meaningful. The 12 C 13 C / ratio should also be measured for our sources to check and account for any deviations from the standard.
## 4.4. Comparison to solar system values
A direct comparison of the nitrogen isotope composition with that of solar system bodies is di ffi cult as possible mechanisms for a separation of the protostellar material into a 15 N-poor gas phase and 15 N-rich solids are not yet understood. The freezeout of 15 N-substituted molecular species, in particular ammonia, onto dust grains and their incorporation into the ice mantles is believed to be important (Rodgers & Charnley 2008b; Wirström et al. 2012), but the detailed processes are still unclear. Existing observations of the nitrogen isotope composition in star-forming regions do not point at a large nitrogen isotope fractionation in ammonia (Gerin et al. 2009; Daniel et al. 2013). Gerin et al. (2009) and Bizzocchi et al. (2013) even argue for a depletion of 15 N in NH3 and N2H , respectively, in the prestel-+ lar core L1544. The comparison is further complicated by the fact that the chemical carrier of the elevated 15 N-signature or the functional group carrying the isotopic anomaly in primitive solar system matter has not yet been identified. Furthermore, the cosmomaterials that can nowadays be sampled may represent only a minor reservoir of the presolar nebula. It may have also underdone significant processing during the evolution of the solar system, such as heating or aqueous alteration, which could a ff ect the isotopic signatures (e.g., Alexander et al. 2007).
Bulk 15 N-enrichments in meteorites are comparable to the values inferred from star-forming regions. However, meteoritic 15 N-hotspots feature enrichments exceeding the one most extreme values for prestellar cores (Hily-Blant et al. 2013a) or our protostar sample. Within errors, the cometary ratios are also in agreement with our HCN results for IRAS 16293A, but lower than the OMC-3 MMS6 and R CrA IRS7B results. Overall, the 14 N 15 N ratios in primitive solar system matter, i.e., meteorites, / comets, and IDPs, are comparable to the ones measured around prestellar cores and protostars (cf. comparison in Fig. 4), but the ambiguities on both the meteoritic and the astrochemical side do not yet allow concluding whether or not the 15 N-signature in meteorites was directly inherited from a 15 N-enriched precursor molecule in the presolar nebula or whether it is the result of a secondary process.
## 4.5. Chemical fractionation or photochemistry?
A possible explanation for the isotopic variations among our sample could be the di ff erent amounts of external irradiation the sources are exposed to. IRAS 16293A, being a relatively isolated source, has the lowest temperatures in the outer envelope, whereas OMC-3 MMS6 and R CrA IRS7B are both exposed to an enhanced radiation field, resulting in warmer outer envelope temperatures (Jørgensen et al. 2006; Lindberg & Jørgen sen 2012). There is a tentative trend toward a decreasing 15 Nenhancement with increasing outer envelope temperature in our sample, but the statistics is too small to draw any conclusions. If the trend can be confirmed, it would argue in favor of the chemical fractionation scenario over isotope-selective photochem- istry. If photodissociation of N2 was the process dominating the nitrogen isotope composition, the highest 15 N-enrichments should be present in the two externally irradiated sources, not IRAS 16293A. However, the single-dish observations only provide a spatially averaged value of the 14 N 15 N ratio. / Because fractionation may not occur everywhere in the envelope, and because the mechanism that dominates the nitrogen isotope composition may even vary throughout the envelope depending on physical parameters such as temperature, irradiation, and opacity, spatially resolved observations are clearly needed to study potential spatial variations in 14 N 15 N ratio. / For instance, photochemistry could determine the 14 N 15 N ratio in the outermost / and maybe innermost parts close to the protostar, where the radiation field is intense, whereas chemical fractionation is expected to be important in cold, shielded regions. Our understanding of the evolution of nitrogen isotopic composition during star- and planet formation would also benefit from a more comprehensive data set with nitrogen isotope ratios measured from a census of molecular tracers and sources in di ff erent evolutionary stages. Nonetheless, our results indicate that enhanced 15 N-values in HCN and HNC are present not just in the prestellar stage, but also around at least some Class 0 protostars. Whether fractionation is still ongoing in these sources or just preserved from the precedent prestellar core phase and what mechanism is causing the 15 N-enrichment in HCN and HNC is not yet clear and needs to be further investigated.
## 5. Conclusions
We have measured the 14 N 15 N isotope ratio in the three Class 0 / protostars IRAS 16293A, R CrA IRS7B, and OMC-3 MMS6 from 15 N- and 13 C-substituted isotopologues of HCN and HNC, observed with the APEX telescope. Optically thin emission, the same excitation temperature for both isotopologues, and a standard carbon isotope ratio of 12 C 13 C / = 69 (Wilson 1999; Milam et al. 2005) were assumed to derive the 14 N 15 N ratios di-/ rectly from the integrated line intensities. The assumption of optically thin emission is supported by the results from H 13 CN(10) hyperfine structure fitting for IRAS 16293A and by non-LTE line radiative transfer modeling for all three sources.
The ratios derived from the HCN and HNC isotopologues are relatively similar, but there is significant variation between the tree sources. The values for IRAS 16293A ( 14 N / 15 N ∼ 160 -240) are in the range of typical prestellar core values ( 14 N 15 N / ∼ 140 -360, Hily-Blant et al. 2013a), whereas the results for R CrA IRS7B ( 14 N 15 N / ∼ 260 -290) indicate that the nitrogen isotopes in this source are less fractionated. For OMC-3MMS6( 14 N / 15 N ∼ 370 -460), a solar and hence unfractionated nitrogen isotope composition cannot be excluded. The di ff erences between the sources possibly stem from the di ff erent outer envelope temperatures, presumably caused by external irradiation heating the envelope.
Our results indicate that fractionation of nitrogen isotopes in HCN and HNC either continues in the Class 0 stage or is at least preserved from the prestellar core phase. Additional tracers should also be studied to learn more about what molecules carry enhanced 15 N and could be responsible for the 15 N-enrichment observed in solar system materials.
Chemical fractionation seems a more likely scenario than N2 self-shielding to explain the nitrogen isotope measurements in our data set because of a tentative trend toward a decreasing 14 N 15 N with decreasing envelope temperature. / However, spatially resolved observations are crucial to pinpoint which mech-
anism causes the observed 15 N-enrichment in some Class 0 protostars.
## References
Abbas, M. M., LeClair, A., Owen, T., et al. 2004, ApJ, 602, 1063 Adande, G. R. & Ziurys, L. M. 2012, ApJ, 744, 194 Aléon, J. 2010, ApJ, 722, 1342 Alexander, C. M. O´D., Fogel, M., Yabuta, H., & Cody, G. D.
Geochim. Cosmochim. Acta, 71, 4380
- Anders, E. & Grevesse, N. 1989, Geochim. Cosmochim. Acta, 53, 197
- Bertin, M., Fayolle, E. C., Romanzin, C., et al. 2013, ApJ, 779, 120
- Birck, J. L. 2004, Rev. Mineral. Geochem., 55, 25
- Bizzocchi, L., Caselli, P., & Dore, L. 2010, A&A, 510, L5 +
- Bizzocchi, L., Caselli, P., Leonardo, E., & Dore, L. 2013, A&A, 555, A109
- Blake, G. A., van Dishoeck, E. F., Jansen, D. J., Groesbeck, T. D., & Mundy, L. G. 1994, ApJ, 428, 680
Bockelée-Morvan, D., Biver, N., Jehin, E., et al. 2008, ApJ,679, L49
- Böhlke, J. K., de Laeter, J. R., De Bièvre, P., et al. 2005, J. P hys. Chem. Ref. Data, 34, 57
- Bonal, L., Huss, G. R., Krot, A. N., et al. 2010, Geochim. Cosmochim. Acta, 74, 6590
- Boogert, A. C. A., Hogerheijde, M. R., Ceccarelli, C., et al. 2002, ApJ, 570, 708 Briani, G., Gounelle, M., Marrocchi, Y., et al. 2009, Proceedings of the National Academy of Sciences, 106
Brinch, C., Jørgensen, J. K., & Hogerheijde, M. R. 2009, A&A,
502, 199
Busemann, H., Young, A. F., O'D. Alexander, C. M., et al. 2006, Science, 312,
727
- Caux, E., Kahane, C., Castets, A., et al. 2011, A&A, 532, A23
Clayton, R. N. 2002, Nature, 415, 860
- Connelly, J. N., Bizzarro, M., Krot, A. N., et al. 2012, Science, 338, 651
Coplen, T. B., Krouse, H. R., & Böhlke, J. K. 1992, Pure Appl. Chem, 64, 907
Dahmen, G., Wilson, T. L., & Matteucci, F. 1995, A&A, 295, 194
- Daniel, F., Gérin, M., Roue , E., et al. 2013, A&A, 560, A3 ff
- Dumke, M. & Mac-Auli ff e, F. 2010, in Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series, Vol. 7737
Dumouchel, F., Faure, A., & Lique, F. 2010, MNRAS, 406, 2488
Fletcher, L. N., Greathouse, T. K., Orton, G. S., et al. 2014, Icarus, 238, 170
Floss,
C.,
Stadermann,
F.
J.,
Bradley,
J.
P., et
al.
2006,
Geochim.
Cos- mochim. Acta, 70, 2371
Fouchet, T., Irwin, P. G. J., Parrish, P., et al. 2004, Icarus, 172, 50
Fouchet, T., Lellouch, E., Bézard, B., et al. 2000, Icarus, 1 43, 223
- Gerin, M., Marcelino, N., Biver, N., et al. 2009, A&A, 498, L9
Green, S. & Thaddeus, P. 1974, ApJ, 191, 653
- Guesten, R. & Ungerechts, H. 1985, A&A, 145, 241
Güsten, R., Nyman, L. Å., Schilke, P., et al. 2006, A&A, 454, L 13
- Hashizume, K., Chaussidon, M., Marty, B., & Robert, F. 2000, Science, 290, 1142
- Heays, A. N., Visser, R., Gredel, R., et al. 2014, A&A, 562, A61
- Hily-Blant, P., Bonal, L., Faure, A., & Quirico, E. 2013a, Icarus, 223, 582
- Hily-Blant, P., Pineau des Forêts, G., Faure, A., Le Gal, R.,& Padovani, M. 2013b, A&A, 557, A65
- Ho man, J. H., Hodges, R. R., McElroy, M. B., Donahue, T. M., & Kolpin, M. ff 1979, Science, 205, 49
Hogerheijde, M. R. & van der Tak, F. F. S. 2000, A&A, 362, 697
Hutsemékers, D., Manfroid, J., Jehin, E., et al. 2005, A&A, 4 40, L21
Ikeda, M., Hirota, T., & Yamamoto, S. 2002, ApJ, 575, 250
- Jehin, E., Manfroid, J., Hutsemékers, D., Arpigny, C., & Zuc coni, J.-M. 2009, Earth Moon and Planets, 105, 167
- Johnstone, D., Boonman, A. M. S., & van Dishoeck, E. F. 2003, A&A, 412, 157 Jørgensen, J. K., Bourke, T. L., Myers, P. C., et al. 2005a, Ap J, 632, 973
Jørgensen, J. K., Bourke, T. L., Nguyen Luong, Q., & Takakuwa , S. 2011, A&A,
534, A100
- Jørgensen, J. K., Johnstone, D., Kirk, H., et al. 2008, ApJ, 6 83, 822
- Jørgensen, J. K., Johnstone, D., van Dishoeck, E. F., & Doty, S. D. 2006, A&A, 449, 609
Jørgensen, J. K., Schöier, F. L., & van Dishoeck, E. F. 2004, A &A, 416, 603 Jørgensen, J. K., Schöier, F. L., & van Dishoeck, E. F. 2005b, A&A, 435, 177 Junk, G. & Svec, H. J. 1958, Geochim. Cosmochim. Acta, 14, 234 Kerridge, J. F. 1993, Reviews of Geophysics, 31, 423 Li, X., Heays, A. N., Visser, R., et al. 2013, A&A, 555, A14 Liang, M.-C., Heays, A. N., Lewis, B. R., Gibson, S. T., & Yung, Y. L. 2007,
ApJ, 664, L115 Lindberg, J. E. & Jørgensen, J. K. 2012, A&A, 548, A24 Lindberg, J. E., Jørgensen, J. K., Brinch, C., et al. 2014, A& A, 566, A74 Linke, R. A., Goldsmith, P. F., Wannier, P. G., Wilson, R. W., & Penzias, A. A. 1977, ApJ, 214, 50
Lis, D. C., Wootten, A., Gerin, M., & Roue ff , E. 2010, ApJ, 710, L49 Loinard, L., Castets, A., Ceccarelli, C., et al. 2000, A&A, 359, 1169
- 2007,
- Lucas, R. & Liszt, H. 1998, A&A, 337, 246
- Lyons, J. R., Bergin, E. A., Ciesla, F. J., et al. 2009, Geochim. Cosmochim. Acta, 73, 4998
- Manfroid, J., Jehin, E., Hutsemékers, D., et al. 2009, A&A, 5 03, 613
- Mariotti, A. 1983, Nature, 303, 685
- Marty, B., Chaussidon, M., Wiens, R. C., Jurewicz, A. J. G., & Burnett, D. S. 2011, Science, 332, 1533
- Mathew, K. J. & Marti, K. 2001, J. Geophys. Res., 106, 1401
- McKeegan, K. D., Aléon, J., Bradley, J., et al. 2006, Science , 314, 1724
Meibom, A., Krot, A. N., Robert, F., et al. 2007, ApJ, 656, L33
- Messenger, S. 2000, Nature, 404, 968
- Milam, S. N. & Charnley, S. B. 2012, in Lunar and Planetary Inst. Technical Report, Vol. 43, Lunar and Planetary Institute Science Conference Abstracts, 2618
- Milam, S. N., Savage, C., Brewster, M. A., Ziurys, L. M., & Wycko ff , S. 2005, ApJ, 634, 1126
- Müller, H. S. P., Schlöder, F., Stutzki, J., & Winnewisser, G. 2005, Journal of Molecular Structure, 742, 215
- Müller, H. S. P., Thorwirth, S., Roth, D. A., & Winnewisser, G. 2001, A&A, 370, L49
- Mumma, M. J. & Charnley, S. B. 2011, ARA&A, 49, 471
- Niemann, H. B., Atreya, S. K., Bauer, S. J., et al. 2005, Nature, 438, 779
- Owen, T., Maha ff y, P. R., Niemann, H. B., Atreya, S., & Wong, M. 2001, ApJ, 553, L77
- Persson, M. V., Jørgensen, J. K., & van Dishoeck, E. F. 2013, A &A, 549, L3
Pickett,
H. M., Poynter, R. L., Cohen, E. A., et al. 1998, J. Quant. Spec. Ra- diat. Transf., 60, 883
- Pineda, J. E., Maury, A. J., Fuller, G. A., et al. 2012, A&A, 544, L7
- Rodgers, S. D. & Charnley, S. B. 2008a, ApJ, 689, 1448
- Rodgers, S. D. & Charnley, S. B. 2008b, MNRAS, 385, L48
- Romano, D. & Matteucci, F. 2003, MNRAS, 342, 185
- Rousselot, P., Pirali, O., Jehin, E., et al. 2014, ApJ, 780, L17
- Schöier, F. L., Jørgensen, J. K., Pontoppidan, K. M., & Lundg ren, A. A. 2006, A&A, 454, L67
- Schöier, F. L., Jørgensen, J. K., van Dishoeck, E. F., & Blake , G. A. 2002, A&A, 390, 1001
- Schöier, F. L., van der Tak, F. F. S., van Dishoeck, E. F., & Black, J. H. 2005, A&A, 432, 369
- Shinnaka, Y., Kawakita, H., Kobayashi, H., Nagashima, M., & Boice, D. C. 2014, ApJ, 782, L16
- Takakuwa, S., Ohashi, N., Bourke, T. L., et al. 2007, ApJ, 662, 431
- Tennekes, P. P., Harju, J., Juvela, M., & Tóth, L. V . 2006, A&A, 456, 1037
Terzieva, R. & Herbst, E. 2000, MNRAS, 317, 563
- van Dishoeck, E. F., Blake, G. A., Jansen, D. J., & Groesbeck, T. D. 1995, ApJ, 447, 760
- Vassilev, V., Meledin, D., Lapkin, I., et al. 2008, A&A, 490, 1157
- Wannier, P. G., Linke, R. A., & Penzias, A. A. 1981, ApJ, 247, 522
- Watanabe, Y., Sakai, N., Lindberg, J. E., et al. 2012, ApJ, 745, 126
- Wielen, R. & Wilson, T. L. 1997, A&A, 326, 139
- Wilson, T. L. 1999, Reports on Progress in Physics, 62, 143
- Wilson, T. L. & Rood, R. 1994, ARA&A, 32, 191
- Wirström, E. S., Charnley, S. B., Cordiner, M. A., & Milam, S. N. 2012, ApJ, 757, L11
- Womack, M., Ziurys, L. M., & Wycko ff , S. 1992, ApJ, 387, 417
- Wong, M. H., Atreya, S. K., Maha ff y, P. N., et al. 2013, Geophys. Res. Lett., 40, 6033
Acknowledgements. The authors thank the referee for constructive comments that helped to improve the article. Research for this project was supported by a Danish National Research Foundation Centre of Excellence grant to M.B. (grant number DNRF97) and from a Lundbeck Foundation Junior Group Leader Fellowship to J.K.J. M.B. further acknowledges funding from the European Research Council under ERC Consolidator grant agreement 616027-STARDUST2ASTEROIDS. The research leading to these results has received funding from the European Commission Seventh Framework Programme (FP 2007-2013) under grant agreement No 283393 (RadioNet3). / The authors would like to thank the Onsala APEX sta / ff for carrying out the observations. Per Bergman and the Onsala APEX sta / ff are also thanked for help assessing potential contamination from the o ff position.
## Appendix A: Observational details
Fig. A.1. Comparison of the H 13 CO (3-2) line from IRAS 16293A ob-+ served with APEX. The spectra resulting from phase 1 (corresponding to o ff position on one side of the source in symmetric wobbler switching mode) and phase 2 (o ff position on the other side) are shown in blue and red, respectively. The spectra have been baseline subtracted and normalized to the average peak intensity of the two spectra. Data from 2012-04-01.
Fig. A.2. Like Fig A.1 but for H 13 CN(3-2). Includes a sinusoidal baseline for standing wave removal.
Fig. A.3. Like Fig A.1 but for R CrA IRS7B. Data from 2012-09-27.
Fig. A.4. Like Fig A.3 but for H 13 CN(3-2).
Fig. A.5. Like Fig A.1 but for OMC-3 MMS6. Data from 2012-11-22.
Fig. A.6. Like Fig A.5 but for H 13 CN(3-2).
Table A.1. Observational details.
| Transition(s) | Date | # scans | time (on) [min] | Remarks |
|------------------------------------------|--------------|--------------|-------------------|----------------------------------------------------------------------------------|
| IRAS 16293A | IRAS 16293A | IRAS 16293A | IRAS 16293A | IRAS 16293A |
| H 13 CN(3-2), HC 15 N(3-2) | 2012-04-01 | 67 | 20 | |
| H 13 CN(4-3), HC 15 N(4-3) | 2012-06-04 | 41 | 12 | |
| H 13 C 15 N(3-2) | 2012-06-13 | 150 | 44 | USB (30 scans), LSB (120 scans) LSB su ff ers from strong spur |
| DCN(3-2) | 2012-06-08 | 29 | 9 | scans 47-48 ignored (LO locked out) |
| DCN(4-3) | 2012-06-07 | 30 | 9 | |
| DCN(5-4), HNC(4-3) | 2012-06-04 | 40 | 12 | baseline ripples |
| D 13 CN(3-2) | 2012-06-08 | 120 | 35 | scans 541-542 ignored (LO locked out) |
| | 2012-06-09 | 41 | 12 | scans 215-218 ignored (LO locked out) |
| | 2012-06-10 | 50 | 15 | |
| | 2012-06-11 | 60 | 18 | |
| HN 13 C(3-2), H 13 CN(3-2), HC 15 N(3-2) | 2013-08-20 | 36 | 21 | H 13 CN and HC 15 N intensities higher than those observed on 2012-04-01 |
| H 15 NC(3-2), HCN(3-2) | 2013-08-19 | 36 | 21 | |
| R CrA IRS 7B | R CrA IRS 7B | R CrA IRS 7B | R CrA IRS 7B | R CrA IRS 7B |
| HN 13 C(3-2), H 13 CN(3-2), HC 15 N(3-2) | 2012-08-16 | 30 | 9 | |
| | 2012-09-27 | 50 | 15 | |
| HN 13 C(4-3), H 13 CN(4-3) | 2012-08-22 | 50 | 15 | |
| | 2012-09-25 | 30 | 9 | |
| H 15 N(3-2)C, HCN(3-2) | 2012-08-16 | 108 | 32 | |
| | 2012-09-27 | 140 | 41 | |
| | 2012-09-29 | 90 | 26 | |
| | 2012-09-30 | 40 | 12 | intensities &line profiles o ff , data discarded |
| | 2012-11-20 | 30 | 9 | |
| | 2012-11-21 | 40 | 12 | |
| | 2012-11-23 | 100 | 29 | |
| | 2012-11-24 | 10 | 3 | |
| | 2012-11-28 | 50 | 15 | |
| HNC(3-2)C | 2012-09-27 | 20 | 6 | |
| OMC3 MMS6 | OMC3 MMS6 | OMC3 MMS6 | OMC3 MMS6 | OMC3 MMS6 |
| HN 13 C(3-2), H 13 CN(3-2), HC 15 N(3-2) | 2012-08-16 | 28 | 8 | |
| | 2012-11-22 | 120 | 35 | |
| HN 13 C(4-3), H 13 CN(4-3) | 2012-09-26 | 30 | 9 | |
| H 15 NC(3-2), HCN(3-2) | 2012-08-16 | 70 | 21 | |
| | 2012-09-25 | 170 | 50 | |
| | 2012-11-24 | 90 | 26 | |
| | 2012-11-25 | 70 | 21 | |
| | 2012-11-26 | 190 | 56 | |
| HNC(3-2)C | 2012-09-27 | 10 | 3 | |
| Elias 29 | Elias 29 | Elias 29 | Elias 29 | Elias 29 |
| HN 13 C(3-2), H 13 CN(3-2), HC 15 N(3-2) | 2012-08-16 | 30 | 9 | intensity on the two days is very di ff erent (factor of ∼ 1 . 5), data not used |
| HN 13 C(4-3), H 13 CN(4-3) | 2012-09-30 | 80 | 24 | intensity on the two days is very di ff erent (factor of ∼ 1 . 5), data not used |
| H 15 N(3-2)C, HCN(3-2) | 2012-08-16 | 50 | 15 | intensity on the two days is very di ff erent (factor of ∼ 1 . 5), data not used |
| | 2012-10-01 | 76 | 23 | intensity on the two days is very di ff erent (factor of ∼ 1 . 5), data not used |
| HNC(3-2)C | 2012-09-30 | 160 | 47 | intensity on the two days is very di ff erent (factor of ∼ 1 . 5), data not used |
| | 2012-10-01 | 50 | 15 | |
Appendix B: Literature values
Table B.1. Measurements of the 14 N 15 / N ratio in solar system materials from the literature. (B) stands for 'bulk' measurements and (HS) for 'hotspot'. Boldface indicates the unit(s) reported in the original publication (if none of the values is in boldface, the original publication reports 15 N 14 / N). Values marked with a star were used for Fig. 4.
| Object | 14 N / 15 N | δ 15 N [ /permil ] | Ref. | Remarks |
|-----------------|----------------------------------|-------------------------------------|-----------|----------------------------------------------------------|
| Sun | 441 ± 6 | - 383 ± 8 | a , /star | Genesis |
| Lunar soils | ≥ 358 | - 240 | b | trapped solar wind |
| Meteorite (CAI) | 424 ± 3 | ≤ - 359 ± 6 | c , /star | Isheyevo, osbornite (TiN) |
| Jupiter | 526 | 423 | d | ISO, NH 3 , 15 N / 14 N = 1 . 9 + 0 . 9 - 1 . 0 × 10 - 3 |
| Jupiter | 435 ± 57 | - - 375 ± 82 | e , /star | Galileo, NH 3 |
| Jupiter | 448 62 | 393 84 | f | Cassini-CIRS, NH |
| Jupiter | ± 450 ± 106 | - ± - 396 ± 142 | g | 3 Cassini-CIRS, NH 3 |
| Saturn | ≥ 357 | - 238 | h | ground-based, NH 3 |
| Titan | 183 ± 5 | ≤ 486 ± 41 | i | Huygens, N 2 , secondary origin |
| Earth (atmo.) | 272 | 0 | j , /star | |
| Venus (atmo.) | 272 ± 54 | 0 ± 200 | k , /star | |
| Mars (rock) | 280 | 30 | l , /star | ALH84001 meteorite |
| Mars (atmo.) | 173 ± 11 | - 572 ± 82 | m | Curiosity, secondary origin |
| Comets | 130 to 220 | 230 to 1100 | n , /star | approx. range, CN |
| Comet | 139 ± 26 | 957 ± 366 | o | 17P / Holmes, HCN |
| Comet | 165 ± 40 | 648 ± 400 | o | 17P / Holmes, CN |
| Comet | 139 ± 38 | 957 ± 535 | p | Comet C / 2012 S1 (ISON), NH 3 |
| Comets | 127 ( 80 - 190 ) | 1142 | q | avg. spectrum (12 comets), NH 3 |
| Comet (B) | 181 to 272 | 0 to 500 | r | Comet 81P / Wild2, Stardust |
| Comet (HS) | 118 to 368 | - 260 to 1300 ± 400 | r | Comet 81P / Wild2, Stardust |
| IDPs (B) | 180 . 7 ± 2 . 3 to 305 . 1 ± 3 . | - 108 ± 9 to 500 ± 20 | s | IDPs 'Kipling', 'Eliot' |
| IDPs (HS) | 118 ± 6 to 171 ± 4 | 590 ± 40 to 1300 ± 120 | s | IDPs 'Kipling', 'Pirandello' |
| Cluster IDPs | 184 to 300 | - 93 ± 4 to 480 ± 40 | t | IDPs 'Porky', 'Dragonfly' |
| Meteorites (B) | 192 to 237 | 146 to 415 | u , /star | |
| Meteorites (HS) | 65 to 193 | 410 ± 130 to 3200 ± 700 | u | |
| Meteorites | 192 to 291 | - 65 . 6 ± 2 . 9 to 415 . 3 ± 1 . 6 | v , /star | insoluble organic matter (IOM) |
| Meteorite (HS) | 50 | 4450 | w , /star | Isheyevo clast |
| Meteorite (HS) | 46 | 4900 300 | x | Isheyevo clast |
±
References. a Marty et al. (2011), b Hashizume et al. (2000), c Meibom et al. (2007), d Fouchet et al. (2000), e Owen et al. (2001), f Abbas et al. (2004), g Fouchet et al. (2004), h Fletcher et al. (2014), i Niemann et al. (2005) j Junk & Svec (1958); Coplen et al. (1992); Böhlke et al. (2005), k Ho man et al. (1979), ff l Mathew & Marti (2001), m Wong et al. (2013), n Manfroid et al. (2009); Jehin et al. (2009); Mumma & Charnley (2011) and references therein, o Bockelée-Morvan et al. (2008), p Shinnaka et al. (2014), q Rousselot et al. (2014), r McKeegan et al. (2006), s Floss et al. (2006), t Messenger (2000), u Busemann et al. (2006), v Alexander et al. (2007), w Bonal et al. (2010), x Briani et al. (2009)
Table B.2. Integrated intensities from the surveys by Caux et al. (2011) and Blake et al. (1994); van Dishoeck et al. (1995).
| Transition | ∫ Tdv [K km / s] | Telescope |
|-----------------------------------------------------------------------------------------------------------------------------------|--------------------|-------------|
| H 13 CN (1 2 -0 1 ) H 13 CN (1 1 -0 1 ) H 13 CN (1 0 -0 1 ) H 13 CN (3-2) H 13 CN (4-3) HC 15 N (1-0) HC 15 N (2-1) HC 15 N (3-2) | 1 . 04 ± 0 . 12 | IRAM |
| | 0 . 75 ± 0 . 09 | IRAM |
| | 0 . 25 ± 0 . 03 | IRAM |
| | 8 . 16 ± 1 . 39 | IRAM |
| | 6 . 90 ± 1 . 24 | JCMT |
| | 0 . 51 ± 0 . 06 | IRAM |
| | 1 . 88 ± 0 . 32 | IRAM |
| | 2 . 57 ± 0 . 45 | IRAM |
| HC 15 N (4-3) | 1 . 49 ± 0 . 27 | JCMT |
| H 13 CN (3-2) | 3 . 45 | JCMT |
| H 13 CN (4-3) | 8 . 13 | CSO |
| HC 15 N (3-2) | 1 . 82 | JCMT |
| HN 13 C (3-2) | 1 . 33 | JCMT | | 10.1051/0004-6361/201423773 | [
"S. F. Wampfler",
"J. K. Jorgensen",
"M. Bizzarro",
"S. E. Bisschop"
] | 2014-08-01T20:00:12+00:00 | 2014-08-01T20:00:12+00:00 | [
"astro-ph.SR",
"astro-ph.EP",
"astro-ph.GA"
] | Observations of nitrogen isotope fractionation in deeply embedded protostars | (Abridged) The terrestrial planets, comets, and meteorites are significantly
enriched in 15N compared to the Sun and Jupiter. While the solar and jovian
nitrogen isotope ratio is believed to represent the composition of the
protosolar nebula, a still unidentified process has caused 15N-enrichment in
the solids. Several mechanisms have been proposed to explain the variations,
including chemical fractionation. However, observational results that constrain
the fractionation models are scarce. While there is evidence of 15N-enrichment
in prestellar cores, it is unclear how the signature evolves into the
protostellar phases. Our aim is to measure the 14N/15N ratio around three
nearby, embedded low-to-intermediate-mass protostars. Isotopologues of HCN and
HNC were used to probe the 14N/15N ratio. A selection of H13CN, HC15N, HN13C,
and H15NC transitions was observed with the APEX telescope. The 14N/15N ratios
were derived from the integrated intensities assuming a standard 12C/13C ratio.
The assumption of optically thin emission was verified using radiative transfer
modeling and hyperfine structure fitting. Two sources, IRAS 16293A and R CrA
IRS7B, show 15N-enrichment by a factor of around 1.5-2.5 in both HCN and HNC
with respect to the solar composition. Solar composition cannot be excluded for
the third source, OMC-3 MMS6. Furthermore, there are indications of a trend
toward increasing 14N/15N ratios with increasing outer envelope temperature.
The enhanced 15N abundances in HCN and HNC found in two Class~0 sources
(14N/15N of 160-290) and the tentative trend toward a temperature-dependent
14N/15N ratio are consistent with the chemical fractionation scenario, but
14N/15N ratios from additional tracers are indispensable for testing the
models. Spatially resolved observations are needed to distinguish between
chemical fractionation and isotope-selective photochemistry. |
1408.0287v1 | ## Consistent Use of Effective Potentials
Anders Andreassen , William Frost , and Matthew D. Schwartz ∗ † ‡
Department of Physics, Harvard University, Cambridge, MA 02138, USA
August 5, 2014
## Abstract
It is well known that effective potentials can be gauge-dependent while their values at extrema should be gauge-invariant. Unfortunately, establishing this invariance in perturbation theory is not straightforward, since contributions from arbitrarily highorder loops can be of the same size. We show in massless scalar QED that an infinite class of loops can be summed (and must be summed) to give a gauge invariant value for the potential at its minimum. In addition, we show that the exact potential depends on both the scale at which it is calculated and the normalization of the fields, but the vacuum energy does not. Using these insights, we propose a method to extract some physical quantities from effective potentials which is self-consistent order-by-order in perturbation theory, including improvement with the renormalization group.
## 1 Introduction
Effective actions provide a powerful organizing framework for quantum field theory. Often one computes an effective action by integrating out a massive particle, usually at fixed order in perturbation theory as in the 4-Fermi theory, but sometimes to all orders in couplings but leading order in 1 m 2 , as in the Euler-Heisenberg Lagrangian [1,2]. In these cases, the effective action is an action just like the original one, but with a subset of its fields. Thus it can be used for the same calculations, such as S -matrix elements, in the same way. An advantage of the effective action approach is that it is often easier to expand in 1 m 2 and then calculate the physical quantity than to perform the calculation before the expansion.
A more subtle use of effective actions occurs when one integrates out all the particles. Integrating out everything of course just produces a number (usually ∞ ). To get functional dependence one can instead integrate out all the particles in the presence of classical sources
∗ [email protected]
† [email protected]
‡ [email protected]
J for each to get a functional W J [ ] of those sources. The Legendre transform of W J [ ] gives an effective action Γ[ φ ] = W J [ ] -Jφ , which we call the 1PI effective action. In the 1PI effective action, the fields are classical background fields. One cannot perform loops using Γ[ φ ] since all the loops have already been performed in integrating out the particles. Γ[ φ ] nevertheless encodes all the classical and quantum physics of the original theory. This Γ[ φ ] is therefore an extremely powerful object. It is also extremely difficult to compute.
These two types of effective action are of course closely related; for example, the 1PI effective action for QED reduces to the Euler-Heisenberg Lagrangian when restricted to a single electron loop and constant electromagnetic fields. It is nevertheless important to understand that the 1PI effective action does not function like the kind of integral over a local Lagrangian one usually uses in quantum field theory. Γ[ φ ] describes fluctuations in the presence of background sources, which are normally absent. The fields φ on which Γ depends are classical fields, functions of space-time, not quantum operators acting on a Hilbert space. The value of the action at generic classical field values φ has no clear physical meaning. Indeed, its value depends on the way the calculation is performed: what gauge is chosen, the normalization of the fields, and even the scale at which the action is computed (see Section 2 below). On-shell quantities, such as S -matrix elements computed from the effective action do not depend on these conventions. An important property of the effective action, or the effective potential to which it reduces for constant fields, is that its minimum gives the energy of the true ground state of a theory. For this reason, effective actions provide unrivaled insight into spontaneous symmetry breaking [3,4].
Despite the unphysical nature of the effective action at generic field values, its gaugedependence had elicited a surprising amount of consternation. Since the gauge-dependence was first observed forty years ago [5], papers are regularly written claiming that the effective potential should be defined in some other way, often through a field redefinition, so that it is manifestly gauge invariant. The easiest way to make the potential gauge invariant is simply to pick a gauge: In the mid 1970's, Dolan and Jackiw argued that only unitary gauge ( ξ = ∞ ) makes sense [6] while S. Weinberg insisted upon Landau gauge [7] ( ξ = 0). Around the same time, Frere and Nicoletopoulos (based on observations of Fischler and Brout [8]) argued that one should use dressed fields which have no field strength renormalization [9], which is similar to Yennie gauge ( ξ = 3). In 1998, Tye and Vtorov-Karevsky argued that one should use an exact nonlinear field redefinition, replacing a scalar doublet φ 1 + iφ 2 by a linear sigma model σe iπ [10]. Others have tried using composite fields, such as φ φ ∗ [11], working in a Hamiltonian formulation [12] or the Vikovisky-DeWitt [13] formalism, where the introduction of gauge-invariance (and therefore gauge non-invariance) is sidestepped completely. In 2014, Nielsen [14] proposed a nonlinear, but perturbative, field redefinition determined by solving a differential equation with a boundary condition at ξ = ξ 0 to remove the dependence on ξ .
The periodic frustration with the gauge-dependence of the effective potential seems to us somewhat misguided. Gauge invariance of unphysical quantities is commonplace. One is not normally bothered by the ξ -dependence of the photon two-point function or the anomalous dimension of a charged scalar field. Similarly, there is no reason to be dismayed about the
gauge-dependence of the effective action. While field redefinitions can treat a symptom of the unphysical nature of the potential, they cannot make it any more physical. In fact, none of papers cited above seem to address the incongruity of both using the freedom in quantum field theory to perform field redefinitions and claiming that a particular field redefinition is preferred. Of course, physical quantities, such as on-shell S -matrix elements must be gauge-invariant. Furthermore, gauge-invariance provides a very useful theoretical consistency check. Not only does it help verify that you did the relevant loop calculations correctly, but it can also verify that the quantity you are calculating could possibly be physical. Thus, for an object as unintuitive as the effective potential, gauge-dependence is in many ways actually desirable.
With these observations in mind, let us consider how gauge invariance of some physical quantities calculated from the effective potential have been established in perturbation theory and beyond. Our focus in this paper is on scalar QED at zero-temperature. With a negative mass-squared for the scalar this model is called the Abelian Higgs model. With zero mass we call it the Coleman-Weinberg model [7]. General non-perturbative demonstrations that certain quantities calculated from the effective potential in these models should be gauge invariant were provided by Nielsen [15] and independently Fukuda and Kugo [16]. Quantities which are expected to be gauge-invariant in these models include the value of the potential at its extrema and the pole masses of the scalar and vector bosons in the spontaneously broken phase. We review some of the main results from these papers in Sections 2 and 3.
Although all of our new results pertain to the Coleman-Weinberg model, we devote Section 3 to the Abelian Higgs model since it illustrates a number of aspects of gaugeinvariance not present when m = 0. We try to provide a useful summary of this model since the literature is scattered and somewhat inconsistent. Also, issues which are not present in the Coleman-Weinberg model, such as unphysical gauge-dependent minimum of the effective potential and the presence of an additional dimensionful scale, provide a broader picture of some subtleties associated with gauge-invariance.
The main result of this paper concerns the extrema of the Coleman-Weinberg effective potential. We demonstrate that the minimum of the renormalization-group improved potential in this model is gauge-invariant order-by-order in perturbation theory. To show this requires first an understanding of the appropriate perturbation expansion. Perturbation theory must respect the relation λ ∼ glyph[planckover2pi1] e 4 which determines the minimum at 1-loop (see Section 4). We calculate all the 2-loop contributions in R ξ gauges proportional to glyph[planckover2pi1] 2 e 6 (Section 5), as well as all the terms like glyph[planckover2pi1] n +2 e 6 e 4 n λ n coming from 3-loop and higher-order graphs (Section 6) which also contribute at order glyph[planckover2pi1] 2 e 6 when λ ∼ glyph[planckover2pi1] e 4 . Demonstrating gauge-invariance in perturbation theory also requires careful consideration of what dimensionful scale the potential at the minimum should be expressed in terms of. The most obvious dimensionful scale v = 〈 φ 〉 is a questionable candidate since it is gauge-dependent itself. These issues are discussed in Section 7, where we show explicitly that the instability scale Λ I defined so that V (Λ ) = 0 I depends on ξ .
Next we augment our perturbative calculation with a proposal for how perturbation theory in which gauge-invariance holds order-by-order can be used in a resummed effective
potential. The challenge is that, first, resummation breaks the strict λ ∼ glyph[planckover2pi1] e 4 power-counting required for gauge-invariance and, second, that the renormalization-group kernel for the field strength renormalization, Γ = ∫ γ d ln µ has gauge-dependence to all orders in any expansion. Our solution exploits the invariance of physical quantities in the effective potential to the scale where the potential is calculated: we simply run the couplings towards the minimum before calculating the potential in perturbation theory. Then the field strength renormalization plays no direct role.
Although the potential at the minimum is gauge invariant, field values are gauge-dependent. For example, the expectation value of the field, v = 〈 φ 〉 , depends on ξ (cf. Eq. (3.19)) as does the instabilitiy scale Λ , defined as the field value where I V (Λ ) = 0 (cf. I Eq. (7.10)). Although the gauge-dependence of field values is often emphasized [15,17-19] there are still a surprising number of papers which try to ascribe physical significance to values of φ . Sometimes the physical question one wants to answer is akin to when higher-dimension operators, for example from Planck scale physics, can significantly affect the potential. We present one gauge-invariant way to deal with these higher-dimension operators in Section 9. Our conclusions and implications of our work are discussed in Section 10.
## 2 Effective actions
We begin with a rapid introduction to effective actions and their gauge dependence. In this section, φ generically denotes all the fields in a given theory (in the rest of the paper φ will denote a specific field in scalar QED).
A very useful object in quantum field theory is the 1PI effective action Γ[ φ ]. When φ is independent of x , the effective action reduces to the effective potential
$$\Gamma [ \phi ] = - \int d ^ { 4 } x V ( \phi ) + \text{derivative and nonlocal terms}$$
glyph[negationslash]
The interactions in the effective action embody a sum over all of the off-shell 1PI graphs in the quantum theory. Since any graph can be produced by sewing together 1P1 graphs, the effective action when used at tree-level can reproduce all the loops in the original theory. Thus the (quantum) effective action, when used classically, reproduces the full quantum physics of a given classical theory. In compensation for this enormous power, one must sacrifice manifest locality: in general, Γ[ φ ] = ∫ d x 4 L [ φ x ( )] for any L . Knowing the effective action amounts to solving a theory. As this is generally impossible, one must resort to an approximation to the effective action and argue that this approximation is valid for the computation of interest.
Although one could compute the 1PI effective action by computing all 1PI graphs in a theory, this is generally inefficient. Instead, one often computes the effective action by relating it to the Legendre transform of the classical action and doing a background field expansion [5]. We first define the functional
$$W [ J ] \equiv - i \ln Z [ J ] = - i \ln \left \lceil \int \mathcal { D } \phi e ^ { i S [ \phi ] + i \int d ^ { 4 } x \phi J } \right \rceil$$
where J x ( ) is an external current. The effective action is the Legendre transform of W J [ ]:
$$\Gamma [ \phi ] = W [ J _ { \phi } ] - \int d ^ { 4 } x J _ { \phi } \phi$$
Here J φ denotes the value of the external current which would make a given φ solve the equations of motion. Equivalently, in the presence of a current J φ , the expectation value of the field is φ J :
$$\phi _ { J } = \langle J _ { \phi } | \phi ( x ) | J _ { \phi } \rangle = - i \frac { 1 } { Z } \frac { \partial Z [ J ] } { \partial J ( x ) } \Big | _ { J = J _ { \phi _ { J } } } = \frac { \partial W [ J ] } { \partial J ( x ) } \Big | _ { J = J _ { \phi _ { J } } }$$
In this equation J φ is the independent variable which determines φ J as the expectation value of the field in its presence. For example, when J = 0, so that no current is turned on, then φ 0 = 〈 φ 〉 is the true vacuum in the theory. The conjugate relation is
$$\frac { \partial \Gamma [ \phi ] } { \partial \phi } \Big | _ { \phi = \phi _ { J } } = - J _ { \phi }$$
Here φ J is the independent variable to be chosen and used to compute J φ . Importantly, this equation implies that the true vacuum of the theory, φ 0 , is an extremum of the effective action. There is an assumed one-to-one correspondence between φ J and J φ , which is why the effective potential should always be convex. 1
The effective action is to be used classically, so that fields are always on-shell, satisfying their equations of motion. When φ is not an extremum of Γ, it represents the expectation value of the field in the presence of a nonzero background current; it is the classical solution to a different system. In other words, Γ[ φ ] encodes how the system responds to external currents for which Eq. (2.4) is satisfied. That this Legendre transform produces the object whose vertices are 1PI graphs in the original theory is not obvious. It is also not obvious that Γ[ φ ] can be computed by shifting the fields φ → φ + φ q and computing 1PI graphs involving φ q with φ held fixed. These three representations of Γ[ φ ] are explained in more detail in [4, 22, 23].
## 2.1 Gauge dependence
The main issue of concern in this paper is the gauge-dependence of the effective action. The gauge-dependence arises because when the source J is nonzero, the system has a nondynamical background charge density. Since there is no corresponding coupling of this charge density to the photon, the Ward identity will consequently be violated [5]. In the special case when J = 0, there is no charged background. Then φ 0 = 0 〈 | φ x ( ) 0 | 〉 is extremal and we have simply
$$\Gamma [ \phi _ { 0 } ] = W [ 0 ] = - i \ln Z [ 0 ] = - i \ln \int \mathcal { D } \phi e ^ { i S [ \phi ] }$$
1 The convexity issue is a thorny one [20,21], as the effective potential in the Standard Model appears to be perturbatively nonconvex. Conveniently, all the effective potentials we consider in this paper are convex.
This expression is gauge-invariant, since gauge-fixing modifies Z [0] only by an infinite constant, which drops out of ln Z [0] (see [16] for a longer discussion with a careful treatment of the divergences in Z [0]). Thus when ∂ Γ[ φ ] ∂φ = 0, then the effective action is gauge invariant by Eq. (2.5).
An alternative way to show gauge invariance at the extremum is by exploring how and when Ward identities are violated in the presence of a background current [15, 16]. In particular, Ref. [15] demonstrates that in Fermi gauges, where the gauge-fixing term is -1 2 ξ ( ∂ A µ µ ) 2 , the effective potential satisfies
$$\left [ \xi \frac { \partial } { \partial \xi } + C ( \phi, \xi ) \frac { \partial } { \partial \phi } \right ] V ( \phi, \xi ) = 0$$
where C φ, ξ ( ) is some precisely well-defined object, independently calculable order-by-order in perturbation theory. What this equation implies is that the gauge dependence of the effective potential can be compensated for by a rescaling of φ . It also apparently implies that at an extremum, where ∂ ∂φ V ( φ, ξ ) = 0, the effective potential is automatically gauge invariant, ∂ ∂ξ V ( φ, ξ ) = 0, consistent with the simple argument in the previous paragraph.
There are a number of unsettling features of the Nielsen identity. First of all, it is somewhat trivial, in the sense that one can always define a function C φ, ξ ( ) so that Eq. (2.7) holds; one simply defines C φ, ξ ( ) by Eq. (2.7). Indeed, this is a shortcut to calculating C φ, ξ ( ). This shortcut makes it clear that the Nielsen identity only guarantees that ∂ ∂ξ V ( φ, ξ ) = 0 at an extremum if C φ, ξ ( ) is finite at the extremum. However, as Nielsen himself observes [15], although C φ, ξ ( ) being infinite is 'unlikely for ξ finite' [p. 180], it is also true that ' C φ, ξ ( ) diverges in the one-loop approximation of the [Abelian] Higgs model' [p. 182]. In fact, the potential at an extremum can be gauge-invariant with C φ, ξ ( ) finite or infinite. It is also possible for C φ, ξ ( ) to be infinite at the extremum to all orders in perturbation theory but finite non-perturbatively.
For a concrete example, as we will see in Section 7, in the Coleman-Weinberg model C φ, ξ ( ) is logarithmically divergent as φ approaches an extremum when the potential is expressed in terms of the expectation value v = 〈 φ 〉 . The origin of this divergence is that v itself is gauge-dependent. If instead one expresses V in terms of the MS scale µ , then C φ, ξ ( ) is finite at the minimum. Thus, as we show in this paper, the value of the potential at the minimum is gauge-invariant order-by-order only in a carefully considered perturbation expansion. We do not find the Nielsen identity, which is a non-perturbative statement, particularly helpful in directing this expansion.
## 2.2 Rescaling and calculation scale invariance
Another feature of the potential at the minimum is that it is independent of field rescaling: φ → κφ for any κ . This is not a deep observation, as the value of any function at a minimum is invariant under rescaling of the dependent variable (see Fig. 1). More generally, the value of the potential at an extremum is independent of any field redefinition, φ → φ ′ ( φ ). These are intuitive features of quantum field theories, since the path integral is also invariant under
Figure 1: Under the rescaling of the dependent variable, a function changes but its values at extrema do not [19]. This elementary mathematical fact explains why the effective potential can depend on the field normalization, but V min does not.
field redefinitions, and the effective action in the true vacuum is defined as a path integral in Eq. (2.6). Indeed, gauge invariance of the action at extrema is a special case of general field redefinition invariance, since one can view gauge-transformations as field redefinitions. In doing so, however, one must allow for the possibility that if with one definition an extremum is at A µ = 0, with another definition it may be with a nonzero and x -dependent expectation value for A µ .
A corollary of the above argument is that the potential away from its extrema does depend on how the field is normalized and defined. This is also obvious from Fig. 1. Away from an extremum, the action describes the system in the presence of a current J . When one rescales the field φ , the Jφ term with J fixed breaks the invariance of the path integral under rescaling. Equivalently, from Eq. (2.5), we see that J κφ = 1 κ J φ so that when a field is rescaled, Γ[ κφ ] gives the least action in the presence of a rescaled current.
A number of authors have proposed that the gauge-dependence of the effective potential can be removed through a field redefinition [8-11,14]. For example, Tye and Vtorov-Karevsky argue that one should replace φ 1 + iφ 2 → σ exp( iπ ) [10]. Then σ is a U(1) singlet and so its source J is neutral and the interaction Jσ in the Lagrangian does not cause the Ward identity to be violated. Although there is nothing wrong with this argument, physical quantities, such as the value of the potential at its minimum, should be independent of field redefinitions. A field redefinition is in a sense similar to a gauge-choice. It does not make the potential away from the minimum any more physical. Moreover, with this nonlinear field redefinition, a renormalizable theory becomes nonrenormalizable and nominally straightforward calculations can become drastically more complicated (try computing β λ at 1-loop in this theory). The point is that physics should be independent of field redefinitions, so one should choose a basis which makes calculations easiest, not one that makes unphysical quantities more comforting.
An interesting and underappreciated fact about the effective potential is that its form depends on the scale where it is computed. The effective potential satisfies an RGE:
$$\left ( \mu \frac { \partial } { \partial \mu } - \gamma _ { i } \phi _ { i } \frac { \partial } { \partial \phi _ { i } } + \beta _ { i } \frac { \partial } { \partial \lambda _ { i } } \right ) V = 0$$
This equation says that the explicit dependence on RG scale µ in the effective potential can be compensated by redefining the couplings λ i , through their β functions, and rescaling the fields φ i based on their anomalous dimensions γ i . One can compute the potential to fixed order at a scale µ 0 , in terms of λ i ( µ 0 ) and µ 0 and then evolve to some other scale µ by solving this equation simultaneously with the RGEs for the coupling constants. Call this method 1. Alternatively, one could evolve the couplings to µ and then compute the effective potential there. Call this method 2. The potentials at µ computed these two different ways will not agree: they differ because in method 1, the rescaling of φ i from µ 0 to µ is included, while in method 2, it is not. We show through an explicit example in Section 8.1 that in the Coleman-Weinberg model, the two methods give potentials which indeed differ by ln µ µ 0 terms at next-to-leading order.
One can easily write a differential equation similar to Eq. (2.7) for the scale-dependence:
$$\left [ \mu _ { 0 } \frac { \partial } { \partial \mu _ { 0 } } - \gamma _ { i } \phi _ { i } \frac { \partial } { \partial \phi _ { i } } \right ] V ( \phi, \mu _ { 0 }, \mu ) = 0$$
where µ 0 is the scale at which the potential is calculated before it is run to the scale µ by solving Eq. (2.8). In a sense, this equation although trivial to derive is more useful than the Nielsen identity, since γ i cannot be infinite (unlike C φ, ξ ( ) in Eq. (2.7)). It immediately indicates that at an extremum, the effective potential is µ 0 -independent. We use this calculation-scale invariance in Section 8.3 to show how a resummed effective potential can be computed so that it is gauge-invariant order-by-order in a particular perturbation expansion.
For another perspective, recall that the vertices of the effective action encode (amputated) 1PI Green's functions. General amputated Green's functions satsify an RGE identical to Eq. (2.8), including the γ term. Thus they also depend on the scale where they are calculated. In converting such Green's functions to S -matrix elements, the LSZ reduction theorem instructs us to multiply by ∏ i √ Z i , where √ Z i is the on-shell wave-function renormalization factor for external leg i . In MS, these Z -factors are not 1 but are gauge and scale-dependent. Indeed, their µ -dependence explicitly cancels the γ term in Eq. (2.8) so that all scale dependence is S -matrix elements is compensated for by rescaling the couplings according to their β -functions. As the β -functions are gauge-invariant in MS, the S -matrix can then be gauge-invariant at any scale.
In summary, the effective potential at general field values depends on an arbitrary calculation scale µ 0 . This is different from the usual dependence on the renormalization scale µ arising from a truncation to fixed order in perturbation theory. The effective potential has that dependence as well, but it depends on µ 0 even non-perturbatively. The value of the potential at extrema is µ 0 -independent. That the µ 0 -dependence drops out at extrema does not require field redefinitions or multiplication by the wavefunction renormalization factors included in going from Green's functions to S -matrix elements.
## 3 Abelian Higgs Model
This paper focuses on scalar QED, a renormalizable theory with two scalar fields φ 1 and φ 2 , a photon, A µ and an O(2) symmetry. If the scalar doublet is given a negative mass-squared in the classical potential, the model is called the Abelian Higgs model. This section is a review of results in [6, 17, 24, 25] with some details filled in and a few added comments. Establishing gauge-invariance in this model is simpler than in the Coleman-Weinberg model for the essential reason that there is spontaneous symmetry breaking already in the classical potential. Consequently, a traditional loop expansion is justified and the first non-trivial check occurs at 1-loop. Nevertheless, the Abelian Higgs model presents interesting features and subtleties complementary to those in the Coleman-Weinberg model, which is why we discuss it here.
We write the Lagrangian as
$$\mathcal { L } = - \frac { 1 } { 4 } F _ { \mu \nu } ^ { 2 } + \frac { 1 } { 2 } ( \partial _ { \mu } \phi _ { 1 } - e A _ { \mu } \phi _ { 2 } ) ^ { 2 } + \frac { 1 } { 2 } ( \partial _ { \mu } \phi _ { 2 } + e A _ { \mu } \phi _ { 1 } ) ^ { 2 } - V _ { 0 } + \mathcal { L } _ { \text{GF} } + \mathcal { L } _ { \text{ghost} }$$
with the classical potential
$$V _ { 0 } ( \phi ) = - \frac { 1 } { 2 } m ^ { 2 } \phi ^ { 2 } + \frac { \lambda } { 2 4 } \phi ^ { 4 }$$
where
$$\phi ^ { 2 } \equiv \phi _ { 1 } ^ { 2 } + \phi _ { 2 } ^ { 2 }$$
Since the theory has a global O(2) symmetry rotating φ 1 and φ 2 , the effective potential should only depend on φ . When m > 2 0, the classical potential is minimized when
$$v _ { \text{cl} } \equiv \langle \phi _ { 1 } \rangle _ { \text{classical} } = \sqrt { \frac { 6 } { \lambda } } m$$
We have used global O(2) symmetry to put the vacuum expectation value (vev) of in φ 1 with 〈 φ 2 〉 = 0. Since we have normalized the vacuum energy so that V 0 (0) = 0, the value of the classical potential at the minimum is V 0 ( v cl ) = -3 2 λ m 4 = -λ 24 v 4 cl .
To compute quantum corrections to the vacuum energy, we need to fix a gauge. In gaugefixing, as always, one looks for a deformation of the theory with the fewest nasty features. To appreciate the relevant tradeoffs, it is helpful to use a three-parameter family ( ξ , Υ and 1 Υ ) of gauges [6]. 2 We write
$$\mathcal { L } _ { \text{GF} } = - \frac { 1 } { 2 \xi } ( \partial _ { \mu } A ^ { \mu } + e \xi \Upsilon _ { 1 } \phi _ { 1 } + e \xi \Upsilon _ { 2 } \phi _ { 2 } ) ^ { 2 }$$
The associated Fadeev-Popov ghost Lagrangian is
$$\mathcal { L } _ { \text{ghost} } = ( \partial _ { \mu } \bar { c } ) ( \partial _ { \mu } c ) - e ^ { 2 } \xi \Upsilon _ { 1 } \phi _ { 2 } \, \bar { c } c + e ^ { 2 } \xi \Upsilon _ { 2 } \phi _ { 1 } \, \bar { c } c$$
where the ghosts c and antighosts ¯ are Grassmann-valued scalar fields. c
For Υ i = 0, these are the Fermi gauges . Fermi gauges are appealing because they do not explicitly break the global O(2) symmetry rotating φ 1 and φ 2 and because the ghosts
decouple. A drawback of Fermi gauges is that there is uncanceled kinetic mixing between the gauge field and the scalars. A more serious problem with Fermi gauges is that they can lead to singular intermediate results, as we will see, which make some calculations difficult.
glyph[negationslash]
The R ξ gauges [26] are Υ 1 = v cl and Υ 2 = 0. Note that even in R ξ gauges, there is still kinetic mixing when φ = v cl and, except for Landau gauge, ξ = 0, the ghosts are not free. In the R ξ gauges, the global O(2) symmetry is explicitly broken. This breaking is not dangerous, since gauge-independent quantities must be independent of the gauge-fixing and should respect the O(2) symmetry. A more troubling fact about Υ 1 = v cl is that the classical minimum v cl is then hard-coded into the Lagrangian, when presumably only the true quantum minimum has any physical significance. It would be even more troubling to allow Υ i to be dynamical, for example with Υ 1 = φ ; such gauges lead S. Weinberg to declare R ξ gauges problematic in all but Landau gauge, ξ = 0 [7] (see [16] for a careful refutation of the argument in [7]). In the 3-parameter family we consider here, Υ i are arbitrary, so v cl does not appear.
Another feature of the Abelian Higgs model with this gauge fixing is that it has a spurious unphysical minimum. The gauge fixing induces a coupling between φ 1 and φ 2 in the classical potential. Thus there is a solution for constant fields with A µ = c = ¯ = 0 and c
$$\phi _ { i } = e \Upsilon _ { i } \sqrt { \frac { 6 m ^ { 2 } } { \lambda e ^ { 2 } ( \Upsilon _ { 1 } ^ { 2 } + \Upsilon _ { 2 } ^ { 2 } ) } + \frac { 6 \xi } { \lambda } }$$
The potential on this solution has the value
$$V _ { \min } = - \frac { 3 } { 2 \lambda } \left [ m ^ { 2 } + e ^ { 2 } \xi ( \Upsilon _ { 1 } ^ { 2 } + \Upsilon _ { 2 } ^ { 2 } ) \right ] ^ { 2 }$$
which is gauge-dependent and unphysical. This solution, found by Dolan and Jackiw [6], led them to argue for unitary gauge, ξ = ∞ . As explained in [16] (see also [25]), the trouble with this solution is not the gauge-dependence, but that, despite satisfying the Euler-Lagrange equations, it does not extremize the action. An extremal solution related to this one has a non-vanishing and position dependent expectation value for A µ ; the gauge-fixing term contributes at spatial infinity so the boundary terms cannot be dropped in deriving the Euler-Lagrange equations. In this sense the 3-parameter gauge fixing provides what Ref. [16] calls a bad gauge . Note that there is nothing inconsistent about bad gauges, they are just rather inconvenient for doing calculations because of the position-dependent extrema. If we stick to solutions close to the gauge-independent classical one, with φ 1 = v cl and φ 2 = 0, the position dependent solutions can be ignored.
We can always use the O(2) symmetry to keep 〈 φ 2 〉 = 0 to all orders. If we do so, then we do not even need to turn on a background field for φ 2 , since it will not contribute to the extremal solutions for φ 1 . To avoid the bad gauges discussed above, we should correspondingly take Υ 1 = 0. A more general O(2)-invariant condition is to only turn on the linear combination which has Υ i · 〈 φ i 〉 = 0. This is essentially what was done in [24]. Then,
glyph[negationslash]
the renormalized 1-loop effective potential in MS with m = 0 is
$$V _ { 1 } ( \phi ) = \frac { \hbar } { 1 6 \pi ^ { 2 } } \left [ \frac { 3 } { 4 } M _ { A } ^ { 4 } \left ( \ln \frac { M _ { A } ^ { 2 } } { \mu ^ { 2 } } - \frac { 5 } { 6 } \right ) + \frac { 1 } { 4 } M _ { B } ^ { 4 } \left ( \ln \frac { M _ { B } ^ { 2 } } { \mu ^ { 2 } } - \frac { 3 } { 2 } \right ) \\ - \frac { 1 } { 2 } M _ { G } ^ { 4 } \left ( \frac { M _ { G } ^ { 2 } } { \mu ^ { 2 } } - \frac { 3 } { 2 } \right ) + \frac { 1 } { 4 } M _ { + } ^ { 4 } \left ( \ln \frac { M _ { + } ^ { 2 } } { \mu ^ { 2 } } - \frac { 3 } { 2 } \right ) + \frac { 1 } { 4 } M _ { - } ^ { 4 } \left ( \ln \frac { M _ { - } ^ { 2 } } { \mu ^ { 2 } } - \frac { 3 } { 2 } \right ) \right ] \ \ ( 3. 9 )$$
where
$$M _ { A } ^ { 2 } = e ^ { 2 } \phi ^ { 2 }$$
$$M _ { B } ^ { 2 } = \frac { \lambda } { 2 } \phi ^ { 2 } - m ^ { 2 } \\ \Phi _ { \Phi } = \Phi _ { \Phi } \Phi _ { \Phi } + \Phi _ { \Phi } \Phi _ { \Phi } \Phi _ { \Phi } \Phi _ { \Phi } \Phi _ { \Phi } \Phi _ { \Phi } \Phi _ { \Phi } \Phi _ { \Phi } \Phi _ { \Phi } \Phi _ { \Phi } \Phi _ { \Phi } \Phi _ { \Phi } \Phi _ { \Phi } \Phi _ { \Phi } \Phi _ { \Phi } \Phi _ { \Phi } \Phi _ { \Phi } \Phi _ { \Phi } \Phi _ { \Phi } \Phi _ { \Phi } \Phi _ { \Phi } \Phi _ { \Phi } \Phi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi_ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ {\Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phis } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi } \Psi _ { \Phi }$$
$$M _ { G } ^ { 2 } = \xi e ^ { 2 } ( \Upsilon _ { 1 } \phi _ { 2 } - \Upsilon _ { 2 } \phi _ { 1 } )$$
are the contributions from A µ , φ and ghosts, respectively and
$$M _ { \pm } ^ { 2 } = \frac { \lambda } { 1 2 } ( \phi ^ { 2 } - v _ { c l } ^ { 2 } ) & + \xi e ^ { 2 } ( \Upsilon _ { 1 } \phi _ { 2 } - \Upsilon _ { 2 } \phi _ { 1 } ) \\ & \pm \frac { 1 } { 2 } \sqrt { \frac { \lambda } { 6 } } \left [ \phi ^ { 2 } - v _ { c l } ^ { 2 } \right ] \left [ \frac { \lambda } { 6 } ( \phi ^ { 2 } - v _ { c l } ^ { 2 } ) - 4 \xi e ^ { 2 } \phi ^ { 2 } + 4 \xi e ^ { 2 } ( \Upsilon _ { 1 } \phi _ { 2 } - \Upsilon _ { 2 } \phi _ { 1 } ) \right ] } \quad ( 3. 1 3 )$$
come from the kinetic mixing.
Then, assuming that the vev of φ 1 gets a perturbative correction v = 〈 φ 1 〉 = v cl + v 1 + · · · for some v 1 of order glyph[planckover2pi1] , we can evaluate the potential at the minimum perturbatively:
$$V _ { \min } = V _ { 0 } ( v ) + V _ { 1 } ( v ) + \cdots$$
$$= V _ { 0 } ( v _ { \text{cl} } ) + v _ { 1 } V _ { 0 } ^ { \prime } ( v _ { \text{cl} } ) + V _ { 1 } ( v _ { \text{cl} } ) + \cdots$$
$$= V _ { 0 } ( v _ { \text{cl} } ) + V _ { 1 } ( v _ { \text{cl} } ) + \dots + \mathcal { O } ( \hbar { ^ { 2 } } )$$
where the omitted terms are O ( glyph[planckover2pi1] 2 ) and V ′ 0 ( v cl ) = 0 has been used. The importance of doing an analytic expansion in glyph[planckover2pi1] with an appropriate truncation before evaluating V min or other physical quantities was emphasized by Patel and Ramsey-Musolf in [19] (see also [17]). Since M 2 + = M 2 -= M 2 G = -ξe 2 Υ 2 v cl when φ 1 = v cl and φ 2 = 0, we find that V ( v ) gauge-invariant to order glyph[planckover2pi1] . Explicitly,
$$V _ { \min } = v _ { \text{cl} } ^ { 4 } \left [ - \frac { \lambda } { 2 4 } + \frac { \hbar } { 1 6 \pi ^ { 2 } } \left ( - \frac { 5 } { 8 } e ^ { 4 } - \frac { 1 } { 2 4 } \lambda ^ { 2 } + \frac { 1 } { 3 6 } \lambda ^ { 2 } \ln \frac { \lambda v _ { \text{cl} } ^ { 2 } } { 3 \mu ^ { 2 } } + \frac { 3 } { 4 } e ^ { 4 } \ln \frac { e ^ { 2 } v _ { \text{cl} } ^ { 2 } } { \mu ^ { 2 } } \right ) + \mathcal { O } ( \hbar { 2 } ^ { 2 } ) \right ] \quad ( 3. 1 7 )$$
Thus the vacuum energy is manifestly gauge-independent.
To determine v 1 , we note that the minimum condition is
$$0 = V _ { 0 } ^ { \prime } ( v ) + V _ { 1 } ^ { \prime } ( v ) + \dots = V _ { 0 } ^ { \prime } ( v _ { c l } ) + v _ { 1 } V _ { 0 } ^ { \prime \prime } ( v _ { c l } ) + V _ { 1 } ^ { \prime } ( v _ { c l } ) + \dots$$
using again that V ′ 0 ( v cl ) = 0, we get v 1 = -V ′ 1 ( v cl ) V ′′ 0 ( v cl ) which gives
$$v _ { 1 } = v _ { \text{cl} } \frac { \hbar } { 1 6 \pi ^ { 2 } } \left [ \frac { 3 e ^ { 4 } } { \lambda } + \frac { \lambda } { 2 } - \frac { 9 e ^ { 4 } } { \lambda } \ln \frac { e ^ { 2 } v _ { \text{cl} } ^ { 2 } } { \mu ^ { 2 } } - \frac { \lambda } { 2 } \ln \frac { v _ { \text{cl} } ^ { 2 } \lambda } { 3 \mu ^ { 2 } } + \frac { e ^ { 2 } \xi } { 2 } \ln \frac { - e ^ { 2 } v _ { \text{cl} } \xi \Upsilon _ { 2 } } { \mu ^ { 2 } } \right ] \\ & + \Upsilon _ { 2 } e ^ { 2 } \xi \frac { \hbar } { 1 6 \pi ^ { 2 } } \left [ - \frac { 1 } { 2 } + \ln \frac { - e ^ { 2 } \xi v _ { \text{cl} } \Upsilon _ { 2 } } { \mu ^ { 2 } } \right ] \ \ ( 3. 1 9 )$$
This expression manifestly depends on both ξ and Υ . Dangers of assigning physical signif2 icance to 〈 φ 〉 have been emphasized in [17,19] and elsewhere.
Note that v 1 is singular in Fermi gauges, Υ i → 0, but not in Landau gauge ξ → 0. The singularity is an infrared divergence associated with the ghosts becoming massless. As pointed out in [15] and [24], the function C φ, ξ ( ) in the Nielsen identity also has an infrared divergence in Fermi gauge. As discussed in Section 2.1, that C φ, ξ ( ) can be infinite calls into question the usefulness of the Nielsen identity - as we have seen, the potential at the minimum is gauge-invariant for any choice of Υ 2 and ξ .
At this point, we can conclude that everything works swimmingly: the vacuum energy is gauge invariant but the vev is not, exactly as foretold by the Nielsen identity. At the risk of being too cautious, it is perhaps worth adding that in the Abelian Higgs model, V min is not actually observable - the true renormalized vacuum energy has an arbitrary subtraction associated with the cosmological constant. Even if the vacuum energy can be measured (for example, by weakly coupling this model to gravity), one still needs a scheme to measure the difference ∆ V = V min -V (0) between this energy and the energy at φ = 0. This difference, as a function of v cl , e and λ is what we have calculated. Even then, although ∆ V is independent of gauge, it is expressed in terms of v cl = √ 6 λ m which is itself unphysical. Indeed, even m is not physical: it is not the mass of the scalar in the vacuum. That mass, which we denote m S , itself gets corrections of order glyph[planckover2pi1] from self-energy graphs. The gauge-dependent part of these corrections at order e λ 2 have been calculated in [24], with the result that d dξ m S = 0. Therefore ∆ V/m 4 S is gauge independent and only a function of the MS couplings e and λ . This ratio is in principle observable, if ∆ V is, and depends only on other observables, e and λ . With this long-winded disclaimer, we can now conclude that quantum corrections in effective potential can make (in principle) testable experimental predictions.
In summary, in the Abelian Higgs model both ∆ V = V min -V (0) and m S are separately independent of the gauge parameters ξ and Υ 2 at order glyph[planckover2pi1] when expressed in terms of the classical expectation value of φ , v cl . That ∆ V and m S are separately gauge invariant is not strictly necessary. All that is required is that the gauge-dependence cancels in the ratio ∆ V/m 4 S which is a calculable function of the observables e and λ . As we will see, in the Coleman-Weinberg model, where the classical theory is scaleless, such considerations are critical. We also found that the expectation value of the field v = 〈 φ 〉 is not gauge invariant at order glyph[planckover2pi1] . In the Fermi gauges, Υ i = 0, v even has an infrared divergence.
## 4 Coleman-Weinberg Model
Now let us turn to the main subject of this paper, scalar QED with a massless scalar, also known as the Coleman-Weinberg model. The Lagrangian is as in Eq. (3.1), with
$$V _ { 0 } = \frac { \lambda } { 2 4 } \phi ^ { 4 }$$
and the gauge-fixing term is
$$\lambda _ { \text{GF} } = - \frac { 1 } { 2 \xi } ( \partial _ { \mu } A ^ { \mu } ) ^ { 2 }$$
These can be thought of as R ξ gauges (since v cl = 0 in this theory) or Fermi gauges. They correspond to Υ 1 = Υ 2 = 0 in the 3-parameter gauge family discussed above, and so the ghosts decouple and can be ignored. There is unavoidable kinetic mixing between φ and ∂ A µ µ in this theory, but as in the Abelian Higgs model, this is an inconvenient but not insurmountable complication.
The renormalized 1-loop effective potential in MS in this theory is the m → 0 limit of Eq. (3.9):
$$V _ { 1 } ( \phi ) = \phi ^ { 4 } \frac { \hbar } { 1 6 \pi ^ { 2 } } \left [ \frac { 3 } { 4 } e ^ { 4 } & \left ( \ln \frac { e ^ { 2 } \phi ^ { 2 } } { \mu ^ { 2 } } - \frac { 5 } { 6 } \right ) + \frac { \lambda ^ { 2 } } { 1 6 } \left ( \ln \frac { \lambda \phi ^ { 2 } } { 2 \mu ^ { 2 } } - \frac { 3 } { 2 } \right ) \\ & + \left ( \frac { \lambda ^ { 2 } } { 1 4 4 } - \frac { 1 } { 1 2 } e ^ { 2 } \lambda \xi \right ) \left ( \ln \frac { \phi ^ { 2 } } { \mu ^ { 2 } } - \frac { 3 } { 2 } \right ) + \frac { 1 } { 4 } K _ { + } ^ { 4 } \ln K _ { + } ^ { 2 } + \frac { 1 } { 4 } K _ { - } ^ { 4 } \ln K _ { - } ^ { 2 } \right ] \ \ ( 4. 3 ) \\ \dots$$
with
$$K _ { \pm } ^ { 2 } = \frac { 1 } { 1 2 } \left ( \lambda \pm \sqrt { \lambda ^ { 2 } - 2 4 \lambda e ^ { 2 } \xi } \right )$$
The relation
$$\tilde { \Phi } _ { + } ^ { 4 } + K _ { - } ^ { 4 } = \frac { \lambda ^ { 2 } } { 3 6 } - \frac { 1 } { 3 } e ^ { 2 } \lambda \xi$$
has been used to simplify the 1-loop potential.
The tree-level potential in this model has only a single minimum, at φ = 0, where the O(2) symmetry is unbroken. For there to be a minimum at 1-loop, the corrections must be large enough to turn over the potential. For λ small, so that the theory is perturbative, this is only possible if the glyph[planckover2pi1] 16 π 2 e 4 term is as large as the tree-level λ 24 piece. So let us assume λ ∼ glyph[planckover2pi1] 16 π 2 e 4 and that there is a minimum at some scale v . Then the condition for the minimum, V ′ ( v ) = 0 provides a precise relationship between λ and e :
$$\lambda = \frac { \hbar } { 1 6 \pi ^ { 2 } } e ^ { 4 } \left ( 6 - 3 6 \ln \frac { e v } { \mu } \right ) + \mathcal { O } ( e ^ { 6 } )$$
As discussed in [4], to understand this equation one must appreciate dimensional transmutation. In scalar QED, the only scale is the scale µ at which the couplings are defined. µ is arbitrary, so we may as well take µ = v . Then Eq. (4.6) reduces to
$$\lambda = \frac { \hbar } { 1 6 \pi ^ { 2 } } e ^ { 4 } \left ( 6 - 3 6 \ln e \right ) + \mathcal { O } ( e ^ { 6 } )$$
This equation should be thought of as a condition on λ = λ µ ( ) and e = ( ): the minimum e µ occurs at the scale v = µ where Eq. (4.7) holds. Of course, v will get corrections and, as we will see, is gauge-dependent (unlike µ ). But at least at 1-loop, this is an acceptable way to think about the minimum in the effective potential in scalar QED.
Since λ and e can be anything, it is natural to wonder whether Eq. (4.7) requires some kind of finite tuning. As explained in [4] it does not. The evolution of e and λ are determined by the β functions:
$$\mu \frac { d } { d \mu } e = \beta _ { e }, \quad \mu \frac { d } { d \mu } \lambda = \beta _ { \lambda },$$
where, at 1-loop,
$$\beta _ { e } = \frac { \hbar } { 1 6 \pi ^ { 2 } } \left ( \frac { e ^ { 3 } } { 3 } \right ) + \cdots$$
$$\beta _ { \lambda } = \frac { \hbar { \ } } { 1 6 \pi ^ { 2 } } \left ( \stackrel { \cdot } { 3 6 e ^ { 4 } - 1 2 e ^ { 2 } \lambda } + \frac { 1 0 \lambda ^ { 2 } } { 3 } \right )$$
The key feature of these equations is that e is multiplicatively renormalized ( e = 0 is a fixed point of the RG flow) while λ can get an additive correction even at λ = 0. What this means is that if λ and e start off small, e runs logarithmically, but λ will grow at ever increasing rate until it hits a Landau pole. Indeed, the exact solutions to the 1-loop RGEs are [4]
$$e ^ { 2 } ( \mu ) = \frac { e ^ { 2 } ( \mu _ { 0 } ) } { 1 - \frac { e ^ { 2 } ( \mu _ { 0 } ) } { 2 4 \pi ^ { 2 } } \ln \frac { \mu } { \mu _ { 0 } } }$$
which expresses e µ ( ) in terms of e at some reference scale µ 0 and
$$\lambda ( \mu ) = \frac { e ^ { 2 } ( \mu ) } { 1 0 } \left [ 1 9 + \sqrt { 7 1 9 } \tan \left ( \frac { \sqrt { 7 1 9 } } { 2 } \ln \frac { e ( \mu ) ^ { 2 } } { C } \right ) \right ]$$
with C an integration constant which can be traded for λ µ ( 0 ). The tangent in λ µ ( ) implies that as e 2 changes by a factor of ≈ 1 2, . λ will go from -∞ to ∞ . In particular, if λ and e are small, there will always be a point where Eq. (4.7) is satisfied.
At the minimum, we find
$$V _ { \min } = v ^ { 4 } \frac { \hbar } { 1 6 \pi ^ { 2 } } e ^ { 4 } \left ( - \frac { 3 } { 8 } \right ) + \mathcal { O } ( e ^ { 6 } )$$
This is gauge-invariant, simply because the e 4 terms in V 1 ( φ ) are gauge-invariant.
The first non-trivial check on the gauge invariance of V ( v ) requires the terms in the effective potential of order glyph[planckover2pi1] 2 e 6 , with λ counting as order glyph[planckover2pi1] e 4 , and ln e and ln λ counting as order e 0 . In scalar QED, each loop comes with a factor of glyph[planckover2pi1] e 2 or glyph[planckover2pi1] λ from the vertices, so glyph[planckover2pi1] e λ 2 terms come from 1-loop graphs and glyph[planckover2pi1] 2 e 6 terms from 2-loop graphs. Thus we need at least the 2-loop Coleman-Weinberg potential. In addition, effective potential calculations involve
an infinite number of background field insertions, which can be conveniently resummed into dressed propagators with e - and λ -dependent masses. These propagators allow for graphs to have extra factors of e /λ 2 in them. For example, a 3-loop graph gives a term proportional to glyph[planckover2pi1] 3 e 10 /λ which also scales like glyph[planckover2pi1] 2 e 6 when λ ∼ glyph[planckover2pi1] e 4 . These higher-loop terms are discussed in Section 6.
## 5 The two-loop Coleman-Weinberg potential
The complete 2-loop potential in scalar QED with gauge-dependence does not appear in the literature, as far as we are aware. The 2-loop potential in Landau gauge is known in the complete Standard Model [27] and for general renormalizable theories [28]. Kang computed the terms in the 2-loop potential at order e 6 that depend on ln φ 2 µ 2 , but with a subtraction scheme slightly different from MS [29]. For our analysis, we require also the µ -independent part of the effective potential at order e 6 . We will also need an infinite series of other graphs, the daisy graphs, which are discussed in Section 6, but let us start with the 2-loop computation. To compute the effective potential at this order, there is a counterterm contribution and a contribution from 2-loop graphs. We work in d = 4 -2 ε dimensions throughout, and the scale µ is the MS renormalization scale with implicitly contains the ln 4 π and γ E factors.
## 5.1 Counterterm contribution
The counterterm contribution can be extracted from the O ( ε ) part of the 1-loop potential multiplying the 1 ε poles in the renormalization factors. In d = 4 -2 ε dimensions, the 1-loop potential can be written as
$$V _ { 1 } = \phi ^ { 4 } \left [ ( 3 - 2 \varepsilon ) f ( e ^ { 2 } ) + f ( \frac { \lambda } { 2 } ) + f ( K _ { + } ^ { 2 } ) + f ( K _ { - } ^ { 2 } ) \right ] + \mathcal { O } ( \varepsilon ^ { 2 } )$$
where
$$f ( x ) = - \frac { x ^ { 2 } } { 6 4 \pi ^ { 2 } } \left [ \frac { 1 } { \varepsilon } + \frac { 3 } { 2 } - \ln \frac { x \phi ^ { 2 } } { \mu ^ { 2 } } + \frac { \varepsilon } { 2 } \left ( \ln \frac { x \phi ^ { 2 } } { \mu ^ { 2 } } - \frac { 3 } { 2 } \right ) ^ { 2 } + \varepsilon \left ( \frac { 5 } { 8 } + \frac { \pi ^ { 2 } } { 1 2 } \right ) \right ]$$
with K ± defined in Eq. (4.4). The 1-loop renormalization factors are
$$\begin{array} { c c c } \bullet \, \cdot \, & \, \cdot \\ Z _ { \phi } & = 1 + \frac { \hbar } { 1 6 \pi ^ { 2 } } \frac { 1 } { \varepsilon } ( 3 - \xi ) e ^ { 2 } + \cdots, \\ Z _ { e } & = 1 + \frac { \hbar } { 1 6 \pi ^ { 2 } } \frac { 1 } { \varepsilon } \frac { e ^ { 2 } } { 6 } + \cdots \\ \lambda Z _ { \lambda } & = \lambda + \frac { \hbar } { 1 6 \pi ^ { 2 } } \frac { 1 } { \varepsilon } \left ( 1 8 e ^ { 4 } - 6 e ^ { 2 } \lambda + \frac { 5 } { 3 } \lambda ^ { 2 } \right ) + \cdots \end{array}$$
Replacing e → Z e e , λ → Z λ λ and φ → √ Z φ φ and expanding to order glyph[planckover2pi1] 2 ε 0 gives the counterterm contribution to the 2-loop potential:
$$V _ { 2 } ^ { c t } = \left ( \frac { \hbar } { 1 6 \pi ^ { 2 } } \right ) ^ { 2 } \phi ^ { 4 } e ^ { 6 } \left [ ( - 1 0 + 6 \xi ) \ln ^ { 2 } \frac { e \phi } { \mu } + \left ( \frac { 1 0 } { 3 } - 4 \xi + \frac { 3 } { 2 } \xi \ln \frac { \lambda \xi } { 6 e ^ { 2 } } \right ) \ln \frac { e \phi } { \mu } \\ - \frac { 5 } { 3 } - \frac { 5 \pi ^ { 2 } } { 1 2 } + 2 \xi + \frac { \pi ^ { 2 } } { 1 6 } \xi - \frac { 3 } { 4 } \xi \ln \frac { \lambda \xi } { 6 e ^ { 2 } } + \frac { 3 } { 1 6 } \ln ^ { 2 } \frac { \lambda \xi } { 6 e ^ { 2 } } \right ] + \mathcal { O } ( e ^ { 4 } \lambda, e ^ { 2 } \lambda ^ { 2 }, \lambda ^ { 3 } ) \ \ ( 5. 4 )$$
## 5.2 2-loop contribution
The 2-loop contribution to the next-to-next-to-leading order effective potential can be computed from Feynman diagrams with dressed propagators (called prototype diagrams in [4]). The most straightforward way to compute the effective potential is to turn on a background field for φ 1 by replacing φ 1 → φ 1 + φ in the classical Lagrangian. After substituting this into the classical Lagrangian and dropping the terms linear in propagating fields φ 1 , φ 2 and A µ , we compute the propagators by diagonalizing the kinetic terms, treating the background field φ as constant. The resulting propagators are:
$$1 \, \cfrac { 1 } { 2 } \, \cfrac { 1 } { 2 } \, \phi ^ { 2 } } _ { \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \, } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 2 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \cdot \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \ } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 0 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, 1 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 3 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1. 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0. 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 1 } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 3 } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 4 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, \cfrac { 1 } { 2 } } \, 2 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \
\, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0 \, 0$$
2
$$2 \xrightarrow { 2 } \ = \ D _ { 2 2 } ( k ) = i \frac { k ^ { 2 } - e ^ { 2 } \xi \phi ^ { 2 } } { ( k ^ { 2 } - \phi ^ { 2 } K _ { + } ) ( k ^ { 2 } - \phi ^ { 2 } K _ { - } ) } \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
2
2
$$\mu \underset { \ } \sim \underset { \ } \sim \sum \nu \ = \ \Delta _ { \mu \nu } ( k ) = - i \left [ \frac { 1 } { k ^ { 2 } - e ^ { 2 } \phi ^ { 2 } } \left ( g _ { \mu \nu } - \frac { k _ { \mu } k _ { \nu } } { k ^ { 2 } } \right ) + \frac { \xi ( k ^ { 2 } - \frac { 1 } { 6 } \lambda \phi ^ { 2 } ) } { ( k ^ { 2 } - \phi ^ { 2 } K _ { + } ) ( k ^ { 2 } - \phi ^ { 2 } K _ { - } ) } \frac { k _ { \mu } k _ { \nu } } { k ^ { 2 } } \right ]$$
with K ± given in Eq. (4.4) and
$$\mu \underset { \longleftarrow } { \sim } 2 \ = \ T _ { \mu 2 } ( k ) = \frac { \xi e \phi } { ( k ^ { 2 } - \phi ^ { 2 } K _ { + } ) ( k ^ { 2 } - \phi ^ { 2 } K _ { - } ) } k _ { \mu }$$
This last propagator comes from the kinetic mixing between φ 2 and A µ in the presence of a background field for φ 1 . The asymmetry between D 11 and D 22 arises because we are only turning on a background for φ 1 for simplicity. The full effective potential of course also depends on background values for φ 2 . However, since the effective potential has a global O(2) symmetry, it is enough to turn on a φ 1 background for the questions we address in this paper.
The propagators in Eqs. (5.5) to (5.8) are effective propagators, where all of the flipping between scalars and longitudinal vector bosons that can occur between interactions has already been resummed. In this way, an infinite set of diagrams is replaced by a single one:
Diagrams where there is one flip between vertices are still allowed (for example, see Fig. 2 below). The hair on the left-hand-side diagrams indicates background field insertions which are implicit in prototype diagrams, even without kinetic mixing.
There are five types of interactions:
$$\underset { j } { \overset { i } { \underset { k } { \sim } } } ^ { l } = - i \frac { \lambda } { 3 } ( \delta _ { i j } \delta _ { k l } + \delta _ { i k } \delta _ { j l } + \delta _ { i l } \delta _ { j k } ) \, \underset { k } { \overset { i } { \underset { k } { \sim } } } ^ { l } = - i \frac { \lambda } { 3 } \phi ( \delta _ { i j } \delta _ { k 1 } + \delta _ { j k } \delta _ { i 1 } + \delta _ { k i } \delta _ { j 1 } ) \, \left ( 5. 1 0 \right )$$
$$\underset { j } { \sim } \overset { i } { \sim } \overset { \mu } { \sim } = 2 i e ^ { 2 } \delta _ { i j } g _ { \mu \nu } \underset { j } { \sim } \overset { i } { \sim } \overset { \mu } { \sim } = e \epsilon _ { i j } ( k _ { i } ^ { \mu } + k _ { j } ^ { \mu } ) \underset { \nu \nu } { \sim } \overset { \mu } { \sim } \overset { j } { \sim } = 2 i e ^ { 2 } \phi \delta _ { j 1 } g _ { \mu \nu } \ \ ( 5. 1 1 )$$
where the solid lines can be either φ 1 or φ 2 . These Feynman rules can also be found in [29] after rescaling λ → λ 3 to match our normalization, with an extended discussion of their derivation in Appendix A of that reference. Diagrams constructed with these Feynman rules were dubbed prototype diagrams in [4] to emphasize that they represent an infinite number of diagrams with arbitrary numbers of background fields insertions.
The effective potential is computed with these Feynman rules by evaluating vacuum energy contributions: diagrams with no external propagating fields. Each such diagram corresponds to an infinite number of background field insertions. One can also compute terms in the effective action which have derivatives, such as corrections to the kinetic term ( ∂ φ µ )( ∂ φ µ ). To do so, one inserts finite momentum p µ into a single background field line. The procedure is described in [30]. The 1-loop effective potential is a special case: the vacuum diagrams are quadratically divergent. A trick to computing it is to add external legs with p µ = 0. Inserting one such leg produces the first derivative of the potential, V ′ ( φ ). With two legs gives the second derivative, V ′′ ( φ ) and so in. Conversely, the second derivative of the potential V ′′ ( φ ) determines the scalar propagator at zero momentum, which in turn can be used to calculate the mass of the scalar at 1-loop. The path of least resistance is often to compute the 1-loop potential using functional determinants [5] and the potential at 2-loops and beyond with prototype Feynman diagrams.
At 2-loops, no additional tricks are needed and we can compute the effective potential directly from the vacuum diagrams with dressed propagators. All of the 2-loop diagrams can be computed using two master integrals. The first is a standard one-loop integral
$$I _ { 1 } ( \Delta ) = \int \frac { d ^ { d } k } { ( 2 \pi ) ^ { d } } \frac { 1 } { k ^ { 2 } - \Delta + i \varepsilon } = \frac { - i } { ( 4 \pi ) ^ { d / 2 } } \frac { 1 } { \Delta ^ { 1 - \frac { d } { 2 } } } \Gamma \left ( \frac { 2 - d } { 2 } \right )$$
The second is the 2-loop scalar vacuum sunset graph with 3 masses [27,28,31]:
$$I _ { 2 } ( A, B, C ) = \int \frac { d ^ { d } k _ { 1 } } { ( 2 \pi ) ^ { d } } \int \frac { d ^ { d } k _ { 2 } } { ( 2 \pi ) ^ { d } } \frac { 1 } { \left ( k _ { 1 } ^ { 2 } - A + i \varepsilon \right ) \left ( k _ { 2 } ^ { 2 } - B + i \varepsilon \right ) \left ( \left ( k _ { 1 } + k _ { 2 } \right ) ^ { 2 } - C + i \varepsilon \right ) } \quad ( 5. 1 3 )$$
Figure 2: There are 12 prototype diagrams which contribute to the 2-loop ColemanWeinberg potential in R ξ gauges. Only the 4 diagrams on the top row contribute at NLO (order glyph[planckover2pi1] 2 ) when λ ∼ glyph[planckover2pi1] .
This integral is fully symmetric in A, B and C . The full result in d dimensions can be found in [31] along with its ε expansion. Writing d = 4 -2 , the expansion is ε
$$& \text{in} \left [ 3 1 \right ] \text{along with its } \varepsilon \expansion. \text{ Writing } d = 4 - 2 \varepsilon, \text{ the expansion is} \\ & I _ { 2 } ( A, B, C ) = \frac { 1 } { ( 4 \pi ) ^ { 4 - 2 \varepsilon } } \frac { \Gamma ( 1 + \varepsilon ) ^ { 2 } } { ( 1 - \varepsilon ) ( 1 - 2 \varepsilon ) } \frac { 1 } { 2 C ^ { 2 \varepsilon } } \\ & \quad \times \left \{ - \frac { 1 } { \varepsilon ^ { 2 } } \left ( A + B + C \right ) + \frac { 2 } { \varepsilon } \left ( A \ln \frac { A } { C } + B \ln \frac { B } { C } \right ) \\ & \quad - A \ln ^ { 2 } \frac { A } { C } - B \ln ^ { 2 } \frac { B } { C } + ( C - A - B + C \lambda ) \ln \frac { A } { C } \ln \frac { B } { C } \\ & \quad - \frac { 1 } { 3 } C \lambda ^ { 2 } - 2 C \lambda \ln \frac { A - B + C - C \lambda } { 2 C } \ln \frac { B - A + C - C \lambda } { 2 C } \\ & \quad + 2 C \lambda L i _ { 2 } \left ( \frac { B - A + C - C \lambda } { 2 C } \right ) + 2 C \lambda L i _ { 2 } \left ( \frac { A - B + C - C \lambda } { 2 C } \right ) \right \} \quad ( 5. 1 4 ) \\ & \text{with}$$
with
This expansion holds if λ 2 ≥ 0 and √ A + √ B ≤ √ C . The expansion in other regions can be obtained by permuting A, B and C .
$$\lambda = \sqrt { \frac { A ^ { 2 } + ( B - C ) ^ { 2 } - 2 A ( B + C ) } { C ^ { 2 } } }$$
There are 4 diagrams that contribute at order glyph[planckover2pi1] 2 e 6 :
$$\mathcal { M } _ { A } & = \bigcup \bigcap _ { \infty, \infty, \infty } ^ { 2 } \Big | _ { \infty, \infty, \infty } ^ { \gamma \lambda \lambda } = - \frac { h ^ { 2 } e ^ { 2 } } { 2 } \int \frac { d ^ { d } k _ { 1 } } { ( 2 \pi ) ^ { d } } \int \frac { d ^ { d } k _ { 2 } } { ( 2 \pi ) ^ { d } } D _ { 2 2 } ( k _ { 1 } ) \Delta _ { \mu \mu } ( k _ { 2 } ) \\ & = \frac { \hbar { ^ { 2 } } \phi ^ { 4 } e ^ { 6 } } { ( 1 6 \pi ^ { 2 } ) ^ { 2 } } \xi \left [ - 1 2 \ln ^ { 2 } \frac { e \phi } { \mu } + \left ( 8 - 3 \ln \frac { \lambda \xi } { 6 e ^ { 2 } } \right ) \ln \frac { e \phi } { \mu } - \frac { 5 } { 2 } - \frac { \pi ^ { 2 } } { 1 6 } - \frac { 3 } { 1 6 } \ln ^ { 2 } \frac { \lambda \xi } { 6 e ^ { 2 } } + \ln \frac { \lambda \xi } { 6 e ^ { 2 } } \right ] + \cdots$$
$$( 5. 1 6 )$$
$$\mathcal { M } _ { B } = \bigcup _ { 1 } ^ { 2 } \ = \frac { i } { 2 } e ^ { 2 } \int \frac { d ^ { d } k _ { 1 } } { ( 2 \pi ) ^ { d } } \int \frac { d ^ { d } k _ { 2 } } { ( 2 \pi ) ^ { d } } ( k _ { 2 } ^ { \mu } - k _ { 1 } ^ { \mu } ) ( k _ { 1 } ^ { \nu } - k _ { 2 } ^ { \nu } ) D _ { 1 1 } ( k _ { 1 } ) D _ { 2 2 } ( k _ { 2 } ) \Delta _ { \mu \nu } ( k _ { 1 } + k _ { 2 } )$$
$$= \frac { \hbar { ^ { 2 } } \phi ^ { 4 } e ^ { 6 } } { ( 1 6 \pi ^ { 2 } ) ^ { 2 } } \left [ ( 2 + 6 \xi ) \ln ^ { 2 } \frac { e \phi } { \mu } - ( 3 + 7 \xi ) \ln \frac { e \phi } { \mu } + \frac { 7 } { 4 } + \frac { \pi ^ { 2 } } { 8 } + \frac { 1 5 } { 4 } \xi + \frac { 3 \pi ^ { 2 } } { 8 } \xi \right ] + \cdots \quad \ ( 5. 1 7 )$$
$$\mathcal { M } _ { C } = \left ( \frac { \binom { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \tilde { \t \tilde { \tilde { \t \tilde { \tilde { \t \tilde { \t \tilde { \t \tilde { \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \ t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \tt \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \t \n } } } } } } } \right ) = - i h ^ { 2 } e ^ { 4 } \phi ^ { 2 } \int \frac { d ^ { d } k _ { 1 } } { ( 2 \pi ) ^ { d } } \int \frac { d ^ { d } k _ { 2 } } { ( 2 \pi ) ^ { d } } \Delta _ { \mu \nu } ( k _ { 1 } ) \Delta _ { \mu \nu } ( k _ { 2 } ) D _ { 1 1 } ( k _ { 1 } + k _ { 2 } )$$
$$= \frac { \ddot { h } ^ { 2 } \phi ^ { 4 } e ^ { 6 } } { ( 1 6 \pi ^ { 2 } ) ^ { 2 } } \left [ ( 1 8 + 6 \xi ) \ln ^ { 2 } \frac { e \phi } { \mu } - ( 2 1 + 7 \xi ) \ln \frac { e \phi } { \mu } + \frac { 4 7 } { 4 } + \frac { 7 \pi ^ { 2 } } { 2 4 } + \frac { 1 5 } { 4 } \xi + \frac { 3 \pi ^ { 2 } } { 8 } \xi \right ] + \cdots$$
$$\mathcal { M } _ { D } & = \left \langle \frac { \binom { \binom { \sim } { 2 } } { \dots } } { \dots } \right \rangle _ { 1 } = - 2 \hslash ^ { 2 } e ^ { 3 } \phi \int \frac { d ^ { d } k _ { 1 } } { ( 2 \pi ) ^ { d } } \int \frac { d ^ { d } k _ { 2 } } { ( 2 \pi ) ^ { d } } ( k _ { 1 } ^ { \mu } - k _ { 2 } ^ { \mu } ) \Delta _ { \mu \nu } ( k _ { 1 } + k _ { 2 } ) T _ { \nu 2 } ( k _ { 2 } ) D _ { 1 1 } ( k _ { 1 } ) \\ & = \frac { \hslash ^ { 2 } \phi ^ { 4 } e ^ { 6 } } { ( 1 6 \pi ^ { 2 } ) ^ { 2 } } \xi \left [ - 1 2 \ln ^ { 2 } \frac { e \phi } { \mu } + 1 4 \ln \frac { e \phi } { \mu } - \frac { 1 5 } { 2 } - \frac { 3 \pi ^ { 2 } } { 4 } \right ] + \cdots$$
with the · · · vanishing as λ → 0. Kang computed some of the logarithmic terms in these amplitudes in [29], and we agree with his results.
Adding the contribution of these graphs to the counterterm contribution in Eq. (5.4) gives the glyph[planckover2pi1] 2 contributions to the effective potential:
$$V _ { 2 } = \left ( \frac { \hbar } { 1 6 \pi ^ { 2 } } \right ) ^ { 2 } e ^ { 6 } \phi ^ { 4 } \left [ \left ( 1 0 - 6 \xi \right ) \ln ^ { 2 } \frac { e \phi } { \mu } + \left ( - \frac { 6 2 } { 3 } + 4 \xi - \frac { 3 } { 2 } \xi \ln \frac { \lambda \xi } { 6 e ^ { 2 } } \right ) \ln \frac { e \phi } { \mu } \\ + \xi \left ( - \frac { 1 } { 2 } + \frac { 1 } { 4 } \ln \frac { \lambda \xi } { 6 e ^ { 2 } } \right ) + \frac { 7 1 } { 6 } \right ] + \cdots \ \ ( 5. 2 0 )$$
where again the · · · vanish as λ → 0.
## 5.3 RGE cross-check
As a cross-check, it is easy to verify that V = V 0 + V 1 + V 2 satisfies the renormalization group equation
$$\left ( \mu \frac { \partial } { \partial \mu } - \gamma \phi \frac { \partial } { \partial \phi } + \beta _ { e } \frac { \partial } { \partial e } + \beta _ { \lambda } \frac { \partial } { \partial \lambda } \right ) V = 0$$
up to order glyph[planckover2pi1] 2 . The anomalous dimensions and β function coefficients in scalar QED up to 2-loop order can be extracted from [32,33]. We find
$$& \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \rangle \\ & \gamma = \frac { \hbar { } } { 1 6 \pi ^ { 2 } } e ^ { 2 } ( \xi - 3 ) + \frac { \hbar { ^ { 2 } } } { ( 1 6 \pi ^ { 2 } ) ^ { 2 } } \left ( \frac { 1 0 } { 3 } e ^ { 4 } + \frac { 1 } { 9 } \lambda ^ { 2 } \right ) + \cdots \\ & \beta _ { \lambda } = \frac { \hbar { } } { 1 6 \pi ^ { 2 } } \left ( 3 6 e ^ { 4 } - 1 2 e ^ { 2 } \lambda + \frac { 1 0 \lambda ^ { 2 } } { 3 } \right ) + \frac { \hbar { ^ { 2 } } } { ( 1 6 \pi ^ { 2 } ) ^ { 2 } } \left ( - 4 1 6 e ^ { 6 } + \frac { 3 1 6 e ^ { 4 } \lambda } { 3 } + \frac { 5 6 e ^ { 2 } \lambda ^ { 2 } } { 3 } - \frac { 2 0 \lambda ^ { 3 } } { 3 } \right ) + \cdots \\ & \beta _ { e } = \frac { \hbar { } } { 1 6 \pi ^ { 2 } } \frac { e ^ { 3 } } { 3 } + \cdots \\ & I l i n o \text{ these expansions } \text{ $ur $2$} \text{on $2$} \text{for their notital indeed satisfies the RCF. $A$ similar}$$
Using these expansions, our 2-loop effective potential indeed satisfies the RGE. A similar cross check was done in Landau gauge for the 2-loop standard model potential in [27] and for general renormalizable theories in [28].
In fact, the RGE could have been used to bootstrap all of the ln φ µ dependence in the 2-loop effective potential. Note however that the ln λ and ln e terms in the 1-loop effective potential are critical to determining the ln φ µ dependence at 2 loops, and these terms are not fixed by RG invariance. In other words, we needed the exact 1-loop potential to determine the 2-loop ln φ µ dependence. Coleman and Weinberg argue [4, App. A2] that one can ignore logarithms of couplings when working in a subtraction scheme with λ ≡ d V 4 dφ 4 ∣ ∣ ∣ µ for a fixed scale µ . This may be true, but since the fourth derivative of the potential is not gauge invariant (even at an extremum), this renormalization condition induces unnecessary complications when studying gauge dependence. Moreover, in MS, logarithms of couplings play an important role and must be kept.
While we are on the subject of RG invariance, note that the MS β function coefficients for e and λ are gauge invariant. This is true to all orders, since the β functions describe the evolution of couplings which can in principle be measured from scattering experiments. The anomalous dimension for φ is gauge-dependent; it describes the evolution of the field strength Z φ for φ , which is not itself observable. We will return to the RG when considering the resummed potential in Section 8.
## 6 Daisy resummation
As pointed out by Nielsen [15, 34], there can be contributions to the effective potential proportional to λ -1 e 10 coming from daisy diagrams , like those in Figure 3. When λ ∼ e 4 , these terms are of the same order as the e λ 2 terms in the 1-loop potential and the e 6 terms in the 2-loop potential. Thus they should be relevant to showing gauge invariance of physical
Figure 3: Example daisy graphs which contribute inverse powers of λ .
quantities at order e 6 in this power counting. Nielsen argued that the reason the scalar mass Kang calculated does not satisfy his identity was due to the absence of these terms [15]. Some time later, Johnston [35] showed how these terms can be summed into a dressed propagator for the scalar fields, suggesting that these terms could be computed and the Nielsen identity restored, although no explicit contributions to the potential were provided. Shortly afterwards Bazeia [36] showed that even at 1-loop, where daisies are irrelevant, the vacuum energy in the Coleman-Weinberg model has gauge-dependence. In this section, we compute all of the daisy graphs relevant at e 6 (and some of the subleading daisy graphs as well to demonstrate their relevance in Landau gauge). In the next section we demonstrate that, after carefully keeping track of the independent variables, the effective potential at its minimum is indeed gauge-invariant.
Before beginning, it is worth remarking that the daisy resummation we perform here is related to, but not identical to, daisy resummation in finite-temperature field theory [3739]. At finite temperature, the resummation of daisy diagrams is necessary to calculate a critical temperature because, in the limit m glyph[lessmuch] T , new infrared divergences arise. Daisy resummation as a solution to infrared problems associated with massless Goldstone bosons has recently been proposed in [40,41]. The relevance of daisy resummation to solving gaugedependence problems at finite temperature has also been discussed [19, 42]. It is therefore not surprising that daisy resummation is relevant to the gauge-dependence problem in the Coleman-Weinberg model. Nevertheless, the relevant calculations have never been done, to our knowledge, which is why we resum the relevant daisy graphs here.
Normally, one does not get inverse powers of coupling constants from Feynman diagrams. Indeed, at any fixed order in perturbation theory, there are always positive powers of couplings, even in effective potential calculations. However, the effective potential always involves summing an infinite number of graphs, namely those with an arbitrary number of background field insertions. It is this infinite sum which gives the ln φ dependence in the effective potential and which can generate inverse powers of couplings. As discussed in Section 5, we simplify the infinite sums by using dressed propagators. For example, we see from Eq. (5.5) that the φ 1 propagator is D 11 = i k 2 -λ φ 2 which has an effective mass m 2 = λ 2 φ 2 .
In the daisy graphs, each photon loop (the petals) gives a factor proportional to e φ 2 2 , each vertex gives a factor of e 2 , and the loop integral over the scalar propagators can give inverse
2
powers of the effective mass. For example, a 4-loop 3-petal daisy will give
$$\left \langle \begin{smallmatrix} \left \langle \right \rangle \\ \left \langle \right \rangle \\ \left \langle \right \rangle \end{smallmatrix} \right \rangle \right \rangle \left \langle \left \langle \right \rangle \right \rangle$$
Here, we have simply done the integral by dimensional analysis, since it is UV and IR finite. This graph therefore contributes at order e 8 in the λ ∼ e 4 power-counting. It is therefore beyond the order we need for the first non-trivial gauge-invariance check of V min (not to mention that this particular loop is itself ξ -independent).
It is not hard to see, using dimensional analysis, that the only graphs which could contribute at order e 6 (with the λ ∼ e 4 power-counting) must involve φ 2 propagators and have petal-type photon loops. The petals factorize off from central disk, and from each other. Thus a daisy graph with n petals has the form
$$I _ { n } = i ^ { n + 1 } \frac { e ^ { 2 n } } { 2 n } A _ { n \phi _ { 2 } } ( A _ { \gamma } ) ^ { n }$$
where A γ includes the photon loop and its counterterm, A nφ 2 is the scalar integral with n φ 2 propagators, and the factor 1 2 n e 2 n comes from the vertex Feynman rule and the symmetry factor for the circle.
Each photon loop gives
$$A _ { \gamma } ^ { \text{loop} } & = \left \{ \sum _ { \substack { \tilde { \sim } \\ \tilde { \sim } \\ 1 6 \pi ^ { 2 } } } ^ { \tilde { \sim } \\ \tilde { \sigma } } ^ { 2 } \tilde { \phi } ^ { 2 } \left [ \frac { 3 } { \varepsilon } + \left ( 1 - 6 \ln \frac { \phi e } { \mu } \right ) + \varepsilon \left ( 1 + \frac { \pi ^ { 2 } } { 4 } - 2 \ln \frac { \phi e } { \mu } + 6 \ln ^ { 2 } \frac { \phi e } { \mu } \right ) + \cdots \right ] \quad ( 6. 3 )$$
The UV pole in this loop is removed by the 1-loop counterterm:
$$A _ { \gamma } ^ { c. t. } = \frac { \hbar } { 1 6 \pi ^ { 2 } } e ^ { 2 } \phi ^ { 2 } \left [ - \frac { 3 } { \varepsilon } \right ]$$
The O ( ε ) in A loop γ can be important if the scalar loop it multiplies is UV-divergent. This only happens for two petals ( n = 2). For n = 2, the scalar loop gives
$$A _ { 2 \phi _ { 2 } } = \ \widetilde { \binom { 2 } { 2 } } \ = \int \frac { d ^ { d } k } { ( 2 \pi ) ^ { d } } D _ { 2 2 } ^ { 2 } = - i \frac { \hbar } { 1 6 \pi ^ { 2 } } \left [ \frac { 1 } { \varepsilon } + \frac { 3 e ^ { 2 } \xi } { \lambda } + \cdots \right ]$$
Since the 1 ε term in A 2 φ 2 has no 1 λ piece, the cross term between it and the O ( ε ) part of A loop γ will not contribute at order e 6 (with λ ∼ e 4 ). Thus, for all the daisies, we can drop the O ( ε ) terms in A loop γ and take
$$A _ { \gamma } = A _ { \gamma } ^ { \text{loop} } + A _ { \gamma } ^ { \text{c.t.} } = \frac { \hbar } { 1 6 \pi ^ { 2 } } e ^ { 2 } \phi ^ { 2 } \left ( 1 - 6 \ln \frac { e \phi } { \mu } \right )$$
For n > 3, the scalar loop is UV and IR finite. We find
$$A _ { n \phi _ { 2 } } = \overbrace { \bigcirc \nolimits } = \int \frac { d ^ { 4 } k } { ( 2 \pi ) ^ { 4 } } D _ { 2 2 } ^ { n } = \frac { i \hbar } { 1 6 \pi ^ { 2 } } \phi ^ { 4 } \frac { e ^ { 2 } \lambda \xi } { 1 2 ( n - 1 ) } \left ( \frac { - 6 i } { \lambda \phi ^ { 2 } } \right ) ^ { n }$$
so that
$$\mathfrak { s o t h a t } I _ { n } = \frac { \hbar { } } { 1 6 \pi ^ { 2 } } \left ( - \frac { 1 } { 2 4 } e ^ { 2 } \lambda \xi \phi ^ { 4 } \right ) \frac { 1 } { n ( n - 1 ) } \left [ \frac { \hbar { e } ^ { 4 } } { ( 1 6 \pi ^ { 2 } ) \, \lambda } \left ( 6 - 3 6 \ln \frac { e \phi } { \mu } \right ) \right ] ^ { n }$$
Each term in this series contributes at order e 6 when λ ∼ e 4 . Thus they are all equally important for checking gauge invariance and we must sum the series. Summing the series is easy enough to do using
$$\sum _ { n = 2 } ^ { \infty } \frac { 1 } { n ( n - 1 ) } x ^ { n } = x + ( 1 - x ) \ln ( 1 - x )$$
which gives
$$V ^ { e ^ { 6 } \text{daises} } = \phi ^ { 4 } \frac { \hbar { } } { 1 6 \pi ^ { 2 } } \left ( - \frac { e ^ { 2 } \lambda \xi } { 2 4 } \right ) \left [ \frac { \widehat { \lambda } ( \phi ) } { \lambda } + ( 1 - \frac { \widehat { \lambda } ( \phi ) } { \lambda } ) \ln { ( 1 - \frac { \widehat { \lambda } ( \phi ) } { \lambda } ) } \right ] \quad \quad ( 6. 1 0 )$$
where
$$\widehat { \lambda } ( \phi ) \equiv \frac { \hbar { e } ^ { 4 } } { 1 6 \pi ^ { 2 } } \left ( 6 - 3 6 \ln \frac { e \phi } { \mu } \right )$$
We have defined ̂ λ φ ( ) so that according to Eq. (4.6) ̂ λ v ( ) = λ at the scale v where the 1-loop potential has its minimum. Remarkably, while each daisy graph with n > 1 is individually power-divergent as λ → 0 with e fixed, the sum of all daisies scales only like ln λ .
Before moving on, it is worth pointing out that daisy resummation is important even in Landau gauge, ξ = 0. In Landau gauge, there is no kinetic mixing and the scalar propagators are
$$D _ { 1 1 } = \frac { i } { k ^ { 2 } - \frac { \lambda } { 2 } \phi ^ { 2 } }, \quad D _ { 2 2 } = \frac { i } { k ^ { 2 } - \frac { \lambda } { 6 } \phi ^ { 2 } },$$
These propagators still have λ -dependent masses and can produce 1 λ dependence from daisy graphs. For example, summing the daisy graphs with φ 1 or φ 2 running in the loop gives
$$V ^ { e ^ { 8 } \text{daisies} } = - \frac { 1 } { 2 } \phi ^ { 4 } \frac { \hbar } { 1 6 \pi ^ { 2 } } \left [ \frac { \lambda ^ { 2 } } { 4 } F ( \frac { 2 \widehat { \lambda } } { \lambda } ) + \frac { \lambda ^ { 2 } } { 3 6 } F ( \frac { 6 \widehat { \lambda } } { \lambda } ) \right ]$$
where
$$F ( x ) = \frac { x ^ { 2 } - 4 x } { 4 8 } - \frac { 1 } { 2 } ( 1 - \frac { x } { 6 } ) ^ { 2 } \ln \left ( 1 - \frac { x } { 6 } \right )$$
These terms are important, but since λ ∼ e 4 , they have effects comparable to terms in the 3loop Coleman-Weinberg potential. Thus, an advantage of Landau gauge is that it postpones the relevance of daisy resummation by one loop. Landau gauge does not however let us ignore the daisy graphs completely.
Two recent papers also observed that resummation of certain graphs to all orders is necessary starting at 3 loops in Landau gauge [40,41]. These two papers are concerned with resolving an infrared divergence problem associated with massless Goldstone bosons starting at 3 loops. While these two papers discuss diagrams similar to the ones here, the problem they solve is different (infrared divergences, not gauge-dependence) and their results are not directly transferable. However, these two papers, along with the earlier work in [35,43], do explain in a more systematic way how daisy and other relevant diagrams can be resummed through a modification of the effective propagators.
In summary, the full Coleman-Weinberg potential up to order e 6 with λ ∼ e 4 is the sum of Eqs. (4.1), (4.3), (5.20) and (6.10). It is helpful to write the result as
$$V = V ^ { L O } + V ^ { N L O } + \cdots$$
where the leading-order (LO) potential
$$V ^ { \text{LO} } = \frac { \lambda } { 2 4 } \phi ^ { 4 } + \frac { \hbar { e } ^ { 4 } } { 1 6 \pi ^ { 2 } } \phi ^ { 4 } \left ( - \frac { 5 } { 8 } + \frac { 3 } { 2 } \ln \frac { e \phi } { \mu } \right )$$
scales as O ( glyph[planckover2pi1] ) when λ ∼ glyph[planckover2pi1] e 4 and the next-to-leading order (NLO) potential, scaling like O ( glyph[planckover2pi1] 2 ), is
$$\boxed { \begin{array} { c } V ^ { \text{NLO} } = \frac { \hbar { e } ^ { 2 } \lambda } { 1 6 \pi ^ { 2 } } \phi ^ { 4 } \left ( \frac { \xi } { 8 } - \frac { \xi } { 2 4 } \ln \frac { e ^ { 2 } \lambda \xi \phi ^ { 4 } } { 6 \mu ^ { 4 } } \right ) \\ + \frac { \hbar { 2 } ^ { e } e ^ { 6 } } { ( 1 6 \pi ^ { 2 } ) ^ { 2 } } \phi ^ { 4 } \left [ ( 1 0 - 6 \xi ) \ln ^ { 2 } \frac { e \phi } { \mu } + \left ( - \frac { 6 2 } { 3 } + 4 \xi - \frac { 3 } { 2 } \xi \ln \frac { \lambda \xi } { 6 e ^ { 2 } } \right ) \ln \frac { e \phi } { \mu } + \xi \left ( - \frac { 1 } { 2 } + \frac { 1 } { 4 } \ln \frac { \lambda \xi } { 6 e ^ { 2 } } \right ) + \frac { 7 1 } { 6 } \right ] \\ + \phi ^ { 4 } \frac { \hbar { e } ^ { 2 } \lambda } { 1 6 \pi ^ { 2 } } \left ( - \frac { \xi } { 2 4 } \right ) \left [ \frac { \lambda ( \phi ) } { \lambda } + ( 1 - \frac { \widehat { \lambda } ( \phi ) } { \lambda } ) \ln \left ( 1 - \frac { \widehat { \lambda } ( \phi ) } { \lambda } \right ) \right ] \\ \end{array} }$$
with ̂ λ φ ( ) is defined in Eq. (6.11). Note that there are tree and 1-loop contributions to the LO potential and that the NLO potential get contributions from 1-, 2- and all higher order loops.
## 7 Gauge invariance of V min
With the 2-loop Coleman-Weinberg potential in hand and the contribution of daisy graphs understood, we can now demonstrate gauge-invariance of the potential at its minimum.
Recall from Eq. (4.13) that at leading order (LO), V LO min = -v e 4 4 3 128 π 2 . One might naturally expect that the next-to-leading order (NLO) contribution could be written as V NLO min = v e C 4 6 for some ξ -independent constant C . Unfortunately, we cannot expect V min to be explicitly ξ -independent when written in terms of v . The problem is that v = 〈 φ 〉 is a field value and therefore gauge-dependent, so infects all dimensionful quantities expressed in terms of it. One alternative is to calculate not V min but the ratio of V min to some other dimension-four quantity, such as m 4 S , where m S is the scalar mass. Indeed, Coleman and Weinberg [4] and later Kang [29] discussed gauge invariance of the dimensionless ratio of the scalar to vector masses. However, we would really like to see that V min is physical on its own. But then, if not v , what are we to express V min in terms of?
An alternative to expressing V min in terms of v is to express it in terms of the renormalization group scale µ . This scale is as physical as the MS couplings: the two are intrinsically connected. To be concrete, let us define the scale µ X as the scale where Eq. (4.7) is satisfied exactly. That is, µ X is defined by the exact relation
$$\lambda ( \mu _ { X } ) \equiv \frac { \hbar } { 1 6 \pi ^ { 2 } } e ^ { 4 } ( \mu _ { X } ) \left \{ 6 - 3 6 \ln [ e ( \mu _ { X } ) ] \right \}$$
Since this relation is exact, we no longer can or need to solve for e 6 terms in the relation between λ and e , as was done in [29]. One can instead now solve for corrections to v
$$v = \langle \phi \rangle = \mu _ { X } + v _ { 1 } + v _ { 2 } + \cdots$$
with v 1 ∼ e µ ( X ) 2 , v 2 ∼ e µ ( X ) 4 , etc., and µ X is defined by Eq. (7.1). Then one can consistently expand V min at µ = µ X :
$$\begin{array} { c } \cdots \cdots & \cdots \cdots & \cdots & \cdots \\ V _ { \min } = V ^ { L O } ( \mu _ { X } ) & \underset { d \phi } { + } V ^ { N L O } \underset { d \phi } { \overset { = 0 } { \overbrace { d } } } V ^ { L O } \Big | _ { \phi = \mu _ { X } } \\ & + V ^ { N N L O } ( \mu _ { X } ) + v _ { 1 } \frac { d } { d \phi } V ^ { N L O } \Big | _ { \phi = \mu _ { X } } + v _ { 2 } & \underset { d \phi } { \overset { d } { \overbrace { d \phi } } } V ^ { L O } \Big | _ { \phi = \mu _ { X } } + \frac { 1 } { 2 } v _ { 1 } ^ { 2 } \frac { d ^ { 2 } } { d \phi ^ { 2 } } V ^ { L O } \Big | _ { \phi = \mu _ { X } } + \cdots \\ \text{where } V ^ { L O } \text{ is the effective nontial truncated to order } h \text{ with } \lambda \sim \hbar { \ } V ^ { N L O } \text{ is truncated to} \end{array}$$
where V LO is the effective potential truncated to order glyph[planckover2pi1] with λ ∼ glyph[planckover2pi1] , V NLO is truncated to glyph[planckover2pi1] 2 and so on, as in Eqs. (6.16) and (6.17).
A convenient feature of setting φ = µ X is that the daisy contribution simplifies. Indeed, from Eq. (6.11) we see that when φ = µ = µ X then ̂ λ µ ( X ) = λ µ ( X ). Then the entire daisy contribution on the last line of Eq. (6.17) reduces to simply
$$V ^ { N L O, \, d a i s i e s } = - \frac { \xi } { 2 4 } \phi ^ { 4 } \frac { \hbar { e } ^ { 2 } \lambda } { 1 6 \pi ^ { 2 } } = - \frac { \xi } { 2 4 } \frac { \hbar { ^ { 2 } } e ^ { 6 } } { ( 1 6 \pi ^ { 2 } ) ^ { 2 } } ( 6 - 3 6 \ln e )$$
Also, as indicated in Eq. (7.3), we can use that V LO ′ ( µ X ) = 0, which was the defining equation for µ X . This simplifies V min to
$$V _ { \min } = V ^ { L O } ( \mu _ { X } ) + V ^ { N L O } ( \mu _ { X } ) + \cdots$$
Plugging in Eqs. (6.16) and (6.17) and using Eq. (7.1) we then find
$$V _ { \min } = - \frac { 3 \hbar { e } ^ { 4 } } { 1 2 8 \pi ^ { 2 } } \mu _ { X } ^ { 4 } + \frac { e ^ { 6 } \hbar { ^ { 2 } } } { ( 1 6 \pi ^ { 2 } ) ^ { 2 } } \mu _ { X } ^ { 4 } \left ( \frac { 7 1 } { 6 } - \frac { 6 2 } { 3 } \ln e + 1 0 \ln ^ { 2 } e \right )$$
which is manifestly gauge-invariant! The daisies have exactly canceled the ξ dependence of the NLO 1-loop and 2-loop contributions.
Next, let us look at a field value expressed in terms of the MS scale µ , to double check that somehow all of its gauge-dependence is not miraculously absent. Consider the value of the field where the potential is zero Λ , which in the Standard Model is sometimes given I an interpretation as an instability scale [44]. Setting V (Λ ) = 0 gives a different relation I between λ and e than V ′ ( v ) = 0 did. The condition on the running couplings so that V LO = 0 at φ = µ = µ I is
$$\lambda ( \mu _ { I } ) = \frac { \hbar } { 1 6 \pi ^ { 2 } } e ( \mu _ { I } ) ^ { 4 } \left \{ 1 5 - 3 6 \ln [ e ( \mu _ { I } ) ] \right \}$$
At NLO, we then find
$$\Lambda _ { I } = \mu _ { I } - \frac { V ^ { N L O } ( \mu _ { I } ) } { V ^ { L O ^ { \prime } } ( \mu _ { I } ) } + \dots = \mu _ { I } - \frac { 3 2 \pi ^ { 2 } } { 3 e ^ { 4 } \hbar { \mu } _ { I } ^ { 3 } } V ^ { N L O } ( \mu _ { I } )$$
To evaluate the daisy contribution, we can no longer use ̂ λ = λ . Instead we now find
$$\widehat { \frac { \lambda } { \lambda } } = \frac { \widehat { \lambda } ( \phi = \mu = \mu _ { I } ) } { \lambda ( \mu = \mu _ { I } ) } = 1 - \frac { 3 } { 5 - 1 2 \ln e }$$
Up to NLO, we then find
$$\begin{array} { c } \cdot & \cdot \\ \Lambda _ { I } = \mu _ { I } - \frac { V ^ { N L O } ( \mu _ { I } ) } { V ^ { L O ^ { \prime } } ( \mu _ { I } ) } + \cdots \\ = \mu _ { I } + \mu _ { I } \frac { \hbar { e } ^ { 2 } } { 1 6 \pi ^ { 2 } } \left \{ - \frac { 7 1 } { 9 } - \frac { 1 1 \xi } { 1 2 } + \left ( \frac { 5 \xi } { 2 } + \frac { 1 2 4 } { 9 } \right ) \ln e - \frac { 2 0 } { 3 } \ln ^ { 2 } e + \frac { 1 } { 4 } \xi \ln \left [ \frac { \ h \xi } { 3 2 \pi ^ { 2 } } ( 5 - 1 2 \ln e ) \right ] \\ & - \frac { \xi } { 1 2 } \left ( - 2 + 1 2 \ln e - 3 \ln \frac { 3 } { 5 - 1 2 \ln e } \right ) \right \} + \cdots & ( 7. 1 0 ) \\ \\ \tau \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \, \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \colon \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \end{array}$$
This instability scale is linearly dependent on the gauge-parameter ξ , and therefore should not be used to draw physical conclusions. The ξ dependence of other field values can be computed in a similar way, confirming that they are indeed unphysical.
## 8 Renormalization group improvement
At this point we have shown that if the effective potential in scalar QED is expressed in terms of the MS couplings e and λ and the scale µ X where they satisfy Eq. (7.1), then the value of the potential at the quantum minimum V min is gauge invariant. We checked this to the first non-leading order, which required the 2-loop potential and the summation of the leading daisy diagrams. A natural question is whether we must express V min in terms of µ X ? Clearly, we should be able to write any physical quantity in terms of any other scale. More importantly, the check required the truncation of the effective potential to order e 6 with λ µ ( X ) ∼ e µ ( X ) 4 . We would like to be able to use effective potentials in other contexts, where the couplings are defined at some other arbitrary scale µ Y and the renormalization group is used to evolve the potential to a scale near its minimum. So how are we to use this unusual truncation in a RG-improved effective potential, where some λ and e dependence is necessarily included to all orders in glyph[planckover2pi1] ? In this section, we show how to answer both these questions, and that the answers are related.
## 8.1 Calculation-scale invariance
First, let us consider how the calculation of V min would change if we had calculated the effective potential with the couplings defined at a scale µ Y instead of µ X . At µ Y , the potential obviously has the same form, Eq. (6.17), but with e and λ interpreted as e Y ≡ e µ ( Y ) and λ Y ≡ λ µ ( Y ) rather than e X ≡ e µ ( X ) and λ X ≡ λ µ ( X ). That is,
$$V = \frac { \lambda _ { Y } } { 2 4 } \phi ^ { 4 } + \frac { \hbar } { 1 6 \pi ^ { 2 } } \phi ^ { 4 } \left [ e _ { Y } ^ { 4 } \left ( - \frac { 5 } { 8 } + \frac { 3 } { 2 } \ln \frac { e _ { Y } \phi } { \mu _ { Y } } \right ) + e _ { Y } ^ { 2 } \lambda _ { Y } \xi \left ( \frac { 1 } { 8 } - \frac { 1 } { 2 4 } \ln \frac { e _ { Y } ^ { 2 } \lambda _ { Y } \xi \phi ^ { 4 } } { 6 \mu _ { Y } ^ { 4 } } \right ) \right ] + \cdots \quad ( 8. 1 )$$
For simplicity, let us assume for now that µ Y is close enough to µ X so that λ Y is still much smaller than e 2 Y . For the daisy contribution evaluated at µ Y , we can then still use ̂ λ ∼ λ , since the difference is higher order. This lets us continue to drop most of Eq. (6.10).
The easiest way to study V with couplings at µ Y is simply to use the renormalization group to move the couplings back to µ X . Then we can recycle the previous analysis. Expanding Eq. (4.11) gives
$$\circ \cdots \\ e _ { Y } = e _ { X } + \frac { \hbar } { 4 8 \pi ^ { 2 } } e _ { X } ^ { 3 } \ln \frac { \mu _ { Y } } { \mu _ { X } } + \cdots \\.. \.$$
Expanding Eq. (4.12) gives
$$\lambda _ { Y } = \lambda _ { X } + \frac { \hbar } { 1 6 \pi ^ { 2 } } \left [ 3 6 e _ { X } ^ { 4 } - 1 2 e _ { X } ^ { 2 } \lambda _ { X } \right ] \ln \frac { \mu _ { Y } } { \mu _ { X } } + \frac { \hbar { ^ { 2 } } } { ( 1 6 \pi ^ { 2 } ) ^ { 2 } } e _ { X } ^ { 6 } \left [ - 4 1 6 \ln \frac { \mu _ { Y } } { \mu _ { X } } - 1 9 2 \ln ^ { 2 } \frac { \mu _ { Y } } { \mu _ { X } } \right ] + \cdots \ ( 8. 3 )$$
Plugging these into Eq. (8.1) we find
$$& \dashrightarrow \circ \dashrightarrow \dashrightarrow \dashrightarrow \dashrightarrow \dashrightarrow \\ & V = \frac { \lambda _ { X } } { 2 4 } \phi ^ { 4 } \\ & \quad + \frac { \hbar } { 1 6 \pi ^ { 2 } } \phi ^ { 4 } \left [ e _ { X } ^ { 4 } \left ( - \frac { 5 } { 8 } + \frac { 3 } { 2 } \ln \frac { e _ { X } \phi } { \mu _ { X } } \right ) + e _ { X } ^ { 2 } \lambda _ { X } \left ( \frac { \xi } { 8 } - \frac { \xi } { 2 4 } \ln \frac { e _ { X } ^ { 2 } \lambda _ { X } \xi \phi ^ { 4 } } { 6 \mu _ { X } ^ { 4 } } + \frac { \xi - 3 } { 6 } \ln \frac { \mu _ { Y } } { \mu _ { X } } \right ) \right ] + \cdots$$
We have not even shown the 2-loop terms, but already one can see that this potential is not identical to what we got from computing V starting at µ = µ X : there is an uncanceled ln µ Y µ X term. Despite this modification of the potential, we still find that
$$V _ { \min } = \frac { e ^ { 4 } \hbar } { 1 6 \pi ^ { 2 } } \mu _ { X } ^ { 4 } \left ( - \frac { 3 } { 8 } \right ) + \frac { e ^ { 6 } \hbar { ^ { 2 } } } { ( 1 6 \pi ^ { 2 } ) ^ { 2 } } \mu _ { X } ^ { 4 } \left ( \frac { 7 1 } { 6 } - \frac { 6 2 } { 3 } \ln e + 1 0 \ln ^ { 2 } e \right ) + \cdots$$
exactly as in Eq. (7.6). That is, the new ln µ Y µ X term has no effect on the value of the potential at the minimum.
That the µ Y dependence would drop out of V min was anticipated in Section 2.2, with µ 0 replaced by µ Y and µ replaced by µ X . We next review and extend that discussion in the context of this concrete example.
## 8.2 Discussion
There is a quick way to see why computing the potential at a different scale affected V ( φ ) but not V min . The effective potential satisfies the RGE in Eq. (5.21):
$$\left ( \mu \frac { \partial } { \partial \mu } - \gamma \phi \frac { \partial } { \partial \phi } + \beta _ { e } \frac { \partial } { \partial e } + \beta _ { \lambda } \frac { \partial } { \partial \lambda } \right ) V = 0$$
This equation says that the explicit µ -dependence in the effective potential is exactly compensated for by a rescaling of the coupling constants, according to their β -functions, and a rescaling of the field φ , according to its anomalous dimension. In writing the potential computed at µ Y in terms of couplings at µ X we only rescaled the couplings, not the field. Indeed, the extra ξ -3 6 ln µ Y µ X term in Eq. (8.4) would be exactly removed if we rescaled the field as φ → κφ with
$$\kappa = 1 + \frac { e ^ { 2 } } { 1 6 \pi ^ { 2 } } ( 3 - \xi ) \ln \frac { \mu _ { Y } } { \mu _ { X } } + \cdots$$
Because the two potentials (calculated at µ X or µ Y ) differ by terms which can be compensated by a field rescaling, the value of V min will be unchanged. To see this, simply note that the extrema of any function f ( x ) are invariant under rescaling of the dependent variable: this rescaling just scales the x -axis (see Fig. 1). Thus the invariance of V min follows as a special case.
This property of the effective potential, that it depends on the scale where it is calculated, may seem unsettling. One might imagine that the ln µ Y µ X terms can be removed by invoking canonical normalization, perhaps by demanding that the kinetic term in the effective action be reset to 1 2 ( ∂ φ µ ) 2 after including all the quantum corrections. One cannot just demand this, however. First of all, in the effective action even the two-derivative kinetic terms have complicated field dependence
$$\Gamma [ \phi ] = \dots + \frac { 1 } { 2 } Z _ { 1 } [ \phi ] ( \partial _ { \mu } \phi _ { i } ) ( \partial _ { \mu } \phi _ { i } ) + \frac { 1 } { 2 } Z _ { 2 } [ \phi ] \frac { \phi _ { i } \phi _ { j } } { \phi ^ { 2 } } ( \partial _ { \mu } \phi _ { i } ) ( \partial _ { \mu } \phi _ { j } ) + Y _ { 1 } [ \phi ] \frac { 1 } { \phi ^ { 4 } } \left ( ( \partial _ { \mu } \phi _ { i } ) ( \partial _ { \mu } \phi _ { i } ) \right ) ^ { 2 } + \dots \dots$$
The functionals Z , Z , Y 1 2 1 etc. can be calculated from prototype diagrams as a power expansion in external momentum p µ [30]. Explicit results at 1-loop for Z 1 and Z 2 can be found in [10] and for Y 1 in φ 4 theory in [30]. So it is not clear how one would go about canonically normalizing φ apart from some byzantine implicit nonlinear field redefinition.
A more elementary objection to re-normalizing the fields is that we have already renormalized them. In MS the field strength renormalization Z 1 is unambiguously fixed, so there is no remaining freedom. One could have chosen a different normalization convention, for example, an on-shell renormalization for which the residue of the propagator at the physical pole is set to 1 exactly. However, since these are conventions, physical quantities must be independent of them. Thus we can work in MS. Indeed, as we have seen V min is independent of the normalization condition.
## 8.3 RG improved effective potential
We have shown that the effective potential depends on the scale µ Y where it is calculated. We have also shown that when µ Y is close to the scale µ X where the couplings satisfy Eq. (7.1), then the value of V min is independent of µ Y . Now suppose µ Y is not close to µ X , so that ln µ Y µ X is not small. Then we cannot work in fixed-order perturbation theory. For example, in the standard model, one normally extracts the MS couplings through perturbative threshold calculations at a scale µ Y ∼ 100 GeV near the weak scale; the equivalent of µ X is a scale near where λ ∼ 0 which is many orders of magnitude higher in energy; so, µ X glyph[greatermuch] µ Y . The minimum of the effective potential is found by solving the RGE for the effective potential to evolve it from µ Y to µ X . Thus the potential at µ = µ X includes terms to all orders in perturbation theory whether it is written in terms of g µ ( Y ) or g µ ( X ). This seems to limit the use of fixed order relations like Eq. (7.1).
In order to combine the λ ∼ e 4 scaling, which is required for gauge-invariance, with resummation, we can exploit the observation from Section 8.2 that V min is independent of rescaling φ . In particular, we see that we do not have to evolve the entire effective potential from µ Y to µ X . Instead, we can just evolve the couplings, and then compute the fixed order potential directly at µ X . The scale µ Y only enters for determining the boundary conditions of the renormalization equations. Since the β functions are all gauge invariant, gauge invariance of the potential at the minimum then automatically follows from the arguments in Section 7.
To be concrete, the resummed effective potential is often written as
$$V [ \phi ] = \frac { 1 } { 4! } e ^ { 4 \Gamma ( \mu = \phi ) } \left [ \lambda _ { \text{eff} } ^ { ( 0 ) } ( \mu = \phi ) + \lambda _ { \text{eff} } ^ { ( 1 ) } ( \mu = \phi ) + \lambda _ { \text{eff} } ^ { ( 2 ) } ( \mu = \phi ) + \cdots \right ] \phi ^ { 4 }$$
with Γ( µ ) = ∫ µ µ Y γ µ ( ′ ) dµ ′ µ ′ and 1 4! λ ( j ) eff ( µ φ ) 4 the order glyph[planckover2pi1] j term in the fixed-order calculation of the effective potential. We are proposing instead that one should use Eq. (7.3):
$$V _ { \min } = V ^ { L O } ( \mu _ { X } ) + V ^ { N L O } ( \mu _ { X } ) + V ^ { N N L O } ( \mu _ { X } ) + v _ { 1 } \frac { d } { d \phi } V ^ { N L O } \Big | _ { \phi = \mu _ { X } } + \frac { 1 } { 2 } v _ { 1 } ^ { 2 } \frac { d ^ { 2 } } { d \phi ^ { 2 } } V ^ { L O } \Big | _ { \phi = \mu _ { X } } + \cdots \ ( 8. 1 0 )$$
where V LO ∼ glyph[planckover2pi1] , V NLO ∼ glyph[planckover2pi1] 2 , v 1 ∼ glyph[planckover2pi1] and so on, using λ ∼ glyph[planckover2pi1] . Using Eq. (8.10), the value of V min will be gauge invariant order-by-order using this expansion, while V min computed using Eq. (8.9) will not be.
Note that this consistent prescription is not equivalent to using dressed fields where the field strength renormalization is absorbed into the field. Such a redefinition was advocated in [9] (based on [8]) as removing all of the gauge-dependence in the effective potential. First of all, that redefinition does not work as advertised: even after absorbing the field strength renormalization, the effective potential is still gauge-dependent [15,16]. Secondly, we are still using the same fields and MS normalizations relevant for S -matrix calculations, which greatly facilitates applications to the Standard Model. Furthermore, since V NLO ( µ X ) is known in the Standard Model [27] (and in general renormalizable theories [28]) and since the daisy graphs begin at NNLO in Landau gauge, Eq. (7.3) can immediately be implemented for many theories of direct phenomenological interest. The effect of using Eq. (8.10) instead of Eq. (8.9) for the Standard Model effective potential is discussed in [45].
## 9 Higher dimension operators
As we have seen, V min is gauge invariant but the vev at the minimum, v = 〈 φ 〉 is not. One reason one would like v to be physical is that it gives a criteria for when higher dimension operators can have a significant effect. For example, one might expect an operator ∆ L = -1 Λ 2 φ 6 to be relevant at field values where this term contributes of the same order as the leading one: Λ ∼ √ 1 λ v ( ) v . Since v is ξ -dependent, this criteria is not meaningful.
The correct way to evaluate the influence of a gauge-invariant higher-dimension operator is to add it to the classical potential and see how V min is affected. So let us add -1 Λ 2 φ 6 to the Lagrangian (that is, we add + 1 Λ 2 φ 6 to the classical potential). First, consider the case when Λ is very small (so this new term is large). Then the sensitive cancellation between λ and e 4 will be severely disrupted and the new quantum minimum will disappear. With the sign given, the only minimum would be at φ = 0 with V min = 0. If we flip the sign, then the potential will be unbounded from below and again there will be no quantum minimum, with V = 0 still the only extremum and clearly gauge invariant.
Now suppose this new term has a very small effect. To be precise, recall that the quantum minimum in the absence of this term occurs at leading order at the scale φ = µ X where Eq. (7.1) is satisfied. Thus if this term is very small, at LO, it shifts the minimum to
$$V _ { \min } = \mu _ { X } ^ { 4 } \left [ - \frac { 3 } { 8 } \frac { e ^ { 4 } \hbar } { 1 6 \pi ^ { 2 } } + \frac { \mu _ { X } ^ { 2 } } { \Lambda ^ { 2 } } + \mathcal { O } ( \frac { \mu _ { X } ^ { 4 } } { \Lambda ^ { 4 } } ) \right ]$$
This is gauge invariant, but trivially so.
The key observation is that to compute subleading corrections when the effect is small, we cannot simply plug the next-to-leading order vev into 1 Λ 2 φ 6 . Doing so would give a gaugedependent correction. Instead, we must carefully re-evaluate our perturbation expansion in the presence of this new term.
If we are careful we must commit to whether this new term is important or not. If we say it is important, then minimally, the new term in Eq. (9.1) must be comparable to the original one. That means that Λ 2 ∼ µ 2 X e 16 π 2 4 glyph[planckover2pi1] . This power-counting means that the 1-loop minimum is not at the scale µ X but rather at a new scale µ Z where
$$\lambda ( \mu _ { Z } ) \equiv \frac { \hbar } { 1 6 \pi ^ { 2 } } e ^ { 4 } ( \mu _ { Z } ) \left [ 6 - 3 6 \ln [ e ( \mu _ { Z } ) ] \right ] - 3 6 \frac { \mu _ { Z } ^ { 2 } } { \Lambda ^ { 2 } }$$
which gives
$$V _ { \min } = \mu _ { Z } ^ { 4 } \left [ - \frac { 3 } { 8 } \frac { e ^ { 4 } \hbar } { 1 6 \pi ^ { 2 } } - \frac { 1 } { 2 } \frac { \mu _ { Z } ^ { 2 } } { \Lambda ^ { 2 } } + \mathcal { O } ( e ^ { 6 } ) \right ]$$
As before, the first nontrivial check will be at order e 6 . To perform this check, one would need to compute the 2-loop graphs with this modification as well as the daisy resummation in the R ξ gauges.
The main implication of the discussion in this section is that arguments about the relevance of higher dimension operators must be refined to have a consistent perturbative expansion. For example, one might traditionally calculate φ min defined as the field value at the minimum in Landau gauge, and compare it to M Pl to see gravitational effects on the effective potential. This comparison is not gauge invariant. Instead, one must actually insert the new operators and see how physical quantities determined by the effective potential change. An example implementation of this procedure for the Standard Model is discussed in [45].
## 10 Conclusions
In this paper we have resolved a long-standing puzzle about gauge dependence in the Coleman-Weinberg model. We have shown that the potential at the minimum is gaugeinvariant at the first non-trivial order in perturbation theory. The invariance is only manifest if an appropriate power counting is used and if V min is expressed in terms of another gaugeinvariant scale, such as a value of the MS parameter µ where the couplings satisfy some relation. To establish gauge invariance required a calculation of the full 2-loop ColemanWeinberg potential in a general gauge to order e 6 as well as an infinite number of daisy diagrams which are also relevant at next-to-leading order.
Besides providing the first explicit check that a potential is gauge invariant at a loopgenerated minimum, our work has implications for how effective potentials should be used consistently. First of all, we showed that one cannot simply truncate to fixed order in the usual loop expansion. Some terms at a given order must be dropped, and other terms from diagrams of arbitrarily high-loop order must be included. Only then will the potential have desired gauge-invariance properties. At leading order in the appropriate power counting, only a finite number of diagrams contribute. At next-to-leading order, the first infinite class of diagrams becomes relevant in a general gauge and must be resummed. A propitious feature of Landau gauge, ξ = 0, is that this resummation is postponed to next-to-next-toleading order, competitive with 3-loop graphs in the normal loop expansion. Nevertheless,
even in Landau gauge at NLO, one must truncate the effective potential appropriately to have self-consistent results. This result is consistent with recent observations in [40,41].
We also discussed how the consistent use we advocate is compatible with an RG-improved effective potential. The key is to run the couplings first to a scale where the 1-loop potential, appropriately truncated, has its minimum. This scale is gauge-invariant. Then one can add corrections to the effective potential around that scale order-by-order in perturbation theory using couplings at that scale. This essentially amounts to dropping the field-strength renormalization contribution to the resummed effective potential. However, we are not performing any field redefinitions. Thus the same threshold corrections can be used to convert observables to MS parameters at the low scale, and then run to the scale near the quantum minimum. The effective potential computed this way is still gauge-dependent. Moreover, even in Landau gauge (or any other gauge), the potential is different if we first run the couplings and then calculate the potential or if we calculate the potential and then run it by solving its RGE. The difference is not an artifact of the truncation in perturbation theory: the exact potentials computed these two ways would differ. However, this difference, like the difference between potentials computed in different gauges, should not affect physical observables. We have shown that indeed it does not affect the value of the potential at the minimum.
Our work definitively resolves the question of how one physical quantity, the true vacuum energy in a theory, can be computed consistently in perturbation theory. We also know that S -matrix elements are gauge-invariant. There are of course many other quantities that effective actions are used to calculate, and it would be very interesting to understand definitively how they can be computed consistently as well. To compute the scalar and vector masses, one would need to resum the daisy contributions to the scalar and vector self-energy graphs at small momentum, supplementing the results of [29]. Presumably these daisies would cancel the contribution from 2-loop effective potential, leaving Kang's result intact. Another interesting quantity is the tunneling rate between two extrema. A traditional tunneling rate calculation depends on unphysical regions of the potential [46,47], away from the minima. However, any rate calculation should also depend on the kinetic terms in the effective action, so there are new opportunities for cancellation. Although there has been some progress in understanding how tunneling rates may be consistent [21,48], there are, to our knowledge, no explicit demonstrations.
At finite temperature, the critical temperature for a phase transition should be physical. The spectrum of gravity waves as the universe cools through this transition should also be physical. While some progress has been made on showing that these quantities can be computed in a gauge-invariant way in perturbation theory [12,19,42,49] the question does not seem to be completely resolved. Other apparently gauge-dependent quantities include the compactification radius in certain extra-dimensional models [34, 50] and the inflation scale in Higgs-inflation models [51].
Often the gauge-dependence in a quantity extracted from the effective potential originates in dependence on field values. For example, the instability scale in the Standard Model Λ , I defined as the value of φ where the potential goes negative [44], is gauge-dependent [52].
The scale where the Standard Model effective potential is maximal, Λ max , is also sometimes used to make physical arguments [53], despite its gauge-dependence. A common criterion for stability in the Standard Model is that the value v of the field at the minimum be less than M Pl . Since v is gauge-dependent, so is this criterion [54]. Even if one asks only that the potential at the minimum, V min be less than 0, which is a gauge-invariant criterion, one would presumably still want to know when this bound is affected by Planck-suppressed or other higher-dimension operators. To this end, we have sketched one possible solution in Section 9: one should add these new operators to the theory and see how V min changes. Since the new operators affect the perturbation expansion, this procedure is not equivalent to replacing φ → v in these operators and seeing how the vacuum energy changes. With these examples, we are optimistic that apparent gauge-invariance of many physical quantities can be eliminated by the self-consistent use of an appropriate perturbation expansion.
## 11 Acknowledgements
We thank M. Reece, M. Ramsey-Musolf, M. Strassler and A. Strumia for helpful discussions. The authors are supported in part by grant DE-SC003916 from the Department of Energy. AA is supported in part by the Stolt-Nielsen Fund for Education of the the AmericanScandinavian Foundation and the Norway-America Association. WF is supported in part by the Lord Rutherford Memorial Research Fellowship.
## References
- [1] W. Heisenberg and H. Euler, 'Consequences of Dirac's theory of positrons,' Z.Phys. 98 (1936) 714-732, arXiv:physics/0605038 [physics] .
- [2] J. S. Schwinger, 'On gauge invariance and vacuum polarization,' Phys.Rev. 82 (1951) 664-679.
- [3] G. Jona-Lasinio, 'Relativistic field theories with symmetry breaking solutions,' Nuovo Cim. 34 (1964) 1790-1795.
- [4] S. R. Coleman and E. J. Weinberg, 'Radiative Corrections as the Origin of Spontaneous Symmetry Breaking,' Phys.Rev. D7 (1973) 1888-1910.
- [5] R. Jackiw, 'Functional evaluation of the effective potential,' Phys.Rev. D9 (1974) 1686.
- [6] L. Dolan and R. Jackiw, 'Gauge Invariant Signal for Gauge Symmetry Breaking,' Phys.Rev. D9 (1974) 2904.
- [7] S. Weinberg, 'Perturbative Calculations of Symmetry Breaking,' Phys.Rev. D7 (1973) 2887-2910.
- [8] W. Fischler and R. Brout, 'Gauge Invariance in Spontaneously Broken Symmetry,' Phys.Rev. D11 (1975) 905.
- [9] J.-M. Frere and P. Nicoletopoulos, 'Gauge Invariant Content of the Effective Potential,' Phys.Rev. D11 (1975) 2332.
- [10] S. H. Tye and Y. Vtorov-Karevsky, 'Effective action of spontaneously broken gauge theories,' Int.J.Mod.Phys. A13 (1998) 95-124, arXiv:hep-th/9601176 [hep-th] .
- [11] W. Buchmuller, Z. Fodor, and A. Hebecker, 'Gauge invariant treatment of the electroweak phase transition,' Phys.Lett. B331 (1994) 131-136, arXiv:hep-ph/9403391 [hep-ph] .
- [12] D. Boyanovsky, D. Brahm, R. Holman, and D. Lee, 'The Gauge invariant effective potential: Equilibrium and nonequilibrium aspects,' Phys.Rev. D54 (1996) 1763-1777, arXiv:hep-ph/9603337 [hep-ph] .
- [13] G.-L. Lin and T.-K. Chyi, 'Gauge independent effective potential and the Higgs mass bound,' Phys.Rev. D60 (1999) 016002, arXiv:hep-ph/9811213 [hep-ph] .
- [14] N. Nielsen, 'Removing the gauge parameter dependence of the effective potential by a field redefinition,' arXiv:1406.0788 [hep-ph] .
- [15] N. Nielsen, 'On the Gauge Dependence of Spontaneous Symmetry Breaking in Gauge Theories,' Nucl.Phys. B101 (1975) 173.
- [16] R. Fukuda and T. Kugo, 'Gauge Invariance in the Effective Action and Potential,' Phys.Rev. D13 (1976) 3469.
- [17] W. Loinaz and R. Willey, 'Gauge dependence of lower bounds on the Higgs mass derived from electroweak vacuum stability constraints,' Phys.Rev. D56 (1997) 7416-7426, arXiv:hep-ph/9702321 [hep-ph] .
- [18] M. Gonderinger, H. Lim, and M. J. Ramsey-Musolf, 'Complex Scalar Singlet Dark Matter: Vacuum Stability and Phenomenology,' Phys.Rev. D86 (2012) 043511, arXiv:1202.1316 [hep-ph] .
- [19] H. H. Patel and M. J. Ramsey-Musolf, 'Baryon Washout, Electroweak Phase Transition, and Perturbation Theory,' JHEP 1107 (2011) 029, arXiv:1101.4665 [hep-ph] .
- [20] E. J. Weinberg and A.-q. Wu, 'Understanding Complex Perturbative Effective Potentials,' Phys.Rev. D36 (1987) 2474.
- [21] A. Strumia and N. Tetradis, 'A Consistent calculation of bubble nucleation rates,' Nucl.Phys. B542 (1999) 719-741, arXiv:hep-ph/9806453 [hep-ph] .
- [22] L. Abbott, 'Introduction to the Background Field Method,' Acta Phys.Polon. B13 (1982) 33.
- [23] M. D. Schwartz, Quantum Field Theory and the Standard Model . Cambridge Univ. Press, 2014.
- [24] I. Aitchison and C. Fraser, 'Gauge Invariance and the Effective Potential,' Annals Phys. 156 (1984) 1.
- [25] J. Do Nascimento and D. Bazeia, 'Gauge Invariance of the Effective Potential,' Phys.Rev. D35 (1987) 2490-2494.
- [26] K. Fujikawa, B. W. Lee, and A. I. Sanda, 'Generalized renormalizable gauge formulation of spontaneously broken gauge theories,' Phys. Rev. D 6 (Nov, 1972) 2923-2943. http://link.aps.org/doi/10.1103/PhysRevD.6.2923 .
- [27] C. Ford, I. Jack, and D. Jones, 'The Standard model effective potential at two loops,' Nucl.Phys. B387 (1992) 373-390, arXiv:hep-ph/0111190 [hep-ph] .
- [28] S. P. Martin, 'Two loop effective potential for a general renormalizable theory and softly broken supersymmetry,' Phys.Rev. D65 (2002) 116003, arXiv:hep-ph/0111209 [hep-ph] .
- [29] J. Kang, 'Gauge Invariance of the Scalar-Vector Mass Ratio in the Coleman-Weinberg Model,' Phys.Rev. D10 (1974) 3455.
- [30] C. Fraser, 'Calculation of Higher Derivative Terms in the One Loop Effective Lagrangian,' Z.Phys. C28 (1985) 101.
- [31] A. I. Davydychev and J. Tausk, 'Two loop selfenergy diagrams with different masses and the momentum expansion,' Nucl.Phys. B397 (1993) 123-142.
- [32] M. E. Machacek and M. T. Vaughn, 'Two Loop Renormalization Group Equations in a General Quantum Field Theory. 1. Wave Function Renormalization,' Nucl.Phys. B222 (1983) 83.
- [33] M. E. Machacek and M. T. Vaughn, 'Two Loop Renormalization Group Equations in a General Quantum Field Theory. 3. Scalar Quartic Couplings,' Nucl.Phys. B249 (1985) 70.
- [34] N. Nielsen, 'Gauge Dependence of Effective Potential and Selfconsistent Dimensional Reduction,' Z.Phys. C33 (1987) 579-582.
- [35] D. Johnston, 'Coleman-Weinberg, Nielsen and Daisies,' Phys.Lett. B186 (1987) 185.
- [36] D. Bazeia, 'Gauge Invariance and Daisies,' Phys.Lett. B207 (1988) 53-55.
| [37] | D. Kirzhnits and A. D. Linde, 'Macroscopic Consequences of the Weinberg Model,' Phys.Lett. B42 (1972) 471-474. |
|--------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [38] | S. Weinberg, 'Gauge and Global Symmetries at High Temperature,' Phys.Rev. D9 (1974) 3357-3378. |
| [39] | M. Carrington, 'The Effective potential at finite temperature in the Standard Model,' Phys.Rev. D45 (1992) 2933-2944. |
| [40] | S. P. Martin, 'Taming the Goldstone contributions to the effective potential,' arXiv:1406.2355 [hep-ph] . |
| [41] | J. Elias-Miro, J. Espinosa, and T. Konstandin, 'Taming Infrared Divergences in the Effective Potential,' arXiv:1406.2652 [hep-ph] . |
| [42] | C. Wainwright, S. Profumo, and M. J. Ramsey-Musolf, 'Gravity Waves from a Cosmological Phase Transition: Gauge Artifacts and Daisy Resummations,' Phys.Rev. D84 (2011) 023521, arXiv:1104.5487 [hep-ph] . |
| [43] | M. Quiros, 'On daisy and superdaisy resummation of the effective potential at finite temperature,' arXiv:hep-ph/9304284 [hep-ph] . |
| [44] | D. Buttazzo, G. Degrassi, P. P. Giardino, G. F. Giudice, F. Sala, et al. , 'Investigating the near-criticality of the Higgs boson,' JHEP 1312 (2013) 089, arXiv:1307.3536 . |
| [45] | A. Andreassen, W. Frost, and M. D. Schwartz, 'Consistent use of the standard model effective potential,' to appear . |
| [46] | J. Espinosa and M. Quiros, 'Improved metastability bounds on the standard model Higgs mass,' Phys.Lett. B353 (1995) 257-266, arXiv:hep-ph/9504241 [hep-ph] . |
| [47] | G. Isidori, G. Ridolfi, and A. Strumia, 'On the metastability of the standard model vacuum,' Nucl.Phys. B609 (2001) 387-409, arXiv:hep-ph/0104016 [hep-ph] . |
| [48] | D. Metaxas and E. J. Weinberg, 'Gauge independence of the bubble nucleation rate in theories with radiative symmetry breaking,' Phys.Rev. D53 (1996) 836-843, arXiv:hep-ph/9507381 [hep-ph] . |
| [49] | D. E. Morrissey and M. J. Ramsey-Musolf, 'Electroweak baryogenesis,' New J.Phys. 14 (2012) 125003, arXiv:1206.2942 [hep-ph] . |
| [50] | G. Kunstatter and H. Leivo, 'On the Gauge Dependence of the One Loop Effective Potential in Selfconsistent Dimensional Reduction,' Phys.Lett. B166 (1986) 321. |
| [51] | J. L. Cook, L. M. Krauss, A. J. Long, and S. Sabharwal, 'Is Higgs Inflation Dead?,' arXiv:1403.4971 [astro-ph.CO] . |
- [52] L. Di Luzio and L. Mihaila, 'On the gauge dependence of the Standard Model vacuum instability scale,' arXiv:1404.7450 [hep-ph] .
- [53] A. Hook, J. Kearney, B. Shakya, and K. M. Zurek, 'Probable or Improbable Universe? Correlating Electroweak Vacuum Instability with the Scale of Inflation,' arXiv:1404.5953 [hep-ph] .
- [54] W. Loinaz and R. Willey, 'Monte Carlo simulation calculation of critical coupling constant for continuum phi**4 in two-dimensions,' Phys.Rev. D58 (1998) 076003, arXiv:hep-lat/9712008 [hep-lat] . | 10.1103/PhysRevD.91.016009 | [
"Anders Andreassen",
"William Frost",
"Matthew D. Schwartz"
] | 2014-08-01T20:00:14+00:00 | 2014-08-01T20:00:14+00:00 | [
"hep-ph"
] | Consistent Use of Effective Potentials | It is well known that effective potentials can be gauge-dependent while their
values at extrema should be gauge-invariant. Unfortunately, establishing this
invariance in perturbation theory is not straightforward, since contributions
from arbitrarily high- order loops can be of the same size. We show in massless
scalar QED that an infinite class of loops can be summed (and must be summed)
to give a gauge invariant value for the potential at its minimum. In addition,
we show that the exact potential depends on both the scale at which it is
calculated and the normalization of the fields, but the vacuum energy does not.
Using these insights, we propose a method to extract some physical quantities
from effective potentials which is self-consistent order-by-order in
perturbation theory, including improvement with the renormalization group. |
1408.0288v3 | ## A new evaluation of the antiproton production cross section for cosmic ray studies
Mattia di Mauro 1 2 , , Fiorenza Donato , Andreas Goudelis , Pasquale Dario Serpico 1 2 2 1 Physics Department, Torino University, and Istituto Nazionale di Fisica Nucleare, Sezione di Torino, via Giuria 1, 10125 Torino, Italy and 2 LAPTh, Univ. de Savoie, CNRS, B.P.110, Annecy-le-Vieux F-74941, France
Theoretical predictions for the cosmic antiproton spectrum currently fall short of the corresponding experimental level of accuracy. Among the main sources of uncertainty are the antiproton production cross sections in cosmic ray inelastic interactions. We analyse existing data on antiproton production in pp scattering, including for the first time the measurements performed by the NA49 Collaboration. We compute the antiproton spectrum finding that in the energy range where data are available (antiproton energies of about 4-550 GeV) different approaches lead to almost equivalent results, with an uncertainty of 10-20%. Extrapolations outside this region lead to different estimates, with the uncertainties reaching the 50% level around 1 TeV, degrading the diagnostic power of the antiproton channel at those energies. We also comment on the uncertainties in the antiproton production source term coming from nuclei heavier than protons and from neutrons produced in pp scatterings, and point out the need for dedicated experimental campaigns for all processes involving antiproton production in collisions of light nuclei.
PACS numbers: 13.85.-t,13.85.Ni98.70.Sa
## I. INTRODUCTION
Cosmic ray (CR) antiprotons are a remarkable diagnostic tool for astroparticle physics. The bulk of the measured flux is certainly consistent with a purely secondary origin in CR collisions onto interstellar medium gas, but additional primary components are not excluded, either of astrophysical origin (see for instance [1]) or of exotic nature, such as dark matter annihilation or decay [2]. At very least, antiprotons provide a consistency check for the current understanding of galactic CR modeling and can narrow down propagation parameters (see e.g. [3-5]).
This tool is however only as sharp as the uncertainties entering the background (i.e. the secondary component) and signal (i.e. the primary component) computations are robust. Statistical and systematic errors reported by the PAMELA collaboration [6] are already at the 10% level up to the 10 GeV scale, below the theoretical error. In a short time, AMS-02 [7] is expected to provide significantly higher precision, calling for a reassessment of the theoretical predictions.
The contribution of different processes to the ¯ secp ondary yield has been studied in the past, see e.g. [810]. In [9], for instance, the uncertainties on the production cross sections were estimated to be ∼ 25%, and already identified as the limiting factor in theoretical predictions (see also [10] for similar considerations). In practice, nuclei heavier than protons and helium only contribute at a few percent level (see e.g. [8]), thus playing a very marginal role, either as projectiles or targets, in the antiproton production. Reactions involving helium (p-He, He-p, He-He) represent a sizable fraction of the total yield, easily reaching ∼ 50% at low energies [9].
While for processes involving helium nuclei no data is available, the situation is different for the proton-proton case, where there are several experimental studies. The latest re-evaluation of the antiproton production yield in pp collsions was reported in [11], while the Tan & Ng parameterization [12] is still largely used, despite being more than 30 years old. The reason is that, until recently, the available dataset was limited to data only collected in the sixties and seventies. In the last decade, however, two more experimental datasets have become available: the BRAHMS data [13] and-more important for the energies of interest for AMS-02 applications-the NA49 results collected at the CERN Super Proton Synchrotron (SPS) [14].
Given the importance of these nuclear data for new measurements in astroparticle physics, it seems thus timely to re-evaluate the antiproton production cross section in pp collisions in light of this newly available information. In this paper we engage ourselves in this task, in order to provide the community with a parametrization for the inclusive antiproton production cross section as well as with a reliable assessment of the corresponding uncertainties that should be taken into account in CR studies.
The outline of the paper goes as follows: in section II we set up the relevant formalism, present the experimental data that will be used subsequently and describe the analysis methods. In section III we present our results. We begin by validating our analysis framework by reproducing existing results in the literature and then move on to evaluate the inclusive antiproton production cross section, first relying solely on the novel NA49 data and then on the full set of available measurements. In section IV we briefly comment on the impact of other contributions entering the secondary antiproton source term, namely of antineutrons and helium nuclei. Finally, in section V we discuss our key results and present our conclusions. Two appendices follow, where we present some-standard but useful-kinematics used in our analysis as well as a new evaluation of the total, elastic and inelastic pp scattering cross sections that we performed for the energy range of
## II. FRAMEWORK, DATA AND METHODS
## A. Theoretical framework
CR protons interact with the interstellar medium (ISM) and may produce secondary antiprotons. Different channels are involved, with the dominant one being the CR proton flux collisions with the target hydrogen gas ( pp ). The corresponding source term is the convolution of the antiproton production cross section dσ p p → ¯ p dE ¯ p ( E , E p ¯ p ) and the interstellar CR proton energy spectrum
$$q _ { \bar { p } } ^ { p p } ( E _ { \bar { p } } ) & = \int _ { E _ { \text{th} } } ^ { + \infty } \frac { d \sigma _ { p \, p \rightarrow \bar { p } } } { d E _ { \bar { p } } } ( E _ { p }, E _ { \bar { p } } ) n _ { H } ( 4 \pi \Phi _ { p } ( E _ { p } ) ) d E _ { p }, \\ \text{when} \ \infty \ \sim \ \text{is} \ \mathbb { m } \ \mathbb { m } \ \text{when} \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m } \ \mathbb { m }$$
where n H is the ISM hydrogen density, Φ p is the CR proton flux, E p and E ¯ p are the CR proton and antiproton energies, and E th the production threshold energy equal to 7 m p . The overall ISM composition is H:He:C=1:0.1:5 × 10 -4 cm -3 [15]. Whenever needed for illustrative purposes, we will fix Φ p to the fit to the preliminary AMS-02 data [16] reported in [17].
The differential cross section dσ pp → ¯ p /dE ¯ p is in turn the integral over the angle ϑ between the incoming proton and the final state antiproton momenta
$$\frac { d \sigma _ { p p \to \bar { p } } } { d E _ { \bar { p } } } ( E _ { p }, E _ { \bar { p } } ) = 2 \pi p _ { \bar { p } } \int _ { \vartheta _ { \min } } ^ { \vartheta _ { \max } } E _ { \bar { p } } \frac { d ^ { 3 } \sigma } { d p _ { \bar { p } } ^ { 3 } } d ( - \cos \vartheta ), \ \ ( 2 ) \quad \text{(with} \quad \\ \text{TeV}, \quad \text{$\dots$}$$
where p ¯ p = √ E 2 ¯ p -m 2 p , ϑ min = 0 ◦ and the expression for ϑ max is given by Eq.(A15) in Appendix A. The integral in Eq.(2) is computed in the galactic frame at fixed antiproton energy E ¯ p . Its integrand represents the 'Lorentz-invariant distribution function' 1 for the process p + p → ¯+ p X , i.e. the inclusive antiproton production. Inclusive cross sections for processes of the form a + b → c + X can be described in terms of the Lorentzinvariant distribution function
$$f ( a + b \to c + X ) = E _ { c } \frac { d ^ { 3 } \sigma } { d p _ { c } ^ { 3 } } = \frac { E _ { c } } { \pi } \frac { d ^ { 2 } \sigma } { d p _ { L } d p _ { T } ^ { 2 } } = \frac { d ^ { 2 } \sigma } { \pi d y d p _ { T } ^ { 2 } } \,, \quad \begin{array} { c } \text{tion c} \\ 2 3 \text{], r} \\ \text{energy} \\ ( 3 ) \text{cover} \end{array}.$$
where p L , p T and y are respectively the longitudinal and transverse momentum and the rapidity of particle c . Traditionally, the independent variables most frequently used to parameterise this quantity are
- · the centre-of-mass (CM) energy √ s = √ 2 m E p ( p + m p ), which is uniquely fixed by the total incident proton energy in the lab frame, E p ;
1 For simplicity, in the rest of this work we will interchange this term with 'cross section'.
- · p T , the antiproton transverse momentum;
- · the so-called 'radial scaling' variable x R , defined as
$$x _ { R } = \frac { E _ { \bar { p } } ^ { * } } { E _ { \bar { p }, \max } ^ { * } } \,, \quad \quad \quad ( 4 )$$
where E ∗ ¯ p is the antiproton energy and E ∗ ¯ max p, is the maximal energy it can acquire, both defined in the CM frame. The maximal antiproton energy is (see Appendix A for details):
$$E _ { \bar { p }, \max } ^ { * } = \frac { s - 8 m _ { p } ^ { 2 } } { 2 \sqrt { s } } \,,$$
which-from the condition E ∗ ¯ max p, ≥ m p -also implies the threshold energy for the incident proton in the lab frame, E p ≥ 7 m p .
The inclusive antiproton production cross section cannot be computed from first principles. Our primary goal in this work is to provide reliable estimates for the magnitude and the uncertainties of the invariant distribution (3). Our results will be presented mostly in the form of suitable fitting functions. However, we also want to test how reasonable the Ansatz of the chosen functional form(s) is. To that purpose, we will also compare the fitted functions to an 'agnostic' spline interpolation of the data, which only requires a smooth, piecewise functional dependence. We will mainly focus on antiprotons with energies ranging from a few GeV and up to O (1) TeV, with the upper limit of this interval corresponding roughly to the highest energy that can be probed by AMS-02 and the lower one to the point where astrophysical uncertainties become so large that they constitute the dominant limiting factor in CR studies, a point which we will also briefly comment upon in section IV.
## B. Experimental data
In order to estimate the inclusive antiproton production cross section, we consider the datasets [13, 14, 1823], reported in Table I in terms of the centre-of-mass energy √ s and ( p T , x R ) regions (often not rectangular!) covered by each experiment. Note that not all experiments provide data in terms of these kinematic variables; in those cases, the data were first converted in terms of { s, p T , x R } . We report in Appendix A the straightforward but somewhat lengthy derivation of the transformation equations. The data are graphically illustrated in Fig.1. In the left panel, the cross section is shown as a function of E LAB ¯ p , for different combinations of p T and x R . In the right panel, the same data are seen in the p T -x R plane. The NA49 data cover wide ranges in both p T and x R , and describe lab antiproton energies from about 8 GeV up to 70 GeV.
Compared to the previous works [11, 12], the analyses of the NA49 [14] and BRAHMS [13] datasets are new to
TABLE I. Datasets used in our analysis along with their corresponding √ s values and ( p T , x R ) regions.
| Experiment | √ s (GeV) | p T (GeV) | x R |
|-----------------------------------|----------------------------------------|-------------------|---------------------|
| Dekkers et al , CERN 1965 [18] | 6 . 1, 6 . 7 | (0 ., 0 . 79) | (0 . 34 , 0 . 65) |
| Allaby et al , CERN 1970 [19] | 6 . 15 | (0 . 05 , 0 . 90) | (0 . 40 , 0 . 94) |
| Capiluppi et al , CERN 1974 [20] | 23 . 3, 30 . 6, 44 . 6, 53 . 0, 62 . 7 | (0 . 18 , 1 . 29) | (0 . 06 , 0 . 43) |
| Guettler et al , CERN 1976 [21] | 23 . 0, 31 . 0, 45 . 0, 53 . 0, 63 . 0 | (0 . 12 , 0 . 47) | (0 . 036 , 0 . 092) |
| Johnson et al , FNAL 1978 [22] | 13 . 8, 19 . 4, 27 . 4 | (0 . 25 , 0 . 75) | (0 . 31 , 0 . 55) |
| Antreasyan et al , FNAL 1979 [23] | 19 . 4, 23 . 8, 27 . 4 | (0 . 77 , 6 . 15) | (0 . 08 , 0 . 58) |
| BRAHMS, BNL 2008 [13] | 200 | (0 . 82 , 3 . 97) | (0 . 11 , 0 . 39) |
| NA49, CERN 2010 [14] | 17 . 3 | (0 . 10 , 1 . 50) | (0 . 11 , 0 . 44) |
this paper. Note that the BRAHMS centre-of-mass energy corresponds to an incident proton energy of roughly 21 TeV in the lab frame, which lies somewhat beyond the energy region of interest for our work. Given the absence of data for incident proton energies above ∼ 200 GeV, however, we have included this dataset since it can help in guiding the high-energy extrapolation of the fit to physical values. It is worth stressing that in the more interesting tens of GeV region for the antiproton laboratory energy, the major impact will be provided by far by the NA49 data.
Another important conceptual issue concerns the possibility to combine data-whose quality and robustness of error assessment is very diverse-in a global fit. There is no simple answer to this question: on one hand there are some systematic effects that are certainly present in the old data and hard to estimate and correct for. A known example is provided by the feed-down effect. A significant fraction of antiproton production (easily of O (20%)) comes from strange hyperon (Λ, Σ) decays, whose decay lengths are comparable to or larger than length scales of current micro-vertex detections or precision tracking. This effect was taken into account in the NA49 data analysis, where the contribution from hyperons has been subtracted from the measured yields . 2 For older experiments, no such correction was performed: while in some cases-as for the CERN ISR-it may be argued that reasonable estimates make the expected correction negligible, for fixed-target experiments covering an extended range of lab momenta the situation is somewhat more complicated. No a priori correction has been applied in the following for this effect, especially since ex-novo simulations of trajectories through the detectors and the collimators would be needed for robust estimates. For a semi-quantitative discussion, we address the reader to [14]. However, in deriving global fits, we shall allow for experiment-dependent renormalizations, which may account (at least in an averaged way) for such a correction, see below.
2 H. G. Fischer, private communication
On the other hand, relying only on contemporary data, notably NA49, means having the invariant cross section at only one point in √ s , i.e. at one beam energy. Then, in order to obtain the general cross section, one has to supplement the data with some additional theoretical assumption, such as the scaling hypothesis [24], namely that the cross section only depends on p T and x R . While this behaviour is expected to be approximately respected, notably at high √ s , its quantitative accuracy can only be gauged by comparison with experimental data. For this reason we decided to apply both strategies and to use either fits or interpolations, to all datasets or to NA49 only, with or without the scaling hypothesis, to assess the importance of these effects.
## C. Method
Our fits were performed with the MINUIT minimization package. Let us denote with k = 1 , . . . , L the different experimental datasets, with i k the i -th point of the dataset k and let C be the vector of the cross section parameters. The fitting procedure consists of varying the values of the cross section parameters C , comparing the theoretical cross section F s ( i k , x i k , p i k ; C ) with the data f i k ( s i k , x i k , p i k ) and finally finding the minimum of χ 2 ( C ) function defined below. This procedure gives the best-fit configuration C best with the corresponding 1 σ errors σ C . We define the χ 2 ( C ) function to be minimised in the fitting procedure in the following way:
$$\chi ^ { 2 } ( \mathbf C ) = \chi ^ { 2 } _ { \text{stat} } ( \mathbf C ) + \chi ^ { 2 } _ { \text{sys} }$$
where
$$\chi _ { s y s } ^ { 2 } = \sum _ { k = 1 } ^ { L } \frac { ( \omega _ { k } - 1 ) ^ { 2 } } { \epsilon _ { k } ^ { 2 } } \,,$$
and
$$\chi _ { \text{stat} } ^ { 2 } ( \mathbf C ) = \sum _ { k = 1 } ^ { L } \sum _ { i _ { k } } \frac { ( \omega _ { k } f _ { i _ { k } } - F ( s _ { i _ { k } }, x _ { R _ { i _ { k } } }, p _ { T _ { i _ { k } } }, \mathbf C ) ) ^ { 2 } } { \omega _ { k } ^ { 2 } \sigma _ { i _ { k } } ^ { 2 } } \,.$$
5
FIG. 1. The data on d σ 3 p p → ¯ p /dp 3 employed in our analysis are reported as a function of E LAB ¯ p (left panel) and in the p T -x R space (right panel). For the data details, see Table I.
In the equations above, glyph[epsilon1] k is a systematic overall scale error of the dataset k (either quoted in the experimental paper, or assumed conservatively to be of the same order of the statistical one if this information is not available, notably for older references); the parameter ω k renormalises the dataset k and is determined consistently by the global fit: of course, large renormalizations with respect to glyph[epsilon1] k are disfavoured by the large penalties to be paid in the global analysis; σ i k is the statistical error on the data point i k , while the factor ( ω f k i k -F /ω ) k is the difference between experimental values f (accounting for a possible renormalization ω k , unique for each dataset) and the fitting function F , which depends on the independent variables described above, and on the vector of fitting parameters C .
This method is the most natural generalization of the unbiased one presented in [25] (see therein equation (3) and section 4) and it has already been successfully used in the past for other astroparticle physics analyses involving combinations of different datasets, as for instance nuclear reaction rates in primordial nucleosynthesis [26].
Passing on to the data interpolations against which we will be comparing our fitting procedure results, one difficulty lies with the fact that 3-dimensional interpolation of scattered data is a non-trivial problem in contemporary numerical analysis, with very few (if any) relevant publicly available tools. In order to tackle this issue, in our interpolations we will be making the assumption that the invariant distribution (3) scales with √ s only through an overall multiplicative dependence on the inelastic pp scattering cross section σ in . Under this assumption, by dividing the experimental data with σ in we obtain a √ s -invariant set of points for which only a 2-dimensional interpolation is needed. Besides, as a by-product of our analysis we have re-evaluated the inelastic cross section as described in Appendix B.
The interpolations were performed by means of the Python routine SmoothBivariateSpline contained in the scipy library, choosing piecewise cubic polynomials as interpolating functions. Note that the routine does not actually perform an exact interpolation, rather finds a compromise between the smoothness of the interpolating function and the closeness to the experimental data.
Estimating a statistically meaningful uncertainty band in this approach is fairly tricky. What we did was to consider each experimental best determination and error as the mean and standard deviation of a gaussian probability distribution of the cross section at that point. We then sampled these distributions accordingly, thus creating a large number of pseudo-experimental points. Each set of points is then interpolated (and, depending on the variable one is interested in, eventually integrated over cos ϑ and the proton incident energy/proton flux) to obtain an 'envelope band' for the quantity of interest. The average between maximum and minimum of the envelope at each point then defines an 'average interpolation curve'.
## III. RESULTS
## A. Validation of fitting method
As a preliminary exercise, and in order to validate our kinematical data conversion and fitting routines, we checked if the fit of Eq. (6) in [11] is reproduced, of course restricting ourselves to the datasets available at the time [18-23]. The parameterization of the invariant cross sec-
tion is in this case:
$$E \frac { d ^ { 3 } \sigma } { d p ^ { 3 } } & = \sigma _ { \text{in} } ( s ) ( 1 - x _ { R } ) ^ { D _ { 1 } } e ^ { - D _ { 2 } x _ { R } } \quad \text{ or } \\ & \quad \left [ D _ { 3 } ( \sqrt { s } ) ^ { D _ { 4 } } e ^ { - D _ { 5 } p _ { T } } + D _ { 6 } e ^ { - D _ { 7 } p _ { T } ^ { 2 } } \right ] \,, \quad ( 9 )$$
where σ in is the total inelastic cross section for pp collisions for which here, and only here, we used the parameterization adopted in [11]
$$\sigma _ { \text{in} } ( s ) = \sigma _ { 0 } \left [ 1 - 0. 6 2 \, e ^ { - \frac { E _ { \text{inc} } ( s ) } { 2 0 0 } } \sin \left ( \frac { 1 0. 9 } { E _ { \text{inc} } ^ { 0. 2 8 } ( s ) } \right ) \right ] \,, \ \ ( 1 0 ) \quad \text{less} ),$$
where E inc ( s ) is the incident kinetic energy in MeV defined as E inc ( s ) = s/ (2 m p ) -2 m p and σ 0 = 44 40 mbarn. . We show in Tab. II our best fit values and 1 σ errors for the cross section pararemeters D i in Eq.(9). A modest disagreement with the results of [11] was initially found only for D 1 and D 2 , which eventually we could attribute to a typo in the fitting parameter values reported in their Table V. If we invert D 1 with D 2 , not only do we obtain a very good agreement with our results, but also a reduced chi-squared χ 2 ν glyph[similarequal] 3 6, the same value the authors quote . in their paper. By insisting in interpreting literally the values of their table V, we would find χ 2 ν glyph[similarequal] 9 9, . clearly inconsistent with the value of 3.6. In Fig. 2 we display the comparison of the best fit and 3 σ uncertainty band of the source term derived with our best fit values of the parameters in Tab. II, and the best fit source term with the results reported in [11] with D 1 and D 2 inverted. The two results are essentially indistinguishable.
FIG. 2. Comparison of the source term for antiproton production in pp collisions as obtained in [11] (see however text for a correction in their table) and in this work, by re-fitting the same datasets with the same functional form, with our nominal 3 σ statistical error band. Vertical dot-dashed lines show the domain of energy actually covered by the experiments analysed.
## B. Analysis of NA49 data
Once our routines validated, we proceed first with fits to the NA49 dataset only. We use the functional form
$$E \frac { d ^ { 3 } \sigma } { d p ^ { 3 } } & = \ \sigma _ { \text{in} } ( s ) ( 1 - x _ { R } ) ^ { C _ { 1 } } e ^ { - C _ { 2 } x _ { R } } \\ & \quad \left [ C _ { 3 } \, e ^ { - C _ { 4 } p _ { T } } + C _ { 5 } \, e ^ { - C _ { 6 } p _ { T } ^ { 2 } } \right ] \,, \quad \ \ ( 1 1 )$$
where σ in ( s ) is defined in Appendix B, Eq.(B2). This functional form is a simplified version of the standard parametrization proposed in [27] (it has four parameters less), which we found to provide an accurate and more compact description of the data. Note that we implicitly assume some form of scaling, in that the only dependence on s is given by the overall multiplication with the inelastic cross section. The best-fit values and the 1 σ errors are reported in Table III, with the corresponding fit having a reduced chi-square χ 2 ν = 1 3 for 137 degrees of freedom. . The comparison between data and fitted function (along with the corresponding 3 σ bands) is presented in Fig. 3. We see that the data are well represented by the fitting function, Eq.(11), for all the p T and x R values.
Next, we checked that the chosen fitting formula does not impose too strong a theoretical bias. To that purpose, as described in paragraph II C, we performed an 'educated' interpolation of the data by dividing the datapoints by σ in ( s NA49 ) and assuming that the resulting function is independent of s . The final function which is obtained by re-multiplying by σ in ( s ) thus still has a dependence on s , albeit a trivial one, via the overall factor σ in ( s ). The comparison between our fitting and interpolating procedures is shown in yellow in Fig. 4. The vertical lines correspond to the equivalent antiproton energy spanned by the NA49 experiment, where an interpolation is meaningful. In order to obtain a reasonable interpolation outside this interval, we supplemented the datasets with 'fake' points at the boundaries of the interpolation grid, with very large errors not to artificially influence the curve, yet sufficient to prevent the numerical routine from being driven to extreme functional form interpolations (for example, negative cross sections). No reasonable error can be however assigned outside the region covered by the data, apart for a lower limit that should be at least as large as the maximum relative width of the yellow band. The fact that the average interpolation curve is always within ∼ 3 σ of the best-fit previously obtained suggests that this 3 σ band is roughly representing the maximum uncertainty (at least where data exist), accounting not only for statistical errors, but also for possible theoretical biases, acting as additional systematics, related to the choice of the fitting function.
## C. Global analysis
Finally, we proceed to the global analysis of all available data on pp → ¯+ p X listed in Table I. In this case,
TABLE II. Best-fit values and 1 σ errors for the parameters D i in Eq.(9) resulting from a fit to the [11] dataset.
| D 1 (error) | D 2 (error) | D 3 (error) | D 4 (error) | D 5 (error) | D 6 (error) | D 7 (error) |
|----------------|----------------------|----------------|------------------|------------------|--------------------|------------------|
| 4 . 22(0 . 66) | 3 . 435(0 . 016) 0 . | 0067(0 . 0014) | 0 . 510(0 . 050) | 3 . 609(0 . 021) | 0 . 0209(0 . 0010) | 3 . 086(0 . 083) |
TABLE III. Best-fit parameters and 1 σ errors to the NA49 data [14] with Eq.(11) .
| C 1 (error) | C 2 (error) | C 3 (error) | C 4 (error) | C 5 (error) | C 6 (error) |
|----------------|--------------------|------------------|----------------|--------------------|------------------|
| 7 . 56(1 . 15) | 0 . 245(0 . 148) 0 | . 0164(0 . 0025) | 2 . 37(0 . 13) | 0 . 0352(0 . 0020) | 2 . 902(0 . 059) |
given that we wish to describe data referring to quite different √ s values and covering different ( p T , x R ) regions, it is expected that we will have to introduce some complication with respect to the previous paragraph. In this spirit, we tried numerous different functional forms, essentially variations of the standard parametrization proposed in [27]. We present here results on our two most successful attempts, which also provide interesting insights on the extrapolation to regions where data are either scarce or altogether unavailable, a point that we shall discuss in more detail in section V A.
As a first step, we used an improved version of Eq.(11) introducing an explicit dependence on s , namely
$$E \frac { d ^ { 3 } \sigma } { d p ^ { 3 } } & = \sigma _ { \text{in} } ( s ) ( 1 - x _ { R } ) ^ { C _ { 1 } } e ^ { - C _ { 2 } x _ { R } } & ( 1 2 ) \quad \text{$non o$} \\ & \quad \left [ C _ { 3 } ( \sqrt { s } ) ^ { C _ { 4 } } e ^ { - C _ { 5 } p _ { T } } + C _ { 6 } ( \sqrt { s } ) ^ { C _ { 7 } } e ^ { - C _ { 8 } p _ { T } ^ { 2 } } \right ]. \quad \text{$omit$} \quad \text{$in $th$}$$
This parametrization of the cross section is similar to the one proposed in [27] except for the absence of a √ s exponent in the (1 -x R ) C 1 term. The fit gives a reduced chi-square of χ 2 ν = 4 16, . with a number of degrees of freedom of 385. The best-fit values and uncertainties are reported in Table IV. We have also checked that considering the exact form as in [27] we obtain an even worse fit to the data ( χ 2 ν = 5 6 with 385 degrees of freedom). .
Motivated by the relatively poor quality of the fit, we tried an extended version of Eq.(12), namely
The improved value of χ 2 ν is obtained at the expense of some rescaling of the datasets. With respect to our best result given by Eq.(13), the measurements reported in [13, 14, 18-23] are renormalised respectively by factors ω k of { 0 87 . , 1 04 1 16 0 98 . , . , . , 0 95 . , 1 13 . , 1 02 1 16 . , . } . Therefore the NA49 data [14], which represent the bulk of the fitting procedure, are renormalised by a negligible ∼ 2% while [13, 18, 19, 23] by more than 10%. Interestingly, the largest renormalization value is 16% for the BRAHMS dataset [13] giving a factor of 1.16, still not particularly significant given the statistical errors, yet perhaps indicative of some 'theoretical error' effects which become more prominent when an agreement over a large energy range is demanded. We display in Fig. 5 the comparison of the cross section best fit and 3 σ uncertainty band according to Eq.(13) with the datasets [14, 18-23]. We omit the comparison with the BRAHMS results, because in this case the cross section has only been measured along a line in the ( p T , x R ) space (see Fig. 1). Nevertheless, the difference between our best fit cross section and the data in [13] is at most ∼ 30%. We see that most of the data are well reproduced by the fitting function of Eq.(13) at all p T values. This is true in particular for the NA49 data, except for a slight overestimation at the lowest p T value. We have however checked that a 20% shift in the differential cross section for p T < 0 15 has a . negligible effect on the antiproton source term (less than 5%).
$$E \frac { d ^ { 3 } \sigma } { d p ^ { 3 } } = \sigma _ { \text{in} } ( s ) ( 1 - x _ { R } ) ^ { C _ { 1 } } e ^ { - C _ { 2 } x _ { R } } \left | [ C _ { 3 } ( \sqrt { s } ) ^ { C _ { 4 } } e ^ { - C _ { 5 } p _ { T } } + \begin{array} { c } \text{We} \\ \text{In this} \end{array} \\ C _ { 6 } ( \sqrt { s } ) ^ { C _ { 7 } } e ^ { - C _ { 8 } p _ { T } ^ { 2 } } + C _ { 9 } ( \sqrt { s } ) ^ { C _ { 1 0 } } e ^ { - C _ { 1 1 } p _ { T } ^ { 3 } } ] \right |, \quad ( 1 3 ) \quad \text{there} \\ \text{points} \quad \text{$1$} \quad \text{$1$} \quad \text{$1$} \quad \text{$1$} \quad \text{$1$} \quad \text{$1$} \quad \text{$1$} \quad \text{$1$} \quad \text{$1$} \quad \text{$1$} \quad \text{$1$} \quad \text{$1$} \quad \text{$1$} \quad \text{$1$} \quad \text{$1$} \quad \text{$1$} \quad \text{$1$} \quad \text{$1$} \quad \text{$1$} \quad \text{$1$} \quad \text{$1$} \quad \text{$1$} \quad \text{$1$}$$
where the absolute value simply prevents the function from becoming negative in some corners of parameter space. Compared to the previous function, this one further contains the exponential of a cubic function of p T and an additional s -dependence. The best-fit parameters for Eq.(13) are reported in Table V. This parametrization yields a somewhat better χ 2 ν = 3 30 for 382 degrees . of freedom. In order to test the validity of the scaling hypothesis, we obtain that the fit to the same data with Eq.13 containing no dependence on s but not in σ in ( s ), gives a reduced χ 2 ν = 4 82. .
We then repeated the interpolation analysis, previously only performed for NA49, for the entire dataset. In this case, the parameter space coverage is such that there is no need to supplement the dataset with 'fake' points, as previously done for the NA49 data alone. The spline method results in this case are, thus, fully datadriven, modulo our implicit assumption concerning the cross section √ s scaling according to an overall factor σ in ( s ).
In Fig. 6 we compare the results obtained for the antiproton source term through our fits according to the equations (13) and (12), as well as the estimate based on our spline interpolation. The energy range where pp data (except for BRAHMS) are available is bracketed by the vertical lines. We see that above 10 GeV, and within the region where experimental data are available, all three
FIG. 3. Comparison between NA49 data with the fitting function of Eq.(11), see Table III, with 3 σ error bands. For clarity, the data and the theoretical curves at each p T value have been multiplied by a factor of 0 9 . n p T , where n p T is the integer counting the p T , from lower to higher ( i.e. : for p T = 0.60 GeV/ c 2 the rescaling is 0 9 . 6 ).
TABLE IV. Best-fit values and 1 σ errors for the parameters C i in Eq.(12) derived with a fit to all datasets.
| C 1 (error) | C 2 (error) | C 3 (error) | C 4 (error) | C 5 (error) | C 6 (error) | C 7 (error) | C 8 (error) |
|---------------|---------------|------------------|---------------|---------------|----------------|---------------|---------------|
| 4.499(0.040) | 3.41(0.11) | 0.00942(0.00083) | 0.445(0.027) | 3.502(0.018) | 0.0622(0.0086) | -0.247(0.049) | 2.576(0.027) |
TABLE V. Best-fit values and 1 σ errors for the parameters C i in Eq.(13) derived with a fit to all datasets.
| C 1 (error) | C 2 (error) | C 3 (error) | C 4 (error) C | 5 (error) C 6 (error) | |
|---------------|---------------|---------------------------|-----------------|---------------------------|----|
| 4.448(0.035) | 3.735(0.094) | 0.00502(0.00036) | 0.708(0.019) | 3.527(0.014) 0.236(0.024) | |
| C 7 (error) | C 8 (error) | C 9 (error) | C 10 (error) | C 11 (error) | |
| -0.729(0.036) | 2.517(0.027) | - 1 . 822(0 . 009)10 - 11 | 3.527(0.022) | 0.384(0.021) | |
methods yield compatible results. At lower and higher energies, however, there is a significant departure of the three estimates. We will discuss the implications of these results in much more detail in section V A.
In order to compare the results derived in this section to previous published proton-proton cross section estimates, we show in Fig. 7 the best fit and 3 σ uncertainty band source term calculated with our results and the best fit source term derived with the parametrizations adopted in [11, 12]. In the range of antiproton kinetic energy where data exist, our 3 σ band is marginally compatible with the parameterization in [11, 12], which is overestimated (underestimated) below (above) about 20 GeV.
fects of nuclear projectiles and targets in the collisions, and the yield coming from antineutron production. An exhaustive treatment of both subjects goes beyond our current purposes. For completeness, however, in the following we summarise the re-scalings of the yield from the pp → ¯ process that are usually adopted in the literature p to account for both the processes, and some of the issues involved.
## IV. CONTRIBUTIONS FROM NEUTRONS AND NUCLEI
In order to obtain the total antiproton source term, two more effects should be taken into account: the ef-
Concerning nuclear enhancements (effects of protonnucleus, nucleus-proton, and nucleus-nucleus collisions), unfortunately very little data are present, notably none for the most important channels which are the ones involving helium. One possible strategy to deduce cross sections for reactions involving helium is to constrain those of nuclear species for which some data are available, and extrapolate from heavier species to lighter ones, see e.g. [11]. Given that helium is quite light, however, it has often been considered reliable to deduce the relevant cross sections from rescaling the pp ones either with semiempirical formulae or via hadronic models, see e.g. [9]. The most recent dedicated studies were performed on
FIG. 4. Comparison between the function Eq.(11) fitted to NA49 data, with 3 σ error band (solid curve with cyan/blue shaded band), and interpolated curve (dashed red), with the interpolation envelope band, yellow/light shading. The dashed vertical lines correspond to the equivalent antiproton energy sampled by the NA49 experiment, where an interpolation is actually justified.
the basis of the Monte Carlo (MC) model DTUNUC in [28] and in [9]. The models implemented in the software are based on the Dual Parton Model [29] and the Gribov-Glauber approach for a unified treatment of soft and hard scattering processes. The former are parameterised according to Regge phenomenology whereas the latter rely on lowest order perturbative QCD. Eventually, DTUNUC formed the basis of/merged into DPMJET-III (see [30] and refs. therein for further details) 3 .
Afit of the nuclei enhancement yield of antiprotons found in [28] compared to the one in pp collision, is given in [10]:
$$Q _ { t o t } / Q _ { p p } = 0. 1 2 ( T _ { \bar { p } } / G e V ) ^ { - 1. 6 7 } + 1. 7 8 \,, \quad ( 1 4 ) \quad \text{anti} \text{.}$$
with the above expression assuming 10% density ratio of He to H nuclei. Note that this ignores different spectral indices and species-dependent spectral breaks, which have been reported by some experiments but have not been confirmed by preliminary results of AMS-02. To gauge their possible effect at high-energies, we address the reader to the brief discussion in [32].
Lacking empirical information for the most relevant channels involving helium, it is hard to assess the accuracy of the previous models. The overall uncertainty (on the total source term yield Q tot ) was estimated in [28] to reach 40%, from the dispersion of predictions based on different prescriptions, but this conclusion is overly pessimistic, since not all the models/evaluations have the same reliability (some were based on obsolete prescriptions, for instance). In [9] the error estimate was closer
3 Recently, some new theoretical evaluation appeared in the Appendix of [31].
to 20 -25%, provided that the pp cross section does not depart from the Tan & Ng parameterization [12] by more than 10%, which seems to be only marginally compatible with our results.
Since all these prescriptions do not include subtle nuclear effects, it is also likely that the uncertainty at low energies (where they are expected to play a larger role) is significantly larger than at high energy. If, in addition, one considers the more complicated astrophysical propagation effects at low-energies (reacceleation, convection, solar modulation) 4 and the need to correctly account for (catastrophic and non-catastrophic) energy losses, themselves affected by errors, it is clear that below a few GeV's the lower the energy the less reliable is the theoretical prediction. Most likely, this energy window cannot be used (but very crudely) for astroparticle physics constraints.
Another correction which is needed to infer the total antiproton flux from σ pp → ¯ p consists in accounting for the antiproton flux coming from antineutron production. Traditionally, it has been assumed that σ pp → ¯ n = σ pp → ¯ p , so that the previous results have been simply multiplied by a factor 2. However, the NA49 collaboration itself [34] has reported an isospin-dependence in studies of secondary yields in np and pp collisions: in pp reactions, there is a significant preference of the positively charged pn ¯ combination over ¯ pn (the opposite being true for neutron projectiles). This results in σ pp → ¯ n = κσ pp → ¯ p with κ glyph[similarequal] 1 5 . around x F ∼ 0 (see also Fig. 3 in [34]; x F is defined in Appendix A), although the effect depends on x F to some extent. Given the still rudimentary knowledge of these effects, an energy independent rescaling κ glyph[similarequal] 1 3 . ± 0 2 should encompass the data and be a . better approximation than the usually assumed κ = 1. It is clear that addressing these issues is of paramount importance for further reducing the uncertainties in the antiproton source term.
## V. DISCUSSION AND CONCLUSIONS
## A. Discussion
We now discuss our findings focusing on the global analysis outlined in Sect. III C. As we can see in Fig. 7, for antiproton energies lying roughly within the interval (10 300) GeV, we find that our results on the antipro-, ton source term from proton-proton scattering are consistent with previous estimates. They are moreover stable with respect to reasonable changes in the parametrization choice and in agreement with data-driven methods. These findings can be understood considering that the
4 It has been clearly shown as different propagation setups can be responsible for a ∼ 50% min-max dispersion in the flux expected at low-energy [33 ? ].
FIG. 5. Differental cross section for antiproton production in pp scattering, as a function of x R , for different p T values. The curves refer to the 3 σ uncertainty band around the best fit obtained with a fit of Eq.(13) to the datasets in Tab. I. The data, from top left to bottom right, are from [14, 18-23]. For the sake of clarity, the data from [19] and [14] and the relevant theoretical curves at each p T value have been rescaled by a factor 0 6 . n p T and 0 9 . n p T , respectively, as described in Fig.3.
FIG. 6. Comparison between fitted function of Eq.(13) with 3 σ band (solid curve with cyan/blue shaded band), of Eq.(12) with 3 σ band (dot-dashed curve with green/light green shaded band) and interpolated curve (dashed red), with the interpolation envelope band, red/orange shading. The dashed vertical lines correspond to the equivalent antiproton energy sampled by the global dataset, where an interpolation is in principle meaningful.
FIG. 7. The best fit and 3 σ uncertainty band source term derived with the fit of Eq.(12) and Eq.(13) to all datasets in Tab. I is shown together with the source term obtained using [11, 12] cross section parametrizations.
majority of the data lie in the T ∈ (10 300) GeV range, , where the most reliable estimates of the distribution in Eq.(3) can be obtained and which, even prior to the NA49 and BRAHMS measurements, were already discreetly populated with data. In this sense, given that the NA49 data are not in contradiction with previous experimental results, it is expected (and verified) that the estimates presented in Tan & Ng [12] and Duperray et al [11] are in good agreement with our findings for this energy range. Moreover, as long as a reasonable functional form is adopted for the invariant distribution, it is more or less bound to predict a comparable source term within this energy range. The small discrepancies of our spline interpolation and fitting approaches could be likely attributed to the fact that the interpolation essentially neglects the scaling violation , while the fits do allow for some flexibility (extra dependence on s ) to accommodate it.
On the other hand, at low and high energies, the relatively small amount of available data essentially implies extrapolations of the fits performed principally for T between 8 and 300 GeV. Consequently, moderately different assumptions can yield significantly different results. This is demonstrated by the fact that adopting two slightly different parameterizations while using the same dataset changes the high-energy source term prediction quite dramatically. Moreover, these findings are insensitive to the inclusion or not of the BRAHMS data in the analysis, which means that the results in [13] are not sufficient to constrain the high-energy behavior of the invariant distribution and, hence, the antiproton source function. This is due to the fact that the data of [13], only cover the exponentially suppressed highp T region (similarly to the ones of [23]), see Fig. 1. In this sense, both the lowenergy and high-energy behavior of the invariant distribution remain highly uncertain. Given that both the spline method and the fit with Eq.(13) demonstrate a similar trend at high energies, we believe that making any conclusive statement concerning the high-energy behaviour of the antiproton inclusive cross section would be risky. This is all the more the case since spline interpolations can be notoriously misleading when extrapolated outside data-covered regions.
Whereas in the low-energy regime this point is not very important, given that in any case the secondary antiproton flux is dominated by huge uncertainties coming from astrophysical sources (solar modulation, propagation parameters, antiproton scattering cross sections), it is plausible that in the region of several hundreds of GeV and higher the main uncertainty is still due to the antiproton production cross section.
We summarise in Table VI the pp -induced source term along with the associated percentage uncertainties resulting from our analysis of the NA49 data according to Eq.(11), our global analysis according to Eqs.(12) and (13), our spline interpolation method of the full dataset, and the previous estimates in [12] and [11], for a few representative values of the antiproton energy. This table simply illustrates the results reported in Figs. 6 and 7: with increasing energy, the different approaches turn from marginally compatible (at the lowest energies, few GeV) to fully compatible until, towards the end of the region for which experimental data are available, they yield very different results.
Concerning the error estimates, we also point out that the nominal 1 σ error band obtained from the χ 2 minimization procedure is underestimated, for several reasons. In some case where χ 2 /dof is close to 1, as in the fit to NA49 data only with a simple fitting formula, we showed how the agreement with an interpolation method is only meaningful if roughly a 3 σ band is used as typical estimate of the error. This is the choice we presented in
TABLE VI. Best-fit values and corresponding percentage relative errors for the pp -induced source term (in GeV -1 cm -3 s -1 ), for some representative antiproton energies and different approaches in the data analysis.
| T (GeV) | Eq.(11) (% error) | Eq.(12) (% error) | Eq.(13) (% error) | spline (% error) | Tan & Ng | Duperray et al |
|-----------|--------------------------|--------------------------|--------------------------|------------------------------|------------------|------------------|
| 5 | 1 . 23 · 10 - 30 (4 . 9) | 1 . 47 · 10 - 30 (6 . 1) | 1 . 67 · 10 - 30 (5 . 4) | 1 . 38 · 10 - 30 (2 . 7) | 1 . 42 · 10 - 30 | 1 . 40 · 10 - 30 |
| 10 | 4 . 31 · 10 - 31 (4 . 2) | 4 . 87 · 10 - 31 (3 . 0) | 5 . 17 · 10 - 31 (4 . 8) | 4 . 34 · 10 - 31 (2 . 5) | 4 . 96 · 10 - 31 | 4 . 74 · 10 - 31 |
| 100 | 1 . 70 · 10 - 33 (5 . 9) | 1 . 82 · 10 - 33 (8 . 7) | 1 . 77 · 10 - 33 (6 . 8) | 2 . 03 · 10 - 33 (3 . 2) | 1 . 82 · 10 - 33 | 2 . 04 · 10 - 33 |
| 500 | 2 . 42 · 10 - 35 (6 . 2) | 2 . 82 · 10 - 35 (9 . 5) | 3 . 39 · 10 - 35 (8 . 8) | 3 . 26 · 10 - 35 (5 . 2) | 2 . 38 · 10 - 35 | 3 . 27 · 10 - 35 |
| 1000 | 3 . 13 · 10 - 36 (6 . 9) | 4 . 16 · 10 - 36 (11) | 6 . 83 · 10 - 36 (10) | 7 . 02 · 10 - 36 (5 . 8) 3 . | 29 · 10 - 36 | 4 . 93 · 10 - 36 |
our plots. A similar prescription was found to be more indicative of the real uncertainty, once global fits were performed. In this case, the inadequacy of the nominal 1 σ error band was already hinted to by the relatively large reduced χ 2 , never smaller than χ 2 ν = 3 30. . We attribute these results to a combination of factors: i) underestimated experimental errors, notably in (some of) the older datasets, due to effects that were neglected as the feed-down we mentioned. ii) inadequacy of any simple functional form tested to describe faithfully the data, especially on a large dynamic range; iii) some sort of more or less implicit analytical extrapolation assumption in order to achieve coverage of the 3-dimensional space ( √ s, p T , x R ) starting from a discrete set of points. Note that this also applies to interpolation techniques, which for instance rely on some theoretical assumptions such as scaling. The situation may be certainly improved if high-quality measurements such as the ones provided by NA49 could be extended to a broader dynamic range.
We also stress that outside the regions where data are available, there is no compelling reason for either one of our results according to equations (12) and (13) to be more realistic than the other. Whereas the agreement of all of our computations at intermediate energies hints that the error estimates there is fairly reliable, this is not at all the case at very low and high energies. A more conservative approach is to assume that in this case the error is dominated by the extrapolation uncertainty, for which a proxy is given by the region spanned by the ensemble of our approaches, amounting to about 50% at 1 TeV.
As a practical summary of our analysis, we report in Fig. 8 an estimate for the uncertainties inherent to the production of antiprotons from inelastic pp scatterings. The results are expressed as the ratio of the antiproton source term in Eq.(1) to a reference value. For the blue and the red bands, this reference value has been fixed to the source term obtained by setting the pp production cross section to the best fit to all the data obtained with Eq. (13) (parameters as in Table V). Outside the vertical bands-delimiting the energy range in which data are available-we extrapolate the production cross section by means of the same formula.
The blue band corresponds to considering parametriza-
FIG. 8. Estimate of the uncertainties in the antiproton source term from inelastic pp scattering. The blue band indicates the 3 σ uncertainty band due to the global fit with Eq.(13), while the red band corresponds to the convolution of the uncertainties brought by fits to the data with Eq.(13), Eq.(12) and with the spline interpolation (see Fig.6.). The orange band takes into account the contribution from decays of antineutrons produced in the same reactions. Vertical bands as in Fig.6. See text for details.
tion (13) alone. By simple inspection we can clearly see that the relevant uncertainty is maximally of the order of 10%. The red band is obtained by convoluting the uncertainty bands resulting from fits through Eqs.(13) and (12) and (within the vertical bands) the spline interpolation. This more conservative approach sizes the uncertainties from 20% at the lowest energies to the extrapolated 50% at 1 TeV.
The most conservative estimate is shown by the orange band, where the additional uncertainty on the antineutron production has been taken into account. In this case, the normalization has been fixed to a source term in which the antineutrons produced in pp scatterings contribute with an energy-independent rescaling factor κ = 1 3 (w.r.t. . 1). The relevant uncertainty band has been derived by shifting the (red) previous convolution by an additional factor to account for the antineutron decay, κ glyph[similarequal] 1 3 . ± 0 2, . as discussed in Sect. IV. The orange band indicates that the antiproton source term may
vary by 30% at 1 GeV, and by up to more than 50% at 1 TeV. In the energy range where scattering data are available (between 4 GeV and 550 GeV), the uncertainty is of O (20 -30%) non-symmetrically around the reference source term, with the ambiguity increasing with energy.
Finally, although we did not include an analysis of the uncertainties due to the contribution of nuclei to the yield, it is obvious that the total relative uncertainty cannot be smaller than the one showed above.
## B. Conclusions
In this work we have performed a re-evaluation of the Lorentz-invariant distribution for inclusive antiproton production in pp collisions in light of the recent results from the NA49 and BRAHMS experiments. We have combined these observations with older measurements at different centre-of-mass energies in order to extract a reliable estimate both for the average value of the cross section and for the current status of the corresponding uncertainties. Our main results have been presented in sections III B, III C and IV and summed up as handy fitting functions.
We have paid special attention in quantifying the extent to which the functional form we adopt introduces theoretical bias, by comparing our primary results to those obtained by using different parametrizations, datadriven estimates or different subsets of the data. We are therefore confident that the uncertainties quoted throughout our paper are robust (cf also the discussion in section V A).
Our findings show that, despite the experimental progress in measuring the inclusive antiproton production cross section, uncertainties persist. In most of the well-constrained intermediate antiproton energy range (few GeV - few hundreds GeV), these uncertainties can be as low as 10 -15%. At higher energies, we have shown that our knowledge is much worse, with extrapolations leading to errors larger than ∼ 50% at 1 TeV.
A complete determination of the antiproton yield from pp scattering must include the antineutron decay contribution, usually assumed to be identical to the antiproton one. A recent measurement has reported a significant isospin dependence, which could amount to a slight increase of the antiproton source term and be accompanied by a non negligible uncertainty in the overall flux.
Last but not least, we remind that a significant contribution to the cosmic antiproton flux is due to the reactions involving helium nuclei, both in the incoming radiation and in the ISM. The relevant cross sections have never been measured, and the corresponding antiproton source term can be estimated only by relying on some theoretical models and extrapolations, introducing an additional uncertainty which is hard to quantify.
We expect that for the few years to come, these uncertainties will continue imposing non-negligible limitations on the interpretation of cosmic antiproton data, expected to be measured by AMS-02 with an unprecedented accuracy. As a side-effect, it appears unlikely that any definite conclusion for dark matter indirect detection could be drawn from a relatively featureless 'excess' in the antiproton yield, expected to be be below a few tens of percent of the overall flux, unless perhaps correlated features are found in several different cosmic ray channels. At very low energies, the situation is further aggravated by the poor knowledge of astrophysical parameters, plus additional nuclear and particle physics uncertainties related e.g. to non-catastrophic energy losses of antiprotons.
Despite current limitations, many of these sources of error are not irreducible, but could be addressed with dedicated experimental campaigns. As we stressed, the key advances would come from planning experiments with an extended coverage in the ( p T , x R ) plane, like NA49, with higher energies, with helium target/projectile, and by providing systematic analyses of anti-neutron yields. Another very useful enterprise would be a re-analysis of existing datasets with the aim of assessing the feed-down corrections associated to different experiments. Indeed, the CR antiprotons must include the contributions from hyperon decays, while the current databases are not uniform in that respect.
Cosmic ray antiprotons offer an important tool to address a number of astrophysical and astroparticle questions, as certified by the resources invested in balloon-borne and space-borne detectors in recent years. We believe that a nuclear and particle physics commitment to measure many of these missing ingredients should be identified as a strategic task, to provide astroparticle missions with the crucial laboratory input to fully exploit their results.
Note Added: During the completion of this work, we became aware of a study by R. Kappl and M. W. Winkler, arXiv:1408.0299. On one side, it focuses on NA49 data interpolation rather than performing a global dataset reanalysis with fitting formulae, like ours. On the other hand, it extends to propagated fluxes including other channels as well. As far as a comparison is possible, our results agree with theirs within quoted errors.
## ACKNOWLEDGMENTS
We would like to thank H. G.Fischer for valuable information about NA49 experiment, D. Maurin for stimulating discussions and for careful reading of the manuscript, and R.H. Bernstein for spotting a typo in Eq. (10) and another one in Table (II) in an earlier version of this manuscript. A.G. would like to acknowledge useful discussions with P. Aurenche on issues in hadron physics. M.D.M. would like to thank Andrea Vittino for interesting and useful discussions on antiprotons secondary production uncertainties. At LAPTh, this activity was developed coherently with the research axes supported
by the ANR Labex grant ENIGMASS. F.D. further acknowledges support from AAP-'LeSAdHE' of University of Savoie.
## Appendix A: Useful kinematics
In order to extract the invariant distribution of Eq.(3), we have transformed all data (when needed) in the CM frame, expressed them in the form ( p T , x R , f ( √ s, p T , x R )) and then performed our fit and interpolation as described in the main text. Here we collect for convenience some useful formulae by means of which these conversions can be done. In what follows, all kinematic variables carrying stars are defined in the CM frame whereas those carrying LAB superscripts are defined in the lab frame.
Note that experiments [18, 19, 23] and [14] are fixed-target experiments, whereas [20-22] and [13] are colliding beam ones. With the exception of NA49 [14], fixed-target experiments give their results in the LAB (or, equivalently in this case, target) frame. All other datasets are given in the CM frame.
The experiments [18] and [19] do not provide results for the quantity defined in (3), rather for d σ/d 2 Ω dp which can be recast as [35]
$$\frac { d ^ { 2 } \sigma } { d \Omega d p } = p _ { \tilde { p } } ^ { L A B, 2 } \frac { d ^ { 3 } \sigma } { d p ^ { 3 } }. \quad \quad ( \text{A1} ) \quad \text{all}$$
This leads to
$$E _ { \tilde { p } } ^ { L A B } \frac { d ^ { 3 } \sigma } { d p _ { \tilde { p } } ^ { 3 } } = \frac { E _ { \tilde { p } } ^ { L A B } } { p _ { \tilde { p } } ^ { L A B, 2 } } \frac { d ^ { 2 } \sigma } { d \Omega d p } \quad \quad ( A 2 ) \quad \text{Th} \quad \text{res} \quad \text{ }$$
which can be computed straightforwardly from the available data using E LAB ¯ p = √ p LAB, 2 ¯ p + m 2 p .
The results are given as a function of the transverse momentum component p T and ϑ LAB , the ¯ emission anp gle with respect to the incoming proton beam in the lab frame. Since p T remains unchanged in the two reference frames, one needs to compute the x R values which the measurements correspond to. We have that
$$E _ { \tilde { p } } ^ { * } = \gamma ^ { C M } ( E _ { \tilde { p } } ^ { L A B } - v ^ { C M } p _ { \tilde { p } } ^ { L A B } \cos \vartheta ^ { L A B } ) \quad ( \mathbf A 3 ) \quad \text{From}$$
where v CM and γ CM are the velocity of the CM reference frame with respect to the lab one and the corresponding Lorentz boost. These are given by
$$v ^ { C M } & = p _ { p } ^ { L A B } / ( E _ { p } ^ { L A B } + m _ { p } ) \quad \text{(A4)} \quad \text{and} \\ & = p _ { p } ^ { L A B } / ( \sqrt { p _ { p } ^ { L A B ^ { 2 } } + m _ { p } ^ { 2 } } + m _ { p } ) \quad \text{$\text{It is}$} \\ \gamma ^ { C M } & = ( E _ { p } ^ { L A B } + m _ { p } ) / \sqrt { s } \,. \quad \text{$\text{For}$}$$
Then, direct use of the definition (4) allows us to compute x R .
Experiments [20, 21] and [14] provide results for f ( p + p → ¯+ p X ) in the CM frame, as a function of p T and the alternative Feynman scaling variable x F defined as
$$x _ { F } = \frac { 2 p _ { L } ^ { * } } { \sqrt { s } } \simeq \frac { p _ { L } ^ { * } } { p _ { L, m a x } ^ { * } } \,, \quad \quad ( A 5 )$$
where p ∗ L is the antiproton longitudinal momentum and p ∗ L,max is its maximum possible longitudinal momentum. It is thus necessary to find the correspondence between the x R and x F scaling variables. First, from p ∗ 2 L = p ∗ 2 -p ∗ 2 T and p 2 = E 2 -m 2 we get
$$E _ { \bar { p } } ^ { * 2 } = p _ { L } ^ { * 2 } + m _ { p } ^ { 2 } + p _ { T } ^ { * 2 }. \text{ \quad \ \ } ( \text{A6} )$$
whereas E ∗ ¯max p is given by (5). Direct use of (A5) and (4) then gives
$$x _ { R } = \frac { \sqrt { x _ { F } ^ { 2 } ( s / 4 ) + m _ { p } ^ { 2 } + p _ { T } ^ { 2 } } } { E _ { \bar { p }, \max } ^ { * } } \,. \quad \quad ( A 7 )$$
Reference [23] provides results for f ( p + p → ¯ + p X ) in the CM frame as a function of p T and ϑ LAB , also providing the values ϑ ∗ of the angle in the CM frame. The p T values remain unchanged in the CM frame. So all that is left is to calculate x R . Since p ∗ ¯ p = p T / sin ϑ ∗ , by using E = √ glyph[vector] p 2 + m 2 p and equation (A13), we can calculate x R through (4).
The BRAHMS experiment [13] chooses to give its results for the invariant cross section
$$\frac { 1 } { 2 \pi p _ { T } } \frac { d ^ { 2 } \sigma } { d p _ { T } \, d y } = E _ { \tilde { p } } \frac { d ^ { 3 } \sigma } { d p _ { \tilde { p } } ^ { 3 } } \quad \quad ( A 8 )$$
as a function of the antiproton transverse momentum p T and the antiproton rapidity in the CM frame, y ∗ , defined as
$$y ^ { * } = \frac { 1 } { 2 } \ln \left ( \frac { E ^ { * } + p _ { L } ^ { * } } { E ^ { * } - p _ { L } ^ { * } } \right ). \quad \quad \left ( A 9 \right )$$
From this definition and E ∗ 2 ¯ p = p ∗ 2 L + p 2 T + m 2 ¯ p we get
$$E _ { \bar { p } } ^ { * } = \sqrt { m _ { \bar { p } } ^ { 2 } + p _ { T } ^ { 2 } } \cosh y. \text{ \quad \ \ } ( A 1 0 )$$
Then, x R can be calculated by means of this relation and (4).
It is maybe useful to report the computation of the maximal antiproton energy E ∗ ¯max p introduced in 5. For a general inclusive reaction a + b → c + X in the CM frame we have
$$\sqrt { s } = E _ { c } ^ { * } + E _ { X } ^ { * } \. \quad \quad \quad ( A 1 1 )$$
By replacing E ∗ c , E ∗ X through E = √ p 2 + m 2 , squaring the resulting relation, reintroducing the energies in the crossed terms, eliminating E X in favour of √ s and E ∗ c , using p ∗ 2 -E ∗ 2 = -m 2 to eliminate p ∗ 2 and solving for E ∗ c we get
$$E _ { c } ^ { * } = \frac { s + m _ { c } ^ { 2 } - m _ { X } ^ { 2 } } { 2 \sqrt { s } } \,, \quad \quad \ ( \text{A12} ) \quad \ \text{with} \quad \text{where}$$
where we introduced for compactness a slight abuse of notation for the X system by assigning it a mass variable m X . In reality m X simply refers to internal energy of the X system.
Now, for s and m p fixed, we see that E ∗ ¯ max p, is obtained through (A12) once the energy m X of the X system becomes minimal. Conservation of baryon number fixes m X, min = 3 m p (production of 3 additional protons at rest). So we get
$$E _ { \bar { p }, \max } ^ { * } = \frac { s - 8 m _ { p } ^ { 2 } } { 2 \sqrt { s } } \,. \quad \quad ( A 1 3 ) \quad \text{ } T A$$
Finally we derive the maximum antiproton angle with respect to the incoming proton in the laboratory frame ϑ LAB max . This quantity can be derived using the condition x R ≤ 1 and the definition of x R in Eq.(4). Then writing E ∗ ¯ p in the laboratory frame (see Eq.(A3)) the condition x R ≤ 1 can be written as
$$x _ { R } = \frac { \gamma ^ { C M } 2 \sqrt { s } ( E _ { \tilde { p } } ^ { L A B } - v ^ { C M } p _ { \tilde { p } } ^ { L A B } \cos \vartheta ^ { L A B } ) } { s - 8 m _ { p } ^ { 2 } } \leq 1 \,. \quad \begin{array} { c } \text{we are} \\ \text{data} \\ \text{in the} \\ \text{$\text{GeV.}$} \\ \text{avail} \end{array}$$
By using the definition of v CM and γ CM given in Eq.(A4) we get
$$\cos \vartheta _ { m a x } ^ { L A B } = \frac { E _ { \tilde { p } } ^ { L A B } ( E _ { p } ^ { L A B } + m _ { p } ) - m _ { p } ( E _ { p } ^ { L A B } - 3 m _ { p } ) } { \sqrt { E _ { \tilde { p } } ^ { L A B } - m _ { p } ^ { 2 } } \sqrt { E _ { p } ^ { L A B } - m _ { p } ^ { 2 } } } \,. \quad \text{ the n} \quad \text{to } \text{ca} ] \\ ( A 1 5 ) \quad \text{in } T a l \\ \text{if} \quad \text{$\#\cdot\cdot\cdot\cdot\cdot\cdot\cdot$} }$$
## Appendix B: Inelastic cross section parameterization
The inelastic proton cross section is defined as the difference between the total pp scattering cross section σ pp tot and its elastic counterpart σ pp el
$$\sigma _ { \text{in} } ^ { \text{pp} } = \sigma _ { \text{tot} } ^ { \text{pp} } - \sigma _ { \text{el} } ^ { \text{pp} }. \quad \quad \ ( B 1 ) \quad \text{ag} _ { \text{el} }$$
In order to estimate σ pp in for our energy region of interest, we employ the experimental data provided by the
[1] P. Blasi and P. D. Serpico, Phys.Rev.Lett. 103 , 081103 (2009), arXiv:0904.0871 [astro-ph.HE].
[2] P. Salati, F. Donato, and N. Fornengo, (2010), arXiv:1003.4124 [astro-ph.HE].
Particle Data Group (PDG) on the total and elastic pp cross sections [36]. We fit this data by means of the highest-ranking parametrization of the total proton cross section suggested by the PDG itself, which reads
$$\sigma _ { t o t } ^ { \text{pp} } & = Z ^ { \text{pp} } + B ^ { \text{pp} } \log ^ { 2 } ( s / s _ { M } ) + Y _ { 1 } ^ { \text{pp} } ( s _ { M } / s ) ^ { \eta _ { 1 } } - Y _ { 2 } ^ { \text{pp} } ( s _ { M } / s ) ^ { \eta _ { 2 } } \\ ( \text{A12} ) \quad. \quad \underset { \theta \theta } { \ e \theta } \quad \theta \theta \quad \theta \quad \theta \quad \dots \quad \theta \quad \dots \quad \theta \quad \dots$$
where B pp = π ( glyph[planckover2pi1] c ) 2 /M 2 , s M = (2 m p + M ) 2 , all energies are given in GeV and all cross sections in mb.
TABLE VII. Fit results for the total and elastic proton scattering cross sections according to Eq.(B2).
| Parameter | Total | Elastic |
|-------------|---------|----------------|
| M | 2.06 | 3.06 |
| Z pp | 33.44 | 144.98 |
| Y pp 1 | 13.53 | 2.64 |
| Y pp 2 | 6.38 | 137.27 |
| η 1 | 0.324 | 1.57 |
| η 2 | 0.324 | -4.65 × 10 - 3 |
Although this parametrization is given for the total proton cross section, noticing the resemblance in the √ s dependence of both the total and the elastic cross section, we employ the same functional form in order to fit the data on both σ pp tot and σ pp el . In both cases, we include in the fitting procedure only data referring to √ s > 5 GeV. Moreover, given the large amount of relevant data available for low √ s values as opposed to the relative scarcity of measurements for √ s > 80 GeV, we include the highest-energy data contained in the sample in order to capture the cross section behaviour over our entire energy range of interest in a reliable manner.
The fitting parameters for both σ pp tot and σ pp el are shown in Table VII. The resulting functions provide very good fits to the available data, with a χ 2 per degree of freedom of 0 78 for the total cross section and 1 5 for the elastic . . case, where statistical and systematic uncertainties have been added in quadrature. With these two parametrizations at hand, we then compute the inelastic cross section through its definition (B1). We have verified that, modulo small differences due to the specific goals of our analysis described before, our results are in very good agreement not only with the PDG results on σ pp tot but also with the function for σ pp el quoted by all LHC experiments (see e.g. [37, 38]) as well as with the results on both σ pp tot and σ pp el by Block and Halzen [39].
[3] D. Maurin, R. Taillet, F. Donato, P. Salati, A. Barrau, et al., (2002), arXiv:astro-ph/0212111 [astro-ph].
[4] C. Evoli, D. Gaggero, D. Grasso, and L. Maccione, JCAP 0810 , 018 (2008), arXiv:0807.4730 [astro-ph].
- [5] F. Donato, D. Maurin, P. Brun, T. Delahaye, and P. Salati, Phys.Rev.Lett. 102 , 071301 (2009), arXiv:0810.5292 [astro-ph].
- [6] O. Adriani et al. (PAMELA Collaboration), Phys.Rev.Lett. 105 , 121101 (2010), arXiv:1007.0821 [astro-ph.HE].
- [7] http://www.ams02.org/ .
- [8] I. Moskalenko, A. Strong, and O. Reimer, Astron.Astrophys. 338 , L75 (1998), arXiv:astroph/9808084 [astro-ph].
- [9] F. Donato, D. Maurin, P. Salati, A. Barrau, G. Boudoul, et al., Astrophys.J. 563 , 172 (2001), arXiv:astroph/0103150 [astro-ph].
- [10] I. V. Moskalenko, A. W. Strong, J. F. Ormes, and M. S. Potgieter, Astrophys.J. 565 , 280 (2002), arXiv:astroph/0106567 [astro-ph].
- [11] R. Duperray, C.-Y. Huang, K. Protasov, and M. Buenerd, Phys.Rev. D68 , 094017 (2003), arXiv:astroph/0305274 [astro-ph].
- [12] L. Tan and L. Ng, Phys.Rev. D26 , 1179 (1982).
- [13] I. Arsene et al. (BRAHMS Collaboration), Phys.Rev.Lett. 98 , 252001 (2007), arXiv:hep-ex/0701041 [hep-ex].
- [14] T. Anticic et al. (NA49 Collaboration), Eur.Phys.J. C65 , 9 (2010), arXiv:0904.2708 [hep-ex].
- [15] J.-P. Meyer, Apjs 57 , 173 (1985).
- [16] H. Haino and the Ams-02 Collaboration, Talk at the 33rd ICRC Conference (2013).
- [17] M. Di Mauro, F. Donato, N. Fornengo, R. Lineros, and A. Vittino, JCAP 1404 , 006 (2014), arXiv:1402.0321 [astro-ph.HE].
- [18] D. Dekkers, J. Geibel, R. Mermod, G. Weber, T. Willitts, et al., Phys.Rev. 137 , B962 (1965).
- [19] J. Allaby, F. Binon, A. Diddens, P. Duteil, A. Klovning, et al., (1970).
- [20] P. Capiluppi, G. Giacomelli, A. Rossi, G. Vannini, A. Bertin, et al., Nucl.Phys. B79 , 189 (1974).
- [21] K. Guettler et al. (British-Scandinavian-MIT Collaboration), Nucl.Phys. B116 , 77 (1976).
- [22] J. Johnson, R. Kammerud, T. Ohsugi, D. Ritchie, R. Shafer, et al., Phys.Rev.Lett. 39 , 1173 (1977).
- [23] D. Antreasyan, J. Cronin, H. J. Frisch, M. Shochet, L. Kluberg, et al., Phys.Rev. D19 , 764 (1979).
- [24] F. Taylor, D. C. Carey, J. Johnson, R. Kammerud, D. Ritchie, et al., Phys.Rev. D14 , 1217 (1976).
- [25] G. D'Agostini, Nucl.Instrum.Meth. A346 , 306 (1994).
- [26] P. D. Serpico, S. Esposito, F. Iocco, G. Mangano, G. Miele, et al., JCAP 0412 , 010 (2004), arXiv:astroph/0408076 [astro-ph].
- [27] A. N. Kalinovskii, M. V. Mokhov, and Y. Nikitin, American Institute of Physics (1989).
- [28] M. Simon, A. Molnar, and S. Roesler, Astrophys. J. 499 , 250 (1998).
- [29] A. Capella, U. Sukhatme, C.-I. Tan, and J. Tran Thanh Van, Phys.Rept. 236 , 225 (1994).
- [30] S. Roesler, R. Engel, and J. Ranft, , 1033 (2000), arXiv:hep-ph/0012252 [hep-ph].
- [31] M. Kachelriess and S. Ostapchenko, Phys.Rev. D86 , 043004 (2012), arXiv:1206.4705 [astro-ph.HE].
- [32] F. Donato and P. D. Serpico, Phys.Rev. D83 , 023014 (2011), arXiv:1010.5679 [astro-ph.HE].
- [33] M. Cirelli, D. Gaggero, G. Giesen, M. Taoso, and A. Urbano, (2014), arXiv:1407.2173 [hep-ph].
- [34] H. Fischer (NA49 Collaboration), Heavy Ion Phys. 17 , 369 (2003).
- [35] E. Byckling and K. Kajantie, (1971).
- [36] http://pdg.lbl.gov/2013/hadronic-xsections/ hadron.html .
- [37] G. Antchev et al. (TOTEM), Europhys.Lett. 101 , 21004 (2013).
- [38] G. Antchev et al. (TOTEM Collaboration), Phys.Rev.Lett. 111 , 012001 (2013).
- [39] M. M. Block and F. Halzen, Phys.Rev. D86 , 014006 (2012), arXiv:1205.5514 [hep-ph]. | 10.1103/PhysRevD.90.085017 | [
"Mattia di Mauro",
"Fiorenza Donato",
"Andreas Goudelis",
"Pasquale Dario Serpico"
] | 2014-08-01T20:00:23+00:00 | 2018-07-26T16:34:11+00:00 | [
"hep-ph",
"astro-ph.HE"
] | A new evaluation of the antiproton production cross section for cosmic ray studies | Theoretical predictions for the cosmic antiproton spectrum currently fall
short of the corresponding experimental level of accuracy. Among the main
sources of uncertainty are the antiproton production cross sections in cosmic
ray inelastic interactions. We analyse existing data on antiproton production
in $pp$ scattering, including for the first time the measurements performed by
the NA49 Collaboration. We compute the antiproton spectrum finding that in the
energy range where data are available (antiproton energies of about 4-550 GeV)
different approaches lead to almost equivalent results, with an uncertainty of
10-20\%. Extrapolations outside this region lead to different estimates, with
the uncertainties reaching the 50\% level around $1$ TeV, degrading the
diagnostic power of the antiproton channel at those energies. We also comment
on the uncertainties in the antiproton production source term coming from
nuclei heavier than protons and from neutrons produced in $pp$ scatterings, and
point out the need for dedicated experimental campaigns for all processes
involving antiproton production in collisions of light nuclei. |
1408.0289v1 | ## The Not-So-Sterile 4th Neutrino: Constraints on New Gauge Interactions from Neutrino Oscillation Experiments
Joachim Kopp 1 2 , ∗ and Johannes Welter 1 †
1 Max Planck Institut f¨r Kernphysik, Saupfercheckweg 1, 69117 Heidelberg, Germany u 2 PRISMA Cluster of Excellence and Mainz Institute for Theoretical Physics, Johannes Gutenberg University, 55099 Mainz, Germany
(Dated: August 5, 2014)
Sterile neutrino models with new gauge interactions in the sterile sector are phenomenologically interesting since they can lead to novel effects in neutrino oscillation experiments, in cosmology and in dark matter detectors, possibly even explaining some of the observed anomalies in these experiments. Here, we use data from neutrino oscillation experiments, in particular from MiniBooNE, MINOS and solar neutrino experiments, to constrain such models. We focus in particular on the case where the sterile sector gauge boson A ′ couples also to Standard Model particles (for instance to the baryon number current) and thus induces a large Mikheyev-Smirnov-Wolfenstein potential. For eV-scale sterile neutrinos, we obtain strong constraints especially from MINOS, which restricts the strength of the new interaction to be less than ∼ 10 times that of the Standard Model weak interaction unless active-sterile neutrino mixing is very small (sin 2 θ 24 glyph[lessorsimilar] 10 -3 ). This rules out gauge forces large enough to affect short-baseline experiments like MiniBooNE and it imposes nontrivial constraints on signals from sterile neutrino scattering in dark matter experiments.
PACS numbers: 14.60.St, 14.60.Pq
## 1. INTRODUCTION AND MOTIVATION
The possible existence of sterile neutrinos (Standard Model singlet fermions) with masses of order eV has been a widely discussed topic in astroparticle physics over the past few years. It is motivated by several anomalous results from short-baseline neutrino oscillation experiments, in particular the excesses of ν e and ¯ ν e events in a ν µ and ¯ ν µ beam respectively observed by LSND [1] and MiniBooNE [2], the apparently lower than expected ¯ ν e flux from nuclear reactors [3-5] (see however [6]) and the deficit of ν e in radioactive source experiments [7, 8]. Global fits [9-11, 11-15] show that these anomalies could be explained if sterile neutrinos with O (eV) mass and O (10%) mixing with ν e and ν µ exist. However, global fits also reveal that it is difficult to reconcile such a scenario with existing null results from other short-baseline oscillation experiments.
Constraints come also from cosmological observations, which slightly disfavor scenarios with extra relativistic degrees of freedom in the early Universe [16]. Cosmology also imposes a tight constraint on the sum of neutrino masses ∑ j m νj < 0 23, . where the sum runs over all neutrino mass eigenstates that are in thermal equilibrium in the early Universe. Note that these constraints would be relaxed if the recent BICEP-2 data on B-modes in the cosmic microwave background [17] is confirmed [18-22].
An interesting scenario that is unconstrained by cosmology is self-interacting sterile neutrinos [23, 24]. If interactions among sterile neutrinos are mediated by a scalar or gauge boson with
∗ Email: [email protected]
† Email: [email protected]
a mass of order MeV or lighter, sterile neutrinos will feel a strong thermal potential in the early Universe which suppresses their mixing with active neutrinos and thus prohibits their production through oscillations. Moreover, if the new interaction couples not only to sterile neutrinos, but also to dark matter, it has the potential to explain several problems with cosmic structure formation at small scales [24, 25].
If a new interaction is shared between sterile neutrinos and ordinary matter (for instance in models with gauged baryon number coupled to sterile neutrinos and in scenarios in which a sterile sector gauge boson mixes kinetically with the photon), interesting signals in direct dark matter searches are possible [26-29]. The increased neutrino-nucleus scattering cross section might even explain some of the excess events observed by several experiments. On the other hand, such scenarios are more challenging for cosmology because of an additional sterile neutrino production mechanism through the gauge interaction. (Note that these constraints are still avoided for instance in scenarios with extra entropy production in the visible sector after sterile neutrino decoupling [30].)
In this paper, we investigate how novel interactions between sterile neutrinos and ordinary matter are constrained by neutrino oscillation experiments at short and long baseline. This topic has been discussed in a previous paper [31], the conclusions of which we will update below. For definiteness, we will focus on scenarios similar to the 'baryonic sterile neutrino' scenario first introduced in [26], where the sterile neutrino couples to gauged baryon number. We emphasize, however, that our results are directly applicable to any theory in which sterile neutrinos interact with Standard Model (SM) fermions through a new gauge force under which ordinary matter carries a net charge. (The last condition excludes models in which the coupling is only through kinetic mixing between the new gauge boson and the photon.) The new gauge current creates a MikheyevSmirnov-Wolfenstein (MSW) potential for sterile neutrinos propagating through ordinary matter and has thus a potentially large impact on neutrino oscillations. Since the mass of the new gauge boson in this model can be as low as 10 MeV (see [27] for detailed constraints) and since constraints on its coupling are weak [26], the strength of the effective interaction can be more than two orders of magnitude larger than the SM weak interactions responsible for the ordinary MSW effect. This implies that resonant enhancement of the oscillation amplitude could be relevant at O (GeV) energies even for relatively large mass squared difference ∆ m 2 41 ∼ eV between the mostly sterile and mostly active mass eigenstates. The model could thus potentially allow an explanation of some of the short-baseline oscillation anomalies with significantly smaller vacuum mixing angles than in sterile neutrino scenarios without new interactions.
The structure of the paper is as follows. In section 2, we briefly review models with new interactions in the sterile sector in general, and the 'baryonic neutrino' model from [26] in particular. We map these models onto an effective field theory and discuss their implications for neutrino oscillations. In particular, we derive approximate analytical formulas for the oscillation probabilities. In section 3, we then present our main numerical results, which will set strong constraints on new forces coupling sterile neutrinos to SM particles. We will summarize and conclude in section 4.
## 2. MODELS AND FORMALISM
## 2.1. New gauge bosons in the sterile neutrino sector
In the following we shortly describe the model proposed in [26, 27], originally introduced to study the impact of a new gauge force in the sterile neutrino sector on dark matter searches. The basic idea is to introduce a fourth left-handed neutrino flavour ν b , sterile under SM interactions, which can have a relatively large coupling to baryons (10 2 -10 3 times larger than the Fermi constant G F ) without being in conflict with current experimental bounds, like for examples constraints coming
from meson decays such as K → πν ν ¯ b b [26]. It can be implemented by introducing a new U (1) B gauge symmetry under which quarks have charge g / b 3 and the baryonic neutrino ν b has charge g ′ b . We will assume g b and g ′ b to be of order 0.1-1. To cancel anomalies, the introduction of additional fermions charged under U (1) B will be necessary, but we assume that these do not mix significantly with SM neutrinos and can be neglected. The baryonic gauge boson X acquires a mass when U (1) B is broken by a new sterile sector Higgs field h b . The relevant part of the Lagrangian after symmetry breaking can be written as [26]
$$\mathcal { L } & \, \supset & \, - \frac { 1 } { 4 } F _ { X, \mu \nu } F _ { X } ^ { \mu \nu } + \frac { 1 } { 2 } m _ { X } ^ { 2 } X _ { \mu } X ^ { \mu } \\ & \, + \bar { \nu } _ { b } \gamma _ { \mu } ( i \partial ^ { \mu } + g _ { b } ^ { \prime } X ^ { \mu } ) \nu _ { b } + \sum _ { q } \bar { q } \left ( i \not \, \psi _ { \text{SM} } + \frac { 1 } { 3 } g _ { b } \gamma _ { \mu } X ^ { \mu } \right ) q + \mathcal { L } _ { m } \,,$$
where q are the SM quark fields, F X,µν ≡ ∂ X µ ν -∂ X ν µ is the field strength tensor of the baryonic vector boson X µ and m X ∼ 1 GeV is its mass. In a seesaw framework, the baryonic neutrino mixes with the SM through the terms
$$\mathcal { L } _ { m } = - \sum _ { \alpha, j } m _ { D } ^ { \alpha j } \bar { \nu } _ { L } ^ { \alpha } N _ { R } ^ { j } - \sum _ { j } m _ { b } ^ { j } \bar { \nu } _ { b L } N _ { R } ^ { j } - \frac { 1 } { 2 } \sum _ { i, j } m _ { R } ^ { i j } \overline { \left ( N _ { R } ^ { i } \right ) ^ { C } } N _ { R } ^ { j } + h. c. \,,$$
with the Dirac mass matrix m D of the active neutrinos ν α L , the Dirac mass vector of the baryonic neutrino m j b and the the Majorana mass matrix m ij R of the heavy right-handed neutrino fields N j R . The flavour index α runs over e , µ and τ , while the indices i and j run over all heavy right-handed neutrino states.
The Lagrangian of equation (1) implies the existence of a new MSW potential that sterile neutrinos experience while propagating in matter. This effect is caused by coherent elastic forward scattering on neutrons and protons and can lead to resonant enhancement of flavour oscillations. Since coherent forward scattering does not involve any momentum transfer, its amplitude can be most easily obtained from the low energy effective Lagrangian of baryonic neutral current interactions
$$\mathcal { L } _ { b, \text{eff} } \, = \, \frac { G _ { B } } { 2 } \left [ \bar { \nu } _ { b } \gamma _ { \mu } \left ( 1 - \gamma _ { 5 } \right ) \nu _ { b } \right ] \left [ \bar { p } \gamma ^ { \mu } p + \bar { n } \gamma ^ { \mu } n \right ].$$
Here, the effective coupling constant is G B ≡ g g /m b ′ b 2 X . By treating neutrons and protons as a static background field [32], we obtain the matter potential for sterile neutrinos
$$V _ { b } & = G _ { B } N _ { \text{ucl} }.$$
The potential for sterile anti-neutrinos has opposite sign. Here, N nucl is the number density of nucleons in the background matter. Note that G B can be either positive or negative, depending on the relative sign of g b and g ′ b . In the following analysis we will use the ratio of the coupling constants
$$\epsilon \equiv \frac { G _ { B } } { \sqrt { 2 } G _ { F } }$$
as a measure for the relative strength of V b compared to the potential V CC that charged current (CC) interactions with electrons induce for electron neutrinos in the SM. The baryonic potential can be written as
$$V _ { b } = V _ { \mathbb { C C } } \cdot G _ { B } / ( \sqrt { 2 } G _ { F } Y _ { e } ) = \epsilon V _ { \mathbb { C C } } / Y _ { e }$$
$$= \epsilon \cdot 7. 5 6 \cdot 1 0 ^ { - 1 4 } \ e V \cdot \left ( \frac { \rho } { g / c m ^ { 3 } } \right ),$$
where Y e is the number of electrons per nucleon.
As mentioned in the introduction, baryonic sterile neutrinos could lead to novel signals in direct dark matter searches thanks to an enhanced sterile neutrino-nucleus scattering rate. Typically, observable effects in current experiments are expected if glyph[epsilon1] glyph[greaterorsimilar] 100 [26-29]. We will see in section 3.2 that such large values of glyph[epsilon1] are largely excluded for eV scale sterile neutrinos with substantial mixing into the active sector.
We wish to stress here that, while we use baryonic sterile neutrinos as a benchmark scenario, our results will apply to any scenario in which sterile neutrinos have new gauge interactions with SM fermions. It is important to keep in mind, though, that models with new forces in the lepton sector are much more tightly constrained than new baryonic interactions (see e.g. [27] for a review).
The mass terms in equation (2) lead to flavour mixing between ν b and the active neutrinos, as can be seen by integrating out the heavy right-handed neutrinos and diagonalizing the resulting mass matrix. In this way, we obtain the 4 × 4 mixing matrix U connecting mass eigenstates | ν i 〉 and flavour eigenstates | ν α 〉 :
$$| \nu _ { \alpha } \rangle = \sum _ { i } U _ { \alpha i } ^ { * } | \nu _ { i } \rangle.$$
Since U is unitary, it can be parametrized by 6 rotation angles θ ij and 3 complex phases δ ij 1
$$U \, = \, R _ { 3 4 } \cdot R _ { 2 4 } ^ { \prime } \cdot R _ { 1 4 } ^ { \prime } \cdot R _ { 2 3 } \cdot R _ { 1 3 } ^ { \prime } \cdot R _ { 1 2 }.$$
Here, R ij describes a rotation matrix in the ij plane, while R ′ ij corresponds to a complex rotation by the angle θ ij and phase δ ij . Given the mixing matrix U and the mass squared difference ∆ m 2 41 between the mostly sterile mass eigenstate ν 4 and the mostly active mass eigenstate ν 1 , one can write down the effective Hamiltonian 2 in flavour space:
$$H _ { \text{eff} } ^ { \text{flavour} } = \frac { 1 } { 2 E } U \begin{pmatrix} 0 & & \\ & \Delta m _ { 2 1 } ^ { 2 } & \\ & & \Delta m _ { 3 1 } ^ { 2 } & \\ & & & \Delta m _ { 4 1 } ^ { 2 } \end{pmatrix} U ^ { \dagger } + \begin{pmatrix} V _ { C C } & & \\ & 0 & & \\ & & 0 & \\ & & & V _ { b } - V _ { N C } \end{pmatrix}.$$
Here, V NC ≡ -√ 2 G n / F n 2 is the contribution from SM neutral current interactions to the MSW potential. It is proportional to the number density n n of neutrons in the background material.
The oscillation probability P ν α → ν β ( ), t i.e. the probability for a neutrino of initial flavour α to be converted into flavour β after traveling a time t , can then be obtained by diagonalizing the effective Hamiltonian according to H flavour eff = ˜ diag( U λ , λ 1 2 , λ 3 , λ 4 ) ˜ U † and inserting the eigenvalues λ i and the effective mixing matrix ˜ U into the well-known formula
$$P _ { \nu _ { \alpha } \to \nu _ { \beta } } = \left | \left \langle \nu _ { \beta } | \nu _ { \alpha } ( t ) \right \rangle \right | ^ { 2 } = \left | \sum _ { j } \tilde { U } _ { \alpha j } ^ { * } \tilde { U } _ { \beta j } e ^ { - i \lambda _ { j } L } \right | ^ { 2 }.$$
1 We omit the Majorana phases here since they do not contribute to neutrino flavour oscillations.
2 Effective means that terms proportional to the unit matrix are omitted because they do not contribute to flavour oscillations. Also note that we assume a definite three-momentum that is the same for all contributing mass eigenstates so that one can approximate E i ≈ | p | + m / 2 i (2 E ). It is well-known that this approximation, though technically unjustified, leads to correct results for neutrino oscillation probabilities [33].
## 2.2. Approximate oscillation probabilities
As a prelude to the numerical fits we are going to present in section 3, we give here approximate analytic expressions for the oscillation probabilities in the baryonic sterile neutrino model and in models with new sterile neutrino-SM interactions in general. Similar calculations have been carried out previously in [31] and we will compare these results to ours in section 2.3.
Our starting point is to assume | ∆ m 2 41 | glyph[greatermuch] | ∆ m 2 31 | , ∆ m 2 21 , which is a good approximation at sufficiently short baselines. Moreover, we neglect the SM MSW potentials V NC (arising from Z exchange diagrams) and V CC (arising from W exchange diagrams) against the baryonic potential V b , which we assume to be much larger. With these approximations, mixing among the three active flavour eigenstates becomes irrelevant. (They can, however, still oscillate into each other through their mixing with ν b .) We also set U τ 4 = 0 for simplicity, following [31]. With these assumptions, diagonalization of the Hamiltonian H flavour eff from equation (10) yields for the eigenvalues λ i
$$\lambda _ { 1 } = \lambda _ { 2 } = 0, \quad \lambda _ { 3 } = \frac { 1 } { 2 } \left ( V _ { b } + \frac { \Delta m _ { 4 1 } ^ { 2 } } { 2 E } - A \right ), \quad \lambda _ { 4 } = \frac { 1 } { 2 } \left ( V _ { b } + \frac { \Delta m _ { 4 1 } ^ { 2 } } { 2 E } + A \right ). \quad \quad ( 1 2 )$$
The elements of the unitary matrix ˜ U are
$$\tilde { U } _ { \mu 1 } = \tilde { U } _ { e 1 } = 0, \quad | \tilde { U } _ { e 2 } | ^ { 2 } = \frac { | U _ { \mu 4 } | ^ { 2 } } { 1 - | U _ { s 4 } | ^ { 2 } }, \quad | \tilde { U } _ { \mu 2 } | ^ { 2 } = \frac { | U _ { e 4 } | ^ { 2 } } { 1 - | U _ { s 4 } | ^ { 2 } },$$
$$| \tilde { U } _ { e 4 } | ^ { 2 } = | U _ { e 4 } | ^ { 2 } \frac { \frac { \Delta m _ { 4 1 } ^ { 2 } } { 2 E } \left [ A + \frac { \Delta m _ { 4 1 } ^ { 2 } } { 2 E } - V _ { b } \right ] } { A \left [ A + \frac { \Delta m _ { 4 1 } ^ { 2 } } { 2 E } + V _ { b } \right ] }, \quad | \tilde { U } _ { \mu 4 } | ^ { 2 } = | U _ { \mu 4 } | ^ { 2 } \frac { \frac { \Delta m _ { 4 1 } ^ { 2 } } { 2 E } \left [ A + \frac { \Delta m _ { 4 1 } ^ { 2 } } { 2 E } - V _ { b } \right ] } { A \left [ A + \frac { \Delta m _ { 4 1 } ^ { 2 } } { 2 E } + V _ { b } \right ] }. \quad ( 1 3 )$$
Here, we have introduced the abbreviation
$$A \, = \, | V _ { b } | \cdot \sqrt { 1 + ( 4 | U _ { s 4 } | ^ { 2 } - 2 ) \, \frac { \Delta m _ { 4 1 } ^ { 2 } } { 2 E V _ { b } } + \left ( \frac { \Delta m _ { 4 1 } ^ { 2 } } { 2 E V _ { b } } \right ) ^ { 2 } }.$$
With these formulas at hand and using the unitarity condition ∑ i ˜ U ∗ αi ˜ U αi = 1 as well as the observation that ˜ U µ 2 ˜ U ∗ e 2 ˜ U ∗ µ 4 ˜ U e 4 is real, it is straightforward to calculate the oscillation probabilities according to equation (11). For α = µ and β = µ , e we obtain
$$P _ { \nu _ { \mu } \to \nu _ { e } } \, = \, & - 4 \frac { | U _ { e 4 } | ^ { 2 } | U _ { \mu 4 } | ^ { 2 } | U _ { s 4 } | ^ { 2 } } { 1 - | U _ { s 4 } | ^ { 2 } } \left ( \frac { \Delta m _ { 4 1 } ^ { 2 } } { 2 E A } \right ) ^ { 2 } \sin ^ { 2 } \phi _ { 1 } + 2 \frac { | U _ { e 4 } | ^ { 2 } | U _ { \mu 4 } | ^ { 2 } } { ( 1 - | U _ { s 4 } | ^ { 2 } ) ^ { 2 } } \left ( 1 + \frac { V _ { b } - V _ { r e s } } { A } \right ) \sin ^ { 2 } \phi _ { 2 } \\ & + 2 \frac { | U _ { e 4 } | ^ { 2 } | U _ { \mu 4 } | ^ { 2 } } { ( 1 - | U _ { s 4 } | ^ { 2 } ) ^ { 2 } } \left ( 1 - \frac { V _ { b } - V _ { r e s } } { A } \right ) \sin ^ { 2 } \phi _ { 3 }, \\ & \quad \, \intertext { a, \dots 2 } \, \, \stackrel { 2 } { \lambda } \, \dots 2 \, \, \stackrel { 2 } { \lambda }$$
$$P _ { \nu _ { \mu } \to \nu _ { b } } \, = \, 4 | U _ { \mu 4 } | ^ { 2 } | U _ { s 4 } | ^ { 2 } \left ( \frac { \Delta m _ { 4 1 } ^ { 2 } } { 2 E A } \right ) ^ { 2 } \sin ^ { 2 } \phi _ { 1 },$$
$$P _ { \nu _ { \mu } \to \nu _ { \mu } } \ = \ 1 - P _ { \nu _ { \mu } \to \nu _ { \epsilon } } - P _ { \nu _ { \mu } \to \nu _ { \theta } },$$
where the oscillation phases are
$$\phi _ { 1 } = \frac { \lambda _ { 4 } - \lambda _ { 3 } } { 2 } L = \frac { L } { 2 } A, \\. \quad \Delta _ { 1 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 } \Delta _ { 2 }$$
$$\phi _ { 2 } = \frac { \lambda _ { 3 } } { 2 } \bar { L } \quad \ = \frac { \bar { L } } { 4 } \left ( V _ { b } + \frac { \Delta m _ { 4 1 } ^ { 2 } } { 2 E } - A \right ),$$
$$\phi _ { 3 } = \frac { \lambda _ { 4 } } { 2 } L \quad \ = \frac { L } { 4 } \left ( V _ { b } + \frac { \Delta m _ { 4 1 } ^ { 2 } } { 2 E } + A \right )$$
and V res is the value of the matter potential at which A takes its minimal value | U s 4 | √ 1 -| U s 4 | 2 ∆ m /E 2 41 . It is given by
$$V _ { \text{res} } = - \frac { \Delta m _ { 4 1 } ^ { 2 } } { 2 E } \left ( 2 | U _ { s 4 } | ^ { 2 } - 1 \right )$$
and corresponds to the new MSW resonance condition. Whether the resonance is in the neutrino or anti-neutrino sector depends on the sign of V b , i.e. the relative sign of the charges g b and g ′ b . With the assumption sin 2 θ 24 < . 0 5 and for V b < 0 ( V b > 0) the resonance condition can be fulfilled only in the neutrino (anti-neutrino) sector. For ∆ m 2 41 = 1 eV , a matter density of 3 g/cm 2 3 and a neutrino energy of 1 GeV, the resonance condition is fulfilled for neutrinos if glyph[epsilon1] = G / B √ 2 G F glyph[similarequal] - × 2 10 3 and for anti-neutrinos if glyph[epsilon1] has opposite sign. For oscillation experiments, we see that matter enhancement of active-to-sterile neutrino oscillations is expected predominantly in high energy ( O (GeV)) experiments and only if the new gauge force is several orders of magnitude stronger than SM weak interactions. For weaker gauge forces, the new resonance moves to higher energies that are only accessible with atmospheric or cosmic neutrinos.
Note that equation (21) has a structure similar to the expression for the standard MSW resonance condition. To see this, consider the matrix element | U s 4 | 2 in the parametrization of equation (9): | U s 4 | 2 = cos 2 θ 14 cos 2 θ 24 cos 2 θ 34 . If cos 2 θ 34 , cos 2 θ 14 ≈ 1, we have V res = -(∆ m / E 2 41 2 ) cos 2 θ 24 . However, unless ∆ m / E 2 41 2 is much larger than V b , oscillations at short baseline cannot be approximately described in an effective two-flavour framework, unlike the 3+1 model without non-standard matter effects. The reason is that, without the extra matter term, three eigenvalues of the Hamiltonian can be set to zero at short baseline, while large V b implies that this is only possible for two of them.
On the other hand, in the limit of very large matter potential, V b glyph[greatermuch] ∆ m / 2 41 (2 E ), the term proportional to sin 2 φ 2 in equation (15) dominates over the terms containing sin 2 φ 1 and sin 2 φ 3 since the latter two are of higher order in ∆ m / 2 41 (2 EV b ). If we furthermore assume the baseline is not too long, in particular (∆ m 2 41 ) 2 / (4 E V 2 b ) · L/ 2 glyph[lessmuch] 1, we can approximate φ 2 ≈ ( L/ 2)(1 -| U s 4 | 2 )∆ m / 2 41 (2 E ) and obtain for the oscillation probability of equation (15) the effective twoflavour formula
$$P _ { \nu _ { \mu } \to \nu _ { e } } \, \approx \, 4 \frac { | U _ { e 4 } | ^ { 2 } | U _ { \mu 4 } | ^ { 2 } } { ( 1 - | U _ { s 4 } | ^ { 2 } ) ^ { 2 } } \cdot \sin ^ { 2 } \left ( \frac { L } { 2 } \left ( 1 - | U _ { s 4 } | ^ { 2 } \right ) \frac { \Delta m _ { 4 1 } ^ { 2 } } { 2 E } \right ) + \mathcal { O } \left ( \left ( \frac { \Delta m _ { 4 1 } ^ { 2 } } { 2 E V _ { b } } \right ) ^ { 2 } \right ).$$
As expected, in the limit of large matter potential V b , the corresponding neutrino ν b decouples from flavour oscillations, P ν µ → ν b ≈ 0 and the ν µ survival probability becomes P ν µ → ν µ ≈ 1 -P ν µ → ν e .
We do not expect that scenarios with large V b can explain the short-baseline anomalies better than conventional models without new interactions. The reactor [3-5] and gallium [7, 8] experiments were too low in energy; in LSND [1], neutrinos traveled mostly through air; MiniBooNE could in principle be sensitive to new matter effects, but resonant enhancement could only explain an anomaly in either the neutrino or the anti-neutrino sector, while the data shows similar deviations from expectations in both sectors. 3 On the other hand, we expect that MiniBooNE-along with long-baseline experiments like MINOS and with solar neutrinos-will impose tight constraints on V b .
3 Note that in earlier MiniBooNE data [34-36], there appeared to be mild tension between the neutrino and antineutrino mode data. This motivated the authors of [31] to consider resonantly enhanced active-sterile neutrino mixing even as a possible explanation of the MiniBooNE anomaly.
## 2.3. Accuracy of analytic approximations
In the following, we discuss the implications of sterile neutrinos with non-standard matter effects in terrestrial long-baseline experiments, taking MiniBooNE and MINOS as examples. In doing so, we also compare our analytic expressions (17) and (15) to a numerical computation in the full four flavour framework and to the results of [31].
To obtain the exact four-flavour oscillation probabilities, we diagonalize the effective Hamiltonian of equation (10) numerically and use the resulting eigenvalues and eigenvectors in equation (11). In doing so, we absorb the neutral current potential V NC into a redefinition of V b . 4 To average out fast oscillations that would not be resolvable by experiments, we also implement a low-pass filter by multiplying each term in the oscillation probability equation (11) by a Gaussian factor [37]. This yields:
$$P _ { \nu _ { \alpha } \to \nu _ { \beta } } \, = \, \sum _ { j, k } \tilde { U } _ { \alpha j } ^ { * } \tilde { U } _ { \beta j } \tilde { U } _ { \alpha k } \tilde { U } _ { \beta k } ^ { * } \exp \left [ - \, i L ( \lambda _ { j } - \lambda _ { k } ) \right ] \, \exp \left [ - \, L ^ { 2 } ( \lambda _ { j } - \lambda _ { k } ) ^ { 2 } \cdot \frac { \sigma _ { f } ( E ) ^ { 2 } } { 2 E ^ { 2 } } \right ],$$
where σ f ( E ) is the energy width of the filter, which is related to the energy resolution of the experiment. This form for the low-pass filter can also be motivated in a wave packet treatment, where the finite energy resolution of the production and detection processes determines the width of the neutrino wave packets (see [38] and references therein). When comparing analytical and numerical results, we also apply such a low-pass filter to the analytic expressions (15) and (17) by replacing the oscillation terms sin 2 φ i according to
↦
$$\sin ^ { 2 } \phi _ { i } \mapsto \frac { 1 } { 2 } \left ( 1 - \cos ( 2 \phi _ { i } ) \cdot \exp \left [ - \left ( 2 \phi _ { i } \right ) ^ { 2 } \frac { \sigma _ { f } ( E ) ^ { 2 } } { 2 E ^ { 2 } } \right ] \right ).$$
In the following, we choose σ f ( E ) = 0 01 GeV. .
In the calculation of the analytical formulas in [31] the eigenvalues λ i are approximated by setting A ≈ V b + ∆ m 2 41 2 E (i.e. taking | U s 4 | = 1 in equation (14)). This leads to λ 1 = λ 2 = λ 3 = 0 and λ 4 = V b +∆ m / E 2 41 2 . The oscillation phases of equations (18)-(20) then become φ 1 = φ 3 = 1 2 L V ( b + ∆ m / E 2 41 2 ) and φ 2 = 0. With this replacements our equation (15) reduces to equations (21)-(22) in [31]. In the limit of large V b we see from equation (22) that this approximation is only valid if L/ 2(1 -| U s 4 | 2 )∆ m / 2 41 (2 E ) glyph[lessmuch] 1.
Since the latter condition is fulfilled in the L/E regime at which the LSND and MiniBooNE experiments are sensitive to ν µ → ν e flavour transitions, the approximation from [31] is applicable there. This can be seen in figure 1, where the transition probabilities for neutrinos (in green) and anti-neutrinos (in blue) are shown for L = 541 m and E = 0 1-3 MeV. We have taken the . model parameters at the best fit point from [31] (which we will show to be in fact excluded by MINOS in section 3.2). Dashed curves correspond to our analytical approximation (equation (15)), which agrees extremely well with numerical results, while dotted curves show the approximation from equations (21)-(22) of [31]. The difference between the neutrino and anti-neutrino sectors originates from the different signs of the matter potential. As expected, glyph[epsilon1] > 0 ( ⇔ V b > 0) leads to a resonant enhancement of the anti-neutrino transition probabilities and a suppression of the neutrino transition probabilities compared to the case glyph[epsilon1] = 0 (black curve). We see that the approximations used in [31] are fairly accurate in the most relevant energy range below 1 GeV, but fail at higher energies.
4 This is only approximately correct if V b glyph[similarequal] V NC and the proton-to-neutron ratio is varying along the neutrino trajectory. Since we are mainly interested in scenarios with V b glyph[greatermuch] V NC , our results are insensitive to this subtlety.
L
L
H
@ D Figure 1: The electron neutrino (green) and anti-neutrino (blue) appearance probability in a model with a large MSW potential in the sterile sector (for instance the baryonic sterile neutrino model from [26]). We use the baseline L = 541 m and the energy range 0.1-3 GeV of the MiniBooNE experiment and take the favored model parameters from [31]: glyph[epsilon1] = G / B ( √ 2 G F ) = 1 3 . · 10 3 ( ⇔ V b = 2 · 10 -10 eV for ρ = 3 g/cm ), 3 ∆ m 2 41 = 0 47 eV , . 2 | U s 4 | 2 = 0 9, . | U e 4 | 2 = | U µ 4 | 2 = 0 05. . (We will see below, that this particular parameter point is in fact excluded by MINOS data, though.) For the standard oscillation parameters, we have used the results of the global fit 'Free Fluxes and RSBL' of [39]. In black, we show also the prediction of a sterile neutrino model without new interactions ( glyph[epsilon1] = 0). Dashed lines correspond to our analytic approximations, which coincide with numerical results (solid curves) in this baseline and energy range, while dotted lines show the results from [31].
e =
0
L
H
H
@
D
@
D
Figure 2: The figure shows the ν µ disappearance probability for the MINOS baseline of L = 735km and energies up to 10 GeV. We show the survival probability P ν µ → ν µ for glyph[epsilon1] = 0 (left panel) and glyph[epsilon1] = 1 3 . · 10 3 ( ⇔ V b = 2 10 · -10 eV for ρ = 3 g/cm ) (right panel) using the best fit parameters from the LSND/MiniBooNE 3 fit of [31]. Solid curves correspond to a numerical calculation in the full four flavour oscillation framework, using for the standard oscillation parameters the values from the fit 'Free Fluxes and RSBL' of [39]. Dashed curves show our analytic approximation, equation (17), while dotted curves correspond to equations 20 and 28 of [31]. The comparison shows that, when the new matter potential V b is switched on ( glyph[epsilon1] > 0), the active-sterile oscillation mode dominates over the standard atmospheric oscillation pattern, an effect which is not captured by the approximations made in [31].
Since standard and non-standard matter effects are most relevant at long baseline ( glyph[greaterorsimilar] few × 100 km), it is important to also study the disappearance probability 1 -P ν µ → ν µ as a function of energy for long-baseline oscillation experiments like MINOS. MINOS has measured P ν µ → ν µ at a baseline of L = 735 km in the energy range 1-50 GeV. The oscillation probabilities for this baseline and energy range are shown in figure 2 for glyph[epsilon1] > 0 (right panel) and also for the Standard Model ( glyph[epsilon1] = 0, left panel). We see that, due to matter-enhanced oscillations inside the earth, a scenario with strong non-standard matter effects leads to very large muon disappearance even at energies as high as 10 GeV, well above the standard oscillation maximum at ∼ 1 5 GeV. This is in conflict . with MINOS data and we therefore expect that MINOS is able to place very strong constraints on new matter effects in the sterile neutrino sector. Figure 2 also implies that the parameters favored in [31] are ruled out by MINOS.
Comparing numerical results (solid lines) to our analytic approximation (dashed lines), we find, as expected, that the approximations of equations (15) and (17) are accurate at ∆ m L/ 2 31 (2 E ) ∼ 1 only if V b is very large. We also see that the analytic approximations from [31] (dotted curves in figure 2) are not applicable at long baseline even for large V b . For example, in the MINOS case, ∆ m L/ 2 41 (2 E ) · eV 2 ∼ 100 for ∆ m 2 41 ∼ eV , the phase 2 φ 2 (see equations (19) and (22)) becomes non-negligible. This is the reason why our conclusions regarding the importance of MINOS data for constraining sterile neutrino matter effects differ from those of [31], where φ 2 has been neglected.
## 3. CONSTRAINTS FROM OSCILLATION EXPERIMENTS
From the analysis in the previous section we expect that the baryonic sterile neutrino model (or models with new sterile neutrino-SM interactions in general) could potentially explain by resonant enhancement an event excess in the MiniBooNE neutrino or anti-neutrino data (but not in both), but is strongly constrained by data from long-baseline experiments. Therefore, we now derive limits on the model using a numerical χ 2 analysis of data from MiniBooNE, MINOS and also solar neutrino experiments.
## 3.1. Analysis method
In our analysis we fix the standard oscillation parameters at their best fit values from the global fit by Gonzalez-Garcia et al. [39] (see table I) and we assume a normal mass ordering. We have checked that our results for inverted ordering are very similar, with only the solar limits becoming somewhat weaker. (We will comment on this in more detail in section 3.2.) For simplicity we set δ 13 = δ 14 = δ 24 = 0 because none of the experiments considered here is sensitive to CP violation in the small V b limit and equations (15)-(17) show that also the leading terms in the oscillation probabilities for large V b are independent of complex phases. We fix the mixing angle θ 34 = 0 since MiniBooNE is not sensitive to this angle and MINOS has only very limited sensitivity [15]. The impact of θ 34 > 0 on the constraints from solar experiments will be discussed in section 3.2. Finally, we set sin 2 2 θ 14 = 0 12 so that the reactor anomaly [3-5] can be explained. We will comment on the . effect of relaxing this assumption also in section 3.2. The constraints we impose on the parameter space are also summarized in table I. The remaining three parameters glyph[epsilon1] = G / B ( √ 2 G F ), ∆ m 2 41 and θ 24 are scanned over the ranges | glyph[epsilon1] | = 1 -32000, ∆ m /eV 2 41 2 = 0 01 . -11 and sin 2 θ 24 = 0 0001 . -1.
We now discuss the details of our fits to MINOS, MiniBooNE and solar neutrino data.
Table I: The parameter values of the baryonic sterile neutrino model that we have fixed in our parameter scan.
| sin 2 θ 12 | sin 2 θ 23 | sin 2 θ 13 | ∆ m 2 21 [eV 2 ] | ∆ m 2 31 [eV 2 ] | δ 13 , δ 14 δ 24 | sin 2 2 θ 14 | sin 2 θ 34 |
|--------------|--------------|--------------|--------------------|--------------------|--------------------|----------------|--------------|
| 0 . 302 | 0 . 413 | 0 . 0227 | 7 . 5 · 10 - 5 | 2 . 473 · 10 - 3 | 0 | 0 . 12 | 0 |
## 3.1.1. MINOS
For MINOS, we use GLoBES [40, 41] to compute the energy dependent oscillation probabilities P near ( E ) for the near detector and P far ( E ) for the far detector numerically. We include a low pass filter according to equation (23) with σ f ( E ) = 0 06 . · E . The matter density ρ along the neutrino trajectory to the far detector is assumed to be constant at its average value
$$\langle \rho _ { \text{far} } \rangle \ = \ \frac { 2 } { L _ { \text{far} } } \int _ { \sqrt { R _ { \oplus } ^ { 2 } - ( L _ { \text{far} } / 2 ) ^ { 2 } } } ^ { R _ { \oplus } } \rho ( r ) \frac { d } { d r } \left ( \sqrt { r ^ { 2 } - R _ { \oplus } ^ { 2 } + ( L _ { \text{far} } / 2 ) ^ { 2 } } \right ) d r.$$
In this expression, which can be understood from geometric arguments, r is the distance of the neutrino from the center of the earth, R ⊕ is the radius of the earth and L far = 735km is the neutrino path length from the source to the far detector [42]. Using the matter density profile from the Preliminary Reference Earth Model (PREM) [43] we obtain 〈 ρ far 〉 ≈ 2 36 g . / cm . 3
For large V b , matter effects can be relevant even in the near detector at a baseline L target = 965m from the target. In computing the average matter density 〈 ρ 〉 near that neutrinos experience on their way to the near detector, we account for the fact that they first travel along the evacuated decay pipe with a length of L pipe = 675m. We estimate 〈 ρ 〉 near ≈ ( L target -L pipe ) /L near · 3 g cm 3 , where L near glyph[similarequal] 763 m is the average distance between the neutrino production vertex and the near detector. It is obtained from the decay length of the neutrinos' parent pions, which have an average energy of 4 -5 GeV [44].
We compute the theoretically predicted event spectrum N osc by multiplying the ratio P far ( E /P ) near ( E ) with the background-subtracted prediction for the MINOS event rate in the absence of oscillations, N no osc :
$$N _ { \text{osc} } ( E ) = \left [ N _ { \text{no osc} } ( E ) - N _ { \text{bg} } ( E ) \right ] \frac { P _ { \text{far} } ( E ) } { P _ { \text{near} } ( E ) }.$$
The no-oscllation rate N no osc ( E ) and the background rate N bg ( E ) are taken from [45], which is similar to [46] but contains data up to 50 GeV. The higher energy data is important to us since it increases the sensitivity at low matter potential V b .
To account for the finite energy resolution of the detector, we fold N osc with the detector response function f ( E,E ′ ), which maps the true event energy E ′ to the reconstructed energy E . Finally, we also add the small experimental background N bg ( E ):
$$N _ { \text{th} } ( E ) = N _ { \text{bg} } ( E ) + \int _ { \text{f} } ( E, E ^ { \prime } ) N _ { \text{osc} } ( E ^ { \prime } ) d E ^ { \prime }.$$
We assume a Gaussian shape for f ( E,E ′ ),
$$f ( E, E ^ { \prime } ) = \frac { 1 } { \sigma ( E ^ { \prime } ) \sqrt { 2 \pi } } \exp \left ( - \frac { ( E - E ^ { \prime } ) ^ { 2 } } { 2 \sigma ^ { 2 } ( E ^ { \prime } ) } \right ),$$
where we choose σ E ( ′ ) = 0 2 GeV . √ E / ′ GeV. This choice allows us to reproduce the oscillated event rates and the constraints on θ 23 and ∆ m 2 31 from [46] with good accuracy. When evaluating
equation (27) numerically, we discretize the integral so that N osc ( E ) needs to be evaluated only at fixed support points E ′ j with a step size of ∆ E ′ j = 0 25 GeV in between. . (We have checked that choosing a smaller value for ∆ E ′ j does not change our results significantly, which implies that possible aliasing effects are under control.) Following the MINOS analysis [45], events are binned for the analysis according to their reconstructed energy E . The rate in th i -th bin is given by
$$N _ { \text{th} } ^ { i } = \int _ { E _ { i } - \Delta E _ { i } / 2 } ^ { E _ { i } + \Delta E _ { i } / 2 } N _ { \text{th} } ( E ) \, d E = N _ { \text{bg} } ^ { i } + \sum _ { j } F ^ { i j } N _ { \text{osc} } ( E _ { j } ^ { \prime } ),$$
where N i bg is the total background in the i -th bin and the elements of the detector response matrix F ij are F ij ≡ = ∫ E i +∆ E / i 2 E i -∆ E / i 2 f ( E,E ′ j ) dE . It is important to note that the F ij need to be computed only once.
From equation (29) we compute χ 2 according to
$$\chi ^ { 2 } = \sum _ { i } \left ( \frac { N _ { t h } ^ { i } - N _ { \exp } ^ { i } } { \sqrt { N _ { \exp } ^ { i } } + 0. 1 \cdot N _ { \exp } ^ { i } } \right ) ^ { 2 },$$
where N i exp is the observed event rate in the i -th energy bin [45] and the sum runs over all energy bins. Note that we have included an additional uncertainty of 10% in order to account for systematic errors without modeling them in detail. Like our choice of σ E ( ′ ) in equation (28), also our simplified treatment of systematic errors has been confirmed by cross-checking our simulations against the results of [15, 46].
In figure 3, we compare our prediction for the oscillated neutrino spectrum in MINOS assuming standard 3-flavour oscillations (blue shaded histogram) to the official MINOS prediction (blue unshaded histogram) and to the data (black points with error bars). We find excellent agreement, which validates our calculations. We also show the MINOS no oscillation prediction (red histogram) which is the starting point for our predictions, as well as the survival probability P ν µ → ν µ (dashed green line; corresponding vertical scale shown on the right).
## 3.1.2. MiniBooNE
As for MINOS, the oscillation probabilities for MiniBooNE are calculated numerically in the full four flavour framework with the help of GLoBES [40, 41], including a low pass filter according to equation (23) with σ f ( E ) = 0 06 . E . Since the MiniBooNE decay pipe is only 50 m long, while the distance from the target to the detector is L = 541 m, we neglect the effect of the finite pion decay length. Instead, we take the matter density to be 〈 ρ 〉 ∼ 3 g / cm 3 along the whole neutrino trajectory.
We use a χ 2 analysis to compare our predicted oscillation probabilities with the experimentally measured probabilities, which are given in [47] as a function of L/E . The data from [47] are shown in figure 4 together with the trivial no-oscillation prediction and with our prediction for the MiniBooNE best fit points in the baryonic sterile neutrino scenario for glyph[epsilon1] > 0 and glyph[epsilon1] < 0.
## 3.1.3. Solar neutrinos
We analyze solar neutrino oscillation data by comparing the measured ν e survival probability P ν e → ν e at different energies to our theoretical predictions. The data points are taken from [48] and include results from Super-Kamiokande, SNO, Borexino and radiochemical experiments.
L
H
GLYPH<144>
L
L
H
GLYPH<144>
L
H
H
@ D @ D Figure 3: The measured and predicted event spectra for the MINOS ν µ (left) and ν µ (right) disappearance measurements. The red histogram is the MINOS prediction assuming no neutrino oscillation [45]. In blue, we show the predicted event spectrum including oscillations according to equation (27), assuming standard three flavour oscillations with the parameters listed in table I. The blue shaded histogram is our prediction, the unshaded histogram is the prediction by the MINOS collaboration. We overlay the survival probability P ν µ → ν µ (dashed green curve and vertical scale on the right).
n
L
L
H
H
GLYPH<144> @ GLYPH<144> D GLYPH<144> @ GLYPH<144> D Figure 4: The measured MiniBooNE ν µ → ν e (left) and ¯ ν µ → ¯ ν e (right) appearance probabilities compared to the predictions of the baryonic sterile neutrino scenario for glyph[epsilon1] > 0 (blue line) and glyph[epsilon1] < 0 (dashed red line) at the MiniBooNE best fit points from table II. Without sterile neutrinos, the appearance probability at the MiniBooNE baseline is approximately zero (solid black line).
In calculating P ν e → ν e , we assume MSW flavour transitions to be fully adiabatic and we account for the fact that solar neutrinos arrive at the earth as an incoherent mixture of mass eigenstates. We obtain P ν e → ν e according to
$$P _ { \nu _ { e } \to \nu _ { e } } = \sum _ { i } | U _ { e i } | ^ { 2 } \cdot | \tilde { U } _ { e i } ( 0 ) | ^ { 2 },$$
where ˜ U ei (0) is the mixing matrix in matter at the center of the Sun ( t = 0) and U ei is the vacuum
GLYPH<144>
mixing matrix. We neglect earth matter effects here, but we have checked that, in the parameter ranges of interest to us, the day-night effect caused by the earth matter is of the order of few per cent, comparable to the day-night effect in the Standard Model. We thus anticipate that our limits would only change marginally if Earth matter effects were included.
In order to verify that the assumption of full adiabaticity is justified, we have examined the adiabaticity parameter γ in the two flavour approximation and we have checked that the adiabaticity condition [32]
$$\gamma ^ { - 1 } = \frac { \sin 2 \theta \frac { \Delta m _ { i j } ^ { 2 } } { 2 E } } { | \lambda _ { i } - \lambda _ { j } | ^ { 3 } } \cdot \left | \frac { d V _ { b } } { d t } \right | \ll 1$$
holds for all relevant mass squared difference ∆ m 2 ij even for large V b and the smallest relevant differences between the Hamiltonian eigenvalues λ i and λ j , which occur at the resonance position. We determine the derivative of the matter potential, ∣ ∣ dV /dt b ∣ ∣ , from the solar density profile of the standard solar model BS'05 (OP) [49].
In figure 5, we compare the measured solar neutrino oscillation probabilities P ν e → ν e to the theoretical predictions for standard three flavour oscillations and for the best fitting baryonic neutrino scenarios with glyph[epsilon1] > 0 (blue) and glyph[epsilon1] < 0 (red).
We observe that for glyph[epsilon1] < 0, a peak-like structure appears in P ν e → ν e , which suggests that mixing of ν e with other flavors is dynamically driven to zero for specific parameter combinations. The peak occurs at parameter points where ∆ m / 2 41 (2 E , ) ∆ m / 2 31 (2 E ) glyph[greatermuch] ∆ m / 2 21 (2 E ), V b and where moreover θ 34 and θ 13 are small. To understand its origin, it is therefore helpful to determine the eigenvalues of the Hamiltonian H flavor eff (see equation (10)) using time-independent perturbation theory, with the zeroth order Hamiltonian given by
$$H _ { \text{eff} } ^ { \text{flavour}, ( 0 ) } \equiv \frac { 1 } { 2 E } U \text{diag} ( 0, 0, \Delta m _ { 3 1 } ^ { 2 }, \Delta m _ { 4 1 } ^ { 2 } ) U ^ { \dagger } \,,$$
and the perturbation being H flavour,(1) eff ≡ H flavour eff -H flavour,(0) eff . In the approximation θ 34 = θ 13 = 0, a set of zeroth order eigenvectors is obviously given by the matrix U (0) ≡ R ′ 24 R ′ 14 R 23 , where, as before, R ij and R ′ ij are real and complex rotation matrices, respectively. Since H flavour,(0) eff has zero as a double eigenvalue, we next have to find eigenvectors of H flavour,(1) eff in the subspace corresponding to this double eigenvalue. In other words, we need to compute U (0) † H flavour,(1) eff U (0) and then diagonalize the upper left 2 × 2 block. It turns out that, if the condition
$$\frac { \Delta m _ { 2 1 } ^ { 2 } } { 2 E } \sin 2 \theta _ { 1 2 } + V _ { b } \cos \theta _ { 2 3 } \sin \theta _ { 1 4 } \sin 2 \theta _ { 2 4 } \simeq 0$$
is fulfilled, this 2 × 2 block is automatically diagonal. This implies that U (0) (1 0 0 0) , , , T glyph[similarequal] (1 , 0 0 0) , , is an approximate eigenvector of H flavour eff . Hence, if (34) holds at the center of the Sun, solar neutrinos are produced in an almost pure ν 1 mass eigenstate. After adiabatic flavour conversion, the resulting ν e admixture is of order cos θ 2 12 , leading to a peak in the observed solar neutrino spectrum at Earth.
## 3.2. Results
In figures 6 and 7 our constraints on the parameter space of baryonic sterile neutrinos are presented as contour plots for glyph[epsilon1] > 0 and glyph[epsilon1] < 0, respectively. We show exclusion limits (lines of constant χ 2 -χ 2 min ) at the 95% and 3 σ confidence levels. In each panel, we keep either glyph[epsilon1] or ∆ m 2 41
L
H
@ D Figure 5: Comparison of the measured solar neutrino oscillation probabilities to our theoretical predictions for standard three flavour oscillations (black) and for the best fit parameter points of the baryonic sterile neutrino model with glyph[epsilon1] > 0 (dotted blue) and glyph[epsilon1] < 0 (dashed red).
| | | glyph[epsilon1] = G B / ( √ 2 G F ) | ∆ m 2 41 [eV 2 ] | sin 2 θ 24 | χ 2 min / d.o.f. |
|-----------|---------------------|---------------------------------------|--------------------|--------------|--------------------|
| MINOS | glyph[epsilon1] > 0 | 16 . 9 | 0 . 014 | 0 . 0024 | 37 . 7 / 49 |
| MINOS | glyph[epsilon1] < 0 | - 19 . 2 | 0 . 037 | 0 . 00083 | 36 . 1 / 49 |
| MiniBooNE | glyph[epsilon1] > 0 | 30634 | 0 . 316 | 0 . 10 | 16 . 1 / 20 |
| MiniBooNE | glyph[epsilon1] < 0 | - 32000 | 0 . 116 | 0 . 75 | 16 . 4 / 20 |
| Solar | glyph[epsilon1] > 0 | 0 . 20 | insensitive | 1 . 0 | 1 . 10 / 3 |
| Solar | glyph[epsilon1] < 0 | - 38 . 0 | 0 . 013 | 0 . 046 | 0 . 41 / 3 |
Table II: Best fit values resulting from our parameter scan for the different experimental data sets. For the MiniBooNE fit with glyph[epsilon1] < 0 analysis the best fit value for glyph[epsilon1] is located outside the boundary of the analysis region, but χ 2 hardly depends on | glyph[epsilon1] | in this region. Also note that the solar best fit in the glyph[epsilon1] > 0 case has glyph[epsilon1] < 1 and is not sensitive to the exact value of ∆ m 2 41 in the interval [0 01 . , 11].
fixed at the value indicated in the plot and show constraints on the remaining two parameters. Moreover, as discussed in section 3.1, we fixed sin 2 2 θ 14 = 0 12. Blue lines correspond to constraints . from solar experiments, black lines are the limits from MINOS and the colored regions show the parameter region preferred by MiniBooNE. The best fit values for glyph[epsilon1] > 0 and glyph[epsilon1] < 0 are listed in table II.
We see that values of | glyph[epsilon1] | glyph[greaterorsimilar] 10 are strongly disfavored by MINOS except in the case of tiny activesterile mixing angles. For such large values of glyph[epsilon1] , the new MSW resonance at ∆ m / 2 41 (2 E ) ∼ V b lies within the MINOS energy range E < 50 GeV and leads to a constraint sin 2 θ 24 glyph[lessorsimilar] 10 -3 . Such small mixing angles are, however, irrelevant for possible explanations of MiniBooNE and other short-baseline anomalies. The MINOS contours also show that in most of the mass range 10 -2 eV 2 glyph[lessorsimilar] ∆ m 2 41 glyph[lessorsimilar] 10 1 eV , values of sin 2 2 θ 24 glyph[greaterorsimilar] 0 01 are excluded, with limits becoming much . stronger at large glyph[epsilon1] .
Solar neutrinos also have some sensitivity to θ 24 , but limits on glyph[epsilon1] vary a lot with sin 2 θ 24 . For intermediate values 0 01 . glyph[lessorsimilar] sin 2 θ 24 glyph[lessorsimilar] 0 1, even values of . | glyph[epsilon1] | as large as few × 10 3 are compatible with solar neutrino data. For glyph[epsilon1] > 0, we notice that solar limits on glyph[epsilon1] are weakest at sin 2 θ 24 ∼ few × 10 -2 .
Figure 6: 95% and 3 σ confidence level constraints on the parameters ∆ m 2 41 , sin 2 2 θ 24 and glyph[epsilon1] (strength of the new MSW potential) of the baryonic sterile neutrino model in the glyph[epsilon1] > 0 case. Blue contours show constraints from solar experiments, black contours are for MINOS and shaded areas correspond to the region preferred by MiniBooNE. We have fixed sin 2 2 θ 14 = 0 12, as motivated by the reactor and gallium anomalies. .
Figure 7: 95% and 3 σ confidence level constraints on the parameters ∆ m 2 41 , sin 2 2 θ 24 and glyph[epsilon1] (strength of the new MSW potential) of the baryonic sterile neutrino model in the glyph[epsilon1] < 0 case. Blue contours show constraints from solar experiments, black contours are for MINOS and shaded areas correspond to the region preferred by MiniBooNE. We have fixed sin 2 2 θ 14 = 0 12, as motivated by the reactor and gallium anomalies. .
e
24
Figure 8: Constraints on glyph[epsilon1] and sin 2 θ 24 from solar neutrinos for fixed sin 2 2 θ 14 = 0 12 (as motivated by the . reactor and gallium anomalies), but for different values of θ 34 . The value of ∆ m 2 41 has been marginalized over in the range 10 -2 ≤ ∆ m 2 41 ≤ 1 1 . · 10 . 1
In this regime, the additional neutrino disappearance due to nonzero θ 14 and θ 24 is partially compensated by V b -induced modifications to the MSW resonance structure. In particular, the 1-4 and 2-4 mixings imply that above the solar MSW resonance, ν 1 -ν 2 mixing is not as strongly suppressed as in the standard case. This reduces the flavour transition probability at energies above the resonance. Note that this effect is related to a sterile neutrino-induced smearing of the atmospheric resonance (which at the center of the Sun lies at about 200 MeV) to the extent that it has a small impact even at energies as low as ∼ 10 MeV. The effect is therefore absent if the neutrino mass ordering is inverted so that the atmospheric resonance lies in the anti-neutrino sector. We have checked that indeed the limits on glyph[epsilon1] from solar neutrino experiments become somewhat weaker in this case. For glyph[epsilon1] < 0, the exclusion contours reveal an allowed 'island' at glyph[epsilon1] ∼ -10 . 3 In the parameter region corresponding to these islands, the non-standard MSW resonance at ∆ m / E 2 41 2 glyph[similarequal] V b mimics the effect of the standard solar resonance. Also, in this parameter region, the atmospheric MSW resonance-modified by the presence of the sterile neutrinos-has a small impact. Therefore, the 'islands' move down by almost an order of magnitude in | glyph[epsilon1] | if the neutrino mass ordering is inverted. The ∆ m 2 41 -independent 'peninsula' at glyph[epsilon1] ∼ -20, is related to the appearance of the peak structure in P ν e → ν e which we discussed in section 3.1.3 and which is independent of the mass ordering.
The allowed parameter region for the measured appearance signal in MiniBooNE is very similar to the one obtained in conventional sterile neutrino scenarios (see for instance the analysis by the MiniBooNE collaboration themselves [2]) with the exception that for large matter potentials, the allowed region is expanded towards lower sin 2 θ 24 and higher ∆ m 2 41 .
We now relax our assumption θ 34 = 0. The main sensitivity to θ 34 is expected to come from solar neutrinos [15] (and from MINOS neutral current measurements, which we did not consider in this work, though). We compare the solar neutrino limits in the sin 2 θ 24 -glyph[epsilon1] plane for different values of θ 34 in figure 8, marginalizing over the sterile neutrino mass in the range 10 -2 ≤ ∆ m 2 41 ≤ 1 1 . · 10 . 1 We see that the constraints on glyph[epsilon1] become somewhat weaker if sin 2 θ 34 ∼ 0 01 and change significantly .
e
Figure 9: The effect of varying θ 14 on the constraints in the sin 2 θ 24 -glyph[epsilon1] plane. The plot in the center reproduces middle panel in the upper row of figure 6, while the left and right panels show similar constraints for smaller and larger θ 14 , respectively.
for larger values of sin 2 θ 34 . This implies that, for large θ 34 , a scenario with strong non-standard matter potential can be consistent with solar data and with MiniBooNE. Nevertheless, such a scenario would still be ruled out by MINOS.
Finally, let us also discuss the effect of choosing sin 2 2 θ 14 different from the value 0.12 preferred by the reactor neutrino anomaly. To this end, we show in figure 9 how the constraints on glyph[epsilon1] and θ 24 for fixed ∆ m 2 41 are modified if sin 2 2 θ 14 is taken a factor of 2 smaller (left panel) or a factor of 2 larger (right panel) than the preferred value. We see that the MiniBooNE preferred region, which is sensitive only to the combination sin 2 2 θ 14 sin 2 θ 24 is simply shifted by a factor of 2. Solar limits are affected in a less trivial way and we find that at large θ 14 , there is even a preference for nonzero θ 24 . Note, however, that the goodness of fit becomes slightly worse as θ 14 is increased: the minimum χ / 2 dof is 0.8/3 for sin 2 2 θ 14 = 0 6 and 2.7/3 for sin . 2 2 θ 14 = 0 24. . Finally, MINOS limits are weakened if θ 14 is large, especially at large glyph[epsilon1] . This happens because a larger mixing between ν e and ν b by unitarity implies more ν µ disappearance.
## 4. CONCLUSIONS
To summarize, we have derived constraints on models with extended sterile neutrino sectors that feature in particular a new gauge interaction between sterile neutrinos and SM particles. As a specific example, we have considered a scenario in which eV-scale sterile neutrinos are charged under gauged baryon number U (1) B . In principle, such interactions could be several orders of magnitude stronger than SM weak interactions, so the Mikheyev-Smirnov-Wolfenstein (MSW) potentials they generate could be significantly larger than the matter potential in standard three-flavour neutrino oscillations.
We have also computed approximate analytic expressions for the relevant oscillation probabilities in matter, improving and extending the expressions previously derived in [31]. We have then numerically analyzed data from the MINOS experiment, from solar neutrino measurements and from MiniBooNE to show that new gauge interactions in the sterile neutrino sector cannot be large unless the active-sterile neutrino mixing is very small. In particular, if the ratio glyph[epsilon1] of the non-standard and standard matter potentials is larger than ∼ 10, MINOS excludes mixing angles down to sin 2 2 θ 24 ∼ 10 -3 . (This limit becomes stronger if θ 14 = 0.)
We conclude that sterile neutrino searches in oscillation experiments are powerful tools to con-
strain certain models with hidden sector gauge interactions. We also conclude that such models do not help to resolve the tension in the global fit to short-baseline oscillation data.
Comparing to the interaction strength required for baryonic sterile neutrinos to yield signals in dark matter detectors [26-29], we conclude that in the case of eV scale sterile neutrinos, baryonic interactions cannot be large enough to be observable in the current generation of experiments. On the other hand, interesting signals may still be possible in future ton-scale experiments.
## Acknowledgments
It is a pleasure to thank Janet Conrad and Georgia Karagiorgi for very helpful discussions.
- [1] A. Aguilar et al. (LSND), Phys. Rev. D64 , 112007 (2001), hep-ex/0104049.
- [2] A. Aguilar-Arevalo et al. (MiniBooNE Collaboration) (2012), 1207.4809.
- [3] T. Mueller, D. Lhuillier, M. Fallot, A. Letourneau, S. Cormon, et al., Phys.Rev. C83 , 054615 (2011), 1101.2663.
- [4] G. Mention, M. Fechner, T. Lasserre, T. Mueller, D. Lhuillier, et al., Phys.Rev. D83 , 073006 (2011), 1101.2755.
- [5] P. Huber, Phys.Rev. C84 , 024617 (2011), 1106.0687.
- [6] A. Hayes, J. Friar, G. Garvey, and G. Jonkmans (2013), 1309.4146.
- [7] M. A. Acero, C. Giunti, and M. Laveder, Phys.Rev. D78 , 073009 (2008), 0711.4222.
- [8] C. Giunti and M. Laveder, Phys.Rev. C83 , 065504 (2011), 1006.3244.
- [9] J. Kopp, M. Maltoni, and T. Schwetz, Phys.Rev.Lett. 107 , 091801 (2011), 1103.4570.
- [10] C. Giunti and M. Laveder, Phys.Lett. B706 , 200 (2011), 1111.1069.
- [11] G. Karagiorgi (2011), 1110.3735.
- [12] C. Giunti and M. Laveder, Phys.Rev. D84 , 093006 (2011), 1109.4033.
- [13] C. Giunti and M. Laveder, Phys.Rev. D84 , 073008 (2011), 1107.1452.
- [14] K. Abazajian, M. Acero, S. Agarwalla, A. Aguilar-Arevalo, C. Albright, et al. (2012), 1204.5379.
- [15] J. Kopp, P. A. N. Machado, M. Maltoni, and T. Schwetz, JHEP 1305 , 050 (2013), 1303.3011.
- [16] P. Ade et al. (Planck Collaboration) (2013), 1303.5076.
- [17] P. Ade et al. (BICEP2 Collaboration) (2014), 1403.3985.
- [18] E. Giusarma, E. Di Valentino, M. Lattanzi, A. Melchiorri, and O. Mena (2014), 1403.4852.
- [19] C. Dvorkin, M. Wyman, D. H. Rudd, and W. Hu (2014), 1403.8049.
- [20] H. Li, J.-Q. Xia, and X. Zhang (2014), 1404.0238.
- [21] J.-F. Zhang, Y.-H. Li, and X. Zhang (2014), 1403.7028.
- [22] M. Archidiacono, N. Fornengo, S. Gariazzo, C. Giunti, S. Hannestad, et al. (2014), 1404.1794.
- [23] S. Hannestad, R. S. Hansen, and T. Tram, Phys.Rev.Lett. 112 , 031802 (2014), 1310.5926.
- [24] B. Dasgupta and J. Kopp, Phys.Rev.Lett. 112 , 031803 (2014), 1310.6337.
- [25] T. Bringmann, J. Hasenkamp, and J. Kersten (2013), 1312.4947.
- [26] M. Pospelov, Phys.Rev. D84 , 085008 (2011), 1103.3261.
- [27] R. Harnik, J. Kopp, and P. A. Machado, JCAP 1207 , 026 (2012), 1202.6073.
- [28] M. Pospelov and J. Pradler, Phys.Rev. D85 , 113016 (2012), 1203.0545.
- [29] M. Pospelov and J. Pradler (2013), 1311.5764.
- [30] C. M. Ho and R. J. Scherrer, Phys.Rev. D87 , 065016 (2013), 1212.1689.
- [31] G. Karagiorgi, M. Shaevitz, and J. Conrad (2012), 1202.1024.
- [32] E. K. Akhmedov (1999), hep-ph/0001264.
- [33] C. Giunti and C. W. Kim, Fundamentals of Neutrino Physics and Astrophysics (OUP Oxford, New York, 2007).
- [34] A. Aguilar-Arevalo et al. (The MiniBooNE Collaboration), Phys.Rev.Lett. 98 , 231801 (2007), 0704.1500.
- [35] A. Aguilar-Arevalo et al. (MiniBooNE Collaboration), Phys.Rev.Lett. 103 , 111801 (2009), 0904.1958.
- [36] A. Aguilar-Arevalo et al. (The MiniBooNE Collaboration), Phys.Rev.Lett. 105 , 181801 (2010), 1007.1150.
- [37] P. Huber, J. Kopp, M. Lindner, M. Rolinec, and W. Winter, Globes manual (2013), URL http: //www.mpi-hd.mpg.de/personalhomes/globes/documentation/globes-manual-3.0.8.pdf .
- [38] M. Beuthe, Phys.Rept. 375 , 105 (2003), hep-ph/0109119.
- [39] M. Gonzalez-Garcia, M. Maltoni, J. Salvado, and T. Schwetz, JHEP 1212 , 123 (2012), 1209.3023.
- [40] P. Huber, M. Lindner, and W. Winter, Comput.Phys.Commun. 167 , 195 (2005), hep-ph/0407333.
- [41] P. Huber, J. Kopp, M. Lindner, M. Rolinec, and W. Winter, Comput.Phys.Commun. 177 , 432 (2007), hep-ph/0701187.
- [42] D. Michael et al. (MINOS Collaboration), Phys.Rev.Lett. 97 , 191801 (2006), hep-ex/0607088.
- [43] D. Anderson, Theory of the Earth (Blackwell Scientific Publications, 1989), ISBN 978-0-865-42123-3.
- [44] M. Diwan, B. Viren, D. Harris, A. Marchionni, J. Morfin, et al. (2004).
- [45] J. de Jong (MINOS Collaboration), Nucl.Phys.Proc.Suppl. 237-238 , 166 (2013).
- [46] P. Adamson et al. (MINOS Collaboration), Phys.Rev.Lett. 110 , 251801 (2013), 1304.6335.
- [47] A. Aguilar-Arevalo et al. (MiniBooNE Collaboration), Phys.Rev.Lett. 110 , 161801 (2013), 1207.4809.
- [48] G. Bellini et al. (Borexino Collaboration) (2013), 1308.0443.
- [49] J. N. Bahcall, A. M. Serenelli, and S. Basu, Astrophys.J. 621 , L85 (2005), astro-ph/0412440. | 10.1007/JHEP12(2014)104 | [
"Joachim Kopp",
"Johannes Welter"
] | 2014-08-01T20:00:29+00:00 | 2014-08-01T20:00:29+00:00 | [
"hep-ph"
] | The Not-So-Sterile 4th Neutrino: Constraints on New Gauge Interactions from Neutrino Oscillation Experiments | Sterile neutrino models with new gauge interactions in the sterile sector are
phenomenologically interesting since they can lead to novel effects in neutrino
oscillation experiments, in cosmology and in dark matter detectors, possibly
even explaining some of the observed anomalies in these experiments. Here, we
use data from neutrino oscillation experiments, in particular from MiniBooNE,
MINOS and solar neutrino experiments, to constrain such models. We focus in
particular on the case where the sterile sector gauge boson $A'$ couples also
to Standard Model particles (for instance to the baryon number current) and
thus induces a large Mikheyev-Smirnov-Wolfenstein potential. For eV-scale
sterile neutrinos, we obtain strong constraints especially from MINOS, which
restricts the strength of the new interaction to be less than $\sim 10$ times
that of the Standard Model weak interaction unless active-sterile neutrino
mixing is very small ($\sin^2 \theta_{24} \lesssim 10^{-3}$). This rules out
gauge forces large enough to affect short baseline experiments like MiniBooNE
and it imposes nontrivial constraints on signals from sterile neutrino
scattering in dark matter experiments. |
1408.0290v1 | ## NGC6522: A typical globular cluster in the Galactic bulge without signatures of rapidly rotating Population III stars
## Melissa Ness 1 /star , Martin Asplund 2 , Andrew R. Casey 3
Max-Planck-Institut f¨r Astronomie, K¨nigstuhl 17, D-69117 Heidelberg, Germany u o 2 Research School of Astronomy and Astrophysics, Australian National University, Cotter Road, Weston Creek, ACT 2611, Australia 3 Institute of Astronomy, University of Cambridge, Madingley Road, Cambridge CB3 0HA
1
Accepted ? Received ?; in original form April 01, 2014
## ABSTRACT
We present an abundance analysis of eight potential member stars of the old Galactic bulge globular cluster NGC6522. The same stars have previously been studied by Chiappini et al. (2011), who found very high abundances of the slow neutron capture elements compared with other clusters and field stars of similar metallicity, which they interpreted as reflecting nucleosynthesis in rapidly rotating, massive Population III stars. In contrast to their analysis, we do not find any unusual enhancements of the neutron capture elements Sr, Y, Ba and Eu and conclude that previous claims result mainly from not properly accounting for blending lines. Instead we find NGC 6522 to be an unremarkable globular cluster with comparable abundance trends to other Galactic globular clusters at the same metallicity ([Fe / H] = -1 15 . ± 0 16). The stars are . also chemically similar to halo and bulge field stars at the same metallicity, spanning a small range in [Y/Ba] and with normal α -element abundances. We thus find no observational evidence for any chemical signatures of rapidly rotating Population III stars in NGC 6522.
Key words: Stars
## 1 INTRODUCTION
As one of the oldest stellar populations in our Galaxy, globular clusters are key to understanding the earliest epochs of galaxy evolution and stellar nucleosynthesis. Globular clusters may therefore contain unique chemical markers of the first generations of stars. All globular clusters indeed show abundance peculiarities compared with the typical field population, such as an O-Na anti-correlation, which is normally interpreted as reflecting two or more generations of stars in the clusters with the first one sharing the chemical composition with normal field stars of the same metallicity (Gratton et al. 2012).
The nature of the polluter of the later generations of stars is still being debated with intermediate mass asymptotic giant branch stars or rapidly rotating massive stars the leading contenders. As the polluters would be part of the first cluster generation, some of which are still present, they would thus not be metal-free Population III stars but Population II stars of the same metallicity as the cluster as a whole. Until recently there have been no convincing observations of Population III nucleosynthesis in any globular
/star [email protected]
cluster star but Chiappini et al. (2011) claim to have found exactly such a smoking gun in NGC 6522.
NGC6522 is a little studied globular cluster in the Milky Way's bulge. Observations of individual stars in this cluster are difficult due to the large crowding of more metalrich bulge stars and foreground disk contamination in this region as well as the heavy interstellar extinction and reddening characteristic of sight lines of the Galactic bulge, especially near the plane. However, this is a particularly important globular cluster because it may be the oldest globular cluster in the Milky Way (Barbuy et al. 2009) and it may contain vital clues to the formation and early evolution of the still poorly understood bulge (e.g. Ness et al. 2013). Chiappini et al. (2011) recently analysed eight potential NGC6522 cluster members and found remarkably high abundances of slow neutron-capture elements, distinct from any other globular cluster observed. The authors demonstrated that such a chemical abundance pattern would be consistent with the nucleosynthesis predicted for rapidly rotating massive, extremely metal-poor or Population III stars, so-called spinstars (Pignatari et al. 2008).
The fact that this cluster is so far unique in carrying such a nucleosynthetic signature may reflect that the very first stars born after the Big Bang formed in the largest overdensities, which subsequently grew with time to become cen-
tres of galaxies. The oldest Galactic stars should therefore be concentrated in the bulge region (Tumlinson 2010), which also experienced the most rapid chemical enrichment due to the intense star formation rate, making it conceivable that any Population III chemical fingerprint may preferentially show up in an old bulge cluster with [Fe / H] ≈ -1 such as NGC6522.
Given the potentially far-reaching implications of the work of Chiappini et al. (2011), we present a re-analysis of the same cluster stars but importantly with a more detailed accounting of blending lines as well as isotopic hyperfine splitting of the s -process abundance diagnostics.
## 2 OBSERVATIONS
The GIRAFFE/VLT data from the program 071.B-0617(A) (PI=A. Renzini) were downloaded from the ESO archive. These observations were taken during May-July 2003 and comprised of two field pointings and three wavelength settings: HR13 (612-640nm), HR14 (638.3-662.6nm) and HR15 (679.7-696.5nm). A more comprehensive description of the data can be found in Barbuy et al. (2009). A number of calibration frames were missing from the ESO archive but were provided by ESO support upon request. We could not locate all of the data corresponding to the total exposure times reported in Barbuy et al. (2009) for unknown reasons. The available data were reduced using Gasgano (version 1.24) with the wavelength calibration verified using skylines and telluric lines. The data were sky subtracted using an automated python routine that determined a median sky from the available sky fibres and removed it from each science frame. For each wavelength setting, separate frames were co-added. The resulting signal-to-noise ( S/N ) ratio per resolution element for the three wavelength settings are listed in Table 1.
The stellar radial velocities were determined from a cross-correlation with a theoretical model atmosphere spectrum (corresponding to T eff = 5000 K, log g = 3 0 [cgs], . [Fe/H]= -1 0) smoothed to the resolving power of the ob-. servations using the routine fxcor in pyraf . Bad pixels and telluric regions were masked out for this procedure. Our measured radial and heliocentric velocities are listed in Table 1. The mean heliocentric velocity we report for these nine candidate NGC6522 stars is 〈 V helio 〉 = -13 4 . ± 4 9 km s . -1 , which is in reasonable agreement with the value quoted by Harris (2010): -21 1 . ± 3 4 km s . -1 . It is possible that some of the candidate stars may not be cluster members in spite of the similarity in velocity and metallicity, although there is some discussion in the literature regarding the actual heliocentric velocity of the cluster (Terndrup et al. 1998). It should be noted that our heliocentric velocities differ from those reported by (Barbuy et al. 2009) based on the same observations for unknown reasons. Our inferred heliocentric velocities from the three different wavelength settings obtained at different times are in excellent agreement with each other ( σ < 0 7 km s . -1 ), which is not true for Barbuy et al. (2009) (see their Table 3).
Table 2. Summary of derived stellar parameters
| Star | T eff | log g | [Fe/H] | ξ turb |
|--------|---------|---------|----------|----------|
| B-008 | 4960 | 2.65 | -1.02 | 2.3 |
| B-107 | 4970 | 1.9 | -1.34 | 2.1 |
| B-108 | 4750 | 2.2 | -1.37 | 1.4 |
| B-118 | 5000 | 2.25 | -1.04 | 2.45 |
| B-122 | 5100 | 2.75 | -0.95 | 1.2 |
| B-128 | 4760 | 2.1 | -1.02 | 1.3 |
| B-130 | 5090 | 2.65 | -1.23 | 2.8 |
| F-121 | 4955 | 2.2 | -1 | 1.4 |
## 3 ABUNDANCE ANALYSIS
After continuum normalisation, we performed a standard 1D LTE abundance analysis using the SMH spectrum analysis program of Casey et al. (2013), which is built around the well-known MOOG code of Sneden (1973). Initial effective temperature guesses were provided by de-reddened 2MASS photometry (Skrutskie et al. 2006) using the Bessell et al. (1998) calibration. Similarly to Barbuy et al. (2009) we discard the star B-134 from further analysis due to contamination of the spectra from a companion/blending star. Spectroscopic stellar parameters of T eff , log g and [Fe/H] were determined from an iterative procedure to achieve excitation and ionisation balance using ∼ 60 Fe i lines and 12 Fe ii lines. The microturbulence parameter ( ξ turb ) was calculated by removing any abundance trends with reduced equivalent width. The final stellar parameters are reported in Table 2. The typical uncertainties in the stellar parameters are σ T ( eff ) = 80K, σ (log g ) = 0 20 and . σ ([Fe / H]) = 0 15, re-. spectively. We report a marginally lower mean metallicity for these eight stars than Barbuy et al. (2009): [Fe / H] = -1 15 . ± 0 16 compared to [Fe . / H] = -1 00 . ± 0 20. .
Individual abundances were determined from the equivalent widths of unblended lines for which reliable oscillator strengths were available from the Gaia-ESO line list (Heiter et al., in prep. ). In the limited wavelength regions of these GIRAFFE observations, we made use of 22 lines of the α -elements (Mg, Si, Ca, Ti), two lines available for Al, two for Na, two for Ba, three for Eu, one for La and one for Y. For the neutron capture elements we performed spectrum synthesis taking care of including isotopic and hyperfine splitting (HFS) and blending lines to ensure the most reliable results. Our mean elemental abundance measurements are given in Table 3 while the full line-by-line results are provided in Table 4 (online only). We estimate the uncertainties on individual abundances from the quadrature sum of the uncertainties due to the noise in the spectra, and the uncertainties in the stellar parameters and the line-to-line dispersion. These are listed in Table 3.
In general we find comparable stellar abundances to Barbuy et al. (2009), with marginally lower [Fe/H] (by 0 15 dex) . and slightly higher [ α /Fe] ratios (by ∼ 0 2 dex). . These differences can likely be traced to the adopted linelists and difference in adopted stellar parameters (our T eff and ξ turb are typically 200 K and 0.6 km s -1 , respectively higher). We also report roughly similar abundances for [Na/Fe], [La/Fe] and [Eu/Fe] (which have mean values of +0 2 dex, +0 4 dex and +0 4 dex) but lower Ba abundance . . .
Table 1. Summary of the measured velocities the nine NGC 6522 candidate members
| Star | Coordinates (J2000) | Grating | S/N | V rad [km s - 1 ] | V helio [km s - 1 ] | 〈 V helio 〉 [km s - 1 ] | σ ( V helio ) [km s - 1 ] | [Fe/H] dex |
|--------|--------------------------|-----------|-------|---------------------|-----------------------|-----------------------------|-----------------------------|--------------|
| B-008 | 18:03:46.04, -30:00:50.9 | HR13 | 88 | -29.9 | -8.8 | -8.6 | 0.3 | -1.02 |
| B-008 | | HR14 | 81 | -8.6 | -8.6 | | | |
| B-008 | | HR15 | 132 | -16.0 | -8.2 | | | |
| B-107 | 18:03:36.59, -30:02:16.1 | HR13 | 93 | -27.0 | -5.9 | -6.0 | 0.4 | -1.34 |
| B-107 | | HR14 | 86 | -6.5 | -6.5 | | | |
| B-107 | | HR15 | 134 | -13.3 | -5.5 | | | |
| B-108 | 18:03:35.18, -30:02:04.9 | HR13 | 156 | -33.1 | -12.1 | -13.0 | 0.7 | -1.37 |
| B-108 | | HR14 | 142 | -13.7 | -13.8 | | | |
| B-108 | | HR15 | 188 | -20.9 | -13.1 | | | |
| B-118 | 18:03:42.24, -30:03:39.9 | HR13 | 87 | -42.9 | -21.9 | -21.9 | 0.1 | -1.04 |
| B-118 | | HR14 | 85 | -22.0 | -22 | | | |
| B-118 | | HR15 | 133 | -29.7 | -21.9 | | | |
| B-122 | 18:03:33.34, -30:01:58.3 | HR13 | 96 | -36.9 | -15.8 | -15.7 | 0.1 | -0.95 |
| B-122 | | HR14 | 92 | -15.6 | -15.7 | | | |
| B-122 | | HR15 | 134 | -23.3 | -15.5 | | | |
| B-128 | 18:03:44.61, -30:02:10.4 | HR13 | 84 | -33.9 | -12.9 | -12.6 | 0.3 | -1.02 |
| B-128 | | HR14 | 83 | -12.7 | -12.7 | | | |
| B-128 | | HR15 | 119 | -19.9 | -12.1 | | | |
| B-130 | 18:03:41.00, -30:03:03.0 | HR13 | 77 | -36.6 | -15.6 | -15.6 | 0.2 | -1.23 |
| B-130 | | HR14 | 71 | -15.3 | -15.4 | | | |
| B-130 | | HR15 | 117 | - 23.7 | -15.9 | | | |
| B-134 | 18:03:41.16, -30:02:21.6 | HR13 | 84 | -39.4 | -18.4 | -18.9 | 0.4 | - |
| B-134 | | HR14 | 81 | -19.3 | -19.4 | | | |
| B-134 | | HR15 | 127 | -26.8 | -18.9 | | | |
| F-121 | 18:03:36.41, -30:02:19.8 | HR13 | 58 | -14.1 | -8.7 | -8.8 | 0.2 | -0.99 |
| F-121 | | HR14 | 93 | 4.3 | -8.6 | | | |
| F-121 | | HR15 | 140 | -7.3 | -9.1 | | | |
compared to Barbuy et al. (2009). The inclusion of hyperfine splitting makes only a small difference to the measurements. Our mean measurement for Ba is [Ba / Fe] = 0 35 . ± 0 2, com-. pared to [Ba / Fe] = 0 5 . ± 0 4 in the original analysis. .
A more striking discrepancy is obtained for Y compared to the analysis of Chiappini et al. (2011). Our mean ratio is [Y / Fe] = 0 25 . ± 0 10 compared to [Y . / Fe] = 1 2 . ± 0 2 in their . analysis. Here the difference is largely due to adopted Y lines. In our analysis, we use a single unblended Y ii line at 679.54 nm, whereas the original analysis relied on the Y ii line at 661.37 nm, which they did not realise is severely blended by in particular a Fe line at 661.38nm and a Ti line at 661.36 nm (see Figure 1). We obtain a higher abundance of [Y/Fe] by about +0.6 dex when fitting for [Y/Fe] at 661.37 nm without accounting for the blending lines.
Finally, we argue that no reliable Sr abundance can be inferred from these individual GIRAFFE spectra. Chiappini et al. (2011) relied on the Sr i 650.40 nm line, which in our opinion is not discernible over the noise of the individual spectra and is blended. This feature, for the median combination of all 8 stars is shown in Figure 2, synthesised with 0 -0.8 dex of [Sr/Fe] enhancement and a fixed [V/Fe] = 0. A good fit is obtained with an [Sr/Fe] = 0.4 given a [V/Fe] = 0. The four Sr lines in the GaiaESO line-list available in this wavelength region (640.85 nm,
Figure 1. The blended Y II feature, which is very similar for all stars, is shown for star B-128. The panel at left shows the synthesis of this line with the contribution of Ti I and Fe I in blue and the contribution of Ti I, Fe I and Y II in red, for the Y value measured for this star shown at right, of [Y/Fe] = 0.4 dex (and [Ti/Fe] = 0.25 as measured from the Ti lines in the spectra). The panel at right shows the best fit abundance of [Y/Fe] = 0.4 dex and syntheses at ± 0.2 dex comparison, for the unblended feature at 679.4nm which we use to derive all of our [Y/Fe] abundances. The parameters of the absorption features are provided in the figures, to indicate the relative line strengths, in order of element, wavelength, excitation potential and log gf
655.02 nm, 654.68 nm, 661.73 nm, 671.91 nm) are also too weak for abundance purposes. As a result we do not quote any [Sr/Fe] values.
Table 3. Summary of measured abundances for individual elements for the eight candidate NGC 6522 stars. The measurement for each star is in the first row and the upper and lower errors of the measurement are in the two following rows.
| star | [Fe/H] | [FeI/H] | [FeII/H] | [NaI/Fe] | [MgI/Fe] | [AlI/Fe] | [SiI/Fe] | [CaI/Fe] | [TiI/Fe] | [YII/Fe] | [BaII/Fe][LaII/Fe] [EuII/Fe] | [BaII/Fe][LaII/Fe] [EuII/Fe] | [BaII/Fe][LaII/Fe] [EuII/Fe] |
|--------|----------|-----------|------------|------------|------------|------------|------------|------------|------------|------------|--------------------------------|--------------------------------|--------------------------------|
| B-008 | -1.02 | -1.02 | -1.03 | 0.68 | 0.23 | 1 | 0.30 | 0.25 | 0.46 | 0.35 | 0.45 | 0.7 | 0.7 |
| B-008 | 0.15 | 0.15 | 0.15 | 0.1 | 0.2 | 0.1 | + 0.1 | 0.2 | 0.1 | 0.2 | 0.1 | 0.15 | 0.15 |
| B-008 | -0.15 | -0.15 | -0.15 | -0.1 | -0.2 | -0.1 | -0.1 | -0.2 | -0.1 | -0.3 | -0.1 | -0.2 | -0.15 |
| B-107 | -1.34 | -1.34 | -1.34 | 0.13 | 0.61 | 0.49 | 0.52 | 0.34 | 0.27 | 0.15 | 0.4 | 0.3 | 0.25 |
| B-107 | 0.15 | 0.15 | 0.15 | 0.25 | 0.2 | 0.25 | +0.15 | 0.15 | 0.1 | 0.25 | 0.1 | 0.2 | 0.15 |
| B-107 | -0.15 | -0.15 | -0.15 | -0.3 | -0.25 | -0.25 | -0.15 | -0.15 | -0.1 | -0.3 | -0.1 | -0.25 | -0.15 |
| B-108 | -1.37 | -1.37 | -1.37 | -0.23 | 0.63 | 0.26 | 0.18 | 0.38 | 0.37 | 0.15 | 0.1 | 0.35 | 0.4 |
| B-108 | 0.15 | 0.15 | 0.15 | 0.15 | 0.15 | 0.15 | +0.15 | 0.15 | 0.15 | -0.05 | 0.15 | 0.25 | 0.15 |
| B-108 | -0.15 | -0.15 | -0.15 | -0.15 | -0.15 | -0.4 | -0.40 | -0.15 | -0.15 | -0.05 | -0.15 | -0.45 | -0.2 |
| B-118 | -1.04 | -1.04 | -1.04 | 0.4 | 0.37 | 1.01 | 0.29 | 0.28 | 0.54 | 0.3 | 0.3 | 0.55 | 0.4 |
| B-118 | 0.15 | 0.15 | 0.15 | 0.1 | 0.1 | 0.1 | +0.1 | 0.15 | 0.15 | 0.2 | 0.1 | 0.1 | 0.15 |
| B-118 | -0.15 | -0.15 | -0.15 | -0.1 | -0.1 | -0.1 | -0.1 | -0.15 | -0.15 | -0.3 | -0.1 | -0.1 | -0.15 |
| B-122 | -0.95 | -0.95 | -0.95 | 0.42 | 0.36 | 0.62 | 0.21 | 0.29 | 0.33 | 0.1 | 0.35 | 0.3 | 0.5 |
| B-122 | 0.15 | 0.15 | 0.15 | 0.1 | 0.15 | 0.1 | +0.15 | 0.15 | 0.1 | 0.25 | 0.1 | 0.1 | 0.15 |
| B-122 | -0.15 | -0.15 | -0.15 | -0.1 | -0.15 | -0.1 | -0.15 | -0.15 | -0.1 | -0.35 | -0.1 | -0.15 | -0.15 |
| B-128 | -1.02 | -1.03 | -1.03 | 0.2 | 0.34 | 0.82 | 0.21 | 0.5 | 0.25 | 0.4 | 0.8 | 0.3 | 0 |
| B-128 | 0.1 | 0.1 | 0.1 | 0.15 | 0.15 | 0.1 | +0.15 | 0.15 | 0.1 | 0.2 | 0.1 | 0.15 | 0.15 |
| B-128 | -0.1 | -0.1 | -0.1 | -0.15 | -0.15 | -0.1 | -0.15 | -0.15 | -0.1 | -0.25 | -0.1 | -0.15 | -0.15 |
| B-130 | -1.23 | -1.24 | -1.24 | 0.46 | 0.34 | 0.64 | 0.32 | 0.35 | 0.4 | 0.44 | 0.1 | 0.45 | 0.9 |
| B-130 | 0.15 | 0.15 | 0.15 | 0.15 | 0.25 | 0.2 | +0.2 | 0.15 | 0.1 | 0.25 | 0.1 | 0.2 | 0.15 |
| B-130 | -0.15 | -0.15 | -0.15 | -0.25 | -0.25 | -0.2 | -0.2 | -0.15 | -0.1 | -0.3 | -0.1 | -0.2 | -0.15 |
| F-121 | -1 | -0.99 | -1.01 | -0.11 | 0.54 | 0.12 | 0.20 | 0.43 | 0.42 | 0.2 | 0.45 | 0.1 | 0.3 |
| F-121 | 0.15 | 0.15 | 0.15 | 0.2 | 0.2 | 0.2 | +0.15 | 0.15 | 0.15 | 0.2 | 0.2 | 0.1 | 0.1 |
| F-121 | -0.15 | -0.15 | -0.15 | -0.3 | -0.15 | -0.2 | -0.15 | -0.15 | -0.15 | -0.25 | -0.2 | -0.1 | -0.15 |
Figure 2. The median spectra from all 8 stars combined for the Sr feature at 650.4nm. The synthesis shows the line profiles for abundance ratios of [Sr/Fe] = 0,0.2,0.4,0.6 for the typical [V/Fe] ratio of [V/Fe] ≈ 0 for the stars. The solid grey lines represent the 1 -σ measurement of the mean of the 8 stars and the dashed lines either side of the line profile show the width of the noise of the combined spectra. The best fit to the line adopting [V/Fe] = 0 is [Sr/Fe] = 0.40.
Figure 4 compares our inferred abundance ratios for NGC6522 to micro-lensed dwarf stars of the bulge from Bensby et al. (2013) as well as typical halo stars (Suda et al. 2008), demonstrating in general a good agreement. The cluster stars show comparable abundances to the bulge for Mg, Si, Ca, Ti, Y and Ba, while the scatter is Na and Al is directly attributable to the Na-Al correlation typical for globular clusters. Our [Y/Fe] and [Ba/Fe] ratios are consistent with other globular clusters at this metallicity (e.g. Johnson et al. 2010). Figure 3 shows the [Y/Ba] of the eight NGC6522 stars compared to field bulge and halo stars. The cluster stars are not anomalous compared to field stars of
1
Figure 3. [Y/Ba] ratios for stars of the Galactic halo from the SAGA database (grey circles) (Suda et al. 2008) and bulge stars of Bensby et al., (2013) (cyan diamonds) compared to the eight candidate NGC6522 stars (blue squares). Typical model predictions for the contribution of the main r-process is shown at 0.16 dex, along the dashed black line and the model predictions for a standard (ST) and reduced efficiency ST/12 (factor 12) AGB models are shown in the red and black crosses, respectively (see Figure 1 from Chiappini et al. (2011).
similar metallicity, neither in [Y/Ba] nor in their scatter, in contrast to the conclusion by Chiappini et al. (2011).
Figure 4. Comparison of Na, Mg, Ca, Si, Ti, Al, Ba and Y abundances measured for the eight stars (blue squares) compared with stars of the bulge (Bensby et al. 2013) (cyan triangles) and halo stars of the SAGA database (grey crosses).
## 4 CONCLUSIONS
## REFERENCES
We have re-analysed available GIRAFFE/VLT spectra of eight members of the bulge globular cluster NGC 6522 from the ESO archive. We have paid particular attention to considering the effects of isotopic hyperfine splitting and blending lines on the derived abundances of the s -process elements. We also use in some cases on more reliable, cleaner lines. As a result our results are systematically and significantly lower than those reported by Chiappini et al. (2011), which did not consider those effects.
In all cases the NGC 6522 stars display elemental abundances typical of field halo and bulge as well as other globular clusters of similar metallicities. We conclude that neither the normal s -process abundances nor the low and fairly constant [Y/Ba] ratios measured in NGC 6522 support a nucleosynthetic origin by rapidly rotating Population III stars. In terms of its chemical composition, NGC 6522 is an unremarkable globular cluster.
## ACKNOWLEDGMENTS
We appreciate helpful and stimulating discussions with Beatriz Barbuy and Cristina Chiappini. The research has received funding from the European Research Council under the European Union's Seventh Framework Programme (FP 7) ERC Grant Agreement n. [321035]. MA acknowledges generous funding from the Australian Research Council (grant FL110100012) MA acknowledges generous funding from the Australian Research Council (grant FL110100012). ARC acknowledges funding from European Research Council grant 320360: The Gaia-ESO Milky Way Survey.
Barbuy B., Zoccali M., Ortolani S., Hill V., Minniti D., Bica E., Renzini A., G´mez A., 2009, A&A, 507, 405 o Bensby T., Yee J. C., Feltzing S., Johnson J. A., Gould A., Cohen J. G., Asplund M., Mel´ndez J., Lucatello S., et al. e
2013, A&A, 549, A147 Bessell M. S., Castelli F., Plez B., 1998, A&A, 333, 231 Casey A. R., Keller S., Alves-Brito A., Frebel A., Da Costa G., Karakas A., Yong D., Schlaufman K., Jacobson H. R., Yu Q., Fishlock C., 2013, ArXiv e-prints Chiappini C., Frischknecht U., Meynet G., Hirschi R., Barbuy B., Pignatari M., Decressin T., Maeder A., 2011, Nature, 472, 454 Gratton R. G., Carretta E., Bragaglia A., 2012, AAPR, 20, 50 Harris W. E., 2010, ArXiv e-prints Johnson J. A., Dong S., Gould A., 2010, ApJ, 713, 713 Ness M., Freeman K., Athanassoula E., Wylie-de-Boer E., Bland-Hawthorn J., Asplund M., Lewis G. F., Yong D., Lane R. R., Kiss L. L., 2013, MNRAS, 430, 836 Pignatari M., Gallino R., Meynet G., Hirschi R., Herwig F., Wiescher M., 2008, ApJ, 687, L95 Skrutskie M. F., Cutri R. M., Stiening R., Weinberg M. D., Schneider S., Carpenter J. M., Beichman C., Capps R., Chester e. a., 2006, AJ, 131, 1163 Sneden C. A., 1973, PhD thesis, THE UNIVERSITY OF TEXAS AT AUSTIN. Suda T., Katsuta Y., Yamada S., Suwa T., Ishizuka C., Komiya Y., Sorai K., Aikawa M., Fujimoto M. Y., 2008, PASJ, 60, 1159 Terndrup D. M., Popowski P., Gould A., Rich R. M., Sadler E. M., 1998, AJ, 115, 1476
Tumlinson J., 2010, ApJ, 708, 1398 | 10.1093/mnras/stu2144 | [
"Melissa Ness",
"Martin Asplund",
"Andrew R. Casey"
] | 2014-08-01T20:00:41+00:00 | 2014-08-01T20:00:41+00:00 | [
"astro-ph.GA"
] | NGC 6522: A typical globular cluster in the Galactic bulge without signatures of rapidly rotating Population III stars | We present an abundance analysis of eight potential member stars of the old
Galactic bulge globular cluster NGC6522. The same stars have previously been
studied by Chiappini et al. (2011), who found very high abundances of the slow
neutron capture elements compared with other clusters and field stars of
similar metallicity, which they interpreted as reflecting nucleosynthesis in
rapidly rotating, massive Population III stars. In contrast to their analysis,
we do not find any unusual enhancements of the neutron capture elements Sr, Y,
Ba and Eu and conclude that previous claims result mainly from not properly
accounting for blending lines. Instead we find NGC6522 to be an unremarkable
globular cluster with comparable abundance trends to other Galactic globular
clusters at the same metallicity ([Fe/H] = -1.15 +/- 0.16). The stars are also
chemically similar to halo and bulge field stars at the same metallicity,
spanning a small range in [Y/Ba] and with normal {\alpha}-element abundances.
We thus find no observational evidence for any chemical signatures of rapidly
rotating Population III stars in NGC 6522. |
1408.0291v1 | ## Pinning, Ordering, and Dynamics of Vortices in Conformal Crystal and Gradient Pinning Arrays
D. Ray, C. Reichhardt, and C. J. Olson Reichhardt
Theoretical Division, Los Alamos National Laboratory, Los Alamos, New Mexico 87545 USA
(Dated: August 5, 2014)
We numerically investigate magnetization, pinning, ordering, and dynamics of vortices interacting with pinning arrangements which have a density gradient. We focus on conformal crystal structures obtained by conformally transforming a spatially uniform periodic array, as well as non-conformal gradient structures and structures with quasiperiodic order. The conformal structures feature a density gradient and local ordering. Using magnetization simulations we find that conformal pinning arrays exhibit enhanced pinning compared to non-conformal gradient arrays as well as compared to random, periodic, and quasiperiodic arrays, for a broad range of fields. The effectiveness of conformal arrays arises from the continuum of length scales introduced into the arrays by the conformal transformation, allowing for a broad range of local commensuration effects. At higher vortex fillings above the range of conformal effectiveness, we show that a non-conformal rectangular gradient array exhibits strong pinning due to a novel commensuration effect and vortex ordering. Using transport simulations where vortices are driven along the gradient and at an angle to the gradient, we confirm the effectiveness of conformal pinning at increasing the critical current. For a rotated drive, the gradient arrays produce a strong vortex guidance effect in the direction perpendicular to the gradient.
PACS numbers: 74.25.Wx,74.25.Uv
## I. INTRODUCTION
One of the most important issues for many applications of type-II superconductors is understanding how to optimize the pinning of magnetic flux vortices, since the system becomes dissipative when vortices are in motion 1 . The applied current at which vortices depin is called the critical current J c , and there has been intense research in tailoring the pinning in a superconductor so as to maximize J c for either specific applied fields or for a broad range of fields. Beyond the direct application to superconductivity, vortex pinning and dynamics provide a very rich system for exploring collective ordering of particles on random or periodic substrates, as well as for studying nonequilibrium phenomena at depinning which arise in a variety of condensed matter systems, such as driven charge density waves , sliding Wigner crystals , atomic 2 3 friction 4 , adhesion , magnetic domain wall motion , and 5 6 driven colloidal systems 7 .
A widely used method for enhancing the pinning and critical current in superconductors is the creation of nanostructured artificial arrays of pinning sites 8-13 , featuring a well-defined number of sites arranged in a desired configuration. These sites pin vortices effectively, but only a limited number of sites can be added to a sample since their creation causes local damage to the superconductor that decreases the material's current-carrying capacity. Therefore, it is of great importance to determine the optimal geometrical arrangement of a fixed number of pinning sites that generates the strongest pinning over the widest range of fields. In random site arrangements, where there is a distribution of interpin spacings, the closely-spaced pins are inefficiently used due to strong short-range intervortex repulsion, while widely-spaced pins permit easy flow of vortices between them 14 . Periodic pinning arrays such as triangular or square arrays, which avoid both these shortcomings, have been studied in detail; the pinning is enhanced compared to random arrays, but only near certain commensurate field values, where the number of vortices is an integer 8-13,15-17 or fractional 18,19 multiple of the number of pinning sites. At commensurability, ordered or partially ordered vortex arrangements form which can have various types of symmetries 9,15-19 ; however, away from commensurability, the vortex structures are partially or totally disordered, allowing for the formation of weak spots and easy vortex flow channels which reduce the depinning threshold 9,15,18 . Rectangular pinning arrays that are periodic with two different length scales have been shown to produce anisotropic vortex structures and transport 20-24 .
There have been a number of efforts to enhance the pinning in periodic arrays under noncommensurate conditions. Systematically diluting a triangular pinning array gives enhanced pinning not only at matching fields, but also at non-integer matching fields corresponding to what the integer matching fields for the non-diluted array would have been 25-30 . Other studies showed that the addition of some disorder to a triangular pinning lattice enhances the pinning at incommensurate fillings, but at the cost of reducing the maximum pinning at the integer matching fields 31 . Asymmetric system geometries, such as funnel geometries 32-34 or periodic constriction geometries 35 which utilize vortex jamming, produce new types of commensurability effects. Vortex artificial ice states can be created using superlattices of square pinning arrangements 36-38 , and exhibit strong matching at certain non-integer fields. Composite pinning lattices based on intermeshed periodic arrays have also been
created 39-41 .
Another approach to pinning enhancement is the use of pinning structures that combine aspects of periodicity and disorder. Studies of quasiperiodic arrays where pinning sites were placed at the vertices of a Penrose tiling 42-48 revealed novel non-integer matching conditions, and showed that the overall pinning can be enhanced over periodic triangular pinning arrays below the first matching field 42,43,45,47 .
Despite these efforts, pinning arrays which rely on commensuration effects have continued to possess the fundamental drawback of only providing enhanced pinning near certain field values, rather than over a broad range of field. In an effort to overcome this shortcoming, a new type of pinning geometry, known as a conformal crystal pinning array, was recently proposed 49 ; an example of such an array is shown in Fig. 1(b). A conformal crystal is constructed by performing a conformal, or angle-preserving, transformation upon a uniform lattice structure such as a hexagonal array, producing a new structure which preserves the local ordering of the original array but now exhibits a density gradient in one direction 49,50 . Conformal crystals were first observed as the approximate ground state structure of a set of repulsively interacting magnetic spheres restricted to move in two dimensions and subjected to a gravitational potential; the distinctive arching appearance of the resulting structure inspired the whimsical nickname of 'gravity's rainbow' 50 . In vortex simulations, conformal pinning arrays produced enhanced pinning compared to an equivalent number of pinning sites arranged randomly with uniform density over a wide range of applied fields, with the enhancement persisting until the applied field was increased beyond the maximum local pinning density present in the array 49 . The conformal array also outperformed uniform periodic arrays except in the immediate vicinity of the matching field. A random array with a density gradient produced pinning only marginally better than a non-gradient random array and substantially worse than the conformal array, indicating that the preservation of local order was as vital to the conformal array's performance as was the introduction of a gradient. The enhancement was observed both in flux gradient-driven simulations 49 , where the vortices enter or exit the system from the edges, and in current-driven simulations 49,51 , where the vortex density is fixed and the vortices are driven across the sample with a uniform Lorentz force. Two subsequent experiments confirmed the enhanced pinning in conformal arrays. In the first 52 , conformal pinning arrays displayed enhanced critical currents over uniform periodic pinning arrays except in the vicinity of certain matching fields. In the second 53 , conformal pinning arrays showed decreased magnetoresistance in transport experiments compared to periodic or diluted periodic structures with the same number of pinning sites.
Pinning arrays with a density gradient not produced by a conformal transformation have also been investi- gated. A pinning array consisting of concentric rectangular 'rings' of pinning sites with a gradually increasing separation between the rings was fabricated and shown to exhibit enhanced pinning compared to a periodic pinning array 54 . Additionally, there have been numerical studies of pinning arrays arranged in circular hyperbolic tessellation structures so that the pinning sites have a gradient 55 ; however, these studies did not include comparisons with periodic or random pinning arrays to check whether the hyperbolic array produces enhanced pinning over the other geometries.
In this work, we use molecular dynamics simulations to study magnetization, depinning, ordering and dynamics of vortices in systems with various types of pinning arrays featuring a density gradient. We investigate the effect of different local orderings in conformal crystal arrays by considering transformed versions of hexagonal, square, quasiperiodic, and Archimedean 56,57 lattices. We also consider two non-conformal systems, a rectangular pinning array with a one-dimensional (1D) gradient, and a random pinning arrangement with a density gradient. This manuscript is organized as follows: in Section II, we describe how conformal arrays are created and enumerate the various pinning arrays which will be analyzed. In Section III, we explain our physics model and the types of simulations we perform. In Section IV, we show the results of magnetization simulations, demonstrating that conformal pinning arrays produce enhanced pinning over non-conformal gradient arrays over a wide range of applied fields, as well as over arrays of uniform density such as the quasiperiodic Penrose array or periodic arrays (aside from a narrow range around matching fields). In Section V, by examining the pattern of occupied pinning sites as the field is increased, we show that conformal arrays are effective due to a local commensuration effect which is an extension of the global matching exhibited by uniform pinning arrays: because the conformal transformation introduces a continuously changing lattice constant into the array, the vortex lattice can progressively match with different slices of the array as the number of vortices is increased. Consequently, we find that the particular type of local order present is relatively unimportant in determining the effectiveness of conformal pinning arrays, as long as some type of ordering is present that would give a matching effect in a uniform array. We also use local commensuration to explain the maximum effective field value for conformal arrays. At higher fields well above this value, we show in Section VI that a rectangular gradient pinning array gives enhanced pinning over the conformal arrays due to a novel commensuration effect which gives rise to ordered vortex chain structures. In Section VII, we examine current-driven vortex dynamics for vortices driven along the gradient direction as well as at varied angles with respect to the gradient. We find that in general, conformal pinning arrays reduce vortex motion compared to the other arrays, over the full range of drive angles. Compared to the random and rectangular gradient arrays which exhibit channels of easy vortex
x
FIG. 1: (a) A semiannular section of a uniform hexagonal lattice. (b) Effect of a conformal transformation upon the region in (a). The resulting structure (conformal hex or 'cHex') has local hexagonal ordering and a density gradient. See text for details of the transformation.
flow, the arching structure of the conformal arrays prevents the formation of such channels for any choice of drive angle. Additionally we find that many gradient arrays have a pronounced vortex guidance effect, where vortices move much more easily perpendicular to the gradient than along the gradient. Finally, in Section VIII, we present our conclusions.
## II. GRADIENT PINNING ARRAYS
## A. Conformal Arrays
Conformal transformations are a well-studied topic from complex analysis 58 . An analytic function w = f ( z ) maps two infinitesimal vectors originating from the same point in the z plane with angle θ between them to two new vectors in the w plane separated by the same angle. We use this property to transform an ordered structure into one which remains locally ordered but which is distorted on a global scale.
In this work, we allow the transformation
$$w = \frac { \pi } { 2 \alpha } + \frac { 1 } { i \alpha } \ln ( i \alpha z )$$
(where α is a constant which we set to π/ 36) to act on a semiannular section of the z -plane specified by Im z ≥ 0 and r in ≤ | z | ≤ r out for some choice of r in and r out . This transformation maps radial lines in the z plane to vertical lines in the w plane, and circular arcs centered at the
FIG. 2: Conformally transformed versions of various uniform (non-gradient) structures: (a) square lattice; (b) Archimedean 33434 lattice (a periodic tiling of the plane using squares and triangles); (c) 5-fold Penrose tiling using thin and thick rhombs.
origin to horizontal lines. Consequently, the semiannular region ABCDEF is mapped to the rectangle abcdef , as shown in Fig. 1. The circular arcs in the z plane are increasingly stretched by the transformation as their radial coordinate decreases: fixing αr out = 1, arc ABC has the same length as line abc , while arc DEF is stretched to the same length as lines def and abc . This increasing stretch causes an ordered structure with uniform density, such as the hexagonal lattice shown in Fig. 1(a), to be mapped to a structure with a density gradient, as shown in Fig. 1(b). The local ordering of the original structure is preserved due to the conformal nature of the mapping: each site in the transformed lattice retains its six nearest neighbors.
The conformally transformed hexagonal structure was studied in previous work 49,51 , where it was termed a 'conformal pinning array' or 'CPA.' Since in the present work we also consider other types of local order, we here refer to this array as a 'conformal hex' array, or 'cHex' for short. Figure 2 illustrates conformal arrays with other types of local order. In Fig. 2(a), a transformed square lattice exhibits local 4-fold order. We refer to this tiling as 'conformal square' or 'cSquare.' Fig. 2(b) shows the transformed version of a so-called 'Archimedean 33434' array. Archimedean tilings of the plane are periodic tilings created using two or more types of regular polygon. Several such tilings exist 57 ; here we consider a particular type, 33434, composed of squares
FIG. 3: Gradient pinning arrays not formed by a conformal transformation. (a) A rectangular gradient pinning array, consisting of rectangles stretched and compressed in one dimension. (b) Random (disordered) pinning with a density gradient matching the conformal arrays.
and triangles. A uniform pinning array based on such a tiling has been studied extensively 56 . We refer to the transformed version as 'conformal 33434' or 'c33434'. Finally, Fig. 2(c) shows the transform of a Penrose tiling with approximate 5-fold symmetry, constructed using thin and thick rhombs 59 . Pinning arrays based upon such Penrose tilings have been simulated previously 42 ; here, we consider transformed Penrose arrays ('conformal Penrose' or 'cPenrose'), as well as a uniform (untransformed) Penrose array. The uniform version of each array type will be referred to, when required, as 'uHex', 'uSquare', 'u33434', and 'uPenrose.'
All pinning arrays fit into a rectangular region of dimension 36 λ × 12 λ , where λ is the London penetration depth, and have a nominal average density of pinning sites equal to ¯ ρ p = 1 0 . /λ 2 (i.e. each array contains approximately 432 pins, with small deviations on the order of 1% due to boundary effects.) This density can be obtained through appropriate choices of the parameter r in and the lattice constant of the original uniform lattice. The local pinning density varies from a maximum of ρ loc p = 2 0 . /λ 2 to a minimum of ρ loc p = 0 3 . /λ 2 .
## B. Arrays produced by Other Methods
The conformal arrays described above all have the common features of local ordering and a density gradient. The local pinning density (or equivalently, the local lattice constant) varies fairly smoothly through the array due to the continuous nature of the conformal transformation. In order to determine the relative importance of each of these two features in maximizing the effectiveness of the pinning, we consider two other array types.
The first alternative array is shown in Fig. 3(a) and consists of a rectangular pinning array with a lattice constant that varies in the gradient direction so as to approximate the density profile of the conformal arrays as closely as possible. We call this array 'rectangular with gradient,' or 'gRect.' The gRect array differs from the conformal arrays in two respects: ( ) the lattice constant i does not vary smoothly through the array, and ( ii ) the original uniform rectangular array from which the gradient array was constructed does not exhibit a commensuration peak. The importance of this second property will be explained in Section V.
The second alternative array, shown in Fig. 3(b), has a density gradient matching that of the conformal arrays, but the pinning sites are arranged randomly subject only to this constraint, so that all local ordering is eliminated. This array, called 'random with gradient' or 'gRandom,' was studied in earlier work 49 and found to be only minimally more effective than uniform random pinning. Thus, we use the gRandom array to establish a baseline of minimum effectiveness for pinning arrays with a density gradient. Notice that this array type can be generated either by using the desired density profile as a bias when generating pinning locations, or by conformally transforming a uniform random array; the transformation does not add order where none was originally present.
## III. SIMULATION AND SYSTEM
We employ two types of simulation geometries in this work. The first is a flux gradient-driven geometry 49,55,56,60,61 where we apply a slowly varying magnetic field to a superconductor with pinning sites and calculate the resulting magnetization loops. Specifically, we consider a two-dimensional ( x -y ) slice of a bulk superconductor subjected to an applied magnetic field H = H ˆ z . We treat the resulting vortices in the material as perfectly stiff and use the London limit in which the vortices are assumed to be pointlike in the x -y plane. We measure all lengths in terms of the penetration depth λ . Our system is a square of side L = 36 λ with periodic boundary conditions in both the x and y -directions. As shown in Fig. 4(a), the pinning is added in a region extending from x = 6 λ to 30 λ which represents the superconducting sample. Vortices and antivortices are added in the unpinned external regions labelled 'A' to the left and right of the sample; the vortex density in these regions represents the applied external magnetic field. For pinning arrays with a gradient, we place two copies of the array in the sample region, oriented so that their dense sides face the external regions, while their sparse sides face each other. The geometry and system size described here were previously shown to be sufficiently large to accurately model both magnetization curves and vortex ordering for conformal crystal pinning arrays 49 , periodic pinning arrays 49,56,61 , and random pinning 49,55,60 .
At a given time during the simulation, the system con-
FIG. 4: Pinning geometries for both types of simulations, illustrated using the cHex array. In each case, the sample is periodic in the x and y directions. (a) Magnetization simulation geometry. The pinned region representing the superconducting sample extends from x = 6 λ to x = 30 λ . Two copies of a gradient pinning array are placed in the pinned region, with their sparse sides facing each other. Vortices and antivortices are added to the unpinned outer regions labeled 'A', which represent an applied external field. (b) Transport simulation geometry. The pinned region spans the entire simulation box. Three copies of a gradient pinning array are placed in the box, all facing the same direction. Vortices are placed randomly and annealed before the simulation begins; during the simulation, a steadily increasing drive is applied.
tains a total of N v vortices and antivortices. As the external field is changed over the course of the simulation, N v varies. The equation of motion for vortex i is given by arises from N p non-overlapping traps each represented by a truncated parabolic pinning potential, so that the pinning force is given by
$$\eta \frac { d \mathbf R _ { i } } { d t } = \mathbf F _ { i } ^ { v v } + \mathbf F _ { i } ^ { p }.$$
Here R i is the location of the vortex and η is the damping constant which we set to unity. The vortex-vortex interaction force is
$$\mathbf F _ { i } ^ { v v } = \sum _ { j \neq i } s _ { i } s _ { j } F _ { 0 } K _ { 1 } ( R _ { i j } / \lambda ) \hat { \mathbf R } _ { i j }$$
/negationslash where F 0 = φ / πµ λ 2 0 2 0 3 , φ 0 = h/ e 2 is the elementary flux quantum, µ 0 is the permittivity of free space, K 1 is the modified Bessel function, R ij = | R i -R j | , and ˆ R ij = ( R i -R j ) /R ij . We measure all forces in terms of F 0 . For computational efficiency, we truncate the interaction force beyond R ij = 6 , which is a good apλ proximation since K x 1 ( ) falls off exponentially for large x . Vortices repel each other, as do antivortices, while a vortex attracts an antivortex; to represent this interaction, we use the sign prefactor s i which is +1 for a vortex and -1 for an antivortex. If a vortex and an antivortex approach each other within a distance smaller than 0 3 . λ , they are both removed from the system to simulate an annihilation event. Most of our simulations involve only vortices; we use antivortices only to generate full magnetization loops, described below. The pinning
$$\mathbf F _ { i } ^ { p } = \sum _ { k = 1 } ^ { N _ { p } } ( F _ { p } R _ { i k } ^ { p } / R _ { p } ) \Theta ( ( R _ { p } - R _ { i k } ^ { p } ) / \lambda ) \hat { \mathbf R } _ { i k } ^ { p },$$
where R p k is the location of pinning site k , R p ik = | R i -R p k | , ˆ R p ik = ( R i -R p k ) /R p ik , Θ is the Heaviside step function, the pinning radius R p is set to 0 12 . λ , and F p specifies the maximum strength of the pinning force.
To construct a full magnetization loop, we begin with an empty sample and gradually add vortices to the unpinned regions marked 'A' in Fig. 4(a). As the vortex density increases in the unpinned region, vortices are pushed into the sample due to the intervortex repulsion. We continue to add vortices until reaching the desired maximum value of the external magnetic field H ; then we begin to add antivortices. As these annihilate with vortices, H decreases back to zero and then becomes negative as the unpinned region fills with antivortices. Finally, we add vortices again to bring H back up to zero. As the simulation progresses, we record the magnetization
$$M = - \frac { 1 } { 4 \pi V } \int d V \left ( H - B \right )$$
where V represents the sample area, H is the average vortex density over the unpinned region, and B is the
(position-dependent) vortex density in the sample region. According to the Bean critical state model 62 , the width of the magnetization loop is proportional to the critical current J c . This width can be determined using only the first quarter of the magnetization loop, and so in many of our simulations, we generate only a quarter loop instead of a full loop by using vortices to raise the external field up to a maximum value and then halting the simulation.
The second simulation geometry we consider is currentdriven rather than flux gradient-driven, and is similar to that used in previous studies of current-driven vortex dynamics in pinned superconductors 48,63,64 . We add an additional force term F d = F F n d 0 ˆ to the vortex equation of motion in Eqn. 1 to represent the Lorentz force exerted by an applied external current J ; here ˆ is a conn stant unit vector in the x -y plane. This force affects all vortices uniformly. For the purposes of an initial anneal, we also add a term F T i representing thermal Langevin kicks to Eqn. 1, where F T i has the properties 〈 F T i 〉 = 0 and 〈 F T i ( ) t F T j ( t ′ ) 〉 = 2 ηk B Tδ ij δ t ( -t ′ ). We modify the system geometry to measure transport properties by removing the external unpinned regions and employing a pinning configuration composed of three copies of a given pinning array facing in the same direction and filling the entire simulation box, as shown in Fig. 4(b). To initialize the system we place N v vortices in randomly chosen nonoverlapping locations in the simulation box and perform a simulated anneal from F T = 3 0 to . F T = 0 while holding F d = 0 before beginning our measurement. Throughout a transport simulation, N v is held constant in order to model a constant magnetic field level. We increase the magnitude of the current-induced driving force F d in small increments of size δF d = 5 × 10 -4 and spend 500 simulation time steps at each value of F d in order to avoid any transient behavior. We measure the average vortex velocity in the x and y directions as a function of drive: 〈 V x 〉 = (1 /N v ) ∑ N v i =1 v i · ˆ and x 〈 V y 〉 = (1 /N v ) ∑ N v i =1 v i · ˆ. y The vortex velocity is proportional to the voltage response, so the velocity-force curve we generate is comparable to an experimentally measurable current-voltage curve.
## IV. MAGNETIZATION
In Fig. 5 we plot the full magnetization loops M vs H/H φ , where H φ is the matching field at which there is one vortex per pinning site, for the conformal pinning array cHex from Fig. 1, the non-conformal gradient arrays gRect and gRandom from Fig. 3, and a uniform random pinning array. For weak pinning sites with F p = 0 2, . Fig. 5(a) shows that the conformal array provides enhanced pinning compared to the other arrays over the entire range of field, as indicated by the fact that the magnetization loop is widest for the cHex array. The rectangular gradient array gRect and cHex have similar values of M up to H/H φ = 1 0, but for higher . H/H φ the conformal array is clearly the most effective at pin-
0.02
0.01
M
0
-0.01
-0.02
F = 0.20
p
(a)
-2
-1
0
1
cHex gRect
gRandom uRandom
2
H/H
φ
0.06
F = 0.90
p
(c)
-2
-1
0
1
cHex gRect
gRandom uRandom
2
0.03
M
0
-0.03
-0.06
H/H
φ
FIG. 5: Full magnetization M vs. external magnetic field H/H φ loops for the conformal cHex array (outer loop), the non-conformal gRect and gRandom arrays (center loops), and uniform random pinning uRandom (inner loop), with F p = (a) 0 2, . (b) 0 55, . and (c) 0 9. . The conformal array shows enhanced pinning over the other arrays in each case, and this is most noticeable when | H/H φ | > . 1 0.
FIG. 6: Magnetization M vs H/H φ for cHex array (top right) compared to various arrays with two length scales in samples with F p = 0 55. . (a) Uniform 33434 (bottom right) and uniform Penrose (center right); (b) conformal 33434 (center right) and conformal Penrose (bottom right).
ning the vortices. The difference between the conformal and random arrays becomes more pronounced when the pinning strength is increased, as indicated for F p = 0 55 . in Fig. 5(b). The ordered gRect array shows slightly stronger pinning than cHex for H/H φ < 1 0, . but falls off dramatically above that value. These effects are even more pronounced for F p = 0 9, shown in Fig. 5(c). . The difference between the uniform random and random gradient arrays increases with increasing F p , but the mere addition of a gradient is not sufficient to give magnetization values that are anywhere close to as large as those obtained with the cHex array, particularly for higher values of H/H φ . The effectiveness of the conformal array persists up to a field H/H φ = 2 0 before . beginning to decline, and the critical current for cHex remains higher than that of gRect up to H/H φ = 2 5. . Above H/H φ = 1 75 a new peak in . M begins to emerge for the gRect sample for strong pinning; this feature is described in more detail in Section VI. We note that experiments in thin-film superconductors with through-hole pinning sites should exhibit strong pinning behavior equivalent to large values of F p 52 . These results show the superiority of conformal gradient pinning over other types of gradient pinning, and also over random pinning of any sort. In Fig. 4(d) of Ref. 49 we have already showed enhanced pinning for a conformal array compared to uniform periodic arrays over a broad range of fields, apart from certain narrow field ranges corresponding to matching conditions.
We next turn to Penrose pinning arrays, since these have been claimed 42,47 to also provide broadly effective pinning. The previous studies suffered from several drawbacks, such as a small sample and array size which diminishes any actual effect of the array's structure upon pinning effectiveness, the use of fractional pin occupancy
FIG. 7: Magnetization M vs H/H φ for cHex array (top right) compared to various arrays with two length scales in samples with F p = 0 90: . uniform 33434 (lower left), uniform Penrose (lower right), conformal 33434 (upper middle right), and conformal Penrose (center right).
as a proxy for critical current, and most importantly a lack of comparisons to other pinning structures besides triangular. Here, we compare a Penrose array generated from a tiling with approximate 5-fold symmetry, composed of thin and thick rhombs, to a conformal array. The magnetization curves for the uniform Penrose array with F p = 0 55 in Fig. 6(a) show a matching peak near . H/H φ = 1 0, similar to that found for a uniform periodic . array. Away from the matching field, the magnetization drops off rapidly, again similar to the behavior of a periodic array. The magnetization of the cHex array is lower than that of the uPenrose array around the matching field, but is higher everywhere else. In particular, the cHex array gives a large enhancement of the magnetization up to and slightly beyond H/H φ = 2 0. .
These results suggest that the quasiperiodic nature of a Penrose array is relatively unimportant as far as pinning effectiveness is concerned, and in fact its behavior is comparable to that of strictly periodic pinning structures. Since the Penrose tiling employs two different types of tiles giving rise to multiple length scales in the pinning structure, we use as a comparison a periodic tiling known as Archimedean 33434, which also uses two types of tiles, and which has been studied in the context of vortex pinning 56,57 . In Fig. 6(a) we show that the magnetization of the uniform 33434 array is broadly similar to that of the uniform Penrose array, with a sharp matching peak in M at H/H φ = 1 followed by a rapid falloff.
Next, in order to examine the role of local order in conformal arrays, we consider the conformally transformed versions of the Penrose and Archimedean arrays which were shown in Fig. 2. We plot magnetization curves for these arrays in Fig. 6(b). Two behaviors are apparent. First, comparing the uniform arrays in Fig. 6(a) to their conformally transformed versions in Fig. 6(b), we see that the conformal transformation improves an array's performance over a broad range of field at the cost of eliminating the original matching peak at H/H φ . Second, we find from Fig. 6(b) that even though vortices naturally arrange themselves in a hexagonal lattice in the absence of pinning, local hexagonal ordering is not necessary to obtain an effective conformal array: the c33434 array
FIG. 8: Occupied and empty pinning sites in a cHex array with F p = 0 55, at . H/H φ = (a) 1.09, (b) 1.55, and (c) 2.01. Filled circles (blue): occupied pins; open circles (red): unoccupied pins. Pin sizes have been exaggerated for clarity. The occupied pinning sites form vertical bands that move toward the sample edges with increasing field.
works as well, and the cPenrose array almost as well, as the cHex array at enhancing pinning.
To verify that these results are robust, we repeat our measurements in samples with an increased pinning strength of F p = 0 90. As shown in Fig. 7, the magnetiza-. tion curves retain all of the behaviors described above. In particular, we confirm that conformal arrays work better than uniform arrays in each case, and that their performance does not depend sensitively on the type of local order present in the original untransformed array. To further test this, a future study could compare cHex and cSquare arrays in magnetization loops which extend up to the third matching field, since a uniform square array has a commensuration peak at the second matching field but not the third, while the uniform triangular array has the opposite response.
## V. LOCAL COMMENSURATION EFFECT IN CONFORMAL PINNING ARRAYS
In order to better understand how the structure of conformal arrays contributes to their enhanced pinning effectiveness over a broad range of field, and also what determines the limits of this range, we image the distribution of occupied pins as the field is increased. We take as our model system the cHex array with F p = 0 55. . In Fig. 8 we show three snapshots of the occupied pinning sites at different field values, where we find vertical bands in which all the pinning sites are filled. The location of these bands moves outwards towards the left and right edges of the sample as the field increases. This is seen more clearly in Fig. 9, where we plot the fraction of occupied pins P occ ( x ) averaged over the vertical direction y as a function of horizontal position x for several different values of H/H φ . Here x = 18 denotes the center of the pinned region, while the sample edges are at x = 6 and x = 30. For H/H φ = 1 09 in Fig. 9(b), we highlight . two occupancy peaks in P occ ( x ) near x = 11 and x = 25.
x
FIG. 9: The spatial distribution of the fraction of occupied pins P occ ( x ), averaged over the y -direction, versus x position for the cHex array with F p = 0 55 at . H/H φ = (a) 0.86, (b) 1.09, (c) 1.33, (d) 1.55, (e) 1.78, (f) 2.01, and (g) 2.22. The arrows in panels (b-g) highlight local regions of very high pin occupancy, indicating that the vortex lattice is locally commensurate with the pinning in these regions. As H/H φ increases, the matching regions move toward the sample edges where the pinning density is greater. At the extreme edges of the sample the maximum local pinning density is ρ loc p = 2 0 . /λ 2 .
These peaks correspond to the bands mentioned above, and signify areas where the vortex lattice locally matches with the pinning. This matching is distinct from the usual matching effect in a uniform pinning array which is global and occurs throughout the entire array. As H/H φ increases, Figs. 9(c-g) show that the occupancy peaks move toward the sample edges. This occurs because at matching, the vortex density must match the pinning density; thus, as the vortex density increases, the matching region must shift to the denser pinning regions that are closer to the edges of the sample. We also observe dips in P occ ( x ) adjacent to the peaks, on the sides closer to the center of the sample. The dips arise due to the mismatch between vortices and pins once the commensuration peak has passed, with interstitial vortices forcing vortices out of the pinning sites. The migration of the local matching peaks continues as the field increases until the peaks reach the maximally dense pinning regions at the sample edges. For the conformal arrays we use in this work, the maximum local pinning density is ρ loc p = 2 0, so .
ρ
FIG. 10: Vortex density ρ v vs x position for the cHex array with F p = 0 55 as the external field is . gradually increased. Dotted lines indicate the sample edges; the sample extends from x = 6 λ to x = 30 λ as shown in Fig. 4(a). The value of H/H φ for each curve is indicated by the average y value in the external regions x < 6 λ and x > 30 λ . The lowest curve has H/H φ = 0 52, the highest curve has . H/H φ = 2 50, . and there is an interval of 100 vortices, or δH/H φ ≈ 0 077, . between illustrated curves. Profiles in red correspond to the magnetic field values shown in Fig. 9; the letters indicate the corresponding panel in Fig. 9. Blue arrows point to the same locations as the arrows in Fig. 9.
we expect that when the vortex density ρ v exceeds this value by a non-negligible amount, local matching will no longer be possible. In Fig. 9(g) where the external field has increased to H/H φ = 2 22, the matching regions have . reached the edges of the sample and can not move any further. Figure 6 shows that the effectiveness of the pinning of the cHex array falls off above this field value.
The vertical bands of occupied pins act as walls hindering vortices from flowing between the outer region and the inner region. These effective walls can maintain a large difference in vortex densities between these regions, giving a large value of M . This is illustrated in Fig. 10, where we plot the vortex density ρ v averaged over y as a function of horizontal position x as the field is increased. The field profiles have a double-slope, nonBean-like shape as noted in previous work 49 . The highslope regions where the vortex density changes rapidly are precisely the regions in which we find bands of occupied pins. We show this by drawing arrows pointing to the labeled highlighted profiles in Fig. 10 at the same locations where they appear in the corresponding pin oc-
9
cupancy plots in Fig. 9; these arrows clearly track the high-slope profile regions. The high-slope regions move towards the edges of the sample as the field increases, and first touch the edges at H/H φ = 2 0. . Above this field, the high-slope regions begin to disappear since local matching can no longer occur, the vortex density equilibrates between the sample and the outside, and there is a corresponding drop in M as seen for cHex in Fig. 6.
In Fig. 11 we summarize the key characteristics of conformal arrays which produce the enhanced pinning. Although the data presented in the above discussion of local commensuration was from the cHex array, the local commensuration mechanism does not require hexagonal ordering; in fact, any local ordering that exhibits matching phenomena should be similarly effective. To show this, we first examine which array types are capable of exhibiting matching phenomena. Figure 11(a) shows the spatially averaged fraction of occupied pinning sites P vs H/H φ for uniform arrays of the hexagonal, square, Penrose, 33434 (Archimedean), rectangular, and random type. The hexagonal, square, and 33434 arrays have a pronounced peak in P occ just above H/H φ = 1 0, indi-. cating that they exhibit robust matching. The Penrose array shows weaker matching, while the random and the rectangular arrays do not exhibit matching at all. In Fig. 11(b) we plot M versus H/H φ for the corresponding conformal or gradient arrays, cHex, cSquare, cPenrose, c33434, gRect, and gRandom. As expected from the arguments above, we find that the conformal versions of the hexagonal, square, and 33434 arrays all show highly enhanced pinning up to H/H φ = 2 0, resulting from ro-. bust local matching. The conformal Penrose array shows a weaker pinning enhancement since the commensuration effects in the original uniform Penrose array are weaker. The rectangular and random gradient arrays give much lower magnetization values than the other arrays because they are not conformal and do not exhibit a local matching mechanism. We conclude that for any uniform array with a well-defined matching condition, the corresponding conformal array will exhibit enhanced pinning over a broad range of fields. For the conformal arrays we consider, this field range is 0 6 . < H/H φ < 2 0, . and it is determined by the local pinning density ρ loc p ( x ) which we plot in Fig. 11(d) for the conformal hex lattice shown in Fig. 11(c). At the minimum external field H/H φ = 0 6, . the local matching occurs at the minimum density region in the center of the sample, where ρ loc p = 0 3. . The local matching moves toward the sample edges as the field increases, until for H/H φ = 2 0 the matching reaches the . sample edges where the pinning density is also ρ loc p = 2 0 . . Future experimental studies can determine the maximum steepness of the pinning gradient which will still permit local matching to occur and be effective at enhancing the pinning.
p
φ
FIG. 11: Local commensuration effects in conformal pinning arrays. (a) Spatially averaged fraction of occupied pins P vs H/H φ in uniform arrays of (central peak, from top to bottom) hexagonal, square, Penrose, 33434, rectangular, and random type. Peaks in P just above H/H φ = 1 0 show that the hexagonal, square, Penrose, and 33434 arrays exhibit a robust matching . effect. (b) Spatial configuration of the pinning sites in a cHex array. (c) Corresponding local pinning density ρ loc p ( x ) for the same (cHex) array. The local pinning densities for the other conformal arrays are nearly identical to what is shown here. The region between the dotted lines indicates the range of ρ loc p values that exist somewhere inside the array. (d) Magnetization curves M vs H/H φ for gradient pinning arrays (from upper left to lower left): cHex, cSquare, cPenrose, c33434, gRect, and gRandom. For the conformal arrays, the magnetization is enhanced whenever the vortex lattice can locally match with a vertical strip of the pinning array. The range of possible local densities allowing matching [highlighted region in panel (c)] thus determines the range of applied fields [highlighted region in panel (d)] at which the conformal arrays provide enhanced values of M . The non-conformal gradient arrays do not enhance pinning by this mechanism.
## VI. RECTANGULAR GRADIENT ARRAY AND COMMENSURATION EFFECTS
dramatically and rises well above M for the cHex array, forming a broad peak centered at H/H φ = 3 5. .
We next consider higher-field ordering and commensurability effects for the rectangular gradient or gRect pinning array, which were briefly noted in Fig. 5(c). In Fig. 12 we plot M versus H/H φ for both the cHex and gRect arrays with F p = 0 9 for fields up to . H/H φ = 5 2. . Here the effectiveness of the pinning for the conformal array clearly diminishes for fields above H/H φ ≈ 2 0. . The cHex array gives enhanced pinning compared to the gRect array for 1 0 . ≤ H/H φ ≤ 2 5; however, above this . field range, M for the gRect array begins to increase
In Fig. 13(a), we plot M versus H/H φ for rectangular gradient arrays with F p = 0 55, . 0.9, and 1 25, . showing that the broad peak in M is a robust feature. In Fig. 13(b) we plot the corresponding fraction of sixfold coordinated vortices P 6 versus H/H φ , where P 6 = (1 /N vs ) ∑ N vs i =1 δ z ( i -6). Here N vs is the number of vortices within the pinned region, the coordination number z i of vortex i is obtained from a Voronoi tesselation, and for a triangular lattice P 6 = 1 0. . Figure 13(c) shows the corresponding fraction of fivefold coordinated vortices P 5 versus H/H φ . Due to the gradient in the sam-
FIG. 12: M vs H/H φ for cHex (lower right) and gRect (upper right) samples with F p = 0 9. . Here the gRect array shows a commensuration peak near H/H φ = 3 5, where the . magnetization is significantly enhanced compared to the cHex array.
FIG. 13: Results for the rectangular gradient array illustrated in Fig. 3(a). (a) M vs H/H φ for F p = 1 25 (upper . curve), 0 9 (middle curve) and 0 55 (lowest curve). . . (b) The corresponding fraction of sixfold-coordinated vortices P 6 vs H/H φ for F p = 1 25 (lower right), 0 9 (middle right), and 0 55 . . . (upper right). There is a small peak in P 6 near H/H φ = 4 0. . (c) The corresponding fraction of fivefold-coordinated vortices P 5 vs H/H φ for F p = 1 25 (upper right), 0 9 (middle right), . . and 0 55 (lower right). .
ple, we generally find P 7 /lessmuch P 5 since vortices are moving into less dense areas, so we do not show P 7 . In Fig. 13(a), M increases with increasing F p when H/H φ ≤ 1 0, drops . to a low value independent of F p for 1 0 . < H/H φ < 2 0, . and increases with increasing F p again for 2 0 . ≤ H/H φ . For H/H φ /lessorsimilar 1 0, vortices that enter the system tend to . get captured by pinning sites, and the capture process is more efficient when F p is higher. As in all computed magnetization curves irrespective of pinning geometry, the initial change in slope of M from positive to negative occurs at the field at which the vortices first reach the center of the sample. Above this field, the vortices begin to enter the interstitial regions of the rectangular gradient array, where they encounter one-dimensional easyflow channels oriented along the y direction, perpendicular to the gradient. The interstitial vortices, although unpinned, experience an effective pinning force due to caging by the surrounding pinned vortices. This effective force is not affected by an increase in F p , which only traps the pinned vortices more firmly. Thus, in the field range 1 0 . < H/H φ < 2 0 . where interstitial pinning is dominant, we find that the value of M is nearly independent of F p . There is a dip in P 6 in Fig. 13(b) just above H/H φ = 1 0, . with a corresponding peak in P 5 in Fig. 13(c), at the transition to the interstitial vortex regime.
For H/H φ > 3 25, Fig. 13 shows that . P 6 begins to increase as M rises to a broad peak centered near H/H φ = 3 75. . There is a small peak in P 6 near H/H φ = 4 1 fol-. lowed by a small decrease in the region where M is also decreasing. The broad peak in M at these higher fields is due to a commensuration effect associated with the formation of ordered vortex chain structures, as illustrated in Fig. 14 where we show the vortex configurations in the gRect array for different field values. Figure 14(a) shows the configuration at H/H φ = 1 74, where there is a min-. imum in M in Fig. 13(a). Here the vortices are largely disordered, and substantial depinning has occurred as indicated by the large fraction of empty pinning sites. At H/H φ = 2 61 in Fig. 14(b), . M begins to increase and the pinning sites fill, but there is still no clear ordering present. When M reaches its maximum at H/H φ = 3 57, . Fig. 14(c) shows that the vortex configuration within the pinning array is now ordered. To highlight this ordering, we mark the number of vortex chains that fit between pinning columns in Fig. 14(c). Between the two leftmost columns, there is one chain of vortices; between the next two columns, there are two chains with the vortices forming a zig-zag structure; between the next two columns, there are three chains forming a local triangular lattice, and similarly in the next two gaps there are four chains. Similar structures, but with different numbers of vortex chains in between pinning columns, appear in Fig. 14(d) at H/H φ = 4 08 where there is a peak in . P 6 in Fig. 13(b), as well as at H/H φ = 4 88 in Fig. 14(e), where there is a . second peak in P 6 in Fig. 13(b).
## VII. CURRENT-DRIVEN RESULTS
In the previous sections, we studied vortex ordering and critical states using quasistatic flux gradient-driven simulations. To examine the vortex dynamics, we use current-driven simulations in which a steadily increasing drive is uniformly applied to the vortices in the sample. We consider driving directions that are aligned with the gradient ( x ) direction as well as drives oriented at some angle θ to this direction. The equations of motion for the
FIG. 14: Ordered and disordered vortex states in rectangular gradient array with F p = 0 90 . at various values of H/H φ . Filled circles: vortices; open circles: pinning sites. The dashed line on the left indicates the edge of the sample. (a) At H/H φ = 1 73, . M reaches a minimum value and the vortices are disordered. (b) At H/H φ = 2 61, . M increases and the number of occupied pins increases. (c) At H/H φ = 3 57, there is an ordered state corresponding to the . peak in M in Fig. 13(a). The numbers indicate the number of ordered vortex columns between columns of pinning sites. (d) At H/H φ = 4 08, we observe an ordered state at the . P 6 peak in Fig. 13(b). (e) At H/H φ = 4 88 we illustrate the ordered . state at the second P 6 peak in Fig. 13(b).
current-driven simulations were described in Section III; the sample geometry is illustrated in Fig. 15.
## A. Drive Aligned with Gradient
We first consider driving vortices along the gradient in the positive x -direction. We define the driving angle θ as the angle the driving direction makes with the positive x axis, so that θ = 0 corresponds to driving along the gradient direction. In Fig. 16(a) we plot 〈 V x 〉 versus F d for systems with F p = 0 55, . θ = 0, and H/H φ = 1 4; we . consider the cHex array shown in Fig. 4(b), the cSquare, cPenrose, gRect, and gRandom arrays shown in Fig. 15
y
y
FIG. 15: Pinning geometries for current-driven simulations with various pinning arrays: (a) cSquare; (b) cPenrose; (c) gRect; (d) gRandom.
FIG. 16: The average vortex velocity in the x -direction 〈 V x 〉 vs F d for current-driven simulations of cHex, cSquare, cPenrose, gRect, gRandom, and uRandom transport arrays with F p = 0 55 and a drive angle of . θ = 0 at different vortex densities. (a) H/H φ = 1 4 for cHex, cSquare, cPenrose, uRan-. dom, gRandom, and gRect, from lower right to upper right. (b) H/H φ = 0 8 . for cSquare, cHex, cPenrose, uRandom, gRandom, and gRect, from lower right to upper right. (c) H/H φ = 2 2 for gRect, cHex, cSquare, cPenrose, uRandom, . and gRandom, from lower right to upper right. In general, the conformal arrays have a higher depinning threshold and lower vortex velocity than the non-conformal arrays. The gRect array exhibits a large depinning threshold at H/H φ = 2 2 due . to a commensuration effect.
and a uniform random array lacking any gradient. The depinning force F c , defined as the magnitude of the drive above which 〈 V x 〉 ≥ 10 -4 , is significantly larger in the conformal arrays than in the non-conformal arrays. Additionally, in the moving phase the vortices travel more slowly through conformal arrays than through the nonconformal arrays over a substantial range of F d . Among the conformal arrays, the cHex array gives the highest F c , with F c for the cSquare array almost as high and slightly smaller for the cPenrose array. At higher F d the curves start to come together as the effectiveness of the pinning at the higher drives is washed out and 〈 V x 〉 ≈ F d . At H/H φ = 0 8, as shown in Fig. 16(b), . 〈 V x 〉 for the cSquare array drops below that of the cHex array at higher drives, indicating enhanced pinning effectiveness in the cSquare array, while the curves for the cHex and cPenrose arrays are very similar. The gRect array has the weakest pinning, followed by the random arrays with and without a gradient.
In Fig. 16(c) we illustrate the behavior at a higher field of H/H φ = 2 2. . At this field, the gRect array exhibits a highly enhanced depinning force, followed by an abrupt transition to steady vortex motion. Inspection of the vortex configurations reveals that the vortices form an ordered vortex chain state of the type discussed in Section VI, which locks them into place and prevents vortex flow. Because our gRect transport and magnetization arrays have somewhat different configurations, they form these ordered states at different field levels; however, the qualitative behavior of gRect remains the same during transport, with the array exhibiting enhanced performance only when it is able to form vortex chain states. The difference between the conformal and non-conformal random arrays is somewhat reduced at H/H φ = 2 2, since . it is above the cutoff field value of H/H φ = 2 0 where . the conformal arrays start to exhibit decreased pinning effectiveness. The conformal arrays still perform better by giving higher values of F c and lower values of 〈 V x 〉 in the moving phase than the random arrays. The cSquare and cHex arrays have almost the same behavior, and are again superior to cPenrose.
These results are completely consistent with the results from our magnetization simulations. The transport simulations confirm that the pinning effectiveness is enhanced in conformal arrays over non-conformal ones, and the fact that the gRandom and uRandom arrays give nearly identical results indicates that it is not merely the gradient in the pinning that matters but also the local ordering. The precise type of local ordering is of lesser importance, although (as with magnetization) the quasiperiodic Penrose ordering is somewhat less effective at enhancing the pinning than periodic orderings.
## B. Driving in Different Directions
Next we consider the effect of changing θ , the orientation of the driving direction, for various gradient pinning
13
FIG. 17: 〈 V x 〉 and 〈 V y 〉 vs F d for current-driven systems with F p = 0 55 at . H/H φ = 1 4, driven at . θ = 30 ◦ with respect to the positive x -axis. (a) cHex array; (b) gRandom array; (c) gRect array. The labeled current levels a-l correspond to the vortex trajectories illustrated in Fig. 18.
arrays. In Fig. 17 we plot the average velocity in the x -direction 〈 V x 〉 and y -direction 〈 V y 〉 as a function of F d for the cHex, gRandom, and gRect arrays, at a drive angle of θ = 30 . For the cHex array in Fig. 17(a), ◦ 〈 V y 〉 is nonzero for 0 05 . < F d < 0 10 . while 〈 V x 〉 remains zero, meaning that vortex flow occurs strictly in the y -direction, perpendicular to the gradient, even though the largest component of the external drive is along the x direction. Once the vortices start to move along the x direction as well, we find a crossing of the 〈 V x 〉 and 〈 V y 〉 curves near F d = 0 15. . In Fig. 18(a,b,c,d) we illustrate the vortex flow in the cHex array at the drive values marked with the letters a-d in Fig. 17(a). At F d = 0 06 and 0 09 in . . Figs. 18(a) and (b), the vortex flow occurs in winding rivers aligned with the y -direction, which pass through the portions of the sample in which the pinning density is lowest. The cHex array exhibits a pronounced guidance or channeling effect since most of the pins are occupied in the higher pinning density regions, creating a barrier to vortex motion, while the local depinning threshold is reduced in areas of lower pinning density. At F d = 0 12, . Fig. 18(c) shows that vortices begin to cross between the easy flow channels, passing through the denser pinning region. This corresponds with the emergence of a nonzero value of 〈 V x 〉 in Fig. 17(a) at point c. At higher drives the amount of motion in the x -direction between easy-flow channels increases, as illustrated in Fig. 18(d) at F d = 0 14. .
/negationslash
The depinning transition falls at a much lower value of F d for the random gradient array, as shown in Fig. 17(b), and unlike in the cHex array there is no vortex guidance effect in which flow only occurs in the y -direction. Instead, 〈 V x 〉 and 〈 V y 〉 both increase smoothly with increasing F d , and there is no crossing of the velocity curves. In Fig. 16 we showed that a random gradient array driven in the gradient direction with θ = 0 ◦ has essentially identical transport properties to a uniform random array, indicating that the addition of a gradient to a random array does not substantially affect vortex flow. Similarly, we find that for θ = 0 ◦ the flow through random arrays is also not affected by the addition of a gradient. In
FIG. 18: Lines: vortex trajectories, dark dots: vortex positions, and light dots: pin positions for selected values of F d in the current-driven systems described in Fig. 17 with F p = 0 55, . H/H φ = 1 4, and drive orientation . θ = 30 . ◦ (a,b,c,d): cHex; (e,f,g,h): gRandom; (i,j,k,l): gRect. F d = (a,e,i) 0.06, (b,f,j) 0.09, (c,g,k) 0.12, and (d,h,l) 0.14.
Fig. 18(e,f,g,h) we show the vortex trajectories for the drives labeled e-h in Fig. 17. At F d = 0 06 and 0 09 in . . Fig. 18(e) and (f), the vortices move in both the x and y directions and form winding channels oriented along the driving direction. Due to the less effective pinning in the gRandom array, the number of moving vortices is much greater than in the cHex array at the same drive, as can be seen by comparing the high trajectory densities in Fig. 18(e,f) to the much sparser trajectories in Fig. 18(a,b). At F d = 0 14 in Fig. 18(h), most of the vor-. tices are moving. At drives higher than those illustrated here, the vortices dynamically reorder in a manner similar to the dynamic reordering transitions observed for vortices driven over uniform random pinning arrays 64,65 .
Figure 17(c) shows 〈 V x 〉 and 〈 V y 〉 versus F d curves for the rectangular gradient array. Here there is a pronounced guidance effect for 0 02 . < F d < 0 11; however, . unlike in the cHex array, 〈 V x 〉 is not strictly zero, indicating that some vortex motion is occurring in the x -direction between the 1D columns of pins since the gRect array is less effective in restricting vortex motion along its gradient. The vortex flow images in Fig. 18(i,j,k,l) correspond to the labeled drives in Fig.17(c). For F d = 0 06 . and 0 09 in Fig. 18(i) and (j), the vortices follow nearly . 1D paths along the y direction with very little winding, while there are a smaller number of flow channels oriented approximately parallel to the x direction that connect some of the 1D paths. The 1D vertical flow channels are concentrated in the least dense portions of the pinning array, and are similar in nature to the 1D flow of interstitial vortices observed in uniform square or rectangular periodic pinning arrays 66 . At F d = 0 12 in Fig. 18(k), there is . a significant increase in the amount of vortex flow along the x -direction, while for F d = 0 14 in Fig. 18(l) the flow . is continuous though the system and oriented with the driving direction. As with the gRandom array, the high trajectory densities indicate pinning which is less effective than in the cHex array. For F d > 0 20 the vortices . dynamically order, as indicated by the onset of a linear regime of 〈 V x 〉 versus F d in Fig. 17(c).
We summarize the transport properties for different drive angles in Fig. 19, where we plot 〈 V x 〉 and 〈 V y 〉 versus θ for different values of F d in the cHex and gRandom arrays. In the absence of pinning, we would obtain 〈 V x 〉 = F d cos( θ ) and 〈 V y 〉 = F d sin( θ ). For the cHex array with F d = 0 09, . shown in Fig. 19(a), 〈 V x 〉 ≈ 0 for all θ , while 〈 V y 〉 monotonically increases with increasing θ . This drive is too weak to depin the vortices along the x -direction, but the vortices can flow in the y -direction through the low density regions as was shown in Fig. 18(b). Fig. 19(b) shows the same quantities for the gRandom array at the same drive F d = 0 09. . Here 〈 V x 〉 is nonzero over the whole range of θ since the depinning threshold in the x -direction is much lower for the gRandom array. There is an asymmetry in the x and y direction responses, as indicated by the fact that 〈 V x 〉 = 0 015 at . θ = 0 ◦ but 〈 V y 〉 = 0 021 at . θ = 90 . ◦ This indicates that the pinning in the moving phase is
FIG. 19: 〈 V x 〉 (circles) and 〈 V y 〉 (squares) vs drive angle θ for the cHex (a,c,e) and gRandom (b,d,f) arrays, at F d = 0 09 . (a,b), F d = 0 14 (c,d), and . F d = 0 20 (e,f). . We fix F p = 0 55 . and H/H φ = 1 4. .
slightly more effective when the vortices are moving along the gradient direction. However, the vortex velocities are much larger in the gRandom array than in the cHex array over the whole range of θ , showing the robust enhanced effectiveness of the conformal pinning.
When the drive in the cHex array is increased to F d = 0 14, Fig. 19(c) shows that the . x -direction depinning threshold is exceeded for θ < 70 , but that the vor-◦ tex motion along the x direction is still greatly suppressed even when the vortices are flowing. There is still a strong vortex channeling effect in the y -direction, as reflected in the large asymmetry between 〈 V x 〉 ( θ ) and 〈 V y 〉 (90 ◦ -θ ). For the gRandom array at F d = 0 14, Fig. 19(d) indi-. cates that this drive magnitude is already large enough to render the pinning mostly ineffective, as also shown in Fig. 18(h). The 〈 V x 〉 and 〈 V y 〉 curves are almost mirrorsymmetric, indicating that the pinning gradient has little effect on the motion of the vortices. At F d = 0 20, . Fig. 19(e) shows that the response of cHex becomes increasingly symmetrical, although there is still a small channeling regime for θ ≥ 80 ◦ where the flow is strictly confined along the y -direction. At higher drives (not shown), the 〈 V x 〉 and 〈 V y 〉 curves for cHex become symmetric. For the gRandom array at F d = 0 20, Fig. 19(f) . indicates that the response is almost completely symmetrical.
## VIII. SUMMARY
We have investigated the magnetization, ordering, and transport for vortices interacting with pinning arrays where there is a gradient in the pinning density. We consider conformal pinning arrays constructed by taking a conformal transformation of a uniform pinning array; the new structure has a density gradient but preserves certain crystalline topological features of the original lattice. In particular, we examine conformal crystals obtained from
uniform hexagonal, square, Archimedean and quasiperiodic arrays. We also investigate selected non-conformal arrays with a pinning gradient, including a rectangular pinning array with a 1D gradient and a random arrangement of pinning sites with a gradient. In general we find that conformal arrays which are transformed versions of uniform pinning with well-defined commensurability peaks give the strongest pinning over the widest range of fields. The hexagonal and square conformal arrays produce the strongest pinning and highest critical currents, followed by the conformal Archimedean and conformal Penrose arrays. The random gradient array gives only a slight enhancement in pinning over a uniform random array, indicating that it is the preservation of the local periodicity in the conformal transformation combined with the gradient that gives rise to the enhanced pinning. We show that for the conformal crystal arrays, a portion of the vortices in the sample is locally commensurate with a portion of the pinning array, and this local commensuration effect gradually moves through the system as the externally applied magnetic field is changed. When the field becomes high enough that the vortex density exceeds the pinning density at the outer edge of the sample, the effectiveness of the conformal arrays breaks down, although they still show a small enhancement in pinning compared to random arrays. The rectangular gradient pinning array has relatively weak pinning compared to the conformal arrays at lower fields; however, for higher fields, it develops a pronounced broad peak due to
- 1 G. Blatter, M.V. Feigelman, V.B. Geshkenbein, A.I. Larkin, and V.M. Vinokur, Rev. Mod. Phys. 66 , 1125 (1994).
- 2 G. Gr¨ner, Rev. Mod. Phys. u 60 , 1129 (1988).
- 3 V. J. Goldman, M. Santos, M. Shayegan, and J.E. Cunningham, Phys. Rev. Lett. 65 , 2189 (1990); G. A. Cs´thy, a D. C. Tsui, L. N. Pfeiffer, and K. W. West, Phys. Rev. Lett. 98 , 066805 (2007).
- 4 A. Vanossi, N. Manini, M. Urbakh, S. Zapperi, and E. Tosatti, Rev. Mod. Phys. 85 , 529 (2013).
- 5 S. Xia, L. Ponson, G. Ravichandran, and K. Bhattacharya, Phys. Rev. Lett. 108 , 196101 (2012).
- 6 D. Ravelosona, D. Lacour, J. A. Katine, B. D. Terris, and C. Chappert, Phys. Rev. Lett. 95 , 117203 (2005).
- 7 A. Vanossi, N. Manini, and E. Tosatti, Proc. Natl. Acad. Sci. USA 109 , 16429 (2012); D. McDermott, J. Amelang, C. J. Olson Reichhardt, and C. Reichhardt, Phys. Rev. E 88 , 062301 (2013).
- 8 M. Baert, V. V. Metlushko, R. Jonckheere, V. V. Moshchalkov, and Y. Bruynseraede, Phys. Rev. Lett. 74 , 3269 (1995).
- 9 K. Harada, O. Kamimura, H. Kasai, T. Matsuda, A. Tonomura, and V. V. Moshchalkov, Science 274 , 1167 (1996).
- 10 S. Goldberg, Y. Segev, Y. Myasoedov, I. Gutman, N. Avraham, M. Rappaport, E. Zeldov, T. Tamegai, C.W. Hicks, and K.A. Moler, Phys. Rev. B 79 , 064523 (2009).
- 11 I. Swiecicki, C. Ulysse, T. Wolf, R. Bernard, N. Bergeal, J. Briatico, G. Faini, J. Lesueur, and J. E. Villegas, Phys.
a novel commensuration effect in which integer numbers of ordered columns of vortices fit between the columns of pinning sites.
We also show that the enhanced pinning by the conformal arrays can be observed in current-driven transport measurements. Compared to uniform random or random gradient arrays, the conformal arrays have higher critical depinning forces, and lower average vortex velocities in the moving phase. When the vortices are driven at different angles with respect to the gradient direction, the conformal arrays guide vortex flow perpendicular to the gradient, in channels that pass through the regions of lower pinning density. The rectangular gradient array exhibits a similar guidance effect, with a strongly 1D flow of vortices in the interstitial regions.
In this work we focus on vortices in gradient arrays; however, we expect similar results for other systems featuring particles moving over substrates where the substrate has some type of gradient.
## Acknowledgments
We thank Boldizs´r Jank´ for useful discussions. This a o work was carried out under the auspices of the NNSA of the U.S. DoE at LANL under Contract No. DE-AC5206NA25396.
- Rev. B 85 , 224502 (2012); R. Cordoba et al, Nature Communications 4 , 1437 (2013).
- 12 J. I. Mart´ ın, M. Velez, J. Nogues, and I. K. Schuller, Phys. Rev. Lett. 79 , 1929 (1997); C. Chiliotte, G. Pasquini, V. Bekeris, J. E. Villegas, C.-P. Li, and I. K. Schuller, Supercond. Sci. Technol. 24 , 065008 (2011).
- 13 U. Welp, X. L. Xiao, V. Novosad, and V. K. Vlasko-Vlasov, Phys. Rev. B 71 , 014505 (2005).
- 14 C. J. Olson, C. Reichhardt, and F. Nori, Phys. Rev. B 56 , 6175 (1997).
- 15 C. Reichhardt, C. J. Olson, and F. Nori, Phys. Rev. B 57 , 7937 (1998).
- 16 C. Reichhardt, G. T. Zim´nyi, R. T. Scalettar, a A. Hoffmann, and I. K. Schuller, Phys. Rev. B 64 , 052503 (2001).
- 17 G. R. Berdiyorov, M. V. Milosevic, and F. M. Peeters, Phys. Rev. Lett. 96 , 207001 (2006); Phys. Rev. B 74 , 174512 (2006).
- 18 C. Reichhardt and N. Grønbech-Jensen, Phys. Rev. B 63 , 054510 (2001).
- 19 S. B. Field, S. S. James, J. Barentine, V. Metlushko, G. Crabtree, H. Shtrikman, B. Ilic, and S. R. J. Brueck, Phys. Rev. Lett. 88 , 067003 (2002); A. N. Grigorenko, S. J. Bending, M. J. Van Bael, M. Lange, V. V. Moshchalkov, H. Fangohr, and P. A. J. de Groot, Phys. Rev. Lett. 90 , 237001 (2003).
- 20 J. I. Mart´ ın, M. V´ elez, A. Hoffmann, I. K. Schuller, and J. L. Vicent, Phys. Rev. Lett. 83 , 1022 (1999); Ajay D. Thakur, S. Ooi, M. Chand, J. Jesudasan, P. Raychaudhuri,
- and K. Hirata, AIP Conf. Proc. 1591 , 1651 (2014).
- 21 M. V´ elez, J.I. Mart´ ın, J.E. Villegas, A. Hoffmann, E.M. Gonz´ alez, J.L. Vicent, and I. K. Schuller, J. Mag. Mag. Mat. 320 , 2547 (2008).
- 22 C. Reichhardt, G. T. Zim´nyi, and N. Grønbech-Jensen, a Phys. Rev. B 64 , 014501 (2001).
- 23 G. Karapetrov, J. Fedor, M. Iavarone, D. Rosenmann, and W. K. Kwok, Phys. Rev. Lett. 95 , 167002 (2005).
- 24 C. J. Olson Reichhardt, A. Lib´l, and C. Reichhardt, Phys. a Rev. B 73 , 184519 (2006).
- 25 T. C. Wu, J. C. Wang, Lance Horng, J. C. Wu, and T. J. Yang, J. Appl. Phys. 97 , 10B102 (2005).
- 26 C. Reichhardt and C. J. Olson Reichhardt, Phys. Rev. B 76 , 064523 (2007).
- 27 C. Reichhardt and C. J. Olson Reichhardt, Phys. Rev. Lett. 100 , 167002 (2008).
- 28 M. L. Latimer, G. R. Berdiyorov, Z. L. Xiao, W. K. Kwok, and F. M. Peeters, Phys. Rev. B 85 , 012505 (2012).
- 29 C. Reichhardt and C. J. Olson Reichhardt, Phys. Rev. B 76 , 094512 (2007).
- 30 M. Kemmler, D. Bothner, K. Ilin, M. Siegel, R. Kleiner, and D. Koelle, Phys. Rev. B 79 , 184509 (2009).
- 31 Y. J. Rosen, A. Sharoni, and I. K. Schuller, Phys. Rev. B 82 , 014509 (2010).
- 32 C. J. Olson Reichhardt and C. Reichhardt, Phys. Rev. B 81 , 224516 (2010).
- 33 N. S. Lin, T. W. Heitmann, K. Yu, B. L. T. Plourde, and V. R. Misko, Phys. Rev. B 84 , 144511 (2011).
- 34 G. Karapetrov, V. Yefremenko, G. Mihajlovi´, J. E. Pearc son, M. Iavarone, V. Novosad, and S. D. Bader, Phys. Rev. B 86 , 054524 (2012).
- 35 K. Yu, M. B. S. Hesselberth, P. H. Kes, and B. L. T. Plourde, Phys. Rev. B 81 , 184503 (2010).
- 36 A. Lib´l, C. J. Olson Reichhardt, and C. Reichhardt, Phys. a Rev. Lett. 102 , 237004 (2009).
- 37 M. L. Latimer, G. R. Berdiyorov, Z. L. Xiao, F. M. Peeters, and W. K. Kwok, Phys. Rev. Lett. 111 , 067001 (2013).
- 38 J. Trastoy, M. Malnou, C. Ulysse, R. Bernard, N. Bergeal, G. Faini, J. Lesueur, J. Briatico, and J. E. Villegas, arXiv:1307.2881.
- 39 A. V. Silhanek, L. Van Look, R. Jonckheere, B. Y. Zhu, S. Raedts, and V. V. Moshchalkov, Phys. Rev. B 72 , 014507 (2005).
- 40 R. Cao, Lance Horng, T. J. Yang, T. C. Wu, J. C. Wang, and J. C. Wu, J. of Appl. Phys. 107 , 09E129 (2010).
- 41 A. Kili¸, K. Kili¸, M. Oluta¸ s, and A. Altinkok, Eur. Phys. c c J. B 86 , 339 (2013).
- 42 V. R. Misko, S. Savelev, and F. Nori, Phys. Rev. Lett. 95 , 177007 (2005); Phys. Rev. B 74 , 024522 (2006).
- 43 A. V. Silhanek, W. Gillijns, V. V. Moshchalkov, B. Y. Zhu, J. Moonens, and L. H. A. Leunissen, Appl. Phys. Lett. 89 , 152507 (2006).
- 44 J. E. Villegas, M. I. Montero, C.-P. Li, and I. K. Schuller, Phys. Rev. Lett. 97 , 027002 (2006).
- 45 M. Kemmler, C. Gurlich, A. Sterck, H. P¨ ohler, M. Neuhaus, M. Siegel, R. Kleiner, and D. Koelle, Phys. Rev.
- Lett. 97 , 147003 (2006).
- 46 R. B. G. Kramer, A. V. Silhanek, J. Van de Vondel, B. Raes, and V. V. Moshchalkov, Phys. Rev. Lett. 103 , 067007 (2009).
- 47 V. R. Misko, D. Bothner, M. Kemmler, R. Kleiner, D. Koelle, F. M. Peeters, and F. Nori, Phys. Rev. B 82 , 184512 (2010).
- 48 C. Reichhardt and C. J. Olson Reichhardt, Phys. Rev. Lett. 106 , 060603 (2011).
- 49 D. Ray, C. J. Olson Reichhardt, B. Jank´, and C. Reicho hardt, Phys. Rev. Lett. 110 , 267001 (2013).
- 50 P. Pieranski, in Phase Transitions in Soft Condensed Matter , edited by T. Riste and D. Sherrington (Plenum, New York, 1989), p. 45; F. Rothen, P. Pieranski, N. Rivier, and A. Joyet, Eur. J. Phys. 14 , 227 (1993); F. Rothen and P. Pieranski, Phys. Rev. E 53 , 2828 (1996).
- 51 D. Ray, C. Reichhardt, C. J. Olson Reichhardt, and B. Jank´ o, Physica C 503 , 123 (2014) .
- 52 Y. L. Wang, M. L. Latimer, Z. L. Xiao, R. Divan, L. E. Ocola, G. W. Crabtree, and W. K. Kwok, Phys. Rev. B 87 , 220501(R) (2013).
- 53 S. Gu´ enon, Y. J. Rosen, A. C. Basaran, and I. K. Schuller, Appl. Phys. Lett. 102 , 252602 (2013).
- 54 M. Motta, F. Colauto, W. A. Ortiz, J. Fritzsche, J. Cuppens, W. Gillijns, V. V. Moshchalkov, T. H. Johansen, A. Sanchez, and A. V. Silhanek, Appl. Phys. Lett. 102 , 212601 (2013).
- 55 V. R. Misko and F. Nori, Phys. Rev. B 85 , 184506 (2012).
- 56 D. Ray, C. Reichhardt, and C.J. Olson Reichhardt, Supercond. Sci. Technol. 27 , 075006 (2014).
- 57 B. Gr¨nbaum and G.C. Shephard, u Tilings and Patterns (W.H. Freeman, New York, 1987).
- 58 See, for example, Z. Nehari, Conformal Mapping. (Dover, New York, 1975 [1952]).
- 59 See, for example, J.-B. Suck, M. Schrieber, and P. H¨ussler a (eds.), Quasicrystals: An Introduction to Structure, Physical Properties, and Applications (Springer, Berlin, 2002).
- 60 C. Reichhardt, C. J. Olson, J. Groth, S. Field, and F. Nori, Phys. Rev. B 52 , 10441 (1995); Phys. Rev. B 53 , R8898(R) (1996).
- 61 C. Reichhardt, J. Groth, C. J. Olson, S. B. Field, and F. Nori, Phys. Rev. B 54 , 16108 (1996).
- 62 C.P. Bean, Phys. Rev. Lett. 8 , 250 (1962); Rev. Mod. Phys. 36 , 31 (1964).
- 63 C. Reichhardt, C. J. Olson, and F. Nori, Phys. Rev. Lett. 78 , 2648 (1997).
- 64 A. E. Koshelev and V. M. Vinokur, Phys. Rev. Lett. 73 , 3580 (1994); M. C. Faleski, M. C. Marchetti, and A. A. Middleton, Phys. Rev. B 54 , 12427 (1996).
- 65 C. J. Olson, C. Reichhardt, and F. Nori, Phys. Rev. Lett. 81 , 3757 (1998).
- 66 C. Reichhardt and C. J. Olson Reichhardt, Phys. Rev. B 79 , 134501 (2009); J. Gutierrez, A. V. Silhanek, J. Van de Vondel, W. Gillijns, and V. V. Moshchalkov, Phys. Rev. B 80 , 140514 (2009). | 10.1103/PhysRevB.90.094502 | [
"D. Ray",
"C. Reichhardt",
"C. J. Olson Reichhardt"
] | 2014-08-01T20:00:54+00:00 | 2014-08-01T20:00:54+00:00 | [
"cond-mat.supr-con",
"cond-mat.soft"
] | Pinning, Ordering, and Dynamics of Vortices in Conformal Crystal and Gradient Pinning Arrays | We numerically investigate magnetization, pinning, ordering, and dynamics of
vortices interacting with pinning arrangements which have a density gradient.
We focus on conformal crystal structures obtained by conformally transforming a
spatially uniform periodic array, as well as non-conformal gradient structures
and structures with quasiperiodic order. The conformal structures feature a
density gradient and local ordering. Using magnetization simulations we find
that conformal pinning arrays exhibit enhanced pinning compared to
non-conformal gradient arrays as well as compared to random, periodic, and
quasiperiodic arrays, for a broad range of fields. The effectiveness of
conformal arrays arises from the continuum of length scales introduced into the
arrays by the conformal transformation, allowing for a broad range of local
commensuration effects. At higher vortex fillings above the range of conformal
effectiveness, we show that a non-conformal rectangular gradient array exhibits
strong pinning due to a novel commensuration effect and vortex ordering. Using
transport simulations where vortices are driven along the gradient and at an
angle to the gradient, we confirm the effectiveness of conformal pinning at
increasing the critical current. For a rotated drive, the gradient arrays
produce a strong vortex guidance effect in the direction perpendicular to the
gradient. |
1408.0292v2 | ## Consistent Use of the Standard Model Effective Potential
Anders Andreassen, ∗ William Frost, † and Matthew D. Schwartz ‡ Department of Physics, Harvard University, Cambridge, MA 02138, USA
The stability of the Standard Model is determined by the true minimum of the effective Higgs potential. We show that the potential at its minimum when computed by the traditional method is strongly dependent on the gauge parameter. It moreover depends on the scale where the potential is calculated. We provide a consistent method for determining absolute stability independent of both gauge and calculation scale, order by order in perturbation theory. This leads to a revised stability bounds m pole h > (129 4 . ± 2 3) GeV and . m pole t < (171 2 . ± 0 3) GeV. . We also show how to evaluate the effect of new physics on the stability bound without resorting to unphysical field values.
An intriguing consequence of the recent discovery of the Higgs boson is that its mass apparently places the Standard Model (SM) near the border between absolute stability and metastability [1, 2]. A renowned plot, Fig. 3 of [2], shows the standard model lying close to the end of a metastability funnel in the Higgs-mass/top-mass plane. This unanticipated tuning has inspired a fair amount of speculation about its possible origin and implications. Stability is normally determined by examining the zerotemperature effective potential V for the SM [1-5]: if this potential has a negative minimum at large field values, the SM is said to be unstable; if the inverse decay rate for tunneling out of the electroweak minimum is larger than the lifetime of the universe, the SM is said to be metastable. While these criteria are physical, the extraction of numerical bounds within a consistent perturbation expansion is not straightforward.
One complication in making physical predictions with V is that effective potentials are not gauge-invariant [6]. Although physical quantities extracted from an effective potential (or more generally from an effective action) must be gauge-invariant, there have been surprisingly few explicit checks [7-10]. The traditional approach is simply to work in Landau gauge where calculations are easiest and to assume that the approximations used are self-consistent.
Progress in understanding the gauge-dependence was made by Nielsen [11] and independently by Kugo and Fukuda [12] in 1975. One result from these papers is that the effective potential satisfies a differential equation:
$$\left ( \xi \frac { \partial } { \partial \xi } + C ( h, \xi ) \frac { \partial } { \partial h } \right ) V ( h, \xi ) = 0 \quad \ \ ( 1 ) \quad \stackrel { \tilde { \text{wav} } } { \text{ term} } \\ \sim \lambda _ { \infty }$$
where ξ is the gauge parameter in Fermi gauges and C h,ξ ( ) is a calculable function. This Nielsen identity says that the gauge-dependence of the effective potential can be compensated for by a rescaling of the scalar field h . Two generic implications are that 1) the value of the field h can never be physical, since any rescaling of the field can be compensated by a gauge-change and 2) the value of V ( h, ξ ) at an extremum in h should be gaugeinvariant. Gauge-dependent quantities then include the value of V at any non-extremal point, and the value of h at any point (extremal or not). It is worth noting that Eq. (1) is not quite as powerful as it might seem, since C h,ξ ( ) can be infinite in perturbation theory [9, 11, 13].
Despite the widespread contentment with Landau gauge, the gauge dependence of the effective potential has occasionally caused some discomfort [14-18]. A handful of papers have proposed field redefinitions to generate a gauge-independent potential [19-21]. This approach purportedly allows the effective potential to be used like a classical potential, assigning physical significance to both field values and the potential at each point. However, it is not clear why removing the gauge-dependence automatically makes the potential physical. Moreover, for the field redefinition to be justified it should leave physical quantities unchanged; in that case, we may as well work with fields that make the calculations easiest.
The effective potential has another feature which has not generally been appreciated: it depends on the scale where it is calculated. To see this, note that V satisfies a renormalization group equation [22]:
$$\left ( \mu \frac { \partial } { \partial \mu } - \gamma h \frac { \partial } { \partial h } + \beta _ { i } \frac { \partial } { \partial \lambda _ { i } } \right ) V = 0 \text{ \quad \ \ } \text{(2)}$$
This equation says that the explicit µ -dependence of the potential can be compensated for by rescaling the couplings according to their β -functions and rescaling the field h according to its anomalous dimension, γ . Thus, if we know the potential at a scale µ 0 in terms of the couplings λ i ( µ 0 ) we can find it at a scale µ by solving this equation. Call this method 1. Alternatively, we could have just computed it at the scale µ to begin with, in terms of the couplings λ i ( µ ). Call this method 2. Methods 1 and 2 do not give the same potential, to any order or to all orders in perturbation theory. They differ by the rescaling of the field h . This is not a problem, since we have already concluded from Eq. (1) that physical quantities extracted from V should be independent of field rescaling. The additional freedom of choosing µ 0 illustrates that even in a fixed gauge or with gauge-invariant composite fields, field values are still unphysical.
The unphysical nature of V may be less unsettling after recalling that the effective potential is the constant-field limit of the 1PI effective action. The vertices of this
action at tree-level produce 1PI correlation functions, which can be gauge and scale dependent and satisfy an RGE like Eq. (2). Gauge-invariant S -matrix elements are related to correlation functions by amputation and (in MS) gauge-dependent wave-function renormalization Z -factors. These factors also compensate the scale dependence, letting the S -matrix satisfy an RGE like Eq. (2) without the γ term.
Fortuitously, the value of the potential at a minimum, V min , (or at any extremum) is both gauge-invariant and independent of the scale where it is calculated, without extra Z -factors. The former invariance follows from Eq. (1) and the latter invariance holds simply because the value of any function at any extremum is invariant under any rescaling of its argument. Since the absolute stability bound in the SM is determined by the condition V min < V EW ≈ 0, with V EW the energy of our vacuum (usually renormalized to zero), the bound should be gauge-independent. Unfortunately, gauge-invariance has only been proven non-perturbatively. Indeed, we find that the stability bound is gauge-dependent at each order in perturbation theory if computed by the traditional approach (see Fig. 1 or [18]). In [13], it was shown how effective potential calculations can be reorganized so that V min is gauge-invariant order-by-order. In this paper, we review this 'consistent approach' and apply it to the SM.
We write the SM effective potential as V ( h ), where in unitary gauge the Higgs doublet is normalized as H = 1 √ 2 ( 0 v EW + h ) . The traditional perturbation approach leads to a renormalization-group-improved effective potential of the form [2]
$$V ( h ) = \frac { 1 } { 4 } h ^ { 4 } e ^ { 4 \Gamma ( h ) } \left [ \lambda _ { \text{eff} } ^ { ( 0 ) } ( \mu = h ) + \lambda _ { \text{eff} } ^ { ( 1 ) } ( \mu = h ) + \cdots \right ] \quad \text{(indep} \\ \dots \dots \quad \text{$e$} h \quad \dots \dots \quad \dots \quad \dots \quad \text{noton}$$
with Γ( h ) ≡ ∫ h m t γ µ ( ′ ) dµ ′ µ ′ and 1 4 λ ( j ) eff ( µ h ) 4 the j -loop fixed-order effective potential.
Since stability is determined by large field values and the potential grows as h 4 , the quadratic term -m h 2 2 in the classical potential can be neglected to excellent accuracy. Then the electroweak minimum is at V EW = 0 and the stability bound is determined as the critical Higgs pole mass for which the potential has another minimum with V min = 0. The physical Higgs mass enters through threshold corrections at the weak scale which convert observables into MS couplings. Currently, the β -functions and γ are known to 3-loop order in general R ξ gauges, the fixed-order potential is known to 2-loop order in Landau gauge ( ξ = 0), and the threshold corrections are known to 2 loops (an alternate scheme is discussed in [23]). Using Eq. (3) and the best available data, Ref. [2] found an absolute stability bound of m pole h > (129 1 . ± 1 5) . GeV. Using equations from [2], with some minor corrections confirmed by its authors, and including tau and bottom contributions, we have re- produced this result. We now update the top mass to m pole t = (173 34 . ± 1 12) GeV, with the central value and . ± 0 76 . GeV of the uncertainty from [24], and an additional 0 82GeV theory uncertainty added in quadrature . due to the ambiguity in converting from a Monte Carlo mass scheme to a pole mass scheme [25, 26]. Also including the 3-loop QCD threshold corrections to λ listed but not used in [2], we update this traditional-approach bound to m pole h > (129 67 . ± 1 5) GeV. .
The gauge-dependence of the stability bound at 1-loop is shown in Fig. 1, to be discussed more below. The reason the stability bound appears gauge dependent is due to an improper use of perturbation theory. The key insight, made long ago by Coleman and Weinberg [22] is that the usual loop expansion is inappropriate for effective potentials near quantum-generated minima. Simply put, the classical potential V 0 ∼ λh 4 can only turn over due to 1-loop corrections of the form V 1 ∼ g 4 glyph[planckover2pi1] 16 π 2 h 4 for some g if λ ∼ g 4 glyph[planckover2pi1] 16 π 2 . Since λ ∼ glyph[planckover2pi1] , each factor of λ in a diagram changes its effective loop order. Thus perturbation theory in glyph[planckover2pi1] may still be appropriate, but since λ ∼ glyph[planckover2pi1] it is not the usual loop expansion.
An additional complication is that the effective potential has terms scaling like inverse powers of glyph[planckover2pi1] . For example, a term ∼ glyph[planckover2pi1] 3 g 10 λ -1 appears at 3-loops; since λ counts as glyph[planckover2pi1] , this term scales like glyph[planckover2pi1] 2 and contributes competitively with the 2-loop terms. Including all relevant terms according to this modified power counting, it was shown in [13] that V min is indeed gauge-invariant in scalar QED. The required terms include the 2-loop effective potential in R ξ gauge as well as an infinite series of 'daisy' loops producing terms in V proportional to g 4 +2 j λ 1 -j .
The consistent method for an order-by-order gaugeindependent calculation of V min presented in [13] translates to the SM as follows. First, we truncate the effective potential to order glyph[planckover2pi1] with λ ∼ glyph[planckover2pi1] power counting. This gives the leading-order (LO) potential:
$$\Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad &\Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad& \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad& & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \ \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { 0 } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { 0 \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { n } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { m } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { p } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { 1 } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { 2 } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { p } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { \ } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^ { q } \text{in} \quad & \Lambda _ { h } ^$$
Note that this potential includes tree-level and 1-loop contributions, and is gauge-invariant. From this, we can solve for the scale h = µ X where dV (LO) /dh = 0. Explicitly µ X is the MS scale where the condition
$$\text{wall} _ { \substack { \text{d of} \\ \iota \ [ 2 ], \quad \\ \text{horse}, \quad \\ - 3 \left ( g _ { 1 } ^ { 2 } + g _ { 2 } ^ { 2 } \right ) ^ { 2 } \ln \frac { g _ { 1 } ^ { 2 } + g _ { 2 } ^ { 2 } } { 4 } - 6 g _ { 2 } ^ { 4 } \ln \frac { g _ { 2 } ^ { 2 } } { 4 } + 4 8 y _ { t } ^ { 4 } \ln \frac { y _ { t } ^ { 2 } } { 2 } \right ] \quad ( 5 ) }$$
$$\Phi _ { 0 } = 4 \quad \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi_ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \quad \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Phi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } +\lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ { 1 } + \lambda _ { 3 2 } \, \Psi _ _ { 1 } + \lambda _ { 3 2 } \, \Psi _ _ { 1 } + \lambda _ { 3 2 } \, \Psi _ _ { 1 } + \lambda _ { 3 2 } \, \Psi _ _ { 1 } + \lambda _ { 3 2 } \, \Psi _ _ { 1 } + \lambda _ { 3 2 } \, \Psi _ _ { 1 } + \lambda _ { 3 2 } \, \Psi _ _ { 1 } + \lambda _ { 3 2 } \, \Psi _ _ { 1 } + \lambda _ { 3 2 } \, \Psi _ _ { 1 } + \lambda _ { 3 2 } \, \Psi _ _ { 1 } + \lambda _ { 3 2 } \, \Psi _ _ { 1 } + \lambda _ { 3 2 } \, \Psi _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ } _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ { _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ = _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ | _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ [ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _._ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ __ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _. _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \, _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \\ } } } } } } } } } } }$$
FIG. 1. Gauge dependence of the absolute stability bound with m pole t = 173 34 GeV. .
is satisfied. For values of m h and m t close to the observed SM values, there are two solutions to this equation: the lower µ X is where V (LO) has a maximum, and the higher µ X where the minimum occurs. In the SM these scales are
$$\mu _ { X } ^ { \max } = 2. 4 6 \times 1 0 ^ { 1 0 } \, \text{GeV} \quad \quad ( 6 ) \quad \text{Thi}$$
$$\mu _ { X } ^ { \min } = 3. 4 3 \times 1 0 ^ { 3 0 } \text{ GeV}. \quad \quad ( 7 ) \quad \text{fro}$$
These numbers and results which follow use m pole h = (125 14 . ± 0 24) GeV, combined from [27, 28]. .
For the potential at the next-to-leading order (NLO), one contribution comes from the glyph[planckover2pi1] 2 terms in the 1-loop potential with λ ∼ glyph[planckover2pi1] scaling:
$$V ^ { ( 1, N L O ) } ( h ) & = \frac { - 1 } { 2 5 6 \pi ^ { 2 } } \left [ \xi _ { B } g _ { 1 } ^ { 2 } \left ( \ln \frac { \lambda h ^ { 4 } ( \xi _ { B } g _ { 1 } ^ { 2 } + \xi _ { W } g _ { 2 } ^ { 2 } ) } { 4 \mu ^ { 4 } } - 3 \right ) \quad \text{Fig. 1} \\ + & \xi _ { W } g _ { 2 } ^ { 2 } \left ( \ln \frac { \lambda ^ { 3 } h ^ { 1 2 } \xi _ { W } ^ { 2 } g _ { 2 } ^ { 4 } ( \xi _ { B } g _ { 1 } ^ { 2 } + \xi _ { W } g _ { 2 } ^ { 2 } ) } { 6 4 \mu ^ { 1 2 } } - 9 \right ) \right ] \lambda h ^ { 4 } \quad ( 8 ) \quad \text{plot we} \\ & \quad \text{rameters} \\ & \quad \text{RGE ev} \\ & \quad \text{.}$$
Another contribution V (2 NLO) , ( h ) comes from the λ 0 and ln λ terms in 2-loop potential. In Landau gauge, these terms are h / 4 4 times what is written as λ (2) eff in Eq. (C.4) of the published version of [2]. Finally, there is the contribution, V ( n> , 2 NLO) ( h ) from 3-loop and higher order graphs proportional to inverse powers of λ . Including all these terms, the potential at each extremum will be gauge-invariant. Conveniently, the higher-loop-order graphs contributing at NLO vanish in Landau gauge ( ξ B = ξ W = 0). Thus the gauge-invariant NLO value of the potential at the minimum is simply
$$V _ { \min } ^ { N L O } = V ^ { ( L O ) } ( \mu _ { X } ) + V ^ { ( 2, N L O ) } ( \mu _ { X } ) \quad \ ( 9 ) \quad \ g a u _ { \xi }$$
To derive this, we consistently truncated to O ( glyph[planckover2pi1] 2 ) and used d dh V (LO) = 0 at h = µ X . Note that this is the RGimproved effective potential: the resummation is implicit in the solution for µ X . At NNLO, an infinite number of loops are relevant, even in Landau gauge [13].
FIG. 2. Gauge dependence of the instability scale Λ I , defined by V (Λ ) = 0, at 1-loop in the traditional approach. I There is no known way to make this scale gauge-invariant.
Using Eq. (9) we find that for absolute stability at NLO, the Higgs pole mass must satisfy
$$\text{ales} \quad & \frac { m _ { h } ^ { \text{pole} } } { \text{GeV} } > ( 1 2 9. 4 0 \pm 0. 5 8 ) + 2. 2 6 \, ( \frac { m _ { t } ^ { \text{pole} } - 1 7 3. 3 4 \, \text{GeV} } { 1. 1 2 \, \text{GeV} } ) \\ ( 6 ) \quad & \tau _ { 1 \colon - 1 } \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \end{cases}$$
This bound is around 275 MeV lower than the bound from the traditional approach in Landau gauge ( m pole h > 129 67 GeV). . The ± 0 58 is . pertubative and α s uncertainty [2]. Since the Higgs mass is known better than the top mass, it perhaps makes more sense to write the bound as
$$\stackrel { \cdot \cdot \cdot } { \text{loop} } \quad & m _ { t } ^ { \text{pole} } \\ & \frac { m _ { t } ^ { \text{pole} } } { \text{GeV} } < ( 1 7 1. 2 2 \pm 0. 2 8 ) + 0. 1 2 \, ( \frac { m _ { h } ^ { \text{pole} } - 1 2 5. 1 4 \, \text{GeV} } { 0. 2 4 \, \text{GeV} } ) \\. \quad \cdot \quad \sum _ { \substack { \text{r} \colon \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \, \text{s} \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \colon \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \end{array} } ( 1 1 )$$
Fig. 1 compares the gauge-dependence of the bound at 1-loop to the LO, NLO and 2-loop bounds. For this plot we have taken the U(1) and SU(2) R ξ gauge parameters equal to ξ t when µ = m t and included their RGE evolution [29]. All bounds include 2-loop thresholds and 3-loop running. We find that the bound at LO is m pole h > 129 69 GeV which is nearly identical to . the Landau gauge 1-loop bound in the traditional approach, m pole h > 129 70 GeV. We do not plot the gauge-. dependence of the 2-loop bound since we have not computed the gauge-dependent 2-loop potential or the daisy contribution. That the bound seems to asymptote to a finite value in unitary gauge ( ξ = ∞ ) may be due to much (but not all) of the gauge-dependence being in the e 4Γ prefactor in Eq. (3) which drops out of the V = 0 condition.
Fig. 2 shows the gauge dependence of the instability scale Λ I , defined by V (Λ ) = 0 [1, 2], and its LandauI gauge value at 2-loops, including 3-loop resummation in both cases. Since the instability scale is a field value, it is not obviously physical. We know of no way to compute it in a consistent and gauge-invariant manner.
Fig. 3 shows the value of V max computed by various approaches. We find approximately exponential dependence of V max (and also V min ) on ξ t in the traditional
FIG. 3. Gauge dependence of the SM potential at its maximum with m pole h = 125 14 GeV and . m pole t = 173 34 GeV. .
$$\text{approach at $1$-loop. Decent fits are }$$
$$\left ( V _ { \max } ^ { 1 - \text{loop}, \, \text{trad.} } \right ) ^ { 1 / 4 } & \approx ( 2. 5 0 \times 1 0 ^ { 9 } \, \text{GeV} ) e ^ { - 0. 0 2 \xi _ { t } + 0. 0 0 0 3 \xi _ { t } ^ { 2 } } \quad \text{FIG.} \quad \text{and } r \\ \left ( - V _ { \min } ^ { 1 - \text{loop}, \, \text{trad.} } \right ) ^ { 1 / 4 } & \approx ( 3. 0 8 \times 1 0 ^ { 2 9 } \, \text{GeV} ) e ^ { 0. 0 0 1 \xi _ { t } - 0. 0 0 0 1 \xi _ { t } ^ { 2 } } \quad \text{lower} \\ \quad \text{The } t \quad \text{ed.} \quad.$$
The consistent gauge-invariant values at NLO are
$$\left ( V _ { \max } ^ { N L O } \right ) ^ { 1 / 4 } & = 2. 8 8 \times 1 0 ^ { 9 } \, \text{GeV} \quad \quad ( 1 3 ) \quad \text{lines} \\ \left ( - V _ { \min } ^ { N L O } \right ) ^ { 1 / 4 } & = 2. 4 0 \times 1 0 ^ { 2 9 } \, \text{GeV} \quad \quad \\ \quad \quad \quad \text{diss}$$
Note that -V min corresponds to an energy density well above the Planck scale. Thus, the potential at the minimum will surely be effected by quantum gravity and possible new physics not included in our calculation. Previous analyses have defined stability to be Planck-sensitive if the instability scale Λ I > M Pl [1, 2]. As we have observed, the instability scale is gauge dependent, so this is not a consistent criterion. An alternative criterion is that new operator, such as O 6 ≡ 1 Λ 2 NP h 6 be comparable to V min when h = 〈 h 〉 . Although O 6 and V min are gaugeinvariant, the value of O 6 at the field value h where the minimum occurs is gauge dependent, so this condition is also unsatisfactory. A consistent and satisfactory criterion was explained in [13]: the new operator must be added to the classical theory and its effect on V min evaluated.
Adding O 6 to the potential, we find that the the potential is still negative at its minimum in the SM even for operators with very large coefficients. For example, taking Λ NP = M Pl = 1 22 . × 10 19 GeV, we find that µ min X = 6 0 . × 10 17 GeV and V min = -(1 1 . × 10 17 GeV) . 4 Comparing to Eq. (13) we see that the energy of the true vacuum is very Planck-sensitive.
More generally, a good fit is given by
$$V _ { \min } = - ( 0. 0 1 \, \Lambda _ { N P } ) ^ { 4 }, \quad \Lambda _ { N P } \gtrsim 1 0 ^ { 1 2 } \, G e V \quad ( 1 4 ) \quad \text{true} \quad.$$
When Λ NP < . 3 6 × 10 12 GeV, V min becomes positive and for Λ NP < 3 1 . × 10 12 GeV the maximum and minimum
FIG. 4. Boundaries of absolute stability (lower band, NLO) and metastability (upper line, LO). The thickness of the lower boundary indicates perturbative and α s uncertainty. The theoretical uncertainty of the metastability boundary is unknown. The elliptical contours are 68%, 95% and 99% confidence bands on the Higgs and top masses: m pole h = (125 14 . ± 0 23) GeV and . m pole t = (173 34 . ± 1 12) GeV. Dotted . lines are scales in GeV at which V min can be lifted positive by new physics.
disappear. Thus the stability of the Standard Model can be modified by new physics at the scale 10 12 GeV.
If we vary the Higgs and top masses in the Standard Model, we can compute the boundary of absolute stability. This bound is shown in Figs. 4 and 5. The dotted lines show where V min becomes positive when in the presence of O 6 for the indicated value of Λ NP . Unexpectedly, we find that three independent conditions (1) that V min goes to zero, (2) that Eq. (5) have no solution, and (3) that V min goes positive when Λ NP = M Pl all give nearly identical boundaries in the m pole h / m pole t plane. Knowing that quantum gravity is relevant at M Pl , we should therefore be cautious about giving too strong of an interpretation of the perturbative absolute stability bound in the SM. We also show in this plot the metastability bound, that the lifetime of our vacuum be larger than the age of the universe. At lowest order this translates to λ ( 1 R ) -1 < -14 53 + 0 153 ln[ . . R GeV] for all R [30]. Since λ µ ( ) is gauge invariant, so is this criterion. Although for the Standard Model this approximation is probably sufficient, it has not been demonstrated that the bound can be systematically improved in a guage-invariant way [31].
In this paper, we have only discussed a single physical feature of the effective action: the value of the effective potential at its extrema. There is of course much more content in the effective action, especially when temperature dependence is included. Unfortunately, many uses of the effective action involve evaluating it for particular field configurations, a procedure that has repeatedly been shown to be gauge-dependent. For example, the
FIG. 5. Same as previous figure but zoomed out. Theory uncertainties on the absolute and metastability bounds are not shown.
gauge-dependence of various quantities associated with the electroweak phase transition were discussed [16] and various predictions of Higgs inflation models [32] in [17].
Since observables such as the gravitational wave spectrum or the size of tensor fluctuations in the cosmic microwave background can in principle be predicted within quantum field theory, it should be possible to at least set up such calculations in a way that does not depend on arbitrary gauge or scale choices in the effective action. Questions which involve other parts of the effective action besides the potential provide new opportunities for cancellation. For example, after the Z -factors are added according to the LSZ reduction theorem, S -matrix elements calculated from the effective action are appropriately invariant. It would be interesting to see perturbative demonstrations of the gauge-invariance of other derived quantities from the effective action.
We thank S. Moch, M. Ramsey-Musolf, M. Reece, M. Strassler and A. Strumia for helpful discussions. The authors are supported in part by grant DE-SC003916 from the Department of Energy. AA is supported in part by the Stolt-Nielsen Fund for Education of the the American-Scandinavian Foundation and the NorwayAmerica Association. WF is supported in part by the Lord Rutherford Memorial Research Fellowship.
∗ [email protected]
† [email protected]
- ‡ [email protected]
- [1] G. Degrassi, S. Di Vita, J. Elias-Miro, J. R. Espinosa, G. F. Giudice, et al. , JHEP 1208 , 098 (2012), arXiv:1205.6497 [hep-ph].
- [2] D. Buttazzo, G. Degrassi, P. P. Giardino, G. F. Giudice, F. Sala, et al. , JHEP 1312 , 089 (2013), arXiv:1307.3536.
- [3] N. Cabibbo, L. Maiani, G. Parisi, and R. Petronzio, Nucl.Phys. B158 , 295 (1979).
- [4] M. Sher, Phys.Rept. 179 , 273 (1989).
- [5] J. Casas, J. Espinosa, and M. Quiros, Phys.Lett. B342 , 171 (1995), arXiv:hep-ph/9409458 [hep-ph].
- [6] R. Jackiw, Phys.Rev. D9 , 1686 (1974).
- [7] J. Kang, Phys.Rev. D10 , 3455 (1974).
- [8] L. Dolan and R. Jackiw, Phys.Rev. D9 , 2904 (1974).
- [9] I. Aitchison and C. Fraser, Annals Phys. 156 , 1 (1984).
- [10] W. Loinaz and R. Willey, Phys.Rev. D56 , 7416 (1997), arXiv:hep-ph/9702321 [hep-ph].
- [11] N. Nielsen, Nucl.Phys. B101 , 173 (1975).
- [12] R. Fukuda and T. Kugo, Phys.Rev. D13 , 3469 (1976).
- [13] A. Andreassen, W. Frost, and M. D. Schwartz, (2014), arXiv:1408.0287 [hep-ph].
- [14] L. Di Luzio and L. Mihaila, JHEP 1406 , 079 (2014), arXiv:1404.7450 [hep-ph].
- [15] D. Boyanovsky, W. Loinaz, and R. Willey, Phys.Rev. D57 , 100 (1998), arXiv:hep-ph/9705340 [hep-ph].
- [16] C. Wainwright, S. Profumo, and M. J. Ramsey-Musolf, Phys.Rev. D84 , 023521 (2011), arXiv:1104.5487 [hepph].
- [17] J. L. Cook, L. M. Krauss, A. J. Long, and S. Sabharwal, Phys.Rev. D89 , 103525 (2014), arXiv:1403.4971 [astroph.CO].
- [18] A. Andreassen, master's thesis, NTNU-Trondheim (2013).
- [19] J.-M. Frere and P. Nicoletopoulos, Phys.Rev. D11 , 2332 (1975).
- [20] S. H. Tye and Y. Vtorov-Karevsky, Int.J.Mod.Phys. A13 , 95 (1998), arXiv:hep-th/9601176 [hep-th].
- [21] N. Nielsen, (2014), arXiv:1406.0788 [hep-ph].
- [22] S. R. Coleman and E. J. Weinberg, Phys.Rev. D7 , 1888 (1973).
- [23] A. Spencer-Smith, (2014), arXiv:1405.1975 [hep-ph].
- [24] ATLAS Collaboration, CDF Collaboration, CMS Collaboration and D0 Collaboration, (2014), arXiv:1403.4427 [hep-ex].
- [25] A. H. Hoang and I. W. Stewart, Nucl.Phys.Proc.Suppl. 185 , 220 (2008), arXiv:0808.0222 [hep-ph].
- [26] S. Moch, S. Weinzierl, S. Alekhin, J. Blumlein, L. de la Cruz, et al. , (2014), arXiv:1405.4781 [hep-ph].
- [27] CMS Collaboration, CMS-PAS-HIG-14-009 (2014).
- [28] G. Aad et al. (ATLAS Collaboration), (2014), arXiv:1406.3827 [hep-ex].
- [29] A. Bednyakov, A. Pikelner, and V. Velizhanin, JHEP 1301 , 017 (2013), arXiv:1210.6873 [hep-ph].
- [30] G. Isidori, G. Ridolfi, and A. Strumia, Nucl.Phys. B609 , 387 (2001), arXiv:hep-ph/0104016 [hep-ph].
- [31] A. Strumia and N. Tetradis, Nucl.Phys. B542 , 719 (1999), arXiv:hep-ph/9806453 [hep-ph].
- [32] A. Barvinsky, A. Y. Kamenshchik, and A. Starobinsky, JCAP 0811 , 021 (2008), arXiv:0809.2104 [hep-ph]. | 10.1103/PhysRevLett.113.241801 | [
"Anders Andreassen",
"William Frost",
"Matthew D. Schwartz"
] | 2014-08-01T20:01:27+00:00 | 2014-08-25T19:39:10+00:00 | [
"hep-ph"
] | Consistent Use of the Standard Model Effective Potential | The stability of the Standard Model is determined by the true minimum of the
effective Higgs potential. We show that the potential at its minimum when
computed by the traditional method is strongly dependent on the gauge
parameter. It moreover depends on the scale where the potential is calculated.
We provide a consistent method for determining absolute stability independent
of both gauge and calculation scale, order by order in perturbation theory.
This leads to a revised stability bounds mH > (129.4 \pm 2.3) GeV and mt <
(171.2 \pm 0.3)GeV. We also show how to evaluate the effect of new physics on
the stability bound without resorting to unphysical field values. |
1408.0293v3 | ## On the Halo Velocity Bias
Matteo Biagetti (1) , Vincent Desjacques (1) , Alex Kehagias (1 2) , and Antonio Riotto (1)
(1) Universit´ de Gen`ve, Department of Theoretical Physics and Center for Astroparticle Physics (CAP), e e 24 quai E. Ansermet, CH-1211 Geneva 4, Switzerland and (2) Physics Division, National Technical University of Athens, 15780 Zografou Campus, Athens, Greece
(Dated: November 27, 2014)
It has been recently shown that any halo velocity bias present in the initial conditions does not decay to unity, in agreement with predictions from peak theory. However, this is at odds with the standard formalism based on the coupled-fluids approximation for the coevolution of dark matter and halos. Starting from conservation laws in phase space, we discuss why the fluid momentum conservation equation for the biased tracers needs to be modified in accordance with the change advocated in Baldauf, Desjacques & Seljak (2014). Our findings indicate that a correct description of the halo properties should properly take into account peak constraints when starting from the Vlasov-Boltzmann equation.
## I. INTRODUCTION
While the existence of a spatial bias between the halos hosting galaxies and the underlying dark matter (DM) distribution has been firmly established since the pioneering work of [1], the presence of a halo velocity bias is still being debated. Clearly, whereas galaxy velocities can be physically biased ( i.e. on an object-by-object basis) owing to differences between the dark matter and baryon velocity fields (see, e.g. , Refs. [2, 3]), by construction halo locally flow with the dark matter (in Einstein's theory of gravity), so that a halo velocity bias can only arise statistically. Namely, it must be the statistical manifestation of a selection effect, which should naturally arise since virialized structures preferentially trace overdense regions of the Universe [4, 5].
A step forward in the understanding of halo bias has been recently made in Ref. [6] where, through N-body simulations, the authors have measured a halo velocity bias b v ( k ),
$$v _ { h } ( \vec { k }, t ) = b _ { v } ( k ) \, v _ { \text{dm} } ( \vec { k }, t ), \ b _ { v } ( k ) = ( 1 - R _ { v } ^ { 2 } k ^ { 2 } ), & & ( \text{I.1} )$$
which appears to remain constant throughout the cosmic time t until virialization. Here, R v is the typical scale of the halo velocity bias
$$R _ { v } ^ { 2 } = \frac { \sigma _ { 0 } ^ { 2 } } { \sigma _ { 1 } ^ { 2 } }, \ \sigma _ { j } ^ { 2 } = \int \frac { d ^ { 3 } k } { ( 2 \pi ) ^ { 3 } } \, k ^ { 2 j } P _ { \text{dm} } ( k ) W ^ { 2 } ( K R ),$$
P dm ( k ) is the DM power spectrum and W x ( ) is a spherically symmetric smoothing kernel. This statistical effect is consistent with the relative suppression of the halo velocity divergence power spectrum at late time subsequently reported in Ref. [7], although this type of measurement is more prone to systematics arising from sparse sampling (see e.g. Refs. [8-10]).
Whereas the finding of Ref. [6] is in full agreement with the peak model, which indeed predicts the existence of a linear, statistical halo velocity bias which remains constant with time [11-13], it seems at odds with the prediction based on the coupled-fluids approximation for the coevolution of DM and halos [14]. The latter is widely used to compute the time evolution of bias [15-17] and is based on the idea of following the evolution over cosmic time and in Eulerian space of the halo progenitors - the so-called proto-halos - until their virialization. While their shapes and topology change as a function of time (smaller substructures gradually merge to form the final halo), their centre of mass moves along a well-defined trajectory determined by the surrounding mass density field. Therefore, unlike virialized halos that experience merging, by construction proto-halos always preserve their identity. Their total number is therefore conserved over time, and one can write a continuity equation and an Euler equation for their number density and velocity, respectively. Nevertheless, this approach predicts that any Eulerian velocity bias rapidly decays to unity [14]
$$b _ { v } ^ { \text{E} } ( k, t ) = 1 + D ^ { - 3 / 2 } ( t ) ( b _ { v } ( k ) - 1 ),$$
where D t ( ) is the linear growth rate normalized to unity at the collapse redshift. To reconcile these two apparently contradictory results, the authors of Ref. [6] argued that the Euler equation for halos should be changed from
$$\dot { \theta } _ { h } + H \theta _ { h } + \frac { 3 } { 2 } H ^ { 2 } \Omega _ { \text{dm} } \delta + \dots = 0, \ \theta _ { h } = \vec { \nabla } \cdot \vec { v } _ { h },$$
which predicts the incorrect behavior (I.3), to
$$\dot { \theta } _ { h } + H \theta _ { h } + \frac { 3 } { 2 } b _ { v } ( k ) H ^ { 2 } \Omega _ { d m } \delta _ { d m } + \dots = 0, \ \theta _ { h } = \vec { \nabla } \cdot \vec { v } _ { h },$$
where H is the Hubble rate and Ω dm parametrizes the abundance of DM and the dots stand for higher-order terms. Their physical interpretation is that the gravitational force acting on DM halos is statistically biased. Together with the Euler equation for DM
$$\dot { \theta } _ { \text{dm} } + H \theta _ { \text{dm} } + \frac { 3 } { 2 } H ^ { 2 } \Omega _ { \text{dm} } \delta _ { \text{dm} } + \dots = 0, \ \theta _ { \text{dm} } = \vec { \nabla } \cdot \vec { v } _ { \text{dm} },$$
one indeed recovers the behavior (I.1) and the Eulerian velocity halo bias does not decay in time to unity.
Another reason why Eq. (I.4) cannot describe the momentum evolution of halos is the fact that it does not differ at all from (I.6). Consequently, it is not possible that Eq. (I.4) describes clustered objects like halos (or peaks), which are different from the smooth DM distribution as they are supposed to be located at points where /vector ∇ δ dm = 0 and /vector where the smoothed density contrast is larger than some value νσ 0 , being ν the peak height. In other words, Eq. (I.4) does not contain any information about the fact we are dealing with peaks.
The goal of this short note is to explain - at a somewhat more fundamental level than done in [6] - why and how one needs to modify the momentum fluid equation for the coupled-fluids of halos and DM to obtain Eqs. (I.5) and (I.6) as advocated in [6]. The paper is organized as follows. In section II we present a derivation of the fluid equation from first principles and deal with the effective force felt by peaks in section III. Finally, section IV contains our conclusions and further comments.
## II. FROM THE KLIMONTOVICH-DUPREE EQUATION TO THE FLUID EQUATION FOR HALOS
Let us start from the single particle phase space density
$$f _ { \text{K} } ( \vec { r }, \vec { p }, t ) = \sum _ { i } \delta _ { \text{D} } \left [ \vec { r } - \vec { r } _ { i } ( t ) \right ] \delta _ { \text{D} } \left [ \vec { p } - \vec { p } _ { i } ( t ) \right ],$$
where 'K' indicate the so-called Klimontovich density [18, 19]. We are following the phase space trajectories of single DM 'particles' without any averaging, instead we are considering only one realization of a universe. We have used the cosmic time t and halos will be eventually identified at the points where maxima of the DM density contrast are located and with a smoothed density contrast larger than some νσ 0 . The Klimontovich density obeys the Klimontovich-Dupree equation
$$\frac { \partial f _ { K } } { \partial t } + \vec { p } \cdot \frac { \partial } { \partial \vec { r } } f _ { K } - \vec { \nabla } \Phi _ { K } \cdot \frac { \partial } { \partial \vec { p } } f _ { K } = 0,$$
where
$$\vec { \nabla } \Phi _ { K } & = G _ { N } \int d ^ { 3 } r ^ { \prime } d ^ { 3 } p ^ { \prime } f _ { K } ( \vec { r } ^ { \prime }, \vec { p } ^ { \prime }, t ) \frac { ( \vec { r } - \vec { r } ^ { \prime } ) } { | \vec { r } - \vec { r } ^ { \prime } | ^ { 3 } }. & & ( \text{II.3} )$$
The Klimontovich density following the trajectories of all the single particles, some coarse graining is needed in order to handle the huge amount of information encoded in this quantity. Macrostates can be identified by averaging in a standard way over a statistical ensemble of microstates with similar phase space density in small volumes containing a sufficient amount of particles.
Denoting this averaging using angle brackets 〈· · · 〉 , the first of the distribution functions is given by
$$\left \langle f _ { \mathbf K } ( \vec { r }, \vec { p }, t ) \right \rangle = \left \langle \sum _ { i } \delta _ { \mathbf D } \left [ \vec { r } - \vec { r } _ { i } ( t ) \right ] \delta _ { \mathbf D } \left [ \vec { p } - \vec { p } _ { i } ( t ) \right ] \right \rangle = f ( \vec { r }, \vec { p }, t ).$$
Therefore, on averaging the Klimontovich equation (II.2) over the ensemble of realizations, we obtain
$$\frac { \partial f } { \partial t } + \vec { p } \cdot \frac { \partial f } { \partial \vec { r } } - \langle \vec { \nabla } \Phi \rangle \cdot \frac { \partial f } { \partial \vec { p } } + \dots = 0,$$
where
$$\langle \vec { \nabla } \Phi \rangle = G _ { N } \int { \mathrm d } ^ { 3 } r ^ { \prime } { \mathrm d } ^ { 3 } p ^ { \prime } f ( \vec { r } ^ { \prime }, \vec { p } ^ { \prime }, t ) \frac { ( \vec { r } - \vec { r } ^ { \prime } ) } { \left | \vec { r } - \vec { r } ^ { \prime } \right | ^ { 3 } } }$$
and the · · · stand for terms arising at higher-order in perturbation theory when correlations introduced by gravitational clustering are taken into account. We will comment on this point later on.
The derivation above highlights that one should compute the averaged (in a statistical sense) gravitational force (per unit mass) 〈∇ /vector Φ . The smooth DM distribution is described by the Vlasov equation, such that the average force 〉 〈∇ /vector Φ 〉 can be replaced by the Poisson equation
$$\nabla ^ { 2 } \Phi = \frac { 3 } { 2 } H ^ { 2 } \delta _ { \text{dm} }.$$
However, in the particular case of density peaks, the average force exerted by the smooth DM component is statistically biased owing to the fact that peaks stream towards (or move apart from) each other more coherently in high (low) density environments. One should remember though that on an object-by-object basis, all the particle species (DM, halos etc.) experience the same force, in agreement with Einstein's equivalence principle.
In order to use Eq. (II.5) to describe the halo phase-space it is necessary therefore to consider the mean shift in 〈∇ /vector Φ 〉 in the vicinity of a peak. Since /vector ∇ Φ correlates only with /vector ∇ δ dm (at linear order), one has to compute the conditional probability of having at a given peak location a given value of /vector ∇ Φ given the fact that at the same point the gradient of the DM density contrast has a given value /vector ∇ δ dm . We will perform this calculation in the next section.
## III. THE EFFECTIVE FORCE FOR PEAKS
In order to understand the change in the force felt by the peaks on average, we simply have to calculate the average force (II.6) subject to the peak constraint, that is, /vector ∇ δ dm = 0 and the corresponding Hessian is negative (since peaks /vector are identified with local maxima of the DM over density field).
We will restrict ourselves at the linear level in perturbation theory in such a way that, at early times, the DM density δ dm and /vector ∇ Φ are Gaussian variables. We can therefore apply the theorem presented, for instance, in Refs. [20, 21] stating that the conditional probability of zero-mean Gaussian variables Y A and Y B is itself a Gaussian variable with mean
$$\left \langle Y _ { B } \left | Y _ { A } \right \rangle \equiv \frac { \left \langle Y _ { B } \otimes Y _ { A } \right \rangle } { \left \langle Y _ { A } \otimes Y _ { A } \right \rangle } Y _ { A }$$
and covariance matrix
$$C ( Y _ { B }, Y _ { A } ) & \equiv \left \langle Y _ { B } \otimes Y _ { B } \right \rangle - \frac { \left \langle Y _ { B } \otimes Y _ { A } \right \rangle } { \left \langle Y _ { A } \otimes Y _ { A } \right \rangle } \right \rangle \\ \quad \cdot \,. & \quad \right \rangle \quad. & \quad. \quad.$$
Let us identify Y A with /vector ∇ δ dm and Y B with /vector ∇ Φ. Therefore, the mean shift in 〈∇ /vector Φ 〉 in the vicinity of a peak is given by
$$\left \langle \vec { \nabla } \Phi \right | \vec { \nabla } \delta _ { \text{dm} } \right \rangle & = \frac { \left \langle \vec { \nabla } \Phi \vec { \nabla } \delta _ { \text{dm} } \right \rangle } { \left \langle \left ( \vec { \nabla } \delta _ { \text{dm} } \right ) ^ { 2 } \right \rangle } \vec { \nabla } \delta _ { \text{dm} } \\ & = - \frac { 3 } { 2 } H ^ { 2 } \frac { \left \langle \delta _ { \text{dm} } ^ { 2 } \right \rangle } { \left \langle \left ( \vec { \nabla } \delta _ { \text{dm} } \right ) ^ { 2 } \right \rangle } \vec { \nabla } \delta _ { \text{dm} } \\ & = - \frac { 3 } { 2 } H ^ { 2 } \frac { \sigma _ { 0 } ^ { 2 } } { \sigma _ { 1 } ^ { 2 } } \vec { \nabla } \delta _ { \text{dm} } \\ & = - \frac { 3 } { 2 } H ^ { 2 } R _ { v } ^ { 2 } \vec { \nabla } \delta _ { \text{dm} }. \quad & ( \text{III.3)} \\ \text{is expression, we have assumed that both $\vec{\nabla} \Phi$ and $\vec{\nabla} \delta _ { \text{dm} }$ are Gaussian-distributed and therefore} \\ of the \text{linear} \text{local} \text{ } \text{$\text{$at}$ mirroration theory} \text{ a modification should be avroated}$$
We reiterate that, in this expression, we have assumed that both /vector ∇ Φ and /vector ∇ δ dm are Gaussian-distributed and therefore the result is valid only at the linear level. At higher-order in perturbation theory a modification should be expected when going to smaller distances.
We now proceed to compute the covariance matrix for /vector ∇ Φ given the constraint. First, we provide the various elements
$$\left \langle \vec { \nabla } \Phi \otimes \vec { \nabla } \Phi \right \rangle \, = \, \frac { \alpha ^ { 2 } } { 3 } \sigma _ { - 1 } ^ { 2 } \, 1 _ { 3 \times 3 }, \\ \left \langle \vec { \nabla } \Phi \otimes \vec { \nabla } \delta _ { d m } \right \rangle \, = \, - \frac { \alpha } { 3 } \sigma _ { 0 } ^ { 2 } \, 1 _ { 3 \times 3 }, \\ \left \langle \vec { \nabla } \delta _ { d m } \otimes \vec { \nabla } \Phi \right \rangle \, = \, - \frac { \alpha } { 3 } \sigma _ { 0 } ^ { 2 } \, 1 _ { 3 \times 3 }, \\ \left \langle \vec { \nabla } \delta _ { d m } \otimes \vec { \nabla } \delta _ { d m } \right \rangle \, = \, \frac { 1 } { 3 } \sigma _ { 1 } ^ { 2 } \, 1 _ { 3 \times 3 }, \\ 3 H ^ { 2 } \Omega _ { d m } / 2. \, The covariance matrix for \vec { \nabla } \Phi \, given the constraint is therefore given by$$
where we have defined α = 3 H 2 Ω dm / 2. The covariance matrix for /vector ∇ Φ given the constraint is therefore given by
$$C \left ( \vec { \nabla } \Phi, \vec { \nabla } \Phi \right ) & = \frac { \alpha ^ { 2 } } { 3 } \left ( \sigma _ { - 1 } ^ { 2 } - \frac { \sigma _ { 0 } ^ { 4 } } { \sigma _ { 1 } ^ { 2 } } \right ) \, 1 _ { 3 \times 3 },$$
a result first derived in Ref. [20]. Even though at the location of the peak the mean shift of the gravitational force vanishes and, therefore, at the peak-by-peak level there is no extra force, its variance is not simply proportional to σ 2 -1 , but it receives a correction when the statistical ensemble average is taken. This effect can be simply captured by replacing the force /vector ∇ Φ experienced by the peaks by
$$\vec { \nabla } \Phi _ { \text{eff} } = \vec { \nabla } \Phi + \frac { 3 } { 2 } H ^ { 2 } \Omega _ { \text{dm} } R _ { v } ^ { 2 } \, \vec { \nabla } \delta _ { \text{dm} },$$
which is precisely the velocity bias relation proposed in [11-13] since the force is proportional to the velocity in linear theory. Indeed, the variance of this shifted quantity has zero mean and variance given by
$$\left \langle \left ( \vec { \nabla } \Phi _ { \text{eff} } \right ) ^ { 2 } \right \rangle = \frac { \alpha ^ { 2 } } { 3 } \left ( \sigma _ { - 1 } ^ { 2 } - \frac { \sigma _ { 0 } ^ { 4 } } { \sigma _ { 1 } ^ { 2 } } \right ),$$
which coincides with Eq. (III.5). Inserting this result into Eq. (II.5) for halos
$$\frac { \partial f _ { \text{h} } } { \partial t } + \vec { p } \cdot \frac { \partial f _ { \text{h} } } { \partial \vec { r } } - \vec { \nabla } \Phi _ { \text{eff} } \cdot \frac { \partial f _ { \text{h} } } { \partial \vec { p } } + \cdots = 0,$$
multiplying now Eq. (II.5) by /vector p , reinstating the expansion of the universe, integrating over the momenta and applying the gradient on both sides, we precisely get Eq. (I.5). Let us stress again that the effective force felt by the halos is of pure statistical origin, the definition of Φ eff being deduced from the covariance rather then from the average of /vector ∇ Φ eff , which is still vanishing (at linear order).
## IV. CONCLUSIONS AND FURTHER COMMENTS
In this short note, we have considered conservation laws in phase space to explain why the gravitational force acting on biased tracers of the large-scale structure is itself biased, as was already noted in Refs. [6, 13]. Consequently, the halo momentum conservation equation is modified accordingly, in agreement with the change advocated in Ref. [6]. This modification is responsible for the time constancy of the halo velocity bias (at the linear level).
We conclude by some remarks. Like in Refs. [6, 11-13], our derivation of the shift in the gravitational force felt by halos has made use of the properties of Gaussian random fields, so it applies at early time and on sufficiently large scales, when linear perturbation theory holds. It would be interesting to understand what happens when non-linearities become important. Along these lines, the work done in Ref. [21] might be useful. Furthermore, our derivation makes also clear that, when dealing with halos, the starting Vlasov-Boltzmann equation is highly non-trivial. Indeed, going to higher-order, not only the average force receives extra contribution, but also the equation (III.8) is modified into [19]
$$\frac { \partial f _ { \text{h} } } { \partial t } + \vec { p } \cdot \frac { \partial f _ { \text{h} } } { \partial \vec { r } } - \vec { \nabla } \Phi _ { \text{eff} } \cdot \frac { \partial f _ { \text{h} } } { \partial \vec { p } } = \vec { \nabla } _ { \vec { p } } \cdot \vec { F } _ { \text{h} },$$
where and
$$f _ { 2 c } ( \vec { r }, \vec { p }, \vec { r } ^ { \prime }, \vec { p } ^ { \prime }, t ) & = f _ { 2 } ( \vec { r }, \vec { p }, \vec { r } ^ { \prime }, \vec { p } ^ { \prime }, t ) - f ( \vec { r }, \vec { p }, t ) f ( \vec { r } ^ { \prime }, \vec { p } ^ { \prime }, t ) & & ( I V. 3 )$$
is the irreducible two-particle correlation function, which has to be computed with the necessary peak constraints. The latter cannot be neglected because it may be much larger than the product of the single-particle terms due to strong gravitational clustering. The vector /vector F h represents a correlated force density and arises from fluctuations in the ensemble averaged gravitational potential due to clustering of the matter distribution. This force density must be computed in the presence of the peak constraint, and generally gives rise to drift forces and diffusion in velocity-space. Added up to the fact that an initial constraint on the halo shifts the mean density and causes the density and velocity to be correlated hence, this changes the halo phase space density. Any theory of bias should therefore account of these effects. Work along these lines is in progress [22].
## Acknowledgments
The research of A.K. was implemented under the Aristeia Action of the Operational Programme Education and Lifelong Learning and is co-funded by the European Social Fund (ESF) and National Resources. A.K. is also partially supported by European Union's Seventh Framework Programme (FP7/2007-2013) under REA grant agreement n. 329083. A.R. is supported by the Swiss National Science Foundation (SNSF), project 'The non-Gaussian Universe' (project number: 200021140236). M.B. and V.D. acknowledge support by the Swiss National Science Foundation.
[1] N. Kaiser, Astrophys. J. Lett. 284 , L9 (1984)
- [2] D. Tseliakhovich, C.M. Hirata, Phys. Rev. D 82 , 083520 (2010), arXiv:1005.2416 [astro-ph.CO].
[3] J. Yoo, U. N. Dalal, U. Seljak, JCAP 1107 , 018 (2011) [arXiv:1105.3732 [astro-ph.CO]].
- [4] J.E. Gunn, J.R. Gott III, Astrophys. J. 176 , 1 (1972).
- [5] W.H. Press, P. Schechter, Astrophys. J. 187 , 425 (1974).
- [6] T. Baldauf, V. Desjacques and U. Seljak, arXiv:1405.5885 [astro-ph.CO].
- [7] E. Jennings, C.M. Baugh, D. Hatt, arXiv:1407.7296 [astro-ph.CO].
- [8] W.J. Percival, M. White, Mon. Not. R. Astron. Soc. 393 , 297 (2009) arXiv:0808.0003 [astro-ph.CO].
- [9] A. Elia, A.D. Ludlow, C. Porciani, Mon. Not. R. Astron. Soc. 421 , 3472 (2012) arXiv:1111.4211 [astro-ph.CO].
- [10] P. Zhang, Y. Zheng, Y. Jing, arXiv:1405.7125
$$\vec { F } _ { \mathrm h } ( \vec { r }, \vec { p }, t ) = C o v \left [ \vec { \nabla } \Phi _ { K } ( \vec { r }, t ), f _ { K } ( \vec { r }, \vec { p }, t ) \right ] = G _ { \mathrm N } \int d ^ { 3 } r ^ { \prime } \, d ^ { 3 } \vec { p } ^ { \prime } \, \frac { ( \vec { r } - \vec { r } ^ { \prime } ) } { | \vec { r } - \vec { r } ^ { \prime } | ^ { 3 } } f _ { 2 c } ( \vec { r }, \vec { p }, \vec { r } ^ { \prime }, \vec { p } ^ { \prime }, t ),$$
- [11] V. Desjacques, Phys. Rev. D 78 , 103503 (2008) [arXiv:0806.0007 [astro-ph.CO]].
- [12] V. Desjacques and R. K. Sheth, Phys. Rev. D 81 , 023526 (2010), arXiv:0909.4544 [astro-ph.CO].
- [13] V. Desjacques, M. Crocce, R. Scoccimarro, and R. K. Sheth, Phys. Rev. D 82 , 103529 (2010), arXiv:1009.3449 [astroph.CO].
- [14] K. C. Chan, R. Scoccimarro and R. K. Sheth, Phys. Rev. D 85 , 083509 (2012) [arXiv:1201.3614 [astro-ph.CO]].
- [15] J. N. Fry, Astrophys. J. 461 , L65 (1996).
- [16] M. Tegmark and P. J. E. Peebles, Astrophys. J. Lett. 500 , L79 (1998), astro-ph/9804067.
- [17] A. Elia, S. Kulkarni, C. Porciani, M. Pietroni, and S. Matarrese, Mon. Not. R. Astron. Soc. 416 , 1703 (2011), arXiv:1012.4833 [astro-ph.CO].
- [18] Yu. L. Klimontovich, The Statistical Theory of Non-Equilibrium Processes in a Plasma (MIT Press, Cambridge, 1967).
- [19] E. Bertschinger, in Cosmology and Large Scale Structure , proceedings of the 1996 Les Houches Summer School, Session LX, ed. R. Schaeffer, J. Silk, M. Spiro, and J. Zinn-Justin (Amsterdam: Elsevier Science), 273.
- [20] J. M. Bardeen, J. R. Bond, N. Kaiser and A. S. Szalay, Astrophys. J. 304 (1986) 15.
- [21] C. -P. Ma and E. Bertschinger, Astrophys. J. 612 , 28 (2004) [astro-ph/0311049].
- [22] M. Biagetti, V. Desjaques, A. Kehagias and A. Riotto, to appear. | 10.1103/PhysRevD.90.103529 | [
"Matteo Biagetti",
"Vincent Desjacques",
"Alex Kehagias",
"Antonio Riotto"
] | 2014-08-01T20:05:07+00:00 | 2014-11-26T14:33:14+00:00 | [
"astro-ph.CO",
"hep-th"
] | On the halo velocity bias | It has been recently shown that any halo velocity bias present in the initial
conditions does not decay to unity, in agreement with predictions from peak
theory. However, this is at odds with the standard formalism based on the
coupled fluids approximation for the coevolution of dark matter and halos.
Starting from conservation laws in phase space, we discuss why the fluid
momentum conservation equation for the biased tracers needs to be modified in
accordance with the change advocated in Baldauf, Desjacques & Seljak (2014).
Our findings indicate that a correct description of the halo properties should
properly take into account peak constraints when starting from the
Vlasov-Boltzmann equation. |
1408.0294v2 | ## LARGE DEVIATION UPPER BOUNDS FOR SUMS OF POSITIVELY ASSOCIATED INDICATORS
MATTHIAS L ¨ WE AND FRANCK VERMET O
Abstract. We give exponential upper bounds for P ( S ≤ k ), in particular P ( S = 0), where S is a sum of indicator random variables that are positively associated. These bounds allow, in particular, a comparison with the independent case. We give examples in which we compare with a famous exponential inequality for sums of correlated indicators, the Janson inequality. Here our bound sometimes proves to be superior to Janson's bound.
## 1. Introduction
Consider a family of indicator random variables ( X i ) i ∈ I with
$$\mathbb { P } ( X _ { i } = 1 ) = p _ { i } = 1 - \mathbb { P } ( X _ { i } = 0 ).$$
Let X = ∑ i ∈ I X i be their sum. Moreover, assume that the random variables are associated (or positively associated), i.e. for all coordinatewise increasing functions
$$f, g \, \colon \mathbb { R } ^ { | I | } \rightarrow \mathbb { R }$$
we have that
$$\text{Cov} ( f ( X _ { i }, i \in I ), g ( X _ { i }, i \in I ) ) \geq 0.$$
Association has first been defined and analyzed in [10]. Associated random variables occur in a variety of situations, e.g. in statistical mechanics in the context of the FKG inequalities, in subgraph count statistics in random graphs or in neural networks. For further application it is worth noting, of course, independent random variables are associated and that, if Y , . . . , Y 1 n are associated, and f i : R n → R is increasing for each i = 1 , . . . n , then also Z , . . . Z 1 n , with Z i = f i ( Y ) are associated (see Theorem 2.1 and (P4) in [10]). Here Y = ( Y , . . . , Y 1 n ). Actually this fact allows to construct and/or discover positive association in many examples.
Date : December 22, 2014.
2000 Mathematics Subject Classification. Primary: 60F10, Secondary: 60C05.
Key words and phrases. Large deviations, positive association, Poisson approximation, Janson inequality.
Example 1.1 (k-Runs) . Let Y , Y 1 2 , . . . Y n + k -1 be i.i.d. random variables with P ( Y i = 1) = p = 1 -P ( Y i = 0) and for k ∈ N let
$$Z _ { i } = \prod _ { j = i } ^ { i + k - 1 } Y _ { i } \\ 1 ^ { \prime } s \text{ starting in } i.$$
be the indicator for a k -run of 1's starting in i. Then, as the Z i are increasing functions of the independent and thus associated random variables Y , . . . , Y 1 n , they are associated. k -runs are frequently used e.g. in sequence alignments in mathematical biology. We will be interested in the total number Z of k -runs in a Bernoulli sequence of a given length.
Example 1.2 (Subgraph-counts) . Let ( Y i,j ) 1 ≤ i<j ≤ n be the i.i.d. Bernoulli random variables occurring in the Erd¨s-R´nyi random graph o e G n, p ( ), i.e. Y i,j = 1, if and only if the edge between i and j is present in the realization of the random graph, and this occurs with probability p . Otherwise Y i,j is 0. Now let K k be an arbitrary fixed complete graph, e.g. a triangle for which k=3. The random variables
$$Z _ { s _ { 1 }, s _ { 2 }, \dots s _ { k } } = \prod _ { 1 \leq i < j \leq k } Y _ { s _ { i }, s _ { j } } \\ o. \, a r o \, n o i r r i c o \, d i f f o r n t \, \text{and} \, \int o.$$
are associated. Here the s i are pairwise different and { s , s 1 2 , . . . s k } ⊆ { 1 , . . . , n } . The Z s 1 ,s 2 ,...s k indicate whether the realized graph contains a K k at the vertices s , . . . , s 1 k . We will be interested in the number of K k 's in a realization of G n, p ( ), thus in the sum of the Z s 1 ,s 2 ,...s k , i.e. Z = ∑ 1 ≤ s 1 <...<s k ≤ n Z s 1 ,...,s k .
$$Z _ { i _ { 1 }, \dots, i _ { k } } = \prod _ { j = 1 } ^ { k } Y _ { i _ { j } } \\ \text{associated. We will be}$$
Example 1.3 (U-statistics) . Let Y , Y , . . . Y 1 2 n be i.i.d. random variables with P ( Y i = 1) = p = 1 -P ( Y i = 0) and for k ∈ N and 1 ≤ i 1 , . . . , i k ≤ n pairwise different let
Then again the Z i 1 ,...,i k are associated. We will be interested in the U-statistics Z = ∑ 1 ≤ i 1 <...<i k ≤ n Z i 1 ,...,i k . Note that the corresponding quantities in Examples 1.1 and 1.2 are structurally similar. However, there Z are so-called incomplete U-statistics, i.e. not all possible random variables Z i 1 ,...,i k are actually considered (in Example 1.2, e.g., one requires that the vertices form a clique, rather than just multiplying any collection of k edges). This leads to 'more independence' among the summands. We will see in Section 3 that the bound we give works particularly well for random variables with many but weak correlations.
Example 1.4 (Random hypergraphs) . The following example has several motivations. On the one hand, it is the most generalization of the famous random graph models G N,p ( ) and G N,M ( ) invented by Erd¨ os and R´ enyi ([9]). Here edges are realized between N points i.i.d. with probability p ( G N,p ( )), or we take one of the graphs with exactly M edges at random with equal probability
( G N,M ( )). These models are readily generalized to hypergraphs in the following way. Given a set of hyperedges, e.g. all possible complete graphs with k vertices that can be constructed on the vertices { 1 , . . . , N } we realize each of them independently at random with probability p or we realize a hypergraph with exactly M such hyperedges at random with equal probability. Such random hypergraphs have been studied e.g. in [7], [3], [4], or [16].
On the other hand, the example can be considered as a mean-field version of the very recent topic of loop-percolation (see e.g. [23]): We cover a graph with a random ensemble of subgraphs and ask for the properties of the thus emerging graph. Finally, our example is also motivated by the analysis of a certain neural network (see [1], [11]). However, it may be regarded as an interesting combinatorial game with a certain similarity to group testing. Take the complete graph K N and in each of a finite number of steps delete the edges of a randomly chosen complete subgraph K k ( k fixed), if they are not already deleted. Let Y i,j denote the indicator of the event, that edge { i, j } has been deleted after n steps. Then the ( Y i,j ) are associated (and so are the ( Z i,j ) = (1 -Y i,j )). We will be interested in the probability that the random K k 's cover all of K N , hence that the random hypergraph we realize is connected.
Other examples for association have been found it statistical mechanics, e.g. in models obeying the FKG inequalities ([17]) or recently in the fuzzy fractional Potts model [14] or in the parabolic Anderson model [15].
Due to their frequent occurrence, deviation inequalities for positively associated random variables have been investigated in a number of recent articles. Already Newman and Wright prove an invariance principle [18] and martingale inequalities [19]. Cox and Grimmett show a Central Limit Theorem [8]. Boutsikas and Koutras present a simple upper bound for the distance between the distribution of the sum of n associated random variables X i and a sum of n independent random variables with the same marginals as X i [6]. Exponential inequalities (also partially together with a resulting Strong Law of Large Numbers) were derived by Oliviera [20], [25], [26], and Yang and Chen [28]. Finally a Law of the Iterated Logarithm was proved by Xing and Yang [27]. A very readable survey over these and more results can be found in [21].
The purpose of the present note is to give a new exponential bound for the probability that a sum of positively associated indicator variables is particularly small or even 0. These bounds will also allow for a comparison with the case of sums of independent indicators with the same distribution. Such a comparison is of interest, since often when positively associated random variables are applied in the literature, it is tacitly assumed that the probabilities of their sum being 0 is well approximated by the independent case (see e.g [1]). Moreover, by positive
association, we have that
$$\mathbb { P } ( \sum _ { i } X _ { i } = 0 ) \geq \prod _ { i } \mathbb { P } ( X _ { i } = 0 ) \\ \text{he product being taken over the same set}$$
(with the sum and the product being taken over the same set of indices), anyway. Thus we always have an obvious bound in one direction by the independent case. One way to obtain the comparison between the dependent and the independent situation is Janson's inequality ([12], [13]). This can be stated as follows: In the above setting, let I be index set of the associated variables ( X i ) i ∈ I
$$\stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \text{$\dots } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim \stackrel { \dots } { \Delta } \sim $ \dots }$$
/negationslash
$$\mathfrak { L } = \frac { 1 } { 2 } \sum _ { \stackrel { i \sim j, } { i \neq j } } \mathbb { E } ( X _ { i } X _ { j } ) \quad \overline { \Delta } = \lambda + 2 \Delta, \\ \ell = \frac { 1 } { 2 } \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \collections \quad........................................................................................................................................................................................................$$
/negationslash where we write i ∼ j if and only if Cov( X , X i j ) = 0. Then the following bounds hold.
Theorem 1.5. ( [12] , [13] , also see [24] ) We have
$$\mathbb { P } ( X = 0 ) \leq e ^ { - \lambda + \Delta }$$
and
$$( 1. 1 )$$
$$\mathbb { P } ( X = 0 ) \leq e ^ { - \frac { \lambda } { \overline { \partial } ^ { 2 } } }.$$
Moreover,
$$\mathbb { P } ( X = 0 ) \leq \exp \left ( \frac { \Delta } { 1 - \max _ { i \in I } E X _ { i } } \right ) \prod _ { i \in I } ( 1 - \mathbb { E } X _ { i } ). \quad \ ( 1. 3 ) \\ \text{the runpose of the resent note. we will be most of all interested in comparing}$$
For the purpose of the present note, we will be most of all interested in comparing our results with (1.3), which can be derived from (1.1), but was first proved by Boppona and Spencer [5].
Another way to compare the dependent and the independent cases is the following result, by Boutsikas and Koutras.
Theorem 1.6. ( [6] ) We have
$$\Psi [ X = 0 ] \leq \prod _ { i = 1 } ^ { n } \mathbb { P } [ X _ { i } = 0 ] + \sum _ { i < j } \text{Cov} ( X _ { i }, X _ { j } ). \quad \text{(1.4)} \\ \text{in theorem also gives an unner bound for the probability that the sum of}$$
Our main theorem also gives an upper bound for the probability that the sum of positively or negatively associated indicator random variables is zero. It can be stated as follows.
Theorem 1.7. Let the X i be positively associated indicators. With the above notation we have for all t > 0
$$\mathbb { P } ( X = 0 ) \leq e ^ { - t | I | } \left ( \prod _ { i \in I } \mathbb { E } [ e ^ { t ( 1 - X _ { i } ) } ] + t ^ { 2 } e ^ { t | I | } \sum _ { i < j } \text{Cov} ( X _ { i }, X _ { j } ) \right ) \quad ( 1. 5 )$$
If the ( X i ) are identically distributed with P ( X i = 1) = p = 1 -P ( X i = 0) this boils down to
$$\mathbb { P } ( X = 0 ) \ & \leq \ e ^ { - t | I | } ( p + ( 1 - p ) e ^ { t } ) ^ { | I | } + t ^ { 2 } \sum _ { i < j } \text{Cov} ( X _ { i }, X _ { j } ) \\ & = \ ( 1 - p ) ^ { | I | } ( 1 + e ^ { - t } \frac { p } { 1 - p } ) ^ { | I | } + t ^ { 2 } \sum _ { i < j } \text{Cov} ( X _ { i }, X _ { j } ) \quad \ ( 1. 6 ) \\ \mathbb { m } \quad \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot.$$
Remarks 1.8. (1) It is obvious that Theorem 1.5, Theorem 1.6 as well as our Theorem 1.7 are tailor-made for the situation of weak correlations, where one is close to the independent situation. If correlations are getting large, then they give useless bounds: The probabilities are bounded from above by numbers larger than 1. An important difference between (1.4) and (1.5) is that we have a factor t 2 in front of the sum of the covariances. As we will see in Section 3, choosing t small, this factor can be decisive when the covariances are too large.
- (2) A quick glance at (1.3) and (1.6) already reveals the major difference between the two bounds: While (1.3) is purely multiplicative, our bound (1.6) also has an additive component. This has a consequence for the applicability of our bounds. Consider e.g. the situation for (1.6): In case that P ( X = 0) is comparable to the independent approximation (1 -p ) | I | and p is much larger than 1 / I | | , both probabilities will typically go to 0 and they will do so at an exponential rate. To have the additive term t 2 ∑ i<j Cov( X , X i j ) in (1.6) smaller than the probability we want to approximate, we therefore need to choose t extremely small, which in turn means that the multiplicative correction in (1 + e -t p 1 -p ) | I | is of order e p I | | which typically is only effective, if p is of order 1 / I | | . Hence in such cases, we expect that (1.3) will yield results that are more useful than (1.6).
We will prove Theorem 1.7 in the following section. Section 3 will be devoted to comparing Theorems 1.5 and 1.7 in the examples mentioned above. We will see that our bound sometimes is better than Janson's inequality and also better than the bound given in Theorem 1.6.
## 2. Proof of Theorem 1.7
As usual the central ingredient of the proof of an exponential inequality as in Theorem 1.7 is an estimate of the moment generating function. Such bound can be found in [22], Lemma 3.1. For the sake of completeness (and also as the original proof is somewhat short), we reprove the statement here.
Lemma 2.1. (see [22] , Lemma 3.1) Let X , X , . . . , X 1 2 n be positively associated random variables, such that X i ≤ κ for all i and some κ > 0 . Then for any t > 0
$$\left | \mathbb { E } \left [ e ^ { t \sum _ { i = 1 } ^ { n } X _ { i } } \right ] - \prod _ { i = 1 } ^ { n } \mathbb { E } \left [ e ^ { t X _ { i } } \right ] \right | & \leq t ^ { 2 } e ^ { n t \kappa } \sum _ { 1 \leq i < j \leq n } \text{Cov} ( X _ { i }, X _ { j } ). \\ \text{for}. \text{The key tool is to use Lemma $3$ in $[17]$. $ This Lemma implies that $fc$}$$
∣ ∣ Proof. The key tool is to use Lemma 3 in [17]. This Lemma implies that for associated random variables X and Y with finite variance, and (real or complex valued) functions f and g with bounded derivatives one has
$$| \text{Cov} ( f ( X ), g ( Y ) | \leq | | f ^ { \prime } | | \, | | g ^ { \prime } | | \text{Cov} ( X, Y )$$
where || · || is the sup-norm. Applying this we obtain due to the boundedness of the the X i
$$| \text{Cov} ( e ^ { t X _ { 1 } }, e ^ { t X _ { 2 } } ) | \leq t ^ { 2 } e ^ { 2 t \kappa } \text{Cov} ( X _ { 1 }, X _ { 2 } )$$
which is precisely the assertion for n = 2. The result now follows by induction: According to our earlier remark, if X , X , . . . , X 1 2 n are associated random variables, then so are X 1 + X 2 + . . . , X n -1 and X n . Thus , if the assertion has already been proven for n -1 we obtain
$$\text{even for $n-1$ we obtain} \\ \left | \mathbb { E } \left [ e ^ { t \sum _ { i = 1 } ^ { n } X _ { i } } \right ] - \prod _ { i = 1 } ^ { n } \mathbb { E } \left [ e ^ { t X _ { i } } \right ] \right | \\ \leq \left | \mathbb { E } \left [ e ^ { t \sum _ { i = 1 } ^ { n } X _ { i } } \right ] - \mathbb { E } \left [ e ^ { t \sum _ { i = 1 } ^ { n - 1 } X _ { i } } \right ] \mathbb { E } \left [ e ^ { t X _ { n } } \right ] \right | + \left | \mathbb { E } \left [ e ^ { t \sum _ { i = 1 } ^ { n - 1 } X _ { i } } \right ] \mathbb { E } \left [ e ^ { t X _ { n } } \right ] - \prod _ { i = 1 } ^ { n } \mathbb { E } \left [ e ^ { t X _ { i } } \right ] \right | \\ \leq \ t ^ { 2 } e ^ { n t k } \sum _ { 1 \leq i \leq n - 1 } \text{Cov} ( X _ { i }, X _ { n } ) + t ^ { 2 } e ^ { n t k } \sum _ { 1 \leq i < j \leq n - 1 } \text{Cov} ( X _ { i }, X _ { j } ) \\ \leq \ t ^ { 2 } e ^ { n t k } \sum _ { 1 \leq i < j \leq n } \text{Cov} ( X _ { i }, X _ { j } ), \\ \text{where in $/2\}$} \text{$the first command is bound by another annotation of $I$} \text{amma}$$
$$\sum _ { \leq i < j \leq }$$
where in (2.1) the first summand is bounded by another application of Lemma 3 in [17] inequality, while the bound on the second summand follows from the induction hypothesis. /square
From this bound the proof of the main theorem follows quickly:
Proof of Theorem 1.7. With the notation of the theorem let Y i = 1 -X i . Note that together with the ( X i ) also the ( Y i ) are associated. Using an exponential
Markov inequality and the previous lemma we obtain for all t > 0
$$\text{markov inequality and the previous remain we over all } t > 0 \\ \mathbb { P } ( X = 0 ) \ & = \ \mathbb { P } ( \sum _ { i \in I } X _ { i } = 0 ) = \mathbb { P } ( \sum _ { i \in I } Y _ { i } = | I | ) = \mathbb { P } ( \sum _ { i \in I } Y _ { i } \geq | I | ) \\ & \leq \ e ^ { - t | I | } \mathbb { E } [ e ^ { t } \sum _ { i \in I } Y _ { i } ] \\ & \leq \ e ^ { - t | I | } \left ( \prod _ { i \in I } \mathbb { E } [ e ^ { t Y _ { i } } ] + t ^ { 2 } e ^ { t | I | } \sum _ { i < j } \text{Cov} ( Y _ { i }, Y _ { j } ) \right ) \\ & = \ e ^ { - t | I | } \left ( \prod _ { i \in I } \mathbb { E } [ e ^ { t ( 1 - X _ { i } ) } ] + t ^ { 2 } e ^ { t | I | } \sum _ { i < j } \text{Cov} ( X _ { i }, X _ { j } ) \right ) \\ \\ \cdots \cdots \, / \, \text{$\cdots$}$$
which is (1.5).
Computing the expectation in the product for identically distributed random indicators with success probability p we arrive at
$$\text{maxors with success } p v o a n i n y \, p \, \text{ we all use } \, \\ \mathbb { P } ( X = 0 ) \ & \leq \ e ^ { - t | I | } \left ( \prod _ { i \in I } ( p + ( 1 - p ) e ^ { t } ) + t ^ { 2 } e ^ { t | I | } \sum _ { i < j } C o v ( X _ { i }, X _ { j } ) \right ) \\ & = \ e ^ { - t | I | } ( p + ( 1 - p ) e ^ { t } ) ^ { | I | } + t ^ { 2 } \sum _ { i < j } C o v ( X _ { i }, X _ { j } ) \\ & = \ ( 1 - p ) ^ { | I | } ( 1 + \frac { p } { 1 - p } e ^ { - t } ) ^ { | I | } + t ^ { 2 } \sum _ { i < j } C o v ( X _ { i }, X _ { j } ) \\ \text{which is } ( 1. 6 ). & \quad \Box$$
which is (1.6).
## 3. Examples
In this section we will discuss the quality of our estimate in the examples given in Section 1. The aim is to see, if and when Theorem 1.7 is better than Theorem 1.5, especially, when (1.6) is more powerful than (1.3), and to gain a general understanding, in which situations it is advisable to apply Theorem 1.7.
We start with
Example 3.1 (k-Runs, Example 1.1 continued) . To keep the situation symmetric and simplify computations, we slightly change the setup. So let Y , Y , . . . Y 1 2 n be i.i.d. random variables with P ( Y i = 1) = p = 1 -P ( Y i = 0) and for k ∈ N let
$$Z _ { i } = \prod _ { j = i } ^ { i + k - 1 } Y _ { i }$$
(where i + k -1 now is to be taken modulo n ). Moreover let Z = ∑ n i =1 Z i . Computing the central quantity in Janson's inequality we obtain
/negationslash
$$\text{trait quamify in Jansons s inequality we odd} \\ \Delta \ & = \ \frac { n } { 2 } \sum _ { \substack { 1 \sim j, \\ 1 \neq j } } \mathbb { E } Z _ { 1 } Z _ { j } = \frac { n } { 2 } \sum _ { j = 1 } ^ { k - 1 } p ^ { k + j } \\ & = \ \frac { n } { 2 } p ^ { k + 1 } \frac { 1 - p ^ { k - 1 } } { 1 - p }$$
Thus (1.3) yields
$$\Phi ( Z = 0 ) \leq ( 1 - p ^ { k } ) ^ { n } \exp \left ( \frac { n p ^ { k + 1 } ( 1 - p ^ { k - 1 } ) } { 2 ( 1 - p ) ( 1 - p ^ { k } ) } \right ). \quad ( 3. 1 ) \\ \intertext { t that more generally the total variation distance between the distribution }$$
(Note that more generally the total variation distance between the distribution of Z and a Poisson distribution was bounded in [2], Theorem 8.F.) This bound becomes effective, i.e. we have that P ( Z = 0) is well approximated by (1 -p k ) n , if p /lessmuch n -1 /k +1 , i.e. if np k +1 → 0. On the other hand our Theorem 1.7 yields for any t > 0
$$\ a u y \ t \succ \Omega \\ \mathbb { P } ( Z = 0 ) \ \leq \ ( 1 - p ^ { k } ) ^ { n } ( 1 + e ^ { - t } \frac { p ^ { k } } { 1 - p ^ { k } } ) ^ { n } + t ^ { 2 } \sum _ { i < j } \text{Cov} ( Z _ { i }, Z _ { j } ) \\ \leq \ ( 1 - p ^ { k } ) ^ { n } ( 1 + e ^ { - t } \frac { p ^ { k } } { 1 - p ^ { k } } ) ^ { n } + t ^ { 2 } \Delta \\ \text{choose \infty } \, \partial \, \ast \, \Lambda ^ { - 1 / 2 } \, \text{th} \, \text{coord} \, \text{command} \, \text{on} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \,\text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th}" \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \text{th} \, \end{cases}$$
Choosing 0 < t /lessmuch ∆ -1 2 / the second summand on the right hand side converges to 0. Then the correction factor (1+ e -t p k 1 -p k ) n for the first summand on the right p k hand side is bounded by e n 1 -p k and particularly converges to 1, if p /lessmuch n -1 /k , which improves the bound from Janson's inequality. Note however that even for much larger p an exponential bound on P ( Z = 0) holds true. As a matter of fact Barbour et al. in [2], Theorem 8.G give the bound
$$| \mathbb { P } ( Z = 0 ) - \exp ( - n ( 1 - p ) p ^ { k } ) | \leq ( 2 k ( 1 - p ) + 1 ) p ^ { k }$$
The recipe seen in the previous example illustrates the general method. In (1.6) one always chooses t (depending on n ) so small that the second summand on the right vanishes, and then we obtain a bound of the form (1 -p ) n e n p 1 -p , where p is the success probability of the random variables in question. This works, if p is so small, that (1 -p ) n is of constant order. However, even if both (1 -p ) n and P ( Z = 0) tend to 0, we may choose t so small, that the second summand in (1.6) vanishes, but we may not be able to guarantee that then p is so small, that e n p 1 -p converges to 1.
Example 3.2 (Subgraph-counts, Example 1.2 continued) . In the situation of Example 1.2 consider the number of triangles in the random graph G n, p ( ). Thus we consider
$$Z _ { s _ { 1 }, s _ { 2 }, s _ { 3 } } = Y _ { s _ { 1 }, s _ { 2 } } Y _ { s _ { 2 }, s _ { 3 } } Y _ { s _ { 3 }, s _ { 1 } }$$
and
$$Z = \sum _ { 1 \leq s _ { 1 } < s _ { 2 } < s _ { 3 } \leq n } Z _ { s _ { 1 }, s _ { 2 }, s _ { 3 } }. \text{
itv that the random graph does n}$$
We bound the probability that the random graph does not contain any triangle, hence that Z = 0. The quantity ∆ in (1.3) can this time be computed as
$$\Delta = \frac { 1 } { 2 } \binom { n } { 3 } 3 n p ^ { 5 } \sim \frac { 1 } { 4 } n ^ { 4 } p ^ { 5 }, \\ \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \ dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots\dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \delta \dots \delta \dots \delta \dots \delta \dots \delta \dots \delta \dots \delta \dots \delta \dots \delta \dots \delta \dots \delta \dots \delta \dots \delta \dots \delta \dots \delta \dots \delta \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \Dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots$$
$$\mathbb { P } ( Z = 0 ) \leq ( 1 - p ^ { 3 } ) ^ { \binom { n } { 3 } } \exp \left ( \frac { n ^ { 4 } p ^ { 5 } } { 4 ( 1 - p ^ { 3 } ) } \right ). \\ \intertext { a t i n d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d o d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d s t }$$
as there are ( ) n 3 triangles, each of which is correlated with 3 n other triangles, and in this case E Z s 1 ,s 2 ,s 3 Z s 4 ,s 5 ,s 6 = p . 5 Thus Janson's inequality gives
If p /lessmuch n -4 5 / , this bound is effective, since the factor is tending to 1. This time our bound works well only in some range of the parameters and for slightly different reasons than in the previous example. From (1.6) we obtain for any t > 0
$$\mathbb { P } ( Z = 0 ) \leq ( 1 - p ^ { 3 } ) ^ { \binom { n } { 3 } } \left ( 1 + e ^ { - t } \frac { p ^ { 3 } } { 1 - p ^ { 3 } } \right ) ^ { \binom { n } { 3 } } + t ^ { 2 } n ^ { 4 } p ^ { 5 }. \\ \intertext { o r a r i m e w h o r a n } \, \text{where $n^{-1}\mathbb{ }n$ was a case of $avc$} \, \text{with the problem mentioned at the} }$$
may obtain a bound of the form (1 -p 3 ) ( n 3 ) ( 1 + p 3 1 -p 3 ) ( 3 ) by choosing t extremely small. This is however useless, as the second factor on the right diverges.
In the regime, where n -1 /lessmuch p we see exactly the problem mentioned at the end of the introduction occurring: The probabilities to be approximated are of order exp( -p n 3 3 ) (at least for p going to 0 with n to infinity) and thus converging to 0 exponentially fast. So even if p /lessmuch n -4 5 / , in which case the second term on the right goes to 0, we need to choose t to go to 0 as well, and we cannot improve the first summand in our bound by choosing t large. If now n -1 /lessmuch p /lessmuch n -4 5 / we n
So let us assume that p = λ/n . Choose t = n α for 0 < α < 1 2 . Then (1.6) becomes
$$\Omega \infty \, \partial \, \partial \, \max \, \Pi = \lambda \, \infty \, \cos \, \iota = \eta \, \infty \, \cos \, 2 \, \Pi \, \infty \, \cos \, \\ \mathbb { P } ( Z = 0 ) \ \leq \ ( 1 - p ^ { 3 } ) ^ { \binom { n } { 3 } } \left ( 1 + e ^ { - n ^ { \alpha } } \frac { p ^ { 3 } } { 1 - p ^ { 3 } } \right ) ^ { \binom { n } { 3 } } + \frac { \lambda ^ { 5 } } { n ^ { 1 - 2 \alpha } } \\ = \ ( 1 - p ^ { 3 } ) ^ { \binom { n } { 3 } } \left ( 1 + e ^ { - n ^ { \alpha } } \frac { p ^ { 3 } } { 1 - p ^ { 3 } } \right ) ^ { \binom { n } { 3 } } + o ( 1 ). \\ \Big. \ \left. \ \left. \ \right. \ \right. \ \right. \ \right. \ \right. \ \right. \ \right. \ \right. \ \right. \ \right. \ \right. \ \right. \ \right. \ \right. \ \right.$$
The multiplicative correction ( 1 + e -n α p 3 1 -p 3 ) ( n 3 ) ≤ exp( λ e 3 -n α ) converges to 1 much faster than exp( 1 n ), the correction from (1.3).
Example 3.3 (U-statistics, Example 1.3 continued) . In the situation of Example 1.3, we define
$$Z = \sum _ { i = \{ i _ { 1 }, \dots, i _ { k } \} \subset \{ 1, \dots, n \} } Z _ { i }.$$
We first compute the central quantity in (1.3):
$$\Delta = \frac { 1 } { 2 } \binom { n } { k } \sum _ { j = 1 } ^ { k - 1 } \binom { n - k } { j } p ^ { k + j } \sim C \frac { 1 } { k! } ( n p ) ^ { k + 1 } \frac { 1 - ( n p ) ^ { k - 1 } } { 1 - n p }, \\. \. \. \. \. \. \. \. \.$$
with C ∈ ]0 , 1[. Thus (1.3) yields
$$\mathbb { P } ( Z = 0 ) \leq ( 1 - p ^ { k } ) ^ { \binom { n } { k } } \exp \left ( C \frac { ( n p ) ^ { k + 1 } ( 1 - ( n p ) ^ { k - 1 } ) } { k! ( 1 - n p ) ( 1 - p ^ { k } ) } \right ) \quad \ \ ( 3. 3 ) \\ \dots \dots \dots \dots \varepsilon \varepsilon \varepsilon \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \vdots \end{array}$$
for some constant C , and it is obvious that this bound is only effective, if p = λ n , for some λ > 0 or even p = (1 o /n ). This is also the situation where Theorem 1.7 gives good bounds. If p = λ/n , we choose t /lessmuch e -λ / k k 2 ! / ∆ to guarantee that the second summand in (1.6) is much smaller than the first. Then, for large λ , (1.6) basically boils down to
$$\mathbb { P } ( Z = 0 ) \leq ( 1 - p ^ { k } ) ^ { \binom { n } { k } } \left ( 1 + \frac { p ^ { k } } { 1 - p ^ { k } } \right ) ^ { \binom { n } { k } } + o ( 1 ). \\ \text{still hot for than the bound } ( 2 \, 3 ) \, \text{ obtained from } ( 1 \, 3 \, 3 \, \text{$$
This is still better than the bound (3.3) obtained from (1.3), since our multiplicative error factor behaves like e λ /k k ! , while the one in (3.3) is essentially e λ 2 k /k ! . However, if λ is getting small, our bound gets even better (compared to (3.3)), due to the additional factor e -t (with t chosen as above) in the multiplicative error term. This especially applies when λ → 0 with n going to infinity.
In the previous examples we always estimated the sum of the covariances occurring in (1.6) by the ∆ from Janson's inequality: One may wonder whether we haven't been giving away too much by such a bound. The answer is no, since in these cases the covariances of the variables are dominated by the expectations of the product. However, this need not be the case in general and it will turn out in such situations our bound is significantly better than Janson's inequality.
Example 3.4 (Random hypergraphs, Example 1.4 continued) . Here we will see, that our inequality is able to outperform Janson's inequality in certain situations. To this end, in the situation of Example 1.4 let Y be the event that all edges of K N are covered by one of the randomly chosen K k 's. With the notation of Example 1.4
$$\mathfrak { J } = \left \{ \sum _ { 1 \leq i < j \leq N } Y _ { i, j } = \binom { N } { 2 } \right \} = \left \{ Z \coloneqq \sum _ { 1 \leq i < j \leq N } Z _ { i, j } = 0 \right \}. \\ \text{Ve want to estimate the probability of } \mathfrak { N } \text{ for large $n$ and fixed $k$ to this en}$$
We want to estimate the probability of Y for large n and fixed k . To this end assume we select n K k 's as subgraphs of K N i.i.d. at random. Then for any edge ( i, j )
$$\mathbb { P } ( Z _ { i, j } = 1 ) = \left ( 1 - \frac { \binom { N - 2 } { k - 2 } } { \binom { N } { k } } \right ) ^ { n }.$$
Indeed, Z i,j = 1, if and only if ( i, j ) is not contained in any of the n independent K k 's and this happens with probability ( 1 -( N -2 k -2 ) ( N k ) ) n =: p N, n, k ( ) =: p . For N large, and k fixed,
$$p & \sim \left ( 1 - \frac { k ( k - 1 ) } { N ^ { 2 } } \right ) ^ { n }. \\ \text{rem 1.5 observe that indeed no}$$
To compute ∆ from Theorem 1.5 observe that indeed none of the Z i,j are uncorrelated, but that their correlation is extremely weak. As a matter of fact, there are two different types of correlations between Z i,j and Z i ,j ′ ′ depending on whether the edges ( i, j ) and ( i , j ′ ′ ) share a vertex or not. Assume ( i, j ) and ( i , j ′ ′ ) share a vertex, e.g. j = j ′ . Then with Y r i,j denoting the indicator that ( i, j ) is covered by the r 'th independent copy of K k
$$\mathbb { e } r \text{th independent copy of } K _ { k } \\ \mathbb { P } ( Z _ { i, j } Z _ { i ^ { \prime }, j } = 1 ) \ = \ \mathbb { P } ( Y _ { i, j } + Y _ { i ^ { \prime }, j } = 0 ) \\ = \ \mathbb { P } \left ( \bigcap _ { r = 1 } ^ { n } \{ Y _ { i, j } ^ { r } + Y _ { i ^ { \prime }, j } ^ { r } = 0 \} \right ) \\ = \ \left ( \frac { \binom { N - 3 } { k } } { \binom { N } { k } } + 3 \frac { \binom { N - 3 } { k - 1 } } { \binom { N } { k } } + \frac { \binom { N - 3 } { k - 2 } } { \binom { N } { k } } \right ) ^ { n }, \\ \text{ifrent terms coming from the five possible cases:} \\ \bullet \text{ choose $k$ indices in } \{ 1, \dots, N \} \, \langle \{ i, i ^ { \prime }, j \},$$
the different terms coming from the five possible cases:
- · choose k indices in { 1 , . . . , N } \ { i, i , j ′ } ,
- · choose i and then k -1 indices in { 1 , . . . , N } \ { i, i , j ′ } ,
- · choose i ′ and then k -1 indices in { 1 , . . . , N } \ { i, i , j ′ } ,
- · choose j and then k -1 indices in { 1 , . . . , N } \ { i, i , j ′ } ,
- · choose i and i ′ and then k -2 indices in { 1 , . . . , N } \ { i, i , j ′ } .
In the same way, if ( i, j ) and ( i , j ′ ′ ) do not share a vertex, we have
Thus
$$\mathbb { e } \, \text{same way, if } ( \mathfrak { l }, \mathfrak { l } ) \, \text{and } ( \mathfrak { l }, \mathfrak { l } ) \, \text{go not sane a vertex, we have} \\ \mathbb { P } ( Z _ { i, j } Z _ { i ^ { \prime }, j ^ { \prime } } = 1 ) \ = \ \mathbb { P } \left ( \bigcap _ { r = 1 } ^ { n } \{ Y _ { i, j } ^ { r } + Y _ { i ^ { \prime }, j ^ { \prime } } ^ { r } = 0 \} \right ) \\ = \ \left ( \frac { \binom { N - 4 } { k } } { \binom { N } { k } } + 4 \frac { \binom { N - 4 } { k - 1 } } { \binom { N } { k } } + 4 \frac { \binom { N - 4 } { k - 2 } } { \binom { N } { k } } \right ) ^ { n } \\ \mathfrak { ; } \\ \hat { \nabla } \quad 1 \quad \bigcap \quad \mathbb { m } / \mathfrak { r } \quad \mathfrak { r } \quad \mathfrak { \dots }$$
/negationslash
$$\begin{array} { r c l } \Delta & = & \frac { 1 } { 2 } \sum _ { ( i, j ), ( i ^ { \prime }, j ^ { \prime } ), } \mathbb { E } ( Z _ { i, j } Z _ { i ^ { \prime }, j ^ { \prime } } ) \\ & = & \binom { N } { 2 } \left ( ( N - 2 ) \left ( \frac { ( N - 3 ) } { ( N ^ { \prime } ) } + 3 \frac { \binom { N - 3 } { k - 1 } } { ( N ^ { \prime } ) } + \frac { \binom { N - 3 } { k - 2 } } { ( N ^ { \prime } ) } \right ) ^ { n } \\ & & + \frac { 1 } { 2 } \binom { N - 2 } { 2 } \left ( \frac { ( N - 4 ) } { ( N ^ { \prime } ) } + 4 \frac { \binom { N - 4 } { k - 1 } } { ( N ^ { \prime } ) } + 4 \frac { \binom { N - 4 } { k - 2 } } { ( N ^ { \prime } ) } \right ) ^ { n } \end{array} \right ) \end{array}$$
Now suppose n = λN 2 for some λ > 0 and that N is large. Observe that for n smaller than this, we cannot hope for any kind of Poisson approximation for the probability in question, since we simply do not have enough edges from the K k 's to cover K N . For n as large as n = λN 2 , or larger, the Poisson approximation becomes more likely, since then the probability not to cover an edge, becomes small.
We will now expand the exponentials in the computation of ∆. Let us start with k = 3, such that only the first of the two summands for ∆ is present. We obtain:
$$\xi, \cfrac { \binom { N - 3 } { k } } { \binom { N } { k } } \ & = \ ( 1 - \frac { k } { N } ) ( 1 - \frac { k + 1 } { N } ) ( 1 - \frac { k + 2 } { N } ) \frac { N } { N - 1 } \frac { N } { N - 2 } \\ & = \ 1 - 3 \frac { k } { N } + \frac { 1 } { N ^ { 2 } } ( 3 k ^ { 2 } - 3 k ) + O ( \frac { 1 } { N ^ { 3 } } ) )$$
and
$$\frac { \binom { N - 3 } { k - 1 } } { \binom { N } { k } } & \ = \ \frac { ( N - k ) ( N - k - 1 ) k } { N ( N - 1 ) ( N - 2 ) } \\ & = \ ( 1 - \frac { k } { N } ) ( 1 - \frac { k + 1 } { N } ) \frac { k } { N } ( 1 + \frac { 1 } { N } + O ( \frac { 1 } { N ^ { 2 } } ) ) ( 1 + \frac { 2 } { N } + O ( \frac { 1 } { N ^ { 2 } } ) ) \\ & = \ \frac { k } { N } ( 1 - 2 \frac { k - 1 } { N } + O ( \frac { 1 } { N ^ { 2 } } ) )$$
as well as
$$& \frac { \binom { N - 3 } { k - 2 } } { \binom { N } { k } } = \frac { ( N - 3 )! } { N! } \frac { ( N - k )! } { ( N - k - 1 )! } \frac { k! } { ( k - 2 )! } = \frac { ( N - k ) k ( k - 1 ) } { N ( N - 1 ) ( N - 2 ) } = \frac { k ( k - 1 ) } { N ^ { 2 } } + O ( \frac { 1 } { N ^ { 3 } } ). \\ & \text{Thus} \\ & \quad \, \int _ { N = 0 } \in \frac { N - 2 } { N - 2 } \in \lambda \ \lambda _ { N = 0 } \in \lambda _ { N = 0 } \ \lambda _ { N = 0 } \ \lambda _ { N = 0 } \ \lambda _ { N = 0 } \ \lambda _ { N = 0 } \ \lambda _ { N = 0 } \ \lambda _ { N = 0 } \ \lambda _ { N = 0 } \ \lambda _ { N = 0 } \ \lambda _ { N = 0 } \ \lambda _ { N = 0 } \ \lambda _ { N = 0 } \ \lambda _ { N = 0 } \ \lambda _ { N = 0 } \ \lambda _ { N = 0 } \ \lambda _ { N = 0 } \ \lambda _ { N = 0 } \ \lambda _ { N = 0 } \ \lambda _ { N = 0 } \ \lambda _ { N = 0 } \ \lambda _ { N = 0 } \ \lambda _ { N = 0 } \ \lambda _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ [ N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } \ \alpha _ { N = 0 } }$$
$$\mathbb { P } ( Z _ { i, j } Z _ { i ^ { \prime }, j } = 1 ) \ & = \ \left ( \frac { \binom { N - 3 } { k } } { \binom { N } { k } } + 3 \frac { \binom { N - 3 } { k - 1 } } { \binom { N } { k } } + \frac { \binom { N - 3 } { k - 2 } } { \binom { N } { k } } \right ) ^ { n } \\ & = \ ( 1 - \frac { 2 } { N ^ { 2 } } k ( k - 1 ) + O ( \frac { 1 } { N ^ { 3 } } ) ) ^ { n }$$
As for n = λN 2 the probability p is of constant order we obtain from Janson's inequality (1.3) for k = 3
$$\stackrel { \dots } { \dots } \stackrel { \dots } { \dots } \stackrel { \dots } { \dots } \stackrel { \dots } { \dots } \stackrel { \dots } { \dots } \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \dots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \\ \mathbb { P } ( Z = 0 ) \ \leq \ ( 1 - p ) ^ { \binom { N } { 2 } } \exp \left ( C N ^ { 3 } ( 1 - \frac { 2 } { N ^ { 2 } } k ( k - 1 ) + O ( \frac { 1 } { N ^ { 3 } } ) ) ^ { n } \right ) \\ \quad \leq \ ( 1 - p ) ^ { \binom { N } { 2 } } \exp \left ( C N ^ { 3 } e ^ { - 1 2 \lambda } \right )$$
for N large enough. This is, of course, a useless bound, since the right hand side is converging to infinity.
On the other hand, for Cov( Z i,j , Z i ,j ′ ) we compute up to terms of order n/N 4
$$\text{Cov} ( Z _ { i, j }, Z _ { i ^ { \prime }, j } ) = \frac { n ( k ^ { 3 } - 3 k ^ { 2 } + 2 k ) } { N ^ { 3 } } + O ( \frac { n } { N ^ { 4 } } ).$$
Thus
$$\sum _ { ( i, j ) \sim ( i ^ { \prime }, j ^ { \prime } ) } \text{Cov} ( Z _ { i, j }, Z _ { i ^ { \prime }, j ^ { \prime } } ) & = C \lambda N ^ { 2 }. \\ \cdot \cdot \cdot \cdot \cdot \cdot \frac { \Xi } { \Xi } \quad \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot.$$
Obviously, the bound (1.4) from Theorem 1.6 by Boutsikas and Koutras is useless in this example, since the sum of the covariances diverges. With the factor t 2 in front of the covariances, our bound (1.6) gives for any t > 0.
$$\mathbb { P } ( Z = 0 ) \leq ( 1 - p ) ^ { \binom { N } { 2 } } ( 1 + e ^ { - t } \frac { p } { 1 - p } ) ^ { \binom { N } { 2 } } + C t ^ { 2 } \lambda N ^ { 2 }.$$
If we now choose t extremely small, e.g t = e -N 3 , it is evident that the second summand on the right is asymptotically negligible with respect to the first. This first summand is estimated as
$$( 1 - p ) ^ { \binom { N } { 2 } } ( 1 + e ^ { - t } \frac { p } { 1 - p } ) ^ { \binom { N } { 2 } } \leq ( 1 - p ) ^ { \binom { N } { 2 } } ( 1 + \frac { e ^ { - 6 \lambda } } { 1 - e ^ { - 6 \lambda } } ) ^ { \binom { N } { 2 } }$$
Thus (1.6) gives
$$\mathbb { P } ( Z = 0 ) & \leq \left ( ( 1 - p ) ( 1 + \frac { e ^ { - 6 \lambda } } { 1 - e ^ { - 6 \lambda } } ) \right ) ^ { \binom { N } { 2 } } \\ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot \ \cdot$$
In particular for λ large, this is not only of the right order, but also very close to the lower bound (1 -p ) ( N 2 ) .
For k ≥ 4 one can compute that the covariances are now of order 1 /N 2 (which we spare ourselves as the computations are similar to ones above) , but as a matter of fact, a similar choice of t leads to a similar result as in the case k = 3.
To conclude, the above examples show that for positive correlations our method gives comparable results to that of Janson's inequality. Even more is true: There are examples where our methods gives useful bounds, while (1.3) does not. These examples occur, if many of the random variables are correlated, but correlation is very weak. In this case ∆ can tend to be so large, that (1.3) does not yield reasonable bounds. Moreover, when the sum of the covariances is too large, the last example illustrates also the situation where our bound gives an interesting approximation of the independent case, while the bound by Boutsikas and Koutras doesn't apply.
## References
- [1] B. K. Aliabadi, C. Berrou, V. Gripon, and X. Jiang. Learning sparse messages in networks of neural cliques. IEEE Transactions on Neural Networks and Learning Systems , August 2012. To appear.
- [2] A. D. Barbour, L. Holst, and S. Janson. Poisson approximation , volume 2 of Oxford Studies in Probability . The Clarendon Press, Oxford University Press, New York, 1992. Oxford Science Publications.
- [3] M. Behrisch, A. Coja-Oghlan, and M. Kang. The order of the giant component of random hypergraphs. Random Structures Algorithms , 36(2):149-184, 2010.
- [4] M. Behrisch, A. Coja-Oghlan, and M. Kang. Local limit theorems for the giant component of random hypergraphs. Combin. Probab. Comput. , 23(3):331-366, 2014.
- [5] R. Boppona and J. Spencer. A useful elementary correlation inequality. J. Combin. Theory Ser. A , 50(2):305-307, 1989.
- [6] M. V. Boutsikas and M. V. Koutras. A bound for the distribution of the sum of discrete associated or negatively associated random variables. Ann. Appl. Probab. , 10(4):1137-1150, 2000.
- [7] A. Coja-Oghlan, C. Moore, and V. Sanwalani. Counting connected graphs and hypergraphs via the probabilistic method. Random Structures Algorithms , 31(3):288-329, 2007.
- [8] J. T. Cox and G. Grimmett. Central limit theorems for associated random variables and the percolation model. Ann. Probab. , 12(2):514-528, 1984.
- [9] P. Erd˝s and A. R´ enyi. On random graphs. I. o Publ. Math. Debrecen , 6:290-297, 1959.
- [10] J. D. Esary, F. Proschan, and D. W. Walkup. Association of random variables, with applications. Ann. Math. Statist. , 38:1466-1474, 1967.
- [11] V. Gripon and C. Berrou. Sparse neural networks with large learning diversity. IEEE Transactions on Neural Networks , 22(7):1087-1096, July 2011.
- [12] S. Janson. Poisson approximation for large deviations. Random Structures Algorithms , 1(2):221-229, 1990.
- [13] S. Janson. New versions of Suen's correlation inequality. In Proceedings of the Eighth International Conference 'Random Structures and Algorithms' (Poznan, 1997) , volume 13, pages 467-483, 1998.
- [14] J. Kahn and N. Weininger. Positive association in the fractional fuzzy Potts model. Ann. Probab. , 35(6):2038-2043, 2007.
- [15] N. Karasik. Results on the Sums of Associated Random Variables Motivated by the Parabolic Anderson Model . ProQuest LLC, Ann Arbor, MI, 2013. Thesis (Ph.D.)-University of California, Irvine.
- [16] M. Karo´ nski and T. /suppress uczak. L The phase transition in a random hypergraph. J. Comput. Appl. Math. , 142(1):125-135, 2002. Probabilistic methods in combinatorics and combinatorial optimization.
- [17] C. M. Newman. Normal fluctuations and the FKG inequalities. Comm. Math. Phys. , 74(2):119-128, 1980.
- [18] C. M. Newman and A. L. Wright. An invariance principle for certain dependent sequences. Ann. Probab. , 9(4):671-675, 1981.
- [19] C. M. Newman and A. L. Wright. Associated random variables and martingale inequalities. Z. Wahrsch. Verw. Gebiete , 59(3):361-371, 1982.
- [20] P. E. Oliveira. An exponential inequality for associated variables. Statist. Probab. Lett. , 73(2):189-197, 2005.
- [21] P. E. Oliveira. Asymptotics for associated random variables . Springer, Heidelberg, 2012.
- [22] B. L. S. P. Rao and I. Dewan. A general method of density estimation for associated random variables. Nonparametric Statistics , 10:405-420, 1999.
- [23] D. Romik. Connectivity Patterns in Loop Percolation I: the Rationality Phenomenon and Constant Term Identities. Comm. Math. Phys. , 330(2):499-538, 2014.
- [24] W.-C. S. Suen. A correlation inequality and a Poisson limit theorem for nonoverlapping balanced subgraphs of a random graph. Random Structures Algorithms , 1(2):231-242, 1990.
- [25] S. H. Sung. A note on the exponential inequality for associated random variables. Statist. Probab. Lett. , 77(18):1730-1736, 2007.
- [26] G. Xing and S. Yang. Notes on the exponential inequalities for strictly stationary and positively associated random variables. J. Statist. Plann. Inference , 138(12):4132-4140, 2008.
- [27] G. Xing and S. Yang. Some exponential inequalities for positively associated random variables and rates of convergence of the strong law of large numbers. J. Theoret. Probab. , 23(1):169-192, 2010.
- [28] S.-c. Yang and M. Chen. Exponential inequalities for associated random variables and strong laws of large numbers. Sci. China Ser. A , 50(5):705-714, 2007.
(Matthias L¨we) o Fachbereich Mathematik und Informatik, University of M¨nster, u Einsteinstraße 62, 48149 M¨nster, Germany u
E-mail address , Matthias L¨we: o [email protected]
(Franck Vermet) Laboratoire de Math´ ematiques, UMR CNRS 6205, Universit´ e de Bretagne Occidentale, 6, avenue Victor Le Gorgeu, CS 93837, F-29238 BREST Cedex 3, France
E-mail address , Franck Vermet: [email protected] | null | [
"Matthias Löwe",
"Franck Vermet"
] | 2014-08-01T20:10:12+00:00 | 2014-12-19T13:09:08+00:00 | [
"math.PR",
"Primary: 60F10, Secondary: 60C05"
] | Large deviation upper bounds for sums of positively associated indicators | We give exponential upper bounds for $P(S \le k)$, in particular $P(S=0)$,
where $S$ is a sum of indicator random variables that are positively
associated. These bounds allow, in particular, a comparison with the
independent case. We give examples in which we compare with a famous
exponential inequality for sums of correlated indicators, the Janson
inequality. Here our bound sometimes proves to be superior to Janson's bound. |
1408.0295v3 | ## Cryogenic Dark Matter Search Detector Fabrication Process and Recent Improvements
A. Jastram a, ∗ , H. R. Harris , R. Mahapatra , J. Phillips a a a , M. Platt , K. a Prasad , J. Sander a a,b , S. Upadhyayula a a Department of Physics & Astronomy, Texas A&M University, College Station, TX 77843, USA b Department of Physics, University of South Dakota, Vermillion, SD 57069, USA
## Abstract
A dedicated facility has been commissioned for Cryogenic Dark Matter Search (CDMS) detector fabrication at Texas A&M University (TAMU). The fabrication process has been carefully tuned using this facility and its equipment. Production of successfully tested detectors has been demonstrated. Significant improvements in detector performance have been made using new fabrication methods/equipment and tuning of process parameters.
## 1. Introduction
The quest for an understanding of Dark Matter is at the forefront of particle physics today. One proposed constituent of Dark Matter is a particle called the WIMP (Weakly Interacting Massive Particle). Many experiments around the world are attempting to directly detect this particle in order to further understand the nature of Dark Matter. CDMS (Cryogenic Dark Matter Search) is one of these experiments, utilizing ground based detectors sensitive to nuclear recoils caused by direct collisions with these particles. These detectors, composed of germanium or silicon, are instrumented to detect deposition of both ionization and phonon energy upon interaction with an incident particle. Using these two signals and clever detector design[1], electron recoils (characteristic
∗ Corresponding author Email address: [email protected] (A. Jastram)
of the dominant background signals) can be discriminated from nuclear recoils (characteristic of a WIMP-like interaction). To continue probing new parameter space, greater sensitivity is needed for each generation of experiment, requiring significant improvements in the detector technology used to reject electron recoils and fabrication throughput. While many of these improvements are being made in readout electronics and analysis procedures, the underlying quality of the initial data, and therefore the quality of the detectors, is a fundamental limit[2]. Detector uniformity and consistency are also important, particularly when scaling up in size and quantity of detectors. Great strides in both of these areas have been made at the TAMU facilities. Serving as a second fabrication facility and now the dominant polishing facility, the increased throughput will allow the next generation experiment to produce the required detector payload on a competitive timescale.
## 2. Fabrication
CDMS detectors are produced using techniques and equipment similar to those in typical semiconductor fabrication processes. The process begins with high quality semiconductor substrates (germanium or silicon) and uses photolithography to etch deposited films into circuit structures (see Figure 1). One significant difference, however, is that the substrates used for CDMS detectors are much thicker, ranging from 10mm in original designs to 33.3mm in current production. For this reason, semiconductor equipment and processing typically used for ∼ 1mm thick substrates have been modified and tuned for these larger detector geometries.
## 2.1. Substrate Materials
Detectors are fabricated on high purity germanium and silicon substrates. For detector quality germanium substrates, 'high purity' equates to impurity levels typically on the order of 10 10 cm -3 . These are grown using the Czochralski Process. Substrates used for this experiment typically have dislocation densities of 1000-7000cm -2 . For silicon detectors, quality is specified/determined
Figure 1: Process flow chart from raw substrate to completion (see Section 2).
by room temperature resistivity. While > 8 kΩ-cm is the specification for acceptable material, typical detector quality substrates have a resistivity of > 20 kΩ-cm and are grown using the Float Zone Process. All detectors currently operating in the SuperCDMS Soudan experiment are 76mm diameter x 25mm thick germanium substrates. The next generation will utilize both silicon and germanium detectors, 100mm diameter by 33.3mm thick[3]. The majority of this R&D, including establishment and tuning of the fabrication process (specifically film characterization and photolithography steps), is performed using low resistivity commercial 'Prime Grade' silicon wafers (75mm and 100mm diameter with SEMI Standard 1 thickness and flats). Being much lower in cost, easier to obtain, and easier to clean/prepare than thick substrates, they are a natural choice for practice and R&D. Once established, fabrication procedures are
1 Standards/specifications available from www.semi.org
then tested/confirmed on thick substrates. Low purity/price thick substrates are used for this before fabricating detector quality substrates.
## 2.2. Alignment and Shaping
To improve uniformity and charge collection performance among the detectors, the substrates are shaped and aligned to a specific crystal axis and orientation. Upon delivery from the vendor, the cylindrical substrates are guaranteed to be aligned within ± 2 ◦ of the target crystal axis, typically [100]. For improved ionization drift/collection, they are subsequently re-shaped with the alignment refined to ± 0 1 . ◦ . For this reason, the substrates, as purchased, are slightly over-sized in all dimensions to account for material loss in re-shaping. Re-shaping consists of aligning and grinding the substrates' faces, followed by grinding the cylindrical sidewall. A custom fixture has been made to allow the surface plane of the substrate to be manipulated with micrometers to precisely dial in the crystallographic axis to the coordinates of the x-ray diffractometer (XRD) used in this alignment process. For more information on the XRD process, see [4]. In this setup, a modified Rigaku DMAX-1BX is used with the x-ray source operated at 30kV and 20mA. First, the face of the substrate is positioned/aligned to the point of initial interference with the x-ray beam (which is set to 2 θ =0 ) with the face parallel to the beam and perpendicular ◦ to the goniometer's θ plane. The goniometer is then set to the Bragg angle of the target crystal axis and a local 2 θ sweep is performed (the width of which is dictated by the alignment tolerance from the vendor). This produces a peak near the Bragg angle which will shift according to aforementioned micrometer adjustments. These adjustments and measurements are made iteratively (gross adjustments at first, followed by fine tuning) until the peak is within the required tolerance of the appropriate Bragg angle. The crystal is then locked into that orientation in the alignment fixture, which is designed such that it can be unmounted from the XRD system and attached to a grinding fixture. This assembly is then placed on a Lapmaster 24C lapping machine (equipped with a 220 grit diamond magnetic plate) such that the substrate feeds into the grind-
ing surface along the crystal axis. After grinding, this surface is measured again using XRD to confirm successful alignment. The second face is then ground parallel to the first using this same fixture. Parallelism of the faces is confirmed using a granite indicator stand. The cylindrical sidewall must then be shaped parallel to the crystal axis. To reduce the chance of chipping during this process, circular plates of glass (1/8' thick with a diameter 0.25' larger than the final substrate diameter) are bonded to each face with a wax bonder using 69 ◦ C quartz wax. The sidewall shape is then defined using a diamond coring fixture. The coring diameter is that of the final substrate specification. The glass plates and quartz wax are then removed. To provide room for interface boards in the detector housings (see Section 2.8) and ensure all crystals are fabricated in a uniform rotational orientation, flats are ground on the sidewall of the substrate. These are located normal to a specific crystallographic direction ([011] in the case of [100] crystals). To perform this alignment, the crystal is loaded into a custom XRD mount with the previously aligned crystal axis normal to the 2 θ plane and the x-ray beam incident upon the sidewall (with the sidewall now positioned to just slightly interfere with the beam while 2 θ =0 ). The goniometer ◦ is then set to the Bragg angle of the desired flat orientation, and the crystal is rotated about its axis (in the 2 θ plane) using a precision rotary table indexer until the diffracted intensity is maximized (locating the orientation to ± 1 ◦ ). Using a custom jig, the two diametrically opposed flats are ground using the Lapmaster 24C. The crystal is then lapped (on the same machine) to its desired thickness.
## 2.3. Heavy Etch
In order to remove substrate surfaces that may have been contaminated by previous processing and/or exposure to radon-containing atmosphere, the substrates are then chemically etched 2 . This process removes the outer layer (up to
2 This etch recipe is based on a process described in [5] modified by Paul Brink and Larry Novak.
∼ 250 µ m) of material, which is assumed to be contaminated/compromised. Silicon substrate etching has not been used by CDMS in the past, but is currently under development for future detectors. Germanium etching is performed in the following solution:
- 1. 3200mL 69% HNO 3
- 2. 640mL 50% HF
- 3. 150mL CH CO H (glacial) 3 2
The substrate is dipped in the etchant using a modified PTFE wafer cassette (used in all subsequent acid processing) and agitated lightly by hand, followed by a dip in de-ionized (DI) water. This is repeated 10 times. It is then placed in a Verteq 1600-55M spin rinse/dryer for a standard rinse/dry process (to be referred to as SRD). The SRD process consists of the following steps:
- 1. 35 seconds @600rpm with N 2 purge and DI spray
- 2. 230 seconds @1600rpm with heated N 2 purge
- 3. 90 seconds @1600rpm with N 2 purge
Following this step, substrates are stored in nitrogen purged cabinets when not being actively processed, reducing subsequent exposure to ambient radon.
## 2.4. Lapping/Polishing
Photolithographic processing of micron scale features requires a smooth, featureless substrate surface. For this reason, the coarsely lapped, heavy-etched detector faces must be polished. This is accomplished via four sequential steps:
- 1. Fine-grit manual lapping
- 2. Surface shaping polish
- 3. Scratch-removing polish
- 4. Final surface treatment polish
Substrates are hand-lapped on a slotted glass lapping plate using 9 µ m alumina polishing powder mixed with DI water to form a paste consistency. This is to remove large features from the surface. The surfaces must then be polished to a
specified flatness with a mirror finish, free of visible features (such as scratches or pits) to facilitate uniform film depositions and prevent circuit defects (see Figure 6) in subsequent processing. Polishing is performed on a dual spindle polishing machine. Control of surface curvature (concavity vs. convexity) is maintained with polisher settings and various sizes of polishing pads surfaced with polyester material in a 1:1 mixture of colloidal alumina polishing compound:DI water. This process is carefully controlled such that the final surface has < µ 2 m of total height deviation across the substrate if convex, < µ 1 m if concave (curvature is measured with a desktop laser interferometer). This is to ensure uniform contact with the photo mask (which can conform slightly to convexity but not concavity) during the photolithography process. Small surface scratches resulting from this step are then removed on the same machine using 'regular nap' polyurethane pads and a fresh mixture of the same polishing slurry. Final surface polishing is performed with 'high nap' polyurethane pads in a colloidal silica polishing compound. Final surface inspection is performed using a stereo zoom binocular microscope, manually confirming a defect free mirror finish.
## 2.5. Cleaning
Before the polished substrates can be processed into detectors, they must be cleaned carefully. This removes surface contaminants as well as any particulates that may cause defects in subsequent processing (see Section 2.8). For this reason, the cleaning is performed in a class 100 clean room. Germanium and silicon substrates require different cleaning processes, germanium's being much more time and labor intensive (another benefit of using silicon wafers for R&D).
## 2.5.1. Germanium
Initial germanium cleaning involves a 5 minute soak in acetone followed by a 5 minute soak in isopropyl alcohol (NOTE: all chemicals used in cleaning and subsequent processing are semiconductor grade). Following a thorough rinse with DI water, the surfaces are manually inspected using a microscope equipped with an LED ring light (especially effective for identifying particles on
the surface as it exposes diffuse features). If particulate count is unacceptable ( ≥ 10cm -2 ), the previous chemical process is repeated, and the crystal is dried using a filtered nitrogen gun. If particulate count is still unacceptable, the substrate is rinsed with methanol and manually wiped with a PVA cleaning brush. If the surface condition is still unacceptable, the methanol and brush wipe is repeated as necessary. Otherwise, the substrate proceeds to the oxide removal step. For this, a mixture of 3:1 DI water:50%HF is prepared in which the substrate is submersed for 5 minutes followed by a 3 minute soak in DI water. This is repeated three times and followed by surface inspection. If particulate count has become unacceptable, a methanol rinse and brush wipe are repeated as necessary. Upon completion, the substrate is placed in the oven at 120 ◦ C for 10 minutes to bake out remaining moisture.
## 2.5.2. Silicon
Silicon substrates also receive chemical cleaning, but have not shown the need for manual particulate removal. In the cleaning process, the substrates are initially doused with methanol then isopropyl alcohol, followed by SRD. To remove metals and organic contaminants, a Piranha clean process is used. This consists of a 20 minute dunk in the following solution (heated to 55 ◦ C):
- 1. 600mL 30% H O 2 2
- 2. 700mL 98% H SO 2 4
- 3. 1050mL 0.250N H SO 2 4
Substrates are then soaked in 55 ◦ C DI for 1 minute, manually agitated once every 15 seconds. To remove the native oxide from the substrates' surfaces, they are then placed in the following solution for 20 seconds:
- 1. 2700mL DI
- 2. 50mL 50% HF
The substrates are then dipped again in 55 ◦ CDI for 1 minute, manually agitated once every 15 seconds. To remove ionic and heavy metal atomic contaminants
from the substrates' surfaces, the substrates are submersed in the following solution, heated to 70 ◦ C, for 15 minutes:
- 1. 1750mL DI
- 2. 325mL 30% H O 2 2
- 3. 300mL 37% HCl
This is followed by SRD and a 5 minute dehydrate in the oven at 120 C. The ◦ cleaning process seals the substrate with a thin oxide layer which is removed in the sputtering system prior to film deposition (see Section 2.6).
## 2.6. Thin Film Deposition/Tuning
The films that form the final circuit and sensors of the detector are deposited using a customized plasma sputtering deposition system. Precise and repeatable process control is vital in the deposition of these films as they dictate the quality of the final circuit features (see Sections 3.1, 3.2, and 3.3). The deposition system used in this process is a Perkin Elmer 4400 Delta with PLC/PC interface automated by Semiconductor Engineering Group, Inc. (SEGI), which has been modified for this fabrication process. It can simultaneously sputter 8 substrates, 100mm in diameter (or 6, 150mm diameter) and is composed of two main vacuum chambers: the load lock and the process chamber, separated by a gate valve. The load lock is the location in which substrates are initially loaded, which is then pumped to 4.0x10 -6 Torr using a turbo-molecular pump (added for this process). This allows the process chamber to stay isolated from atmospheric contaminants at all times (specifically when the substrate is transferred through the gate valve into this chamber). To further reduce contaminants, a pre-coat of Aluminum is sputtered in the process chamber as a getter. This process removes traces of O 2 and H O (see Figure 2) as well as other contaminants 2 which can alter film characteristics, further improving process stability. The substrates are then transferred into the process chamber, which is subsequently pumped back to a base pressure of 9.0x10 -7 Torr. The process chamber is cylindrical ( ∼ 1m in diameter) and can simultaneously accommodate three different
targets of sputtering materials. The chamber is equipped with aluminum, tungsten, and silicon targets, all of which are 99.999% pure. After substrates are loaded in the load lock, the rest of the deposition process is entirely automated, including everything from rotation and height settings of the table on which the substrates sit to setting/maintaining the chamber and plasma conditions. Table height settings are customized for each substrate thicknesses to maintain a constant target-substrate distance. Other settings of particular note are the voltage and power supplied to the target (DC or RF), DC bias applied to substrates during deposition, flow of Argon into the chamber, and optional RF pre-etch.
The process chamber is equipped with an RGA (Residual Gas Analyzer) which measures the contents of the gas in the process chamber. The RGA can be used in two modes: plotting the entire spectrum at once, showing peaks at various masses corresponding to contaminants in the chamber (see Figure 2), or plotting the levels of a chosen contaminant over time. A high capacitance valve is implemented to allow the RGA to operate at high vacuum levels as well as in-process levels ( ∼ 10mTorr). This allows in-situ analysis of any possible gas contaminants during the deposition process.
All of the above devices/processes allow the minimization of contaminants and maximization of control and repeatability in the deposition process. Using this system, three thin film layers (designated as the trilayer) are sputtered sequentially on both faces of each substrate: 40nm amorphous silicon (a-Si), 300nm aluminum, and 30nm tungsten. The a-Si layer underlies all final metal circuit structures and is used to protect the substrate surface from aluminum and tungsten etchant chemicals, as well as improving the ionization collection boundary. The aluminum layer forms the phonon collection structures (see Figure 3 and Section 3.1) as well as the circuit lines/electrodes connecting the sensors on the detector. The tungsten layer serves as a cap layer, preventing the aluminum surface from oxidizing (see Section 3.3) and preventing back-sputtering of the aluminum during the subsequent deposition. Before each deposition begins, the target to be used is pre-sputtered for 25 seconds to clean its surface with its shutter closed (to prevent sputtered material from depositing on substrates).
Before all depositions, an aluminum shadow mask is manually placed on the substrate face which covers the outer ∼ 1mm of the exposed surface, preventing deposition on this region (see Section 3.5) and the substrate's sidewall surface. The steps and parameters used for the trilayer deposition are the following:
- 1. 10 minute RF etch, 350 W RF, 50 sccm Ar, 10 mTorr
- 2. 16 minute a-Si deposition, 500 W RF, 50 sccm Ar, 8 mTorr
- 3. 7m18s aluminum deposition, 2.5 kW DC, 40 sccm Ar, 10 mTorr
- 4. 36 second tungsten deposition, 2.5 kW DC, 40 sccm Ar, 8 mTorr
The films are then patterned photolithographically and chemically etched, forming the majority of the detector circuit (described in Section 2.7). After chemical etching, a 40nm layer of tungsten is sputtered on each face of the substrate. This layer forms the transition edge sensors (TES's) of the detector (see Sections 3.2 & 3.3). The steps/parameters used for this deposition are as follows:
- 1. 10 minute RF etch, 350 W RF, 50 sccm Ar, 10 mTorr
- 2. 51 second tungsten deposition, 2.5 kW DC, 40 sccm Ar, 8 mTorr, 100V DC bias delivered to substrate
## 2.7. Photolithography
A three step photolithographic process is used to define the circuit features on the substrates. The original process from which this was adapted is described in [6] and [7]. The first step defines the aluminum structures (circuit lines and phonon collecting fins [see Section 3.1]). The second step defines the tungsten TES features (see Section 3.2), and the third defines the a-Si structure and substrate trenching regions (see Section 3.4). The aluminum and tungsten are etched with chemicals, while the a-Si is plasma etched via an RIE (Reactive Ion Etch) process. In all three steps, an etch resistive mask of photoresist is used to protect the features while the exposed films are etched. The photoresist mask pattern is formed via UV transfer/exposure using a master template mask. Chemical processing of the substrates is performed in a class 100 UV-free clean room.
Figure 2: Example RGA spectra of partial pressures in the SEGI process chamber at various atomic masses before ( top ) and after ( bottom ) aluminum getter deposition (see Section 2.6), demonstrating the efficacy of this process in reducing oxygen and water vapor levels.
## 2.7.1. Trilayer Patterning
Student Version of MATLAB After the trilayer deposition (see Section 2.6), a Solitec 5110-SJ spin coater is used to spin coat Shipley Microposit S1811 photoresist on both faces of the substrate. To create and maintain a vacuum seal between the substrate and spin coater's chuck, 0.032' thick elastomer skirts are stretched around the substrate sidewall. These are removed and discarded after the spin coat process (see Section 3.5 for more information on the photoresist layer/process). The substrate is then placed in the oven (in a PTFE cassette, covered with aluminum foil to protect the fresh photoresist from particulates) at 120 ◦ C to soft bake the photoresist (see Table 1 for bake times). After baking, the crystal is
Figure 3: Image of a single phonon sensor and magnified inset of TES line and 'waterfall' region (arrow indicates waterfall boundary, see Figure 9 for closeup and Section 3.3 for description). Each color corresponds to the exposed film on the final structure: Blue=Aluminum (mask #1, trilayer mask), Pink=Tungsten (mask #2, TES mask), Green=a-Si (mask #3, 'trench' mask) (see Section 2.7). The central vertical line is the ∼ 2 µ m wide TES, and the large aluminum 'fins' are the phonon absorbing structures (see Section 3.1). Note: a-Si underlies all metal features.
allowed to cool to room temperature. The cooling process is accelerated with a gentle stream of filtered nitrogen gas on each face. The substrate is then ready for the photolithographic mask transfer/exposure process. An OAI 206-094735 contact aligner with a 350W Hg g-line UV lamp is used to expose each face for 5.3s at 8.15mW/cm 2 using mask #1, the trilayer mask (see Figure 3). Special care must be taken not to scratch the backside photoresist layer when placing the substrate on the stage (and when flipping the substrate for exposure of the second face). The UV intensity is confirmed before each exposure using an OAI 0308 UV meter tuned to 436nm (g-line). The pattern is then developed using Shipley Microposit MF-319 developer, mildly agitated by hand, until completion. This is judged by eye, typically taking 70-90 seconds. The substrate is gently rinsed in DI water after development, then proceeds to SRD. The photoresist is then inspected to confirm successful development (robust replication of mask structure). The substrates are then returned to the oven at 120 ◦ C to hard bake the photoresist (see Table 1 for bake times). They are then cooled to room temperature with the assistance of gentle nitrogen gas flow. The tungsten layer is etched for 6m30s using 30% H O 2 2 with 2 gentle manual agitations at 1
Figure 4: Detector patterning process (not to scale). Individual film layers are shown throughout the deposition/photolithography process (see Section 2.7). Gray=Substrate, Green=a-Si, Blue=Aluminum, Pink=Tungsten, and Red-brown=Photoresist.
minute intervals, followed by SRD. The aluminum layer is etched using Cyantek Al-11. This typically consists of 5-6 iterations of the following: 45s Al-11 dunk
with constant gentle agitation, followed by a 15s DI rinse. The aluminum etch leaves a slight overhang of the tungsten cap layer, due to the isotropic nature of the reaction (see Figure 5). Intermittent DI rinses are used to control the temperature of the exothermic etch reaction[8], improving etch uniformity and reducing the undercut/overhang issue. When all exposed aluminum appears to have vanished, the substrate receives an additional 15s of Al-11 etch to ensure no aluminum remains, then proceeds to SRD. To remove the overhang feature, another tungsten etch is performed (see Section 3.3). This consists of a 10 minute submersion in 30% H O , with 2 gentle agitations every 2 minutes, fol2 2 lowed by SRD. At this point, the circuit pattern is carefully inspected to confirm successful etching and preserved photoresist integrity. The photoresist layer is removed using a 20 minute dip in Shipley PRX-127 at 45 ◦ C, with 2 gentle agitations every 5 minutes, followed by SRD. As a final cleaning precaution, the substrate is submerged in Baker PRS-1000 for 10 minutes at 45 ◦ C, followed by SRD. Etched features are then inspected (and again after each subsequent photolithography/etch cycle), monitoring for defects and critical circuit feature dimensions. The substrate is then placed in the SEGI under vacuum overnight to boil off any moisture before the following tungsten (TES layer) deposition.
## 2.7.2. TES Patterning
The second deposition, that which forms the TES tungsten layer, is then performed (see Section 2.6). After this deposition, the substrate receives the same spin coat, soft bake, alignment, exposure, develop and hard bake process as previously mentioned. The mask used for this layer (mask #2, see Figure 3) defines the TES structures on the circuit. After hard bake and cooling, the tungsten is etched in 30% H O 2 2 for 12 minutes, with 2 gentle agitations every 2 minutes, followed by SRD. It should be noted that this step etches all tungsten not covered by the mask, including the tungsten cap layer from the mask #1 structures. Therefore, anywhere that masks #1&2 coincide, all four film layers remain. Otherwise, mask #2 defines structures with only TES tungsten on top of a-Si (see Figure 4 and Section 3.3). The photoresist is then inspected
for integrity and removed with the same PRX-127 and PRS-1000 process as before (aside from PRX-127 time reduction to 15 minutes). The substrate is then placed under vacuum overnight to remove moisture (improving adhesion of subsequent photoresist coat).
## 2.7.3. a-Si Patterning and 'Trenching'
The last photolithography step defines the a-Si structure with mask #3, using the same spin coat, soft bake, alignment, exposure, develop, hard bake, and cooling process as previous steps. After hard bake and cooling, the a-Si is etched in a modified Tegal 903C reactive ion etcher using 8 iterations of the following etch and cool down steps:
- 1. 18 second etch, 400 W RF @ 13.56 MHz, 18 sccm SF6, 50 sccm He, 1100 mTorr
- 2. 7 minute purge/cool down, 50 sccm He, 900 mTorr (limited by Helium MFC)
See Section 3.4 for more information on this etch. The photoresist is then removed with the same PRX-127 and PRS-1000 process as before (with original PRX-127 time of 20 minutes).
Figure 5: Diagram depicting the tungsten overhang issue caused by the isotropic aluminum etch process (not to scale). Figure from [9].
Table 1: Bake times for various substrate sizes. Thin (practice) wafers soft bake for 1m50s on a 115 ◦ C hot plate, and hard bake for 15 minutes in the oven at 120 ◦ C.
| Size (Dia. x Thickness) | Soft Bake | Hard Bake |
|---------------------------|-------------|-------------|
| 76mm x 10mm | 20m | 1h30m |
| 76mm x 25mm | 25m | 2h |
| 100mm x 33.3mm | 28m | 2h20m |
## 2.8. Inspection/Surgery/Mounting
It is possible for defects to arise in the photolithography process which can prevent a detector from operating as desired. For this reason, every element of every detector circuit is manually inspected using a microscope. This step is crucial to successful detector fabrication as micron scale defects can knock out an entire sensor channel. Defects of concern include areas of missing metal, causing breaks in the circuit continuity as well as metal films that did not etch properly, causing shorts (see Figure 6). In the case of open circuits due to breaks in metal continuity, a Kulicke & Soffa 4523AD wire bonder (with a DewyL Tool MCSOE-1/16-750-45-C-2025-M wedge and 0.00125' diameter 99% Al 1% Si wire) is used to connect the isolated metal regions with wire bonds. The circuit is designed with extra metal pads (bonding locations) to make this task easier. In the case of a defect causing a short circuit, repairs can be made using one of two options:
- 1. Manually abrading the film with the wedge of the wire bonder to eliminate the unwanted electrical connection (an auxiliary wedge should be used for this, preventing damage to the bonding wedge)
- 2. Using a localized droplet of the proper chemical etchant to remove the metal. This can also be used during the photolithography process (depending on the nature/location of the defect), utilizing the protective photoresist mask, thus minimizing chemical damage to nearby structures.
The detectors are then mounted in hexagonal OFHC copper housings (see Figure 14) which include Detector Interface Boards (DIBs). The detector elec-
trodes have large bond pads near the DIBs, used to wire bond the detector channels to copper traces on the DIBs. This provides a feed-through to the outside of the grounded housing as well as a rigid connection to external hardware.
Figure 6: Left) Example of an un-etched section of Aluminum, causing a possible short. Right) Example of photoresist failure allowing unwanted aluminum etching, breaking circuit continuity, and requiring a 'surgery' wire bond. Both examples were likely caused by particulates on the detector surface during early processing.
## 3. Process Tuning, Results, and Improvements
## 3.1. Aluminum Film
To efficiently read out phonon energy, the phonons are first absorbed in the superconducting aluminum 'fins' (see Figure 3) where the energy is used to dissociate Cooper pairs which split into pairs of quasi-particles. These quasiparticles must diffuse through the aluminum to reach the tungsten TES where they are essentially trapped due to the tungsten's lower superconducting bandgap value (see Section 3.3). A crucial property of the aluminum is a high quasiparticle diffusion length. This length is inhibited by impurities found in the aluminum.
Quasi-particle diffusion length is important to monitor and control. To quantify the quality of the aluminum film, a measurement of the residual resistivity ratio (RRR) is performed. This is the ratio of the film's resistivity at room temperature to its resistivity at 4K. A higher RRR value corresponds to a higher quality film (i.e. one with fewer impurities) [10]. A RRR value of 10 has been
deemed sufficient for a well functioning device in these detectors, but films have been tuned using the SEGI to routinely achieve a value of ∼ 16.
## 3.2. T c Tuning
Optimal detector readout relies on the TES sensors being held at specific temperatures in their superconducting-transition resistance curves. The second tungsten deposition forms the TES layer. Consequently, this deposition must be carefully tuned to produce tungsten of a uniform, consistent, and preciselycontrolled critical temperature (T ). c Critical temperatures of thin tungsten films are largely dictated by the ratio of α to β phase in the material. This is due to the fact that the α -W exhibits a T c of 15 mK [11] while β -W can have T 's ranging from 1 to 4 K [12]. c Utilizing this and the fact that the two phases have different crystallographic structures (and therefore, different Bragg angles), provides a technique of roughly estimating the T c of a given sample at room temperature using XRD[12] (see Figure 7). This technique is useful for tuning film samples to have high α β : ratios (T 's closer to the desired range), c but in this range, the ratio becomes so heavily α -dominated that differences in β concentrations become indistinguishable, making T c predictions difficult. To finely tune deposition parameters to the ∼ 80mK target, a dilution refrigerator is used to physically measure the resistance transition as the sample is cooled past its T c and again as it warms up. With this feedback, depositions with different sputtering power, substrate bias, and argon pressure were produced and tested, creating films of varying T 's. c In this process, a correlation was established connecting room temperature resistivity of the films to their T c (see Figure 7, right ), allowing recipes to be roughly tuned and chosen with simple room temperature measurements (sheet resistance measured with a 4point probe, corrected for film thickness to calculate resistivity)[12]. Using these processes, a recipe was chosen to produce films possessing the desired T . c Current experimentation with devices of varying T 's rely heavily on the c resistivity-T c correlation, saving significant time and money required for dilution refrigerator tests.
Figure 7: Left) XRD spectra showing discrimination between samples of differing α β : concentrations and their resulting T c 's. Also marked are the locations of the peaks of pure α and β phase films. An algorithm fitting two Gaussian functions (centered at these values) provides an estimate of phase ratios, and therefore T c 's, of the films. Right) Plot showing correlation of critical temperature vs. sheet resistance of similar thickness films (40 ± 4nm) [9]. These room temperature characterization methods allow film deposition parameters to be tuned without the time or monetary expense of dilution refrigerators (see Section 3.2).
In addition to depositing films with carefully tuned, repeatable T 's, c the SEGI has demonstrated the ability to produce films with much higher T c uniformity across the substrate surface. Previous systems have had large T c gradients across the face of detectors, beyond an acceptable limit (see Figure 8). To correct this issue, T c distributions must first be mapped (requiring detector testing in a dilution refrigerator), followed by ion implantation of 56 Fe (specifically into the TES's) to correct for the measured T c gradient, a process described in [13]. Films deposited in the SEGI, however, have demonstrated uniformities as good or better than typical post-implant samples from other systems. This 'as-delivered' uniformity circumvents a full round of millikelvin testing (T c mapping) and ion implantation, increasing throughput rates. The consistency and uniformity of films produced by the SEGI may allow the test process to largely avoid T c testing, aside from periodic verification. Circuit continuity tests can be accomplished at higher temperatures (up to ∼ 1K), meaning these detectors may be able to avoid dilution refrigerator testing as a whole during high throughput periods. With improved production throughput rates, the
bottleneck is shifted from fabrication to testing, exaggerating the importance of these consistency and uniformity improvements.
Figure 8: Left) Example of W T c variation (in mK) across a sample substrate face from the previous CDMS deposition system. Middle) T c variation of same film after ion implant compensation (see Section 3.2). Figure adapted from [14]. Right) T measurements (see Table c 2) from the 8 phonon channels of G9F, a detector fabricated at TAMU, demonstrating T c uniformity without ion implantation. All T 's are in mK. c
## 3.3. Conformal Film Deposition
Controlling the fabrication quality of the aluminum-tungsten interface (to maximize quasi-particle diffusion into the TES) is important because phonons absorbed in the aluminum only contribute to the measured phonon signal if they are able to drift into the tungsten. When quasi-particles drift from the aluminum 'fins' to the overlapping TES structure (see Figure 3), they must first drift into the intermediate tungsten cap layer. Since this intermediate tungsten layer is deposited immediately after the aluminum layer without breaking vacuum , no oxide is able to form between the two. Without this cap layer, an oxide forms on the aluminum surface before the TES layer deposition and inhibits the diffusion of quasi-particles from one film to the other. While the cap layer does oxidize slightly, the oxidation is easily removed with the RF etch which precedes the TES film deposition, forming a more favorable interface between the two tungsten layers.
The quasi-particle propagation from the 'fins' to the TES's is aided by the bandgap disparity arising from the aluminum and tungsten films' contrasting
| Channel | T c (mK) | R n (Ω) |
|-----------|------------|-----------|
| A1 | 75 | 0.64 |
| B1 | 75 | 0.59 |
| C1 | 77 | 0.59 |
| D1 | 76 | 0.59 |
| A2 | 75 | 0.69 |
| B2 | 74 | 0.56 |
| C2 | 73 | 0.54 |
| D2 | 76 | 0.56 |
Table 2: Critical temperatures and 'normal' resistance values (R n ) for the 8 phonon channels of detector G9F (see Section 3.2). R n is the resistance of the channel while the aluminum is superconducting, but the tungsten is normal (held at a temperature significantly above its T ). Note: Channels A1 and A2 are outer channels (see Figure 8) and have higher R c n values due to their sensor layout.
T 's (aluminum's T c c of ∼ 1.2K equates to a gap energy of 0.18meV compared to the tungsten's gap energy of ∼ 25 µ eV)[9]. Due to the magnitude of disparity in band gap energies, a process of quasi-particle multiplication can even occur at this boundary[15].
Because it is deposited over an already etched structure, the TES film must maintain continuity while stepping down ∼ 330nm (the initial aluminum + tungsten layer) from the initial tungsten cap layer to the a-Si layer. Discontinuity in this region severs the phonon collection structure from the TES line (see Figures 3, 9, and 10), preventing signal readout. To avoid this issue, the TES film must be a conformal layer closely following the topology, particularly the sidewall, of the trilayer structures. This region is designated as the 'waterfall' region (see Figure 10 for examples of this feature exhibiting both poor and good continuity). To prevent this problem, the 'overhang etch' has been implemented into the process. This etch was tuned by performing many iterations of circuit fabrication on practice wafers with various overhang etch times using SEM imaging for feedback. Once established, the process was confirmed with thick substrates,
again using SEM imaging.
Figure 9: Close-up and SEM image of 'waterfall' boundary (see Section 3.3). Location on phonon sensor and perspective are indicated by the arrow, referencing Figure 3.
Figure 10: Left) SEM images of overhang discontinuity issue. Middle) Conformal sidewall deposition utilizing overhang etch, resulting in robust film continuity (see Section 3.3). Right) SEM image of FIB-prepared cross-section of waterfall feature, showing conformal tungsten deposition. FIB image courtesy of Evans Analytical Group.
## 3.4. a-Si Etch and 'Trenching' of iZIP Detectors
Discrimination between background and signal events using these detectors relies on a calibrated ratio of energy measured in the ionization channels versus that measured in the phonon channels. Charge carriers produced by events near the faces of a detector often fail to drift through the entire crystal to the appropriate electrode. This results in a reduced ionization collection signal, causing the event to be improperly identified in subsequent analyses. A new circuit design has been implemented to combat this. The design is called the i ZIP ( interleaved Z-sensitive Ionization and Phonon detector)[16][17][18], and as
the name suggests, it utilizes interleaved electrodes on each surface. The interleaved electrodes alternate from ground to +2V on one face and from ground to -2V on the other (see Figure 11). This is in contrast to previous designs[1] where one face is held at ground potential while the other is voltage biased. The interleaved design produces a very uniform field in the bulk but local regions of high field intensity near the surface. This causes the carriers (electrons and holes) produced near the surface to both be collected by the adjacent surface, with relatively little charge drifting to the opposite face. Therefore, any events with significant disparities in charge collection from one face to the other (i.e. failing the charge-symmetry requirement) are considered to be surface events. This procedure has been demonstrated to be very successful and is the design currently operating in SuperCDMS Soudan[19]. To fully realize the potential of this technique, detectors should be able to hold higher biases (producing stronger local surface fields) without breakdown. Limitations arise, however, as the electrode spacing is ∼ 1mm, and current begins to leak across the surface as voltage is increased, eventually resulting in breakdown. To reduce this problem and allow higher bias voltage, a trench is etched into the surface of the substrate, between the electrodes. For this purpose, the a-Si etch step is extended by ∼ 700%. Because the gas used to etch the a-Si also etches the substrate material, this extra time allows etching of the substrate itself. The process has been tuned such that a trench of ∼ 1 µ m in depth is created between the electrodes and has been shown[20] to produce detectors that can hold much higher bias voltages without the problems mentioned previously (see Section 4).
## 3.5. Photoresist Studies
The tuning of the photoresist layer is of utmost importance as it dictates not only the geometries of the final detector circuit structure, but whether or not the deposited films survive the fabrication process at all. For this reason, much time was spent investigating the photoresist layer and photolithographic processing of this layer. The cross-section of the developed photoresist pattern is controlled with the UV exposure, which can result in angled sidewalls (inward
iZIP Charge Signal
Thursday, March 28, 13
4
Figure 11: Left) Diagram of iZIP detector geometry/design used in SuperCDMS Soudan showing alternating biased charge collecting rails and 0V phonon rails (see Section 3.4). Figure from [19]. Right) Simulation of iZIP internal field lines, showing strong tangential electric fields at the surface and a uniform drift field in the detector bulk, a method proven to discriminate against the previously problematic surface events. Figure from [21].
or outward), changing the width of the film etched below (see Figure 12). A dedicated study of the UV exposure (varying UV power and time, with SEM feedback) was performed to prevent these problems from affecting our circuit features.
In previous CDMS detector designs, Shipley Microposit S1813 photoresist[22] has been used for the photolithographic processing. This chemical was chosen for multiple reasons:
- 1. Resistance to etchant chemicals used in this process
- 2. Ability to reproduce sub-micron line-widths
- 3. Viscosity to match our desired layer thickness ( ∼ 1.4 µ m) with rotational speeds that produce optimum uniformity (3500-5500rpm [22])
- 4. Compatibility with metal ion-free developers [22]
However, as substrates of larger mass joined the production line, it became desirable to decrease spin coating speeds (causing less strain on the spin coater and less risk to the substrates). For this reason, Shipley Microposit S1811[22] is now used for its lower viscosity while still possessing the other characteristics
Figure 12: SEM image of a photoresist feature's cross-section. Due to improper UV exposure, this feature's sidewalls angle outwards, protecting a wider area of film from etching underneath, resulting in widening of circuit features. Note: This layer's thickness of 1.66 µ m corresponds to a previous spin coat recipe, using S1813 photoresist (see Section 3.5).
mentioned above. Using S1811, the spin coat process is performed at 2300 rpm (as opposed to 4000 rpm required for S1813) for 60 seconds, producing a 1.4 µ m layer on each face. These parameters (along with those of the UV exposure mentioned previously) were tuned using feedback from SEM images confirming faithful reproduction of mask feature line-widths and robust cross-sections after exposure and development.
Another photoresist issue that was studied and successfully remedied is that of a so-called 'edge bead'. After the spin coat process, a thick bead of photoresist can be seen around the edge of the substrate surface. Thicker than the
nominal layer, this bead does not receive enough UV exposure and developing to be properly patterned/removed prior to etching. While there are no vital circuit features in this region, it prevents any films underneath from being etched away, leaving a metal band (which could potentially cause shorts) around the edge of the detector after the photoresist is removed. It is for this reason that the previously mentioned shadow mask is implemented, preventing deposition in this region (see Section 2.6). This procedure has proven to be a low cost yet highly effective method of combating the edge bead problem, with negligible impact on detector patterning at radial extremities.
## 4. Results to Date
Using this process at the dedicated TAMU fabrication facility, detectors have been produced of the size and design of those in SuperCDMS Soudan. Test data from detector G9F, one of the first of these produced at TAMU, can be seen in Figure 13, demonstrating pulses from operational phonon sensor channels as well as the 356 keV photopeak from a Ba-133 calibration source. In addition, this detector showed unparalleled TES T c uniformity without ion implant compensation (see Figure 8 and Table 2). However, it showed an inability to hold adequate bias voltage, leading to further tuning of the trenching process (see Section 3.4). The following detector, G10F (using the improved trenching process), demonstrated more than adequate ability to hold bias, showing no signs of leakage up to ± 5V (the limit of the test stand). Specification standards used to rate SuperCDMS detectors categorize this detector as 'very good'. Subsequent testing showed functional charge performance up to 9V [20], much higher than required for the experiment.
Detectors produced at this facility have demonstrated performance that meets or exceeds the requirements for this experiment, certifying this location as an integral fabrication facility for SuperCDMS SNOLAB detectors. 100 mm x 33.3 mm thick science quality detectors were successfully produced at this facility in early 2013 (see Figure 14) and are currently awaiting testing.
Figure 13: Left) Phonon pulses from detector G9F, fabricated at TAMU. Right) Calibration spectrum from detector G9F, clearly showing the 356 keV Ba 133 peak (see Section 4).
## 5. Conclusion
A robust, repeatable fabrication procedure has been established, demonstrated, and improved at the TAMU fabrication facility. Increased throughput as well as improvements made in the process itself are expected to contribute substantially to the success of the next generation SuperCDMS SNOLAB experiment. In particular, increased fabrication efficiency, improved TES T c consistency and uniformity, increased bias voltage ability due to substrate trenching, and improved signal collection from overhang studies will improve detector success rates, reducing fabrication and testing costs.
## 6. Acknowledgments
The authors would like to thank the CDMS team at SLAC/Stanford (especially Blas Cabrera, Paul Brink, Richard Partridge, Matt Cherry, Astrid Tomada and John Mark Kreikebaum), Sunil Golwala, Betty Young, Jae Woo Suh, the CDMS team at the University of Minnesota (especially Hassan Chagani, Priscilla Cushman, Matt Fritts, Tommy Hofer, Allison Kennedy, Vuk Mandic, Roxanne Radpour, David Strandberg, and Anthony Villano), and the University of California, Berkeley 75 µ W dilution refrigerator test team.
Figure 14: 100mm x 33.3mm detectors fabricated at TAMU, currently awaiting testing.
This work was supported by the Department of Energy (Contracts DE-FG0213ER41918, DE-AC03-76SF00098, DE-FG02-92ER40701, DE-FG02-94ER40823, and DE-SC0004022), the National Science Foundation (Grant Nos. PHY-1102842 and NSF-0919599), and funding from Texas A&M University.
## References
- [1] D. S. Akerib, J. Alvaro-Dean, M. S. Armel-Funkhouser, M. J. Attisha, L. Baudis, D. A. Bauer, J. Beaty, P. L. Brink, R. Bunker, S. P. Burke, et al., First results from the Cryogenic Dark Matter search in the Soudan Underground Laboratory, Physical Review Letters 93 (21) (2004) 211301.
- [2] CDMS II Collaboration, et al., dark matter search results from the CDMS II experiment, Science 327 (5973) (2010) 1619-1621.
- [3] J. Sander, Z. Ahmed, A. J. Anderson, S. Arrenberg, D. Balakishiyeva, R. B. Thakur, D. A. Bauer, D. Brandt, P. L. Brink, R. Bunker, et al., SuperCDMS status from Soudan and plans for SNOLab, in: WORKSHOP ON
DARKMATTER,UNIFICATIONANDNEUTRINOPHYSICS:CETUP* 2012, Vol. 1534, AIP Publishing, 2013, pp. 129-135.
- [4] B. E. Warren, X-ray Diffraction, Dover Publications, Mineola, NY, USA, 1990.
- [5] P. J. Holmes, P. Handler, The electrochemistry of semiconductors, Journal of The Electrochemical Society 109 (8) (1962) 228C-228C.
- [6] P. Brink, Z. Ahmed, D. Akerib, C. Bailey, D. Balakishiyeva, D. Bauer, J. Beaty, R. Bunker, B. Cabrera, D. Caldwell, et al., Supercdms detector fabrication advances, in: THE THIRTEENTH INTERNATIONAL WORKSHOP ON LOW TEMPERATURE DETECTORSLTD13, Vol. 1185, AIP Publishing, 2009, pp. 655-658.
- [7] P. Brink, A. Anderson, D. Balakishiyeva, D. Bauer, J. Beaty, D. Brandt, B. Cabrera, H. Chagani, M. Cherry, J. Cooley, et al., Detector fabrication yield for supercdms soudan, Journal of Low Temperature Physics (2014) 1-7.
- [8] K. R. Williams, R. S. Muller, Etch rates for micromachining processing, Microelectromechanical Systems, Journal of 5 (4) (1996) 256-269.
- [9] K. Prasad, Search for lightly ionizing particles using CDMS-II data and fabrication of cdms detectors with improved homogeneity in properties, Ph.d. Dissertation, Texas A&M University.
- [10] C. James, The residual resistance of metals, Physics Education 2 (6) (1967) 315.
- [11] E. Lassner, W.-D. Schubert, Tungsten: properties, chemistry, technology of the elements, alloys, and chemical compounds, Springer, 1999.
- [12] A. E. Lita, D. Rosenberg, S. Nam, A. J. Miller, D. Balzar, L. M. Kaatz, R. E. Schwall, Tuning of tungsten thin film superconducting transition
temperature for fabrication of photon number resolving detectors, Applied Superconductivity, IEEE Transactions on 15 (2) (2005) 3528-3531.
- [13] B. A. Young, T. Saab, B. Cabrera, J. J. Cross, R. M. Clarke, R. A. Abusaidi, Measurement of T c suppression in tungsten using magnetic impurities, Journal of Applied Physics 86 (12) (1999) 6975-6978.
- [14] GEODM S4 Review, CDMS Internal Note.
- [15] N. E. Booth, Quasiparticle trapping and the quasiparticle multiplier, Applied physics letters 50 (5) (1987) 293-295.
- [16] M. Pyle, Optimizing the design and analysis of cryogenic semiconductor dark matter detectors for maximum sensitivity, Ph.d. Dissertation, Stanford University.
- [17] P. L. Brink, B. Cabrera, J. P. Castle, J. Cooley, L. Novak, R. W. Ogburn, M. Pyle, J. Ruderman, A. Tomada, B. A. Young, J. Filippini, P. Meunier, N. Mirabolfathi, B. Sadoulet, D. N. Seitz, B. Serfass, K. M. Sundqvist, D. S. Akerib, C. N. Bailey, M. R. Dragowsky, D. R. Grant, R. Hennings-Yeomans, R. W. Schnee, First test runs of a dark-matter detector with interleaved ionization electrodes and phonon sensors for surface-event rejection, Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment 559 (2) (2006) 414 -416, proceedings of the 11th International Workshop on Low Temperature Detectors LTD-11 11th International Workshop on Low Temperature Detectors. doi:http://dx.doi.org/10.1016/j.nima.2005.12.026 .
- URL http://www.sciencedirect.com/science/article/pii/ S0168900205024162
- [18] M. Pyle, B. Serfass, P. L. Brink, B. Cabrera, M. Cherry, N. Mirabolfathi, L. Novak, B. Sadoulet, D. Seitz, K. M. Sundqvist, et al., Surface electron rejection from ge detector with interleaved charge and phonon channels,
in: The Thirteenth International Workshop on Low Temperature DetectorsLTD13, Vol. 1185, AIP Publishing, 2009, pp. 223-226.
- [19] R. Agnese, A. J. Anderson, D. Balakishiyeva, R. B. Thakur, D. A. Bauer, A. Borgland, D. Brandt, P. L. Brink, R. Bunker, B. Cabrera, et al., Demonstration of surface electron rejection with interleaved germanium detectors for dark matter searches, Applied Physics Letters 103 (16) (2013) 164105.
- [20] B. Shank, D. Q. Nagasawa, B. Cabrera, M. Cherry, B. A. Young, Charge transport asymmetry in cryogenic high purity germanium, Journal of Low Temperature Physics (2014) 1-7 doi:10.1007/s10909-014-1111-y . URL http://dx.doi.org/10.1007/s10909-014-1111-y
- [21] S. Hertel, Advancing the search for dark matter: from CDMS II to SuperCDMS, Ph.d. Dissertation, Massachusetts Institute of Technology.
- [22] Shipley, Shipley Microposit S1800 Series Photo Resists Data Sheet (May 2012).
URL s1800seriesDataSheet.pdf http://nanofab.ece.cmu.edu/resources/ | 10.1016/j.nima.2014.10.043 | [
"Andrew Jastram",
"Rusty Harris",
"Rupak Mahapatra",
"James Phillips",
"Mark Platt",
"Kunj Prasad",
"Joel Sander",
"Sriteja Upadhyayula"
] | 2014-08-01T20:13:45+00:00 | 2014-09-26T16:05:02+00:00 | [
"physics.ins-det",
"hep-ex"
] | Cryogenic Dark Matter Search Detector Fabrication Process and Recent Improvements | A dedicated facility has been commissioned for Cryogenic Dark Matter Search
(CDMS) detector fabrication at Texas A&M University (TAMU). The fabrication
process has been carefully tuned using this facility and its equipment.
Production of successfully tested detectors has been demonstrated. Significant
improvements in detector performance have been made using new fabrication
methods/equipment and tuning of process parameters. |
1408.0296v2 | ## Solvable Relativistic Hydrogenlike System in Supersymmetric Yang-Mills Theory
Simon Caron-Huot
Niels Bohr International Academy and Discovery Center, Blegdamsvej 17, Copenhagen 2100, Denmark and Institute for Advanced Study, Princeton, New Jersey 08540, USA ∗
Johannes M. Henn Institute for Advanced Study, Princeton, New Jersey 08540, USA † (Dated: January 19, 2015)
The classical Kepler problem, as well as its quantum mechanical version, the hydrogen atom, enjoys a well-known hidden symmetry, the conservation of the Laplace-Runge-Lenz vector, which makes these problems superintegrable. Is there a relativistic quantum field theory extension that preserves this symmetry? In this Letter we show that the answer is positive: in the nonrelativistic limit, we identify the dual conformal symmetry of planar N = 4 super Yang-Mills theory with the well-known symmetries of the hydrogen atom. We point out that the dual conformal symmetry offers a novel way to compute the spectrum of bound states of massive W bosons in the theory. We perform nontrivial tests of this setup at weak and strong coupling and comment on the possible extension to arbitrary values of the coupling.
PACS numbers: 11.15.Pg,11.55.Bq
The classical two-body (or Kepler) problem, with Hamiltonian
$$H = \frac { p ^ { 2 } } { 2 \mu } - \frac { \lambda } { 4 \pi } \frac { 1 } { | x | } \,, \quad \quad ( 1 ) \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \,$$
is well known to possess a nonobvious conserved vector which makes it superintegrable. This Laplace-RungeLenz vector is expressed as
$$\vec { A } = \frac { 1 } { 2 } \left ( \vec { p } \times \vec { L } - \vec { L } \times \vec { p } \right ) - \mu \frac { \lambda } { 4 \pi } \frac { \vec { x } } { | x | } \,, \quad \ \ ( 2 ) \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
\quad \quad \quad$$
where glyph[vector] L = glyph[vector] x × glyph[vector] p is the angular momentum. Physically, its conservation accounts for the fact that the orbits of the 1 / x | | central potential form closed ellipses which do not precess with time.
The same Hamiltonian is relevant for the quantum mechanical description of the hydrogen atom, with glyph[vector] x and glyph[vector] p replaced by operators. As was pointed out early on by Pauli, the Laplace-Runge-Lenz vector in the above form is also conserved quantum mechanically; i.e., it commutes with the Hamiltonian. The symmetry group is enlarged from SO (3) rotations to SO (4). This gives rise to a simple algebraic way of calculating the spectrum, which automatically accounts for its degeneracies [1].
In real hydrogen atoms, both this symmetry and its associated degeneracies are approximate due to relativistic effects whose size are of order m α e 4 , where α is the finestructure constant and m e the electron mass. Is there a relativistic quantum field theory which has an exact symmetry generalizing the conservation of the LaplaceRunge-Lenz vector? In this Letter we will show that such a system exists and use the additional symmetry to facilitate the calculation of its spectrum.
To understand how to formulate the symmetry (2) relativistically, it will be helpful to recall the classic work by
Wick and Cutkosky [2]. These authors studied the relativistic Bethe-Salpeter equation for a bound-state wave function ψ ,
$$\sup _ { \intertext { \text{tor} } } & \psi ( p ) = \int \frac { - 4 i \lambda \, m _ { 1 } m _ { 3 } \, \psi ( q ) \, d ^ { 4 } q / ( 2 \pi ) ^ { 4 } } { ( p - q ) ^ { 2 } \left [ ( q - y _ { 1 } ) ^ { 2 } + m _ { 1 } ^ { 2 } \right ] \left [ ( y _ { 3 } - q ) ^ { 2 } + m _ { 3 } ^ { 2 } \right ] } \,, \ \ ( 3 )$$
where ( y 3 -y 1 ) µ = P µ is the total four-momentum of the bound state and ( q -y 1 ) µ and ( y 3 -q ) µ are the momenta of its two constituents. This is a natural relativistic generalization of the Schr¨dinger equation and arises as the o approximation to electron-proton scattering where one retains only all planar ladder diagrams and treats the photon as a spin-0 particle.
Wick and Cutkosky noticed that the equation is invariant under a larger symmetry than the expected SO (3) rotations. In modern language, their findings may be summarized by the statement that Eq. (3) is covariant under the transformations
$$\colon \text{form} \\ \text{mutes} \quad \delta _ { \xi } p ^ { \mu } & = 2 ( \xi \cdot p ) p ^ { \mu } - p ^ { 2 } \xi ^ { \mu } \,, \ \delta \psi ( p ) = - 2 ( \xi \cdot p ) \, \psi ( p ) \,, \\ \text{arged} \quad \delta _ { \xi } y _ { i } ^ { \mu } & = 2 ( \xi \cdot y _ { i } ) y _ { i } ^ { \mu } - ( y _ { i } ^ { 2 } + m _ { i } ^ { 2 } ) \xi ^ { \mu } \,, \ \delta m _ { i } = 2 ( \xi \cdot y _ { i } ) m _ { i } \,.$$
These transformations have a simple interpretation as conformal transformations of the momentum space of the theory. Following recent literature, we will refer to them as dual conformal transformations. Noticing that Eq. (3) is also invariant under translations of ( p, y i ) as well as under Lorentz transformations, one may see that the equation is covariant under a full SO (4 , 2) group.
The transformations (4) can be used to relate solutions which correspond to different masses. In fact, they imply that the dynamics depends only on the combination [2]
$$u = \frac { 4 m _ { 1 } m _ { 3 } } { - s + ( m _ { 1 } - m _ { 3 } ) ^ { 2 } } \,. \quad \quad ( 5 )$$
This provides a generalization of the concept of reduced mass to this particular relativistic setup. The remaining nontrivial predictions of the SO (4 2) symmetry arise , from the subgroup which preserves the masses and y i .
This subgroup is six dimensional since SO (4 2) is 15 , dimensional and ten constraints are imposed, but only nine are independent. Explicitly, in a rest frame where y 1 = 0 and y 3 = ( P , glyph[vector] 0 0), and setting m 1 = m 3 = m = 2 µ without loss of generality, we find that the nontrivial generators are the following combinations of (4), Lorentz boosts, and translations:
$$\delta ^ { \prime } _ { \vec { \xi } } \vec { q } & = \vec { \xi } \left ( q _ { 0 } ^ { 2 } - \vec { q } ^ { 2 } - P ^ { 0 } q ^ { 0 } + m ^ { 2 } \right ) + 2 \vec { \xi } \cdot \vec { q } \vec { q } \quad \text{fig.\ } \\ \delta ^ { \prime } _ { \vec { \xi } } q ^ { 0 } & = \vec { \$$
with a similar transformation for p . By construction, the points ( y , m i i ) are invariant under this symmetry.
These transformations can be interpreted more easily by taking the nonrelativistic limit of the model. It is well known that Eq. (3) reduces to the Schr¨dinger equation o in this limit, a fact which can be demonstrated by approximating the frequency integration by its residue on the 1 / q ( 2 + m 2 ) propagator. Substituting the value of q 0 on the residue, q 0 ≈ m + glyph[vector] q 2 / (2 m ), the transformation (6) is reduced to
$$\delta ^ { \prime } _ { \vec { \xi } } \vec { q } = - \vec { \xi } \left [ m ( P ^ { 0 } - 2 m ) + \vec { q } \, ^ { 2 } \right ] + 2 \vec { \xi } \cdot \vec { q } \, \vec { q } \,. \quad ( 7 )$$
It is easy to see that this is the canonical transformation generated by the Laplace-Runge-Lenz vector (2) [20]. This demonstrates that the symmetries (6), which arose from SO (4 2) conformal transformations in momentum , space, are nothing but a relativistic generalization of the Laplace-Runge-Lenz vector. For more details on the interpretation of the latter, we refer to [3].
Unfortunately, the Wick-Cutkosky model does not define a consistent relativistic theory, as the ladder approximation is not unitary and lacks multiparticle effects. Remarkably, a consistent quantum field theory generalizing the above symmetry does exist. It has been observed that maximally supersymmetric Yang-Mills theory ( N = 4 SYM) with gauge group SU N ( c ), which has a superconformal symmetry, also has a dual momentum space version of this symmetry, in the planar limit [4]. (The planar limit, which we are going to work in, is defined by N c → ∞ , with the 't Hooft coupling λ = g 2 N c held fixed.) As far as we are aware, this is the only known example of a four-dimensional quantum field theory with such a symmetry.
In the usual formulation, this is a theory of massless particles. However, massive particles can be introduced in a natural way via the Higgs mechanism. This allows us to discuss the scattering of massive W bosons. Their masses can be freely adjusted by varying scalar field expectation values. Let us focus on the four-particle scattering amplitude depicted in Fig. 1. This amplitude is finite in the ultraviolet, due to the finiteness
FIG. 1: Four-point amplitude in N = 4 SYM with nontrivial scalar vacuum expectation values. Thick lines correspond to massive W bosons, while dashed lines correspond to massless particles.
of N = 4 SYM, as well as in the infrared, thanks to the particle masses. Dual conformal symmetry implies that the dependence on the kinematical invariants and masses is as follows, for the symmetry breaking pattern SU N ( c ) → SU N ( c -4) × U (1) 4 [5]:
$$A _ { 4 } ( s, t, m _ { 1 }, m _ { 2 }, m _ { 3 }, m _ { 4 } ) = A _ { 4 } ^ { \text{tree} } \times M ( u, v ) \,, \quad ( 8 )$$
where, as a generalization of Eq. (5),
$$\underset { 2 0 ] { \langle \, \downarrow \, \downarrow } { u } = \frac { 4 m _ { 1 } m _ { 3 } } { - s + ( m _ { 1 } - m _ { 3 } ) ^ { 2 } } \,, \quad v = \frac { 4 m _ { 2 } m _ { 4 } } { - t + ( m _ { 2 } - m _ { 4 } ) ^ { 2 } } \,. \quad ( 9 ) \\ \quad \, _ { - \, \, \,$$
In the remainder of this Letter we wish to discuss implications of the structure (8) which, as we have seen, is intimately tied to the Laplace-Runge-Lenz vector, regarding the spectrum of the theory.
As depicted in Fig. 1, the W bosons interact by exchanging massless gauge fields from the unbroken part of the gauge group. One can readily see that the interaction is attractive, so they will form bound states. At weak coupling these are similar to hydrogen states. As in the Wick-Cutkosky model, we may use Eq. (9) to restrict to the case m 1 = m 3 = m .
The exact dual conformal symmetry ensures that the spectrum organizes into complete SO (4) multiplets, nonperturbatively at any coupling λ . The total degeneracy at principal quantum number n , including supersymmetry, is 256 n 2 . To extract the spectrum from the amplitude, we will benefit from relativity by making use of Regge theory [6]. The latter instructs us to group the highest-spin state at each energy E n into a trajectory j ( s ), where j is the spin:
$$\mathbf s \, \text{al-} \quad j ( s _ { n } ) + 1 = n \ \text{when} \ \mathbf s _ { n } = E _ { n } ^ { 2 } \quad ( n = 1, 2, \dots ) \,. \ ( 1 0 )$$
The analytic continuation of the function j ( s ) then determines the behavior of the amplitude in the ultrarelativistic limit t → ∞ with s < 0 fixed, through M ∼ t j ( s )+1 [provided only that j ( s ) remains the leading trajectory
FIG. 2: Different limits of the four-point amplitude that are equivalent thanks to dual conformal symmetry. The double lines denote Wilson lines.
in that region]. Conversely, if one knows j ( s ) by some means, Eq. (10) can be used to determine the spectrum. A traditional way to calculate Regge trajectories perturbatively is to sum logarithmically enhanced graphs. For example, at the leading-logarithmic accuracy, the ladder integrals shown in Fig. 2 dominate and exponentiate in a simple way. The exponent, the gluon Regge trajectory j ( s ), is given by a two-dimensional bubble integral. In principle this calculation could be carried out to subleading orders as well (see, e.g., [8, 9]).
The relativistic Laplace-Runge-Lenz symmetry offers a novel and easier way to calculate the Regge trajectory j ( s ). Through Eq. (9), we see that the limit t →∞ of the amplitude, with all other variables held fixed, is equivalent to the limit m 4 → 0. In this limit the amplitude is known to become infrared divergent and its leading terms are governed by the anomalous dimension Γ cusp of a Wilson loop with a cusp [7], M ∼ ( m 4 ) Γ cusp ( φ ) . Equating the exponents in the two asymptotic limits using Eq. (9), we thus find that
$$j ( s ) + 1 = - \Gamma _ { c u s p } ( \phi ) \ \text{ where } \ s = 4 m ^ { 2 } \sin ^ { 2 } \frac { \phi } { 2 } \,. \ \ ( 1 1 ) \quad \text{but } w \quad \text{limit}$$
This relation has been derived and used previously in Refs. [8, 9], to which we refer the reader for more details. A similar relation is known to give the infrared-divergent part of the gluon trajectory as m 2 → 0 [10], but we stress that in planar N = 4 SYM, Eq. (11) holds for the complete function of s/m 2 .
We wish to combine this relation with Eq. (10) as a means to obtain the spectrum of hydrogenlike bound states in this theory. At the lowest order the cusp anomalous dimension is readily calculated by a one-loop graph that corresponds to one rung in the Wilson line integral of Fig. 2 [21],
$$\Gamma _ { \text{cusp} } ( \phi ) = - \frac { \lambda } { 8 \pi ^ { 2 } } \, \phi \, \tan \frac { \phi } { 2 } + \mathcal { O } ( \lambda ^ { 2 } ) \,. \quad ( 1 2 )$$
To obtain the spectrum we need to solve Eq. (10) or, equivalently, Γ cusp ( φ n ) = -n . From Eq. (12) we see that, since λ is small, the solution can occur only for φ close to π . In this region we have
$$\Gamma _ { \text{cusp} } ( \pi - \delta ) \approx - \frac { \lambda } { 4 \pi \delta } \,, \quad \quad ( 1 3 ) \quad \text{to}$$
so that δ n ≈ λ/ (4 πn ). Converting to an energy using Eqs. (10) and (11), we thus find
$$E _ { n } - 2 m = - \frac { \lambda ^ { 2 } m } { 6 4 \pi ^ { 2 } n ^ { 2 } } + O ( \lambda ^ { 3 } ) \,. \quad \quad ( 1 4 )$$
This is the well-known hydrogenlike spectrum associated with Eq. (1), as expected, giving a first confirmation of the method.
Because the present hydrogenlike system is embedded in a relativistic quantum field theory we expect the spectrum to be sensitive to a rich set of multiparticle effects. For example, one expects large logarithms to arise from so-called ultrasoft virtual particles, in an analogy with the computation of the Lamb shift in QED. These are modes which are infrared compared to the atomic radius, but not compared to the binding energies. In fact, as we will see, closely related effects do appear in the computation of Γ cusp at the next order, which makes a nontrivial resummation necessary.
To carry out this resummation systematically, we borrow methods used in the study of the heavy quark static potential in QCD [12]. But first we will need to use conformal symmetry one more time, now in the coordinate space of the theory. Through radial quantization, conformal symmetry equates the anomalous dimension Γ cusp ( φ ) to the energy of a pair of static heavy quarks on S 3 × R , where the 'time' r ∈ R is the radial distance from the cusp and δ is the distance between the two quarks on the sphere [9, 13]. Combined with the duality (11), we thus have a relation between dynamical quarks in flat space, and static quarks in the curved space S 3 × R . Such relations (in flat space) are generic in the large mass limit, but we wish to stress that here we are not taking such a limit and we are discussing the full, relativistic system. This mapping to the cylinder S 3 × R helps apply standard methods because one is now computing a static potential.
In the regime δ ∼ λ relevant to the bound states, there are two important length scales on the cylinder: the small size of the pair and the (unit) radius of the sphere, the latter being comparable to the singlet-adjoint energy splitting λ/ (4 πδ ). This second fact signals the need for a resummation of perturbation theory. This was carried out to the next-to-leading order in Ref. [9], whose results we borrow:
$$\Gamma _ { \text{cusp} } ( \pi - \delta ) = & \frac { - \lambda } { 4 \pi \delta } \left ( 1 - \frac { \delta } { \pi } \right ) + \frac { \lambda ^ { 2 } } { 8 \pi ^ { 3 } \delta } \log \frac { \epsilon _ { \text{uv} } } { 2 \delta } \\ \text{r,} \\ \text{t,} \\ \text{se} \quad \text{Here } \epsilon _ { \text{uv} } \text{ is a small ultraviolet regular, which cancels}$$
Here glyph[epsilon1] uv is a small ultraviolet regulator, which cancels against a divergence of the integral. In fact we were able to perform the latter analytically. Upon equating the left-hand side to minus an integer we obtain the following
correction to Eq. (14):
$$\left ( E _ { n } - 2 m \right ) | _ { \lambda ^ { 3 } } = & \frac { - \lambda ^ { 3 } m } { 6 4 \pi ^ { 4 } n ^ { 2 } } \left [ S _ { 1 } ( n ) + \log \frac { \lambda } { 2 \pi n } - 1 - \frac { 1 } { 2 n } \right ], \, ( 1 6 ) \quad \begin{array} { c } 5 \\ 4 \\ \end{array} \right ]$$
for n = 1 2 3 , , , . . . , and where S 1 ( n ) = ∑ n k =1 1 k is the harmonic number.
Let us discuss this equation. First, we note that the size of the correction is uniformly bounded as a function of n , and therefore for small λ it is always smaller than the leading term given in Eq. (14). This demonstrates that the perturbative expansion is under control.
Second, we notice the nonanalytic dependence on the coupling through the log λ term. This originates from the ultrasoft modes alluded to earlier and is conceptually similar to the ( m α e 5 log α ) contribution to the Lamb shift in QED. It appears earlier by two powers of the coupling in the present model because ultrasoft scalar exchanges are not dipole suppressed.
↦
Third, the square bracket becomes constant at large n . Its value is in perfect agreement with replacing the coupling λ in Eq. (14) with the (flat space) static potential, λ → λ + λ 2 2 π 2 ( log λ 2 π + γ E -1 ) + O λ ( 3 ) [12], as it should be.
Finally, we wish to mention that we have verified Eq. (16) against a direct next-to-leading order calculation of the spectrum using conventional methods [14]. This confirms, in a nontrivial way, that the method based on Eqs. (10) and (11) provides the correct spectrum.
The duality (11) can also be verified at strong coupling. The cusp anomalous dimension was obtained in semianalytic form in Ref. [15] while the spectrum was obtained in Ref. [16] by solving numerically a differential equation [22], both using the AdS/CFT correspondence. The two formulations appear very different and we were not able to find an analytic match between them. Nonetheless, when we evaluated numerically the two formulas throughout the range 0 < E < 2 m (corresponding to 0 < φ < π ), we found perfect agreement within numerical accuracy.
In Fig. 3 we show the next-to-leading order trajectory at weak coupling [23], as well as the strong coupling formula taken from either one of Refs. [15, 16]. The spectrum is obtained from the curves by solving j n ( s n ) = n -1 [see Eq. (10)]. With increasing coupling the ground state becomes more tightly bound, as expected. The reader should not attribute a deep meaning to the agreement of the two curves at λ = 10 and large spin; this is simply due to the fact that the weak and strong coupling extrapolations of the flat space static potential turn out to cross roughly at this value. The difference in shape between the two curves offers one measure of the current uncertainties at intermediate coupling. At small s the slope is known exactly [17].
As a final application, the Laplace-Runge-Lenz symmetry allows us to extend the conventional SO (3) partial
/Cross
/Cross
/Cross
/Cross
/Cross
/Cross
/Cross
1
FIG. 3: Regge trajectories of Hydrogen-like states in N = 4 SYM theory for λ = 5 10 10 30 100 (bottom to top). , , , , The solid-blue lines use the weak-coupling formulas while the dashed-red lines use the largeλ formulas (see text). The bound states (crosses) are obtained by equating j to an integer. The inset shows the same curves with the total energy in units of mass on the horizontal axis.
wave decomposition of the four-particle amplitude so as to account for the contribution of full SO (4) multiplets, reducing the complexity of the expansion. By analyzing the three-loop results from Ref. [18] in this way, we found evidence that the first subleading power correction in the high-energy limit is controlled by a single Regge pole or, equivalently, via Eq. (11), a single operator of dimension
$$\Gamma _ { 1 } ( \phi ) = & 1 + \lambda / ( 4 \pi ^ { 2 } ) + \mathcal { O } ( \lambda ^ { 2 } ) \,. \quad \ \ ( 1 7 )$$
Details of the analysis and the full three-loop trajectory will be reported elsewhere [14]. This simplicity hints at further structure in the dynamics of this model, which does not directly follow from the Laplace-Runge-Lenz symmetry but which the latter may help uncover.
To conclude, we mention that the cusp anomalous dimension in N = 4 SYM has recently been reformulated in terms of a system of integral equations which embody the integrability of this theory [19]. Combined with the present results, this could lead to an exact determination of the spectrum at finite coupling in this interacting quantum field theory.
Acknowledgments J. M. H. is supported in part by DOEGrant No. DE-SC0009988. S. C. H.'s work was supported in part by NSF Grant No. PHY-1314311. Both authors acknowledge support from the Marvin L. Goldberger Fund.
∗ Electronic address: [email protected]
† Electronic address: [email protected]
[1] W. Pauli, Z. Phys. 36 , 336 (1926).
- [2] G. C. Wick, Phys. Rev. 96 1124 (1954); R. E. Cutkosky, Phys. Rev. 96 1135 (1954).
- [3] H. Goldstein, Classical Mechanics (Addison-Wesley, Reading, MA, 1951); V. Fock, Z. Phys. 98 145 (1935); M. Bander and C. Itzykson, Rev. Mod. Phys. 38 330 (1966).
- [4] J. M. Drummond, J. Henn, V. A. Smirnov, and E. Sokatchev, J. High Energy Phys. 01 (2007) 064 [arXiv:hep-th/0607160]; L. F. Alday and J. M. Maldacena, J. High Energy Phys. 06 (2007) 064 [arXiv:0705.0303 [hep-th]]; J. M. Drummond, J. Henn, G. P. Korchemsky, and E. Sokatchev, Nucl. Phys. B826 , 337 (2010) [arXiv:0712.1223 [hep-th]].
- [5] L. F. Alday, J. M. Henn, J. Plefka, and T. Schuster, J. High Energy Phys. 01(2010) 077 [arXiv:0908.0684 [hep-th]]; S. Caron-Huot and D. O'Connell, J. High Energy Phys. 08 (2011) 014 [arXiv:1010.5487 [hep-th]]; T. Dennen and Y. -t. Huang, J. High Energy Phys. 01 (2011) 140 [arXiv:1010.5874 [hep-th]].
- [6] V. N. Gribov, The Theory of Complex Angular Momenta: Gribov Lectures on Theoretical Physics, Cambridge Monographs on Mathematical Physics (Cambridge University, Cambridge, England, 2003); S. Donnachie, G. Dosch, P. Landshoff, and O. Nachtmann, Pomeron Physics and QCD, Cambridge Monographs on Particle Physics, Nuclear Physics and Cosmology Vol. 19 (Cambridge University, Cambridge, England, 2002).
- [7] G. P. Korchemsky and A. V. Radyushkin, Phys. Lett. B 279 359 (1992) [hep-ph/9203222].
- [8] J. M. Henn, S. G. Naculich, H. J. Schnitzer, and M. Spradlin, J. High Energy Phys. 04 (2010) 038 [arXiv:1001.1358 [hep-th]].
- [9] D. Correa, J. Henn, J. Maldacena, and A. Sever, J. High Energy Phys. 05 (2012) 098 [arXiv:1203.1019 [hep-th]].
- [10] I. A. Korchemskaya and G. P. Korchemsky, Phys. Lett.
- B 387 346 (1996) [hep-ph/9607229].
- [11] A. M. Polyakov, Nucl. Phys. B 164 (1980) 171.
- [12] A. Pineda, Phys. Rev. D 77 021701 (2008) [arXiv:0709.2876 [hep-th]].
- [13] N. Drukker and V. Forini, J. High Energy Phys. 06 (2011) 131 [arXiv:1105.5144 [hep-th]].
- [14] S. Caron-Huot and J. M. Henn (to be published).
- [15] N. Drukker, D. J. Gross, and H. Ooguri, Phys. Rev. D 60 125006 (1999) [arXiv:hep-th/9904191].
- [16] M. Kruczenski, D. Mateos, R. C. Myers, and D. J. Winters, J. High Energy Phys. 07 (2003) 049 [arXiv:hepth/0304032].
- [17] D. Correa, J. Henn, J. Maldacena, and A. Sever, J. High Energy Phys. 06 (2012) 048 [arXiv:1202.4455 [hep-th]].
- [18] S. Caron-Huot and J. M. Henn, J. High Energy Phys. 06 (2014) 114 [arXiv:1404.2922 [hep-th]].
- [19] D. Correa, J. Maldacena, and A. Sever, J. High Energy Phys. 08 (2012) 134 [arXiv:1203.1913 [hep-th]]; N. Drukker, J. High Energy Phys. 10 (2013) 135 [arXiv:1203.1617 [hep-th]].
- [20] The canonical transformation generated by 2 glyph[vector] ξ A · glyph[vector] reproduces Eq. (7) up to a time shift δ t glyph[vector] ξ = - · glyph[vector] ξ glyph[vector] x , consistent with the fact that we considered only the timeindependent Bethe-Salpeter equation (3).
- [21] The formula differs from the corresponding QCD one [11] due to the coupling of the Wilson lines to scalars.
- [22] Reference [16] considered N = 4 SYM coupled to a N = 2 matter multiplet, but their specific string solutions apply to our setup as well.
- [23] To be precise, the plot shows the two-loop fixed-order result of Ref. [9], plus the difference between the resummed expression (15) and the fixed-order result expressed only as a function of the one-loop Γ cusp in Eq. (12), so as to make it well behaved for all values of the angle. | 10.1103/PhysRevLett.113.161601 | [
"Simon Caron-Huot",
"Johannes M. Henn"
] | 2014-08-01T20:14:06+00:00 | 2015-01-16T13:12:48+00:00 | [
"hep-th"
] | Solvable Relativistic Hydrogenlike System in Supersymmetric Yang-Mills Theory | The classical Kepler problem, as well as its quantum mechanical version, the
Hydrogen atom, enjoy a well-known hidden symmetry, the conservation of the
Laplace-Runge-Lenz vector, which makes these problems superintegrable. Is there
a relativistic quantum field theory extension that preserves this symmetry? In
this Letter we show that the answer is positive: in the non-relativistic limit,
we identify the dual conformal symmetry of planar $\mathcal{N}=4$ super
Yang-Mills with the well-known symmetries of the Hydrogen atom. We point out
that the dual conformal symmetry offers a novel way to compute the spectrum of
bound states of massive $W$ bosons in the theory. We perform nontrivial tests
of this setup at weak and strong coupling, and comment on the possible
extension to arbitrary values of the coupling. |
1408.0297v1 | ## Optically controlled waveplate at a telecom wavelength using a ladder transition in Rb atoms for all-optical switching and high speed Stokesmetric Imaging
Subramanian Krishnamurthy 1 , Y. Tu 1 , Y. Wang 1 , S. Tseng , and M.S. Shahriar 1 1,2,* 1 Department of EECS, Northwestern University, Evanston, IL 60208,USA 2 Department of Physics and Astronomy, Northwestern University, Evanston, IL 60208,USA
*[email protected]
Abstract: We demonstrate an optically controlled waveplate at ~1323 nm using the 5S1/2-5P1/2-6S1/2 ladder transition in a Rb vapor cell. The lower leg of the transitions represents the control beam, while the upper leg represents the signal beam. We show that we can place the signal beam in any arbitrary polarization state with a suitable choice of polarization of the control beam. Specifically, we demonstrate a differential phase retardance of ~180 degrees between the two circularly polarized components of a linearly polarized signal beam. We also demonstrate that the system can act as a Quarter Wave plate. The optical activity responsible for the phase retardation process is explained in terms of selection rules involving the Zeeman sublevels. As such, the system can be used to realize a fast Stokesmetric Imaging system with a speed of nearly 5 MHz. When implemented using a tapered nano fiber embedded in a vapor cell, this system can be used to realize an ultra-low power all-optical switch as well as a Quantum Zeno Effect based all-optical logic gate by combining it with an optically controlled polarizer, previously demonstrated by us. We present numerical simulations of the system using a comprehensive model which incorporates all the relevant Zeeman sub-levels in the system, using a novel algorithm recently developed by us for efficient computation of the evolution of an arbitrary large scale quantum system.
© 2012 Optical Society of America
OCIS codes : (020.4180) Multiphoton processes; (020.1670) Coherent optical effects
## References and links
- 1. S. E. Harris and Y. Yamamoto, 'Photon switching by quantum interference,' Phys. Rev. Lett. 81 (17), 3611-3614 (1998).
- 2. R. G. Beausoleil, W. J. Munro, D. A. Rodrigues, and T. P. Spiller, 'Applications of electromagnetically induced transparency to quantum information processing,' J. Mod. Opt. 51 (16-18), 2441-2448 (2004).
- 3. A. M. C. Dawes, L. Illing, S. M. Clark, and D. J. Gauthier, 'All-optical switching in Rubidium vapor,' Science 308 (5722), 672-674 (2005).
- 4. M. Bajcsy, S. Hofferberth, V. Balic, T. Peyronel, M. Hafezi, A. S. Zibrov, V. Vuletic, and M. D. Lukin, 'Efficient All-optical switching using slow light within a hollow Fiber,' Phys. Rev. Lett. 102 (20), 203902 (2009).
- 5. V. Venkataraman, P. Londero, A. R. Bhagwat, A. D. Slepkov, and A. L. Gaeta, 'All-optical modulation of four-wave mixing in an Rbfilled photonic bandgap fiber,' Opt. Lett. 35 (13), 2287-2289 (2010).
- 6. Subramanian Krishnamurthy, Y. Wang, Y. Tu, S. Tseng, and M. S. Shahriar, 'High efficiency optical modulation at a telecom wavelength using the quantum Zeno effect in a ladder transition in Rb atoms,' Opt. Express 20 (13), 13798-13809 (2012).
- 7. S. M. Spillane, G. S. Pati, K. Salit, M. Hall, P. Kumar, R. G. Beausoleil, and M. S. Shahriar, 'Observation of nonlinear optical interactions of ultralow levels of light in a tapered optical nanofiber embedded in a hot Rubidium vapor,' Phys. Rev. Lett. 100 (23), 233602 (2008).
- 8. K. Salit, M. Salit, Subramanian Krishnamurthy, Y. Wang, P. Kumar, and M. S. Shahriar, 'Ultra-low power, Zeno effect based optical modulation in a degenerate V-system with a tapered nano fiber in atomic vapor,' Opt. Express 19 (23), 22874-22881 (2011).
- 9. Subramanian Krishnamurthy, Y. Wang, Y. Tu, S. Tseng 1 , and M. S. Shahriar, 'Optically controlled polarizer using a ladder transition for high speed Stokesmetric Imaging and Quantum Zeno Effect based optical logic,' Opt. Express 21 (21), 24514-24531 (2013)
- 10. T. Nee and S. F. Nee, 'Infrared polarization signatures for targets,' Proc. SPIE 2469 , 231-241 (1995).
- 11. S. Y. Lu and R. A. Chipman, 'Interpretation of Mueller matrices based on polar decomposition,' J. Opt. Soc. Am. A 13 (5), 1106-1113 (1996).
- 12. X. Liu, S. C. Tseng, R. Tripathi, A. Heifetz, Subramanian Krishnamurthy, and M. S. Shahriar, 'White light interferometric detection of unpolarized light for complete Stokesmetric optical coherence tomography,' Opt. Commun. 284 (14), 3497-3503 (2011).
- 13. X. Liu, A. Heifetz, S. C. Tseng, and M. S. Shahriar, 'High-speed inline holographic Stokesmeter imaging,' Appl. Opt. 48 (19), 38033808 (2009).
- 14. S. Weilandy, L. Alexander, and L. Gaeta, 'Coherent control of the polarization of an optical field,' Phys. Rev. Lett. 81 (16), 3359-3362 (1998).
- 15. M. Xiao, Y. Li, S. Jin, and J. Gea-Banacloche, 'Measurement of dispersive properties of electromagnetically induced transparency of Rubidium atoms,' Phys. Rev. Lett. 74 (5), 666-669 (1995).
- 16. Shujing Li, Bo Wang, Xudong Yang, Yanxu Han, Hai Wang, Min Xiao, and K. C. Peng, 'Controlled polarization rotation of an optical field in multi-Zeeman-sublevel atoms', Phys. Rev. A 74 , 033821 (2006).
- 17. D. A. Steck, 'Alkali D line data,' http://steck.us/alkalidata/rubidium87numbers.pdf
- 18. M. S. Shahriar, Y. Wang, Subramanian Krishnamurthy, Y. Tu, G. S. Pati, and S. Tseng , 'Evolution of an N-level system via automated vectorization of the Liouville equations and application to optically controlled polarization rotation,' J. Mod. Optics 61 (4) , 351-367 (2014)
## 1. Introduction
All-optical switching is important for optical communication and quantum information processing [1-5]. We recently demonstrated a ladder type modulator and proposed a novel scheme (employing high pressure buffer gas to broaden the atomic transitions) for a high-speed modulator (tens of GHz) that can be used to modulate signals at 1323 nm, using a control beam at 795 nm [6]. In addition, we have also described experimental schemes to realize such modulators at ultra-low powers (~40 pW) using a tapered nano fiber (TNF) [7, 8]. However, these modulators cannot be used as directional switches, as required in an optical communication network. In this paper, we report on the realization of an optically controlled waveplate and describe how this can be used as an all-optical switch in the telecommunication band. In combination with an optically controlled polarizer, recently demonstrated by us [9], the system can also be used to produce a Quantum Zeno Effect (QZE) based all-optical logic gate.
The optically controlled waveplate also has potential applications in Stokesmetric Imaging (SI). It is well documented that in many situations of interest, features indiscernible via conventional imaging become highly resolved under SI [10-13]. In a typical SI scenario, a target is illuminated by fully or partially polarized light. The light scattered or reflected by the target is then analyzed using a Stokesmeter, which determines the magnitude of each of the four Stokes parameter components. Stokesmeters, in their simplest form, are comprised of a combination of polarizers and wave-plates with different orientations. A key problem with the existing SI systems is that the polarizers and waveplates cannot be turned on or off or reoriented rapidly. The free space version of the optically controlled waveplate that is described in this paper has the potential to operate at speeds of ~5 MHz and thus holds the promise of making very high speed SI practical.
Rest of the paper is organized as follows. In section 2, we describe an ideal schematic for an optically controlled waveplate, using only four levels. A discussion of the non-idealities in the system and a comprehensive model that includes all the relevant Zeeman sub-levels are presented in section 3. We describe the experimental set-up in section 4 and a Jones matrix analysis of the system is provided in section 5. Experimental results and numerical simulations are presented in section 6. In section 7, we discuss the future work that will be pursued. Finally, in section 8, we present our conclusions.
## 2. Schematic of optically controlled waveplate and all-optical switch
In an atomic system involving ladder type transitions, the presence of two different frequencies open up the possibility of controlling the behavior of the probe (upper leg) polarization by careful design of the pump parameters (lower leg). In particular, it is possible to make the vapor cell act as a waveplate. The mechanism for producing controlled polarization rotation, is illustrated schematically in Fig. 1, using a cascaded atomic transition involving four levels where direct excitation to the upper level from the ground state is dipole forbidden. We consider the mF=0 Zeeman sublevel of a certain hyperfine level into which the atoms have been optically pumped, as the ground state. The lower leg is excited by a right-circularly polarized ( +) beam (control beam), tuned a few GHz below the transition frequency to the intermediate level. The control beam, therefore, produces an off-resonant excitations to only the mF=1 Zeeman sublevel in the intermediate state. The signal beam (probe), applied between the intermediate level and the upper level is chosen to be linearly polarized and hence has two components: + and -. The -component sees the effect of the atoms and, because of the detuning, it sees only a real susceptibility with virtually no absorption, while the + component of the probe does not see any effect of the pump. The parameters of the control beam can be tuned to achieve the condition for a phase-shift for the -component only, so that at the output the linear polarization is rotated by 90 degrees.
Fig.1. Schematic illustration of an optical switch using an optically controlled waveplate. See text for details.
It should be noted that the above schematic can only be used as a polarization rotator for a linear polarization of the probe. For example, it can be seen that it is impossible to obtain a purely circularl polarization of the probe, as it would require complete absorption on one of the legs of the upper transition. In fact, a simple Jones matrix analysis of the above system (in the absence of absorption) shows that the polarization state of the probe is 𝑠𝑖𝑛 𝜙 𝑥 ̂ + 𝑐𝑜𝑠 𝜙 𝑦 ̂ , where 𝜙 is the phase difference introduced between the two legs of the signal beam. In order for the system to behave as a quarter wave plate, for example, one needs to apply a linearly polarized pump. As shown in Fig. 2, in this scenario, both the mF = +1 (|2>) and mF = -1 (|3>) states get coupled to the mF = 0 (|1>) state and the effect produced is not transparent in the simple schematic discussed above. However, by a suitable rotation of the intermediate states, one can reduce the process to a similar scenario, where for any arbitrary polarization of the pump, only one of the intermediate states is coupled, while the other is decoupled. Explicitly, any arbitrary polarization of the pump can be represented as 𝛼𝜎 ̂ + +𝛽𝜎 ̂ -, where 𝛼 and 𝛽 can be complex. This couples |1> to α*|2>+ β*|3> (denoted as |+>), where '*' denotes complex conjugation. The orthogonal state, |-> = α|3>β|2>, is not coupled by the pump, as can be verified by explicit calculation. A similar analysis can be carried out for the probe field as shown in Fig. 2. Now the situation is formally identical to the one presented in Fig. 1, where one of the legs sees the effect of the pump while the other does not.
For illustrative purposes, let us consider a probe polarized along the 𝑦 ̂ direction. Denoting 𝜎 ̂ -= 𝑥 ̂ - 𝑖𝑦/√2 ̂ and 𝜎 ̂ + = -𝑥 ̂ - 𝑖𝑦/√2 ̂ , one obtains 𝑦 ̂ = 𝑖(𝜎 ̂ -+𝜎 ̂ + )/√2 . Decomposing 𝑦 ̂ in terms of the rotated basis and introducing a phase shift ϕ on the field coupling the |+> state to |4>, one obtains the following expression for the polarization state ( 𝑝̂) of the probe
$$\hat { p } = i \left [ ( \alpha ^ { * } + \beta ^ { * } ) ( \alpha \hat { \sigma } _ { - } + \beta \hat { \sigma } _ { + } ) e ^ { i \Phi } + ( \alpha - \beta ) ( \alpha ^ { * } \hat { \sigma } _ { + } + \beta ^ { * } \hat { \sigma } _ { - } ) \right ] / \sqrt { 2 }$$
$$= i \left [ \left ( ( | \alpha | ^ { 2 } + \beta ^ { * } \alpha ) e ^ { i \phi } + ( | \beta | ^ { 2 } - \beta ^ { * } \alpha ) \right ) \hat { \sigma } _ { - } + \left ( ( | \beta | ^ { 2 } + \alpha ^ { * } \beta ) e ^ { i \phi } + ( | \alpha | ^ { 2 } - \alpha ^ { * } \beta ) \right ) \hat { \sigma } _ { + } \right ] / \sqrt { 2 } \, \quad \ \ ( 1 )$$
Using appropriate values of α,β and ϕ, one can obtain circular polarization. For example, consider a pump polarized linearly at 45 , for which α=i/(i-1) and β=1/(i-1). If we now set ϕ=π/2, one obtains 0 𝜎 ̂ + polarization for the probe, as can be verified by explicit substitution in equation 1. In its most general form, the probe can also be in some arbitrary polarization state. All one needs to do is to resolve it in terms of the new rotated orthogonal states and introduce a phase shift on one of the legs, leaving the other unaffected. By a suitable choice of the pump polarization, one can place the probe in any desired polarization state.
In our set-up, we utilized the 5S1/2-5P1/2-6S1/2 cascade system in 87 Rb atoms, with F=2 as the ground state. The pump and probe beams are at 795 nm and 1323nm respectively. Theoretical and experimental investigations of an optically controlled waveplate using ladder transitions in Rb have been carried out previously [14, 15]. However, both of these works employ the EIT effect where the upper leg is excited by a strong control field while the lower leg is probed by a weak optical field, and thus has fundamentally different characteristics than the system we have considered. Of course, the primary reason for choosing the upper leg as the probe is the need for an all-optical switch at a telecommunication wavelength.
In Fig. 1 and Fig.2, we showed a simplified set of energy levels in order to explain the basic process behind an optically controlled waveplate. In practice, however, it is extremely difficult to realize such an ideal system. We first note that it is virtually impossible to optically pump all the atoms into the , 5S1/2, F=2, mF = 0 Zeeman sublevel. Hence, Zeeman sub-levels other than mF= ± 1 at the 5P1/2 manifold also get coupled with the pump and probe optical fields. Furthermore, it is generally necessary to take into account both hyperfine levels (F'=1 and F'=2) in the 5P1/2 manifold to account for Doppler broadening and power broadening. Thus, the full set of energy levels that need to be considered is quite large, and the actual model employed for our system is discussed in the next section. For the remainder of the paper, the hyperfine levels in the ground state are indicated by unprimed alphabets (F), those in the 5P1/2 level are primed (F') and those in the 6S1/2 level are double-primed (F').
## 3. Comprehensive model used for numerical simulation
In previous analyses of similar systems, a simple model consisting of only the relevant hyperfine levels transitions was employed [14, 15]. Other works [16] consider some of the Zeeman sub-levels, but make use of few assumptions to eliminate some of the density matrix elements to arrive at a somewhat approximate result. In our model, we considered all the Zeeman sub-levels which explicitly interact with an optical field (all sub-levels of the F=2, F'=1,2 and F'=1 hyperfine levels), while the F=1 hyperfine level and the 5P3/2 level were only considered as population transfer levels and hence all their sub-levels were lumped together as a single level. The full set of energy levels that we have incorporated in our model are shown in Fig. 2. The transition strengths [17] indicated are expressed as multiples of the weakest transition, which in our case is the transition from the F=2, mF=0 sub-level to the F'=1, mF=1 sub-level, for example. We assume that the control beam is tuned below the F=2 F'=1 transition while the signal beam is detuned by an amount δs from the F'=1 F'=1 transition. Due to the Doppler width and power broadening, the F'=2 hyperfine level also interacts with both the control and the signal optical fields (indicated by dashed lines), albeit at a large detuning, and these interactions have been taken into account in our model. However, we ignored the coherent coupling between F=1 and the 5P1/2 manifold, because of the large frequency difference between F=1 and F=2 (~6.8GHz for 87 Rb).
Fig. 2 Model used for numerical computation. See text for more details.
All the Zeeman sub-levels in the 5P1/2 (6S1/2) manifold are assumed to decay at the same rate, a ~5.75 MHz ( b ~3.45 MHz). We also assume a nominal cross-relaxation rate ( g ~0.01 MHz) between the ground states. The decay rate between any two Zeeman sub-levels was calculated by using the fact that it is proportional to the square of the matrix element of the corresponding transition, and that the sum of all such decays rates from the decaying level must equal the net decay from that level. For example, consider mF=0, F'=2 sub-level which decays at a rate a . The transition strengths for the σ+, σ-, and π-transitions to the F=2 (F=1) are in the ratio √3:√3:0 (1:1:2). Thus, the net decay rate between the mF=0, F'=2 sub-level to the mF=-1,+1 and 0 states of the F=2 level were computed to be a /4, a /4 and 0 respectively and the decay to the F=1 level was computed to be a /2, since all the hyperfine levels in the F=1 state are lumped together as a single state in our model. We have also considered the sourcing of atoms into the ground states from the 6S1/2 manifold via the 5P3/2 state. A detailed calculation, taking into account the various branching ratios into and from all the hyperfine levels of the 5P3/2 state was used to determine these 'effective decay rates' directly from the 6S1/2 states to the ground states. Table 1 shows these 'effective decay rates' from each of the Zeeman sub-levels in the 6S1/2 manifold to the ground states. The ratio between the rates of decay into the 5P1/2 and 5P3/2 states from a given upper level was decided by the ratio of the explicit values of the transition strength of the D1 and D2 lines [17].
Table 1. Effective Decay rates between excited states and ground states
| | | F'=2 | F'=2 | F'=2 | F'=2 | F'=2 | F'=1 | F'=1 | F'=1 |
|-----|---------|---------|----------|----------|----------|---------|-----------|-----------|-----------|
| | | m F =-2 | m F =-1 | m F =0 | m F =1 | m F =2 | m F =-1 | m F =0 | m F =1 |
| | m F =-2 | 0.68852 | 0.19426 | 0.05055 | 0 | 0 | 0.2361 | 0.09722 | 0 |
| | m F =-1 | 0.19426 | 0.47296 | 0.190277 | 0.07583 | 0 | 0.1667 | 0.11805 | 0.04861 |
| F=2 | m F =0 | 0.05055 | 0.190277 | 0.45166 | 0.190277 | 0.05055 | 0.104167 | 0.125 | 0.104167 |
| | m F =1 | 0 | 0.07583 | 0.190277 | 0.47296 | 0.19426 | 0.04861 | 0.11805 | 0.1667 |
| | m F =2 | 0 | 0 | 0.05055 | 0.19426 | 0.68852 | 0 | 0.09722 | 0.2361 |
| | m F =-1 | 0.04722 | 0.03333 | 0.02083 | 0.009722 | 0 | 0.21296 | 0.1226875 | 0.1088 |
| F=1 | m F =0 | 0.01944 | 0.023611 | 0.025 | 0.023611 | 0.01944 | 0.1226875 | 0.199 | 0.1226875 |
| | m F =1 | 0 | 0.009722 | 0.02083 | 0.03333 | 0.04722 | 0.1088 | 0.1226875 | 0.21296 |
We used the Liouville equation, which describes the evolution of the density matrix in terms of a commutator between the density matrix and the Hamiltonian augmented by the phenomenological determined decay rates, to obtain the steady-state solution. The usual method of vectorizing the density matrix and then inverting the coefficient matrix thus obtained, is not easy to handle as the size of the coefficient matrix is very large (400*400). In order to overcome this problem, we made use of a novel algorithm, recently developed by us [18] which would compute the said coefficient matrix automatically in a very efficient manner, given the Hamiltonian and the source matrix. While averaging over the Doppler profile, we used the supercomputing cluster at Northwestern (QUEST) to perform our computations. Using 64 cores and computing the steady state solution for 512 values of detuning, each averaged over 800 points of the Doppler profile, we obtained the steady-state solution for our 22-level system in 3-4 minutes.
## 4. Experimental set-up
The experimental configuration is illustrated schematically in Fig. 4. Briefly, beams from two tunable lasers (one at 795 nm, and the other at 1323 nm) are combined with a dichroic mirror (DCM). A part of the 795 light is sent to a reference vapor cell for saturated absorption spectroscopy and locking. The combined beams are sent through a vapor cell, shielded from magnetic fields with -metal. The cell is heated using bifiliarly wounded wires that do not add any magnetic fields. After passing through the cell, another DCM is used to split the light into two parts, and each frequency is detected with a separate detector. The control beam at 795 nm, initially polarized linearly, is passed through a quarter-wave-plate in order to produce circular polarization. The polarization of the signal beam, at 1323 nm, is controlled separately with a half-wave-plate. Ideally, the 1323 nm laser would also be locked at a particular frequency but this laser was found to be stable, so that locking it was not necessary.
Fig. 4. Experimental Set-up
In order to analyze the polarization of the signal beam after passing through the cell, we inserted an analyzer before detector A, consisting of a voltage-controlled liquid crystal retarder (LCR), whose fast axis is placed at 45 degrees to the initial polarization direction (say 𝑦 ̂) of the signal beam, followed by a polarizer with its axis orthogonal to initial polarization of the signal beam. Having a orthogonal polarizer at the output is, in general, not enough, as one cannot determine if the observed signal is a manifestation of polarization rotation alone or a combination of both rotation and absorption, unless of course the detector signal is at least as large as the far off-resonant signal. However, it can serve as a diagnostic tool in identifying the regions of large phase shift as the probe is scanned across the 6S1/2 manifold. The LCR produces a phase retardance between its orthogonal axes that depends non-linearly on the amount of voltage applied to the LCR controller which can determined from a calibration curve provided by the manufactured and verified independently by us. During our experiment, the control voltage to the LCR is scanned linearly from 0 V to 10 V, with 2 V and 8V corresponding to a phase shifts of approximately π and 0, respectively for the wavelength that we are using. This particular arrangement of the analyzer provides us with a very large set of data points (corresponding to the LCR scan) from which to obtain the values of absorption coefficients and the phase rotations, thus making it more robust against noise in the system. The signal observed at the detector A can be ascertained by performing a Jones matrix analysis of the entire system, which is discussed next.
## 5. Jones matrix analysis
Let Ein be the signal field amplitude before the Rb cell and let 𝑦 ̂ be its polarization direction. Let 𝜎 ̂ + = -( 𝑥 ̂+𝑖𝑦 ̂ √2 ) and 𝜎 ̂ -= ( 𝑥 ̂-𝑖𝑦 ̂ √2 ) be the unit vectors corresponding to right (RCP) and left circular polarizations (LCP) respectively. Then, Ein can be represented in the circular polarization basis as 𝐸 ⃗ 𝑖𝑛 = 𝐸0𝑦 ̂ = 𝐸 𝑖𝑛+ 𝜎 ̂ ++𝐸𝑖𝑛-𝜎 ̂ -, where Ein+ = Ein-= 𝑖 𝐸 0 √2 / and E0 is some arbitrary value. The field amplitude after the cell can be represented, in its most general form as 𝐸 ⃗ ⃗ ⃗ 𝑎𝑓𝑡𝑒𝑟 𝑐𝑒𝑙𝑙 = 𝐸𝑖𝑛+ 𝑒 (-𝛼++𝑗𝜙+) 𝜎 ̂ ++𝐸𝑖𝑛-𝑒 (-𝛼-+𝑗𝜙- ) 𝜎 ̂ -, where (𝛼+, 𝜑 +) and (𝛼-, 𝜑 -) are the attenuations and phase rotations for the RCP and LCP part of the signal beam respectively after passing through the
Rb vapor cell. With 𝑥 ̂ and 𝑦 ̂ as the basis for the Jones vector representation and after some algebraic manipulation, we find that the field amplitude after the cell can be represented as
$$E _ { \text{after cell} } = & \frac { i E _ { 0 } } { 2 } e ^ { ( - \alpha _ { - } + j \phi _ { - } ) } \begin{bmatrix} - e ^ { ( - \alpha _ { d } + j \phi _ { d } ) } + 1 \\ - i \left ( e ^ { ( - \alpha _ { d } + j \phi _ { d } ) } + 1 \right ) \end{bmatrix}, \text{where} \begin{array} { c } \alpha _ { d } = \alpha _ { + } - \alpha _ { - } \\ \phi _ { d } = \phi _ { + } - \phi _ { - } \end{array} \\ \intertext { e a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a n d a }$$
Thus, α d and ϕ d represent the differential absorption and phase rotation between the RCP and LCP parts of the signal beam. If θ represents the phase retardation produced by the LCR, then the Jones matrix for the LCR is given by
$$J _ { \alpha } = \begin{bmatrix} \exp ( i \theta / 2 ) & 0 \\ 0 & \exp ( - i \theta / 2 ) \end{bmatrix} _ { \alpha }$$
and the Jones matrix for the LCR whose axis is rotated by 45 0 is given by JLCR45=R -1 (45 0 )JLCRR(45 0 ) where R(45 0 ) represents the rotation matrix for 45 and is given by 0
$$R ( 4 5 ) = \begin{bmatrix} \frac { 1 } { \sqrt { 2 } } & \frac { 1 } { \sqrt { 2 } } \\ \frac { - 1 } { \sqrt { 2 } } & \frac { 1 } { \sqrt { 2 } } \end{bmatrix} \\ \text{evice has to be some more.}$$
Finally, the polarizer with its axis parallel to the 𝑦 ̂ axis has the Jones matrix representation
$$J _ { X P o l } = \begin{bmatrix} 1 & 0 \\ 0 & 0 \end{bmatrix} _ { \cdot }$$
Thus, the Jones vector for the signal observed at the detector A would be
Jout =JYPol* JLCR45*Eaftercell
Performing the calculations, we find that the intensity as seen by the detector A is given by,
$$I = \frac { E _ { 0 } } { 4 } e ^ { - 2 \alpha _ { - } } ( 1 + e ^ { - 2 \alpha _ { d } } + ( 1 - e ^ { - 2 \alpha _ { d } } ) \sin \theta - 2 e ^ { - \alpha _ { d } } \cos \phi _ { d } \cos \theta ) \\ \varepsilon \cap \varphi _ { d } \colon \, T \colon \, \sin \alpha _ { d } \cap \varphi _ { d } \colon \, \varepsilon \cap \varphi _ { d } \colon \, \varepsilon \cap \varphi _ { d } \colon \, \varepsilon \cap \varphi _ { d } \colon \, \varepsilon \cap \varphi _ { d } \colon \, \varepsilon \cap \varphi _ { d } \colon \, \varepsilon \cap \varphi _ { d } \colon \, \varepsilon \cap \varphi _ { d } \colon \, \varepsilon \cap \varphi _ { d } \colon \, \varepsilon \cap \varphi _ { d } \colon \, \varepsilon \cap \varphi _ { d } \colon \, \varepsilon \cap \varphi _ { d }.$$
In Fig. 5(a), I is plotted for different values of ϕ d (in degrees) and α d =0 as the LCR phase retardance θ varies from 0 (left-end) to π (center) and back to 0 (right-end). As is evident from the figure, the signature for increasing differential phase rotation is the upward shift of the minimum and downward shift of the maximum of the curve until ϕ d = 90 0 , at which point the signal is perfectly flat. For greater values of ϕ d , the shape of the curve gets inverted until ϕ d = π. On the other hand, for non-zero values of α d and ϕ d =0, the minima of the curves get shifted inwards, the curves slope upward on either side of the minima and the central part of the curve is flattened out, as shown in Fig. 5(b). For nonzero values of both α d and ϕ d , the interpretation is not so straight-forward and one has to use 3 data points and invert the expression for I to obtain their values. The algebra is somewhat involved and we present only the final result -
$$y = e ^ { \alpha _ { d } } = \sqrt { \frac { I _ { 2 } ^ { 2 } ( C _ { 1 } - C _ { 3 } - S _ { 3 - 1 } ) + I _ { 1 } I _ { 2 } ( C _ { 3 } - C _ { 2 } - S _ { 2 - 3 } ) + I _ { 2 } I _ { 3 } ( C _ { 2 } - C _ { 1 } - S _ { 1 - 2 } ) } { I _ { 2 } ^ { 2 } ( C _ { 3 } - C _ { 1 } - S _ { 3 - 1 } ) + I _ { 1 } I _ { 2 } ( C _ { 2 } - C _ { 3 } - S _ { 2 - 3 } ) + I _ { 2 } I _ { 3 } ( C _ { 1 } - C _ { 2 } - S _ { 1 - 2 } ) } } \ a n d \\ \cos \phi _ { d } = \frac { ( I _ { 2 } - I _ { 1 } ) \left ( y + \frac { 1 } { y } \right ) + ( I _ { 1 } S _ { 2 } - I _ { 2 } S _ { 1 } ) \left ( \frac { 1 } { y } - y \right ) } { 2 \left ( I _ { 2 } C _ { 1 } - I _ { 1 } C _ { 2 } \right ) }, \\ \text{Where } C _ { j } = \cos \left ( \theta _ { j } \right ), \, S _ { j } = \sin \left ( \theta _ { j } \right ), \, \text{and} \, S _ { j - k } = \sin \left ( \theta _ { j } - \theta _ { k } \right ), \, \text{where } \theta _ { j } \text{ is the phase rotations produced for some}$$
Where Cj= cos (θj), Sj= sin (θj), and Sj-k = sin (θj -θk), where θj is the phase rotations produced for some voltage, and Ij is the corresponding intensity seen by the detector.
Fig. 5 Ideal Output seen by detector for different values of a) ϕ d with α + = α -= 0 and b) α -with ϕ d = α += 0
## 6. Results
The exact spectroscopic details of the 5P1/2-6S1/2 transitions depend critically on the control beam intensity and detuning, the temperature of the cell and probe detuning. Thus, as a diagnostic tool to identify regions of high phase shift, an orthogonal polarizer was placed after the Rb cell. As the probe laser was scanned across the 6S1/2 manifold, the detuning of the control beam was varied in order to maximize the transmission through the orthogonal polarizer over the largest possible bandwidth. Then, with the control beam and signal lasers positioned at the detunings previously determined, the LCR control voltage was scanned in order to estimate the differential phase shift and attenuation.
Figure 6(b) shows the transmission through an orthogonal polarizer as the probe laser was scanned over ~5 GHz. The control beam was right circularly polarized and the signal beam was vertically polarized. The pump was placed at a detuning of about 1.2 GHz. The temperature of the cell was maintained around 130 o Celsius and pump power was about 600 mW, obtained from a Ti-Sapphire laser. The probe laser was about 1 mW obtained from a fiber coupled semiconductor laser. Both beams were focused to a spot size of about 50 μm near the center of the Rb cell. The frequency scan was then stopped and with the probe laser fixed at the regions of high phase shift, the LCR control voltage was scanned linearly from 10V to 0V and back up to 10V.The blue trace (normalized from 0 to 1) corresponds to the situation when the control beam was blocked and can thus be treated as the reference signal, corresponding to 0 phase retardance. When the control beam is unblocked, our system acts as an optically controlled waveplate and the red trace is obtained. Comparing with the theoretical plots, one can see that the phase shifts in Fig. 6(a) and Fig. 6(c) correspond to ϕ d ~ 160 0 and ϕ d ~ 180 0 , respectively with α d ~0. The values obtained are consistent with those obtained using the analytical expressions.
Fig. 6. Observed experimental data showing differential phase rotation of ~ 60 0 and almost no differential absorption. See text for more details
Fig. 7 shows the result obtained from numerical simulations using the model presented in section 2. We perform our calculations by setting Γa to unity and rescaling all parameters in units of Γa. The pump is resonant with the F=1 to F'=2 transition and the probe detuning (δs) ranges from -1200 𝛤 𝑎 to 1200 𝛤 𝑎 . Fig. 7(a) and 7(b) show the phase shift of the RCP and LCP parts of the signal beam introduced by the Rb medium and Fig. 7(c) shows the difference between them. Fig. 7(d)-7(f) show the corresponding figures for attenuation. As is evident from the figure, at a pump detuning of ~1.2 GHz and for Rb density of 10 12 /cm 3 and cell length of 7.5 cm, we can produce a phase shift of about 180 with 0 minimal differential absorption. The relevant parameters used for this particular simulation are as follows- the decay rates Γa, Γb and Γg are 2π*5.75 sec -1 , 2π*3.45 sec -1 and 2π*0.1 sec -1 respectively. The separation Δ, between F'=1 and F'=2 is 2π*814.5 sec -1 ( = 141.4𝛤 𝑎 ) and the Rabi frequencies have been chosen to be Ω𝑝 = 100𝛤 𝑎 , and Ω𝑠 = 0.1𝛤 𝑎 . The expressions used to calculate the attenuation and additional phase retardance introduced by the Rb medium are given by
## Phase Shift:
$$\text{max sum} \colon & \\ \phi _ { + } & = k L \frac { \beta _ { + } } { 2 } R e ( a _ { 1 3, 4 } \rho _ { 1 3, 4 } + a _ { 1 4, 5 } \rho _ { 1 4, 5 } + a _ { 1 2, 7 } \rho _ { 1 2, 7 } + a _ { 1 3, 8 } \rho _ { 1 3, 8 } + a _ { 1 4, 9 } \rho _ { 1 4, 9 } ) \\ \phi _ { - } & = k L \frac { \beta _ { - } } { 2 } R e ( a _ { 1 2, 5 } \rho _ { 1 2, 5 } + a _ { 1 3, 6 } \rho _ { 1 3, 6 } + a _ { 1 2, 9 } \rho _ { 1 2, 9 } + a _ { 1 3, 1 0 } \rho _ { 1 3, 1 0 } + a _ { 1 4, 1 1 } \rho _ { 1 4, 1 1 } )$$
Attenuation:
$$\alpha _ { + } & = \ k L \beta _ { + } I m \left ( a _ { 1 3, 4 } \rho _ { 1 3, 4 } + a _ { 1 4, 5 } \rho _ { 1 4, 5 } + a _ { 1 2, 7 } \rho _ { 1 2, 7 } + a _ { 1 3, 8 } \rho _ { 1 3, 8 } + a _ { 1 4, 9 } \rho _ { 1 4, 9 } \right ) / 2 \\ \alpha _ { - } & = \ k L \beta _ { - } I m \left ( a _ { 1 2, 5 } \rho _ { 1 2, 5 } + a _ { 1 3, 6 } \rho _ { 1 3, 6 } + a _ { 1 2, 9 } \rho _ { 1 2, 9 } + a _ { 1 3, 1 0 } \rho _ { 1 3, 1 0 } + a _ { 1 4, 1 1 } \rho _ { 1 4, 1 1 } \right ) / 2$$
And
$$\beta _ { \pm } = b _ { m i n } ^ { 2 } \frac { 3 n _ { a t o m } \Gamma \lambda ^ { 3 } } { 4 \pi ^ { 2 } \Omega _ { m i n } },$$
where 𝑘 is the wavevector of signal beam, L is the length of the cell, natom is the density of Rb atoms, 𝛺𝑚𝑖𝑛 is the Rabi frequency for the weakest probe transition and the various aij's are the ratios of the Rabi frequency ( 𝛺𝑖𝑗 ) of the |i>-|j> transition to 𝛺𝑚𝑖𝑛. For example, 𝑎12,7 = 𝛺12,7 𝛺14,9 / = √6 𝑏𝑚𝑖𝑛 . 2 is the fraction of the atoms (<1) that decay along the transition corresponding to 𝛺𝑚𝑖𝑛 , among all allowed decay channels from the decaying level. In our model, the amplitudes for all possible transitions from |14> are in the ratio 1:1:1: √3: √6 and hence the fraction of atoms that decay along the different channels are in the ratio 1:1:1:3:6. Thus, 𝑏𝑚𝑖𝑛 2 = 1/(1 + 1 + 1 + 3 + 6) = 1/12 .
Fig. 7. Numerical simulation of 15-level system showing phase shift and attenuation of the RCP and LCP parts of the probe beam as a function of probe detuning. Here δc~1.2 GHz, natom~10 12 /cm 3 and Ωmin=100Γa. See text for more details.
It is possible to produce arbitrarily large phase rotation by increasing the temperature and thus the density of Rb atoms. However, since the difference in phase rotation is so large, the bandwidth over which the device can operate as a halfwave plate, for example, becomes narrower. Thus, for wide bandwidth operation, we would need to operate in the parameter space where we not only have ϕ d ~180 0 , but the slope should also be very small, as shown in Fig. 7(c). We also investigated the waveplate effect under a co-propagating geometry. We found that the absorption line shapes for the co-propagating geometry are broader and shallower compared to the counter-propagating geometry, the reasons for which have been clearly elucidated in Ref 9. As a consequence, under identical conditions, the phase rotation produced is smaller. This was confirmed experimentally, but the results have been omitted for the sake of brevity.
The system can also be used as a quarter-wave plate by using a pump that is linearly polarized at 45 to the 0 direction of polarization of a linearly polarized probe, as explained earlier [9]. Fig. 8a shows the result for such a situation with the output polarizer parallel to the direction of polarization of the probe. Here, the probe is vertically polarized and the pump is linearly polarized at 45 , is at a detuning of δc~1 GHz and has a power of ~600 mW. Fig. 8b 0 shows the corresponding simulation results for a vertical and circularly polarized probe after passing through our analyzing system.
Fig 8. System behaving as Quarter Wave Plate using pump polarized at 45 0
Finally, for certain frequencies, it was observed that that the absorption profile changed dramatically from 0 to approximately 80% when the polarization of the pump was changed by 3-4 degrees, by using a half-wave plate. Fig. 9 shows the result obtained for pump power ~ 600mW and a detuning of ~1 GHz. Such a device could be an inexpensive tool for detecting the polarization of an optical field with high degree of accuracy. Again, the exact location of such behavior depends critically on the density of Rb, the pump power and detuning and the temperature.
## 7. Future Work
In a closely related experiment, we have also demonstrated an optically controlled polarizer, where a control laser at 795 nm is used to realize a polarizer for the signal beam at 1323 nm. Once the optically controlled waveplate and polarizer are combined in the same cell, it should be possible to demonstrate a Quantum Zeno Effect (QZE) based all optical logic gate [9]. Such a gate, realized with a TNF embedded in vapor cell, is expected to require pump power as low as a few nanowatts and thus has potential applications in quantum information processing. In a TNF, the typical mode area is ~0.2 μm 2 [7,8]. Thus, assuming a saturation intensity of 3mW/cm 2 , a Rabi frequency of 10γa would correspond to a power of only ~1 nW. Unlike the free space case, the maximum speed of operation in the TNF system would be limited by the transit time broadening (~60 MHz) rather than the natural linewidth of the 5P manifold (~6 MHz). To see why, note that the speed is limited by the rate at which atoms in the intermediate state relax to the ground state [6] and for a TNF system, this rate is effectively determined by the transit time. A more thorough investigation is needed to identify the parameter space for optimum operation of both the polarizer and waveplate effect simultaneously. We have indicated some of the ways to improve the performance of the optically controlled polarizer [6]. Significant changes to the set-up are required in order to implement the improved scheme both in free space and in the TNF system. Efforts are underway in our laboratory towards making these modifications and we intend to report on the progress and results in the near future.
## 8. Conclusions
To summarize, we have demonstrated an optically controlled waveplate at ~1323 nm using a ladder transition in a Rb vapor cell. We are able to place the probe in any arbitrary polarization by controlling the properties of the pump such as pump power, detuning and polarization. The process is explained in terms of the selection rules for the Zeeman sublevels. The waveplate has applications in Stokesmetric Imaging and optical switching. Using a comprehensive model which incorporates all the relevant Zeeman sub-levels in the system, we identify the parameters for optimal performance. A novel algorithm to compute the evolution of large scale quantum system enabled us to perform this computation. When combined with an optically controlled polarizer, recently demonstrated by us, and using a tapered nanofiber system, such a waveplate can be used to realize a Quantum Zeno Effect based ultra-low power, all-optical logic gate for the telecom band. We have identifed modifications to the system that are necessary in order to achieve a greater phase shift and also discussed some issues related to implementation in the TNF system.
This work was supported in part by AFOSR Grant # FA9550-10-01-0228, and the DARPA ZOE program under Grant # W31P4Q-09-1-0014. | 10.1364/OE.22.028898 | [
"Subramanian Krishnamurthy",
"Y. Tu",
"Y. Wang",
"S. Tseng",
"M. S. Shahriar"
] | 2014-08-01T20:23:48+00:00 | 2014-08-01T20:23:48+00:00 | [
"quant-ph",
"physics.optics"
] | Optically controlled waveplate at a telecom wavelength using a ladder transition in Rb atoms for all-optical switching and high speed Stokesmetric Imaging | We demonstrate an optically controlled waveplate at ~1323 nm using the
5S1/2-5P1/2-6S1/2 ladder transition in a Rb vapor cell. The lower leg of the
transitions represents the control beam, while the upper leg represents the
signal beam. We show that we can place the signal beam in any arbitrary
polarization state with a suitable choice of polarization of the control beam.
Specifically, we demonstrate a differential phase retardance of ~180 degrees
between the two circularly polarized components of a linearly polarized signal
beam. We also demonstrate that the system can act as a Quarter Wave plate. The
optical activity responsible for the phase retardation process is explained in
terms of selection rules involving the Zeeman sublevels. As such, the system
can be used to realize a fast Stokesmetric Imaging system with a speed of
nearly 5 MHz. When implemented using a tapered nano fiber embedded in a vapor
cell, this system can be used to realize an ultra-low power all-optical switch
as well as a Quantum Zeno Effect based all-optical logic gate by combining it
with an optically controlled polarizer, previously demonstrated by us. We
present numerical simulations of the system using a comprehensive model which
incorporates all the relevant Zeeman sub-levels in the system, using a novel
algorithm recently developed by us for efficient computation of the evolution
of an arbitrary large scale quantum system. |
1408.0299v1 | ## The Cosmic Ray Antiproton Background for AMS-02
Rolf Kappl , Martin Wolfgang Winkler a b a Bethe Center for Theoretical Physics & Physikalisches Institut der Universit¨t Bonn, a Nußallee 12, 53115 Bonn, Germany
b
Deutsches Elektronen-Synchrotron DESY,
Notkestraße 85, D-22607 Hamburg, Germany
## Abstract
The AMS-02 experiment is measuring the cosmic ray antiproton flux with high precision. The interpretation of the upcoming data requires a thorough understanding of the secondary antiproton background. In this work, we employ newly available data of the NA49 experiment at CERN, in order to recalculate the antiproton source term arising from cosmic ray spallations on the interstellar matter. We systematically account for the production of antiprotons via hyperon decay and discuss the possible impact of isospin effects on antineutron production. A detailed comparison of our calculation with the existing literature as well as with Monte Carlo based evaluations of the antiproton source term is provided. Our most important result is an updated prediction for the secondary antiproton flux which includes a realistic assessment of the particle physics uncertainties at all energies.
## Contents
| 1 | Introduction | Introduction | 1 |
|-----|---------------------------------------------------|-----------------------------------------------------------------------|-----|
| 2 | Data-driven vs. Monte Carlo Approach | Data-driven vs. Monte Carlo Approach | 2 |
| | 2.1 | Experimental Situation . . . . . . . . . . . . . . . . . . . . . . . | 2 |
| | 2.2 | Monte Carlo Generators . . . . . . . . . . . . . . . . . . . . . . . | 3 |
| 3 | Antiproton Production in Proton Proton Scattering | Antiproton Production in Proton Proton Scattering | 4 |
| | 3.1 | Invariant Cross Section and Radial Scaling . . . . . . . . . . . . . | 4 |
| | 3.2 | Contributions to the Inclusive Cross Section . . . . . . . . . . . . | 5 |
| | | 3.2.1 Prompt Antiproton Production . . . . . . . . . . . . . . . | 5 |
| | | 3.2.2 Hyperons . . . . . . . . . . . . . . . . . . . . . . . . . . . | 7 |
| | | 3.2.3 Antineutrons and Isospin Effects . . . . . . . . . . . . . . | 8 |
| | 3.3 | Corrections to Radial Scaling . . . . . . . . . . . . . . . . . . . . | 10 |
| | 3.4 | Analytic Approximation . . . . . . . . . . . . . . . . . . . . . . . | 12 |
| 4 | Antiproton Production in p A and AA Scattering | Antiproton Production in p A and AA Scattering | 13 |
| | 4.1 | Empirical Description . . . . . . . . . . . . . . . . . . . . . . . . | 14 |
| | 4.2 | Cross Section for p He and HeHe Scattering . . . . . . . . . . . . | 15 |
| 5 | The Secondary Antiproton Flux | The Secondary Antiproton Flux | 15 |
| | 5.1 | The Source Term . . . . . . . . . . . . . . . . . . . . . . . . . . . | 15 |
| | 5.2 | Propagation in the Diffusion Model . . . . . . . . . . . . . . . . . | 18 |
| | 5.3 | Prediction for the Antiproton Flux and Comparison with Data . | 19 |
| 6 | Conclusion | Conclusion | 20 |
## 1 Introduction
Cosmic rays provide us with invaluable information about the most energetic processes happening in our universe. They also play an important role in the search for particle dark matter whose pair-annihilations might leave traces in the spectra of cosmic rays. In this light, the upcoming release of the antiproton measurement by the AMS-02 experiment [1] at the international space station, is eagerly expected.
The dominant contribution to the antiproton flux in our galaxy - the so-called secondary antiproton flux - arises from the spallation of primary cosmic rays on the interstellar matter in the galactic disc. In order to identify a subdominant signal on top of the astrophysical background, the latter must be modeled precisely. This holds in particular as the antiproton signal from the most commonly considered dark matter candidates has a very similar spectral shape as the background flux.
The prediction of the secondary antiproton flux requires knowledge of the differential antiproton production cross section in proton proton and proton nucleus collisions. Existing
parameterizations [2,3] are mainly based on experimental data from the 1970s and early 80s. In this work, we revise those earlier parameterizations by making use of new precision data from the NA49 experiment at CERN [4,5]. Our approach also extends previous calculations in a number of ways:
- · In a collider experiment - depending on detector setup and kinematics - antiprotons stemming from the decay of long-lived hyperons could partially escape detection. This would induce a systematic error in the measured antiproton production cross section. In this work we treat the antiprotons from hyperon decay separately and determine their production from the measured phase-space distribution of the parent hyperons.
- · A standard assumption in previous cosmic ray studies is the equal production of antiprotons and antineutrons in pp scattering. More recently, it was noted that isospin effects may induce an asymmetry in the number of pn ¯ and np ¯ pairs leading to a preference for antineutron production [6,7]. We investigate the possible impact of isospin effects on the antiproton source term.
- · We improve the calculation of proton helium and helium helium scattering by making use of the simple empirical model introduced in [5].
Our approach is to construct the total inclusive antiproton production cross section directly from the experimental data and well-established scaling arguments. For each step in our construction, we are able to include a realistic estimate of the related uncertainty. The resulting cross section is then translated into the secondary source term which we compare with previous calculations in the literature. As a further cross-check, we determine the source term with different Monte Carlo event generators.
Finally we use the two-zone diffusion model [8, 9] for describing the propagation of the antiprotons. As a result, we obtain a fully realistic prediction for the secondary antiproton flux, which includes a realistic assessment of the uncertainties in antiproton production down to lowest energies.
## 2 Data-driven vs. Monte Carlo Approach
The main uncertainty in the secondary antiproton source term is inherited from uncertainties in the inclusive cross sections which enter its determination. While dedicated attempts to evaluate the antiproton production cross sections using Monte Carlo generators have been performed [911], it is very difficult to estimate the uncertainties in the underlying hadronization models. In particular the low energy regime is a cause of concern as there exists no reliable theoretical description of soft hadronic processes. In this work, we therefore base our determination of the antiproton cross sections on experimental data. Nevertheless, several Monte Carlo tools will be used for the sake of comparison.
## 2.1 Experimental Situation
The dominant contribution to the antiproton source term arises from the scattering of cosmic ray protons with hydrogen in the galactic disc. One may think that in the LHC era proton proton scattering can be modeled with high precision. However, current collider experiments have been designed for the discovery of new physics. Detectors like CMS and ATLAS do
not cover the high-rapidity region of the phase space, where most antiprotons are produced. Further, the energy scale relevant for cosmic ray experiments is considerably below the energy scale of operating colliders. AMS-02 is expected to detect antiprotons up to kinetic energies of several 100 GeV, which descend from primary cosmic rays with energies E glyph[similarequal] 10 -10000 GeV. This corresponds to CMS energies √ s glyph[similarequal] 5 -100 GeV of two colliding nucleons - far below LHC energies.
The major part of proton proton scattering data in the relevant energy window were collected in the 1970s and early 80s. The most recent parameterization of the invariant antiproton cross section by Duperray et al. [3] fits these data with χ / 2 d.o.f. ∼ 4 [3]. While one would naively consider this a relatively poor fit, it is hardly possible to improve the level of consistency. The reason is that these data sets show a considerable scatter which complicates the determination of the differential cross section. Systematic errors in the old published data e.g. due to uncertainties in the beam luminosity are difficult to estimate. Certainly, a major source of error arises from the so-called 'feed-down problem': About 1 / 4 of the antiprotons in hadron collisions are produced via the decay of strange hyperons. Hyperons have a macroscopic decay length in the range cm to m, depending on their boost. One can only speculate which fractions of antiprotons from hyperon decay where properly identified by the old detectors. Clearly the fraction of escaping antiprotons depends on the boost of the parent hyperons and details of the detector setup. In particular, it varies between different regions of the phase space and between different experiments.
Fortunately, the situation has drastically improved due to the NA49 experiment at CERN. NA49 was a fixed target experiment with an incoming proton beam of energy E p = 158 GeV at the Super Proton Synchrotron. While the experiment was completed already in 2002, data on antiproton production were published only recently [4]. They provide a high phasespace coverage and drastically exceed the precision of previous data sets. The 'feed-down problem' is absent as the antiprotons descending from hyperon decay have been identified using micro vertex detection and precision tracking. Therefore, we have decided to attempt a new determination of the antiproton source term based on the NA49 data. Different from previous approaches, we do not try to find a parameterization of the differential antiproton cross section, but we directly use the NA49 data as an input.
Apart from proton proton scattering, processes involving helium contribute considerably (in total about 40%) to the antiproton flux. Unfortunately, experimental data on p He scattering do not exist in the relevant energy window. Therefore, one has to extrapolate the helium cross sections from the measured cross sections of heavier nuclei. Fortunately, there exist new precision data on proton carbon scattering by NA49 [5]. The collaboration has introduced a simple empirical model which relates the antiproton cross section in proton nucleus scattering with the cross section in proton proton scattering. This model was shown to describe p C scattering with good precision over the whole phase space and shall be used for p He and HeHe scattering in this work.
## 2.2 Monte Carlo Generators
As a possible alternative to the data-driven approach, Monte Carlo tools have been used to determine the secondary antiproton flux as well as a possible primary component from dark matter annihilations. In this light, it is important to investigate the range of applicability of the underlying hadronization models. In this work, we will employ the Monte Carlo generators PYTHIA 8.1 [12], DPMJET-III [13] and GEANT4 [14] for an independent determination of
the differential antiproton production cross sections. For the simulation with PYTHIA we took into account all inelastic soft QCD processes. In the case of DPMJET, we used the implemented standard PHOJET model without elastic collisions. GEANT was developed as a detector simulation, but we adjusted the code to trace the event chain of single inelastic collisions. For the hadronization, we chose the build-in FTFP model which is based on the FRITIOF description of string fragmentation. The tool ROOT [15] was used for data analysis and procession.
PYTHIA only deals with proton proton interactions, therefore it can only be used to determine the dominant component of the antiproton source term. Subleading components from processes involving helium can be obtained by use of DPMJET and GEANT. As is pointed out in the documentation of the Monte Carlo generators, none of the three tools is suited for the low energy regime, where the hadronization models break down. At higher energies, however, reasonable agreement between the data-driven and Monte Carlo based evaluation of the antiproton source term is expected.
## 3 Antiproton Production in Proton Proton Scattering
Proton proton scattering is the dominant source of antiprotons in our galaxy. In hadronic collisions antiprotons are promptly produced due to the factorization of the colliding partons. Additionally, antiprotons descend as decay products of long-lived intermediate states like antineutrons or hyperons. Before we discuss the different contributions to the inclusive antiproton production cross section, we shall turn to the energy scaling of the cross section.
## 3.1 Invariant Cross Section and Radial Scaling
We are interested in the inclusive production of a hadron h in the reaction pp → h + X , where X stands for the sum of the remaining final state particles. For this we introduce the Lorentz invariant cross section
$$f _ { p p \rightarrow h } = E _ { h } \frac { d ^ { 3 } \sigma } { d p _ { h } ^ { 3 } } = \frac { E _ { h } } { \pi } \frac { d ^ { 2 } \sigma } { d p _ { L } d p _ { T } ^ { 2 } } \,,$$
where E h is the energy of the detected hadron and d σ/dp 3 3 h the differential cross section with respect to the three-momentum p h . The longitudinal and transverse components of p h are denoted by p L and p T respectively. It is useful to express the invariant cross section in terms of p T and a scaling variable. The radial and Feynman scaling variables are defined as
$$x _ { R } = \frac { E ^ { * } } { E _ { \max } ^ { * } } \,, \quad x _ { f } = \frac { p _ { \L } ^ { * } } { \sqrt { s } / 2 } \,,$$
where E ∗ and p ∗ L denote the energy and longitudinal momentum of h in the center of mass frame. The maximal energy is determined as E ∗ max = ( s -M 2 X + m 2 h ) / (2 √ s ) with M X being the minimal mass of the recoiling particles X .
In [16, 17] a large set of experimental data was analyzed. It was shown that the invariant cross section approaches a radial scaling limit
$$f _ { p p \rightarrow h } ( \sqrt { s }, x _ { R }, p _ { T } ) \, \longrightarrow \, f _ { p p \rightarrow h } ( x _ { R }, p _ { T } )$$
for √ s glyph[greaterorsimilar] 10 GeV independent of the nature of the final state hadron. This is an enormous simplification as - within the radial scaling regime - the cross section at all center of mass
energies can be deduced from the cross section at one energy. If, instead, one expresses the invariant cross section in terms of the scaling variable x f , it also approaches a scaling limit [18], however only at considerably higher energies [17]. Therefore, the use of x R as a scaling variable is clearly preferred.
From a theoretical viewpoint, constituent exchange models predict a power law behavior of the invariant cross section [19-21]
$$f _ { p p \to h } \varpropto ( 1 - x ) ^ { n } \,.$$
Here x denotes the light cone fraction of the considered hadron which is not directly observable, but can be approximated as x glyph[similarequal] x R [22]. The fragmentation power n is determined by the dimensional counting rule. It increases with the minimal number of spectator quarks necessitated by quantum number requirements. In particular, n is expected to be larger for antibaryon production compared to meson production due to baryon number conservation. Quantitatively, the quark exchange model predicts the fragmentation power n = 9 for antiproton production in pp collisions [22].
## 3.2 Contributions to the Inclusive Cross Section
In cosmic ray physics, one is ultimately interested in the total number of antiprotons irrespective of their origin. However, in a collider experiment, antiprotons appearing macroscopically displaced from the initial collision vertex may escape detection. While antineutrons decay far outside any contemplable detector, the situation is more subtle for the strange hyperons Λ and ¯ ¯ which have decay lengths comparable to typical detector scales. It was mentioned previously Σ that this leads to the so-called 'feed-down problem' as the fraction of escaping antiprotons from late decays remains dubious in the old experimental data.
In this work, we mainly rely on antiproton data from the NA49 experiment. In contrast to previous experiments, antiprotons from hyperon decay have systematically been identified. This allows us to split the antiproton production cross section into individual components which we discuss separately. The total invariant antiproton cross section in pp collisions can be written as
$$f _ { p p } = f _ { p p \rightarrow \bar { p } } + f _ { p p \rightarrow \bar { n } } \,.$$
Here f pp → ¯ n accounts for the antiprotons from very late decaying antineutrons. This contribution is not directly accessible to experiments and has to be determined by use of symmetry arguments. We further split
$$f _ { p p \rightarrow \bar { p } } = f _ { p p \rightarrow \bar { p } } ^ { 0 } + f _ { p p \rightarrow \bar { p } } ^ { \bar { \Lambda } } \,,$$
where f 0 pp → ¯ p and f ¯ Λ pp → ¯ p denote the contributions from prompt hadronization and from the weak decay of strange hyperons, respectively.
## 3.2.1 Prompt Antiproton Production
The NA49 collaboration has performed a precision measurement of inclusive antiproton production in pp scattering [4]. Only the antiprotons from prompt hadronization are included in the data, while antiprotons from hyperon decay were systematically rejected. The beam energy E p = 158 GeV corresponds to √ s = 17 4 GeV - well within the radial scaling regime. . We can thus, in principle, use the NA49 data to determine the cross section at all energies √ s > 10 GeV.
D
GLYPH<144>
@
Figure 1: Measured invariant antiproton cross section expressed in terms of the radial scaling variable and eyeball fit with extrapolation (see text).
The collaboration provides the invariant differential cross section f 0 pp → ¯ p in bins of x f and p T . Additionally, an eyeball fit to the data is specified. As Feynman scaling is only reached at considerably higher energies, it is convenient to translate x f into x R . In figure 1, we depict a subset of the NA49 data and the corresponding eyeball fit in terms of the radial scaling variable. Different from previous works, we do not attempt to find a parameterization for the invariant cross section, but simply use the fit given by the collaboration (after expressing it in terms of x R ).
Despite the large phase-space coverage of NA49, we still have to use some extrapolation. The region at x R glyph[similarequal] 0 -0 1 . is kinematically not accessible at NA49 as x R > m /E p ∗ max by definition. Further, as the cross section decreases rapidly towards large x R and p T , no data exist in this regime due to the limited statistics. In the covered region of phase space the invariant cross section shows virtually perfect power law behavior f 0 pp → ¯ p ∝ (1 -x R ) n with n = 7 5 at all . p T . We assume that this behavior continues outside the covered region of x R and include an uncertainty in the fragmentation power n = 7 5 . ± 1. This power law form is expected in theoretical hadronization models. As we will show explicitly later, our extrapolation is also consistent with experimental data taken at higher energies. We further observe an exponential decrease of the invariant cross section with the transverse mass m T = √ p 2 T + m 2 p which we assume to continue at p T > . 1 5GeV. We note, however, that this assumption is not of relevance
as the cross section is negligibly small at high p T .
Statistical errors in the NA49 data reside at the level of ∼ 5%. Their effect on the antiproton source term is, however, washed out by the phase space integration. They can be neglected compared to other sources of uncertainty. However, we shall take into account systematic errors which can affect the overall normalization of the cross section. Using a conservative linear error propagation, the NA49 collaboration has estimated the systematic uncertainty to be 6 5% which we adopt in the following. .
## 3.2.2 Hyperons
The NA49 data set [4] includes only the promptly produced antiprotons, while antiprotons from hyperon decay were systematically rejected. In order to determine f ¯ Λ pp → ¯ p , we can use the measured phase-space distribution of the parent hyperons. In figure 22 of [23], the collaboration has published the differential multiplicity of Λ-hyperons ¯
$$\left ( \frac { d n } { d x _ { f _ { \bar { \Lambda } } } } \right ) _ { p p \to \bar { \Lambda } } = \frac { 1 } { \sigma _ { p p } ^ { \text{inel} } } \, \left ( \frac { d \sigma } { d x _ { f _ { \bar { \Lambda } } } } \right ) _ { p p \to \bar { \Lambda } } \,.$$
where σ inel pp denotes the total inelastic cross section in pp collisions. The hyperons decay into antiproton and pion with a branching fraction Br( ¯ Λ → ¯ p π ) = 0 64 [24]. . We can make use of the fact that m Λ glyph[similarequal] m p + m π , which implies that the momentum of Λ is distributed between ¯ the decay products according to their masses. Therefore we can express the Feynman variable x f of the antiproton in terms of the Feynman variable of the parent hyperon
$$x _ { f } \simeq \frac { m _ { p } } { m _ { \Lambda } } \ x _ { f _ { \bar { \Lambda } } } \,.$$
This allows us to predict the multiplicity distribution of antiprotons from hyperon decay
$$\left ( \frac { d n } { d x _ { f } } \right ) _ { p p \to \bar { p } } ^ { \bar { \Lambda } } = \text{Br} ( \bar { \Lambda } \to \bar { p } \, \pi ) \times \frac { m _ { \Lambda } } { m _ { p } } \, \left ( \frac { d n } { d x _ { f _ { \bar { \Lambda } } } } \right ) _ { p p \to \bar { \Lambda } } \, + \left ( \bar { \Sigma } \text{ contribution} \right ) \,.$$
Note that there is a subdominant contribution to n ¯ Λ pp → ¯ p from Σ hyperons. ¯ By symmetry arguments the ratio of produced Σ ¯ -/ ¯ can be estimated to be 0 33 [4] Λ . 1 . We set the branching fraction Br( ¯ Σ -→ ¯ p π ) = 0 52 [24] and assume that the antiprotons from Σ-decay follow the . ¯ same x f -distribution as the antiprotons from Λ-decay. ¯
In figure 2, we depict the differential multiplicity of antiprotons from hyperon decay. The data points were obtained from the measured differential multiplicity of Λ [23] by use of ¯ (9) (including the contribution from Σ). For comparison, we also provide the distribution obtained ¯ with the GEANT Monte Carlo, which is in reasonable agreement with the data. The differential multiplicity of the promptly produced antiprotons scaled by a factor of 0 2 and 0 3 is also shown. . . The latter was determined by integrating the invariant cross section over transverse momenta
$$\cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \\ \left ( \frac { d n } { d x _ { f } } \right ) _ { p p \to \tilde { p } } ^ { 0 } = \frac { \pi } { \sigma _ { p p } ^ { \text{inel} } } \, \frac { \sqrt { s } } { 2 } \, \int d p _ { T } ^ { 2 } \, \frac { f _ { p p \to \tilde { p } } ^ { 0 } } { E _ { \bar { p } } } \,.$$
It can be seen that the antiprotons from hyperon decay and those from prompt hadronization exhibit a very similar momentum distribution. Unfortunately, error bars are not available for
1 According to [4], the ratio of produced Σ ¯ -/ Σ + is expected to be 0 8 ¯ . Λ Λ. / If one uses the measured multiplicities of Λ and Σ + from [23] this leads to the estimate of Σ ¯ -/ ¯ Λ glyph[similarequal] 0 33. .
Figure 2: Multiplicity distribution of hyperon-induced antiprotons integrated over transverse momenta as extracted from the NA49 data (blue circles). The prediction from the GEANT Monte Carlo is shown for comparison (green dashed line). The multiplicity distribution of promptly produced antiprotons scaled by 0 2 . -0 3 is indicated by the blue band. .
the hyperon data. But we presume that within the current precision, both distributions match up to a scaling factor. In the following we take the uncertainty band in f ¯ Λ pp → ¯ p such that all data points of figure 2 are included. This leads to the estimate
$$\frac { f _ { p p \to \bar { p } } ^ { \bar { \Lambda } } } { f _ { p p \to \bar { p } } ^ { 0 } } = 0. 2 - 0. 3 \,.$$
We do not require this ratio to be strictly momentum-independent. Rather, we assume that it only varies between 0 2 and 0 3 over the whole phase space. . .
## 3.2.3 Antineutrons and Isospin Effects
In previous cosmic ray studies, it was assumed that there is no distinction between ¯ and ¯ p n production in pp collisions. Serious concerns about this hypothesis have been raised in [6]. There, data from proton collisions at different CMS energies were analyzed. It was noted that the net proton density, defined as the number of protons minus antiprotons per inelastic event, revealed a considerable energy dependence. At increasing √ s , this quantity was found to grow dramatically towards low x f . Baryon number conservation requires that baryons and antibaryons are produced in pairs. Under the assumption that the numbers of ¯ pn and pn ¯ pairs match, the above defined net proton density should correspond to the non-pair produced protons. The mentioned steep increase seems to indicate a problem with baryon number conservation - given the assumption of equal numbers of ¯ pn and pn ¯ pairs holds. The authors of [6] conclude that there must be an asymmetry between ¯ and ¯ production. p n
While the inclusive process pp → ¯ + n X is not directly observable, it is instructive to consider the flipped reaction np → ¯+ p X . We will decompose the antiproton multiplicity into a projectile and a target component
$$\left ( \frac { d n } { d x _ { f } } \right ) _ { n p \to \bar { p } } = \left ( \frac { d n } { d x _ { f } } \right ) _ { n p \to \bar { p } } ^ { \text{pro} } \, + \, \left ( \frac { d n } { d x _ { f } } \right ) _ { n p \to \bar { p } } ^ { \text{tar} } \,.$$
Ł
Figure 3: Differential antiproton multiplicity in pp and np collisions. The difference between the two distributions can possibly be related to an isospin effect (see text).
The validity of this decomposition requires the independence of target and projectile factorization, such that the total multiplicity arises from the simple superposition of both contributions. This assumption has been experimentally verified in other hadronic interactions (e.g. by comparing πp and pp scattering [25]).
In the forward ( x f > 0) and backward ( x f < 0) region, the multiplicity is dominated by projectile and target factorization respectively. However, there appears a small feed-over at | x f | glyph[lessorsimilar] 0 1, where both contributions slightly leak into the 'wrong' hemisphere. One can define . the target overlap function F tar ( x f ) to project out the target contribution of the multiplicity
$$\left ( \frac { d n } { d x _ { f } } \right ) _ { h p \to \bar { p } } ^ { \text{tar} } = F _ { \text{tar} } ( x _ { f } ) \, \left ( \frac { d n } { d x _ { f } } \right ) _ { p p \to \bar { p } },$$
where h denotes an arbitrary baryon or meson projectile. The overlap function was found to be independent of the transverse momentum p T and of √ s if expressed in terms of x f [25]. We take F tar ( x f ) from table 14 in [5], the projectile overlap function is simply given as F tar (1 -x f ) and fulfills the relation F tar ( x f ) + F pro ( x f ) = 1. Using overlap functions to project out the target and projectile components, we can express
$$\left ( \frac { d n } { d x _ { f } } \right ) _ { n p \to \beta } = F _ { \text{pro} } ( x _ { f } ) \, \left ( \frac { d n } { d x _ { f } } \right ) _ { n n \to \beta } + F _ { \text{tar} } ( x _ { f } ) \, \left ( \frac { d n } { d x _ { f } } \right ) _ { p p \to \beta } \, \right ) \,.$$
$$\cdot \cdot \cdot \\ = ( N _ { \text{IS} } \ F _ { \text{pro} } ( x _ { f } ) + F _ { \text{tar} } ( x _ { f } ) ) \, \left ( \frac { d n } { d x _ { f } } \right ) _ { p p \to p } \,.$$
In the second step, we made use of the fact that proton and neutron can be understood as doublet under an isospin symmetry which implies n nn → ¯ p = n pp → ¯ n . Further, we defined
$$\left ( \frac { d n } { d x _ { f } } \right ) _ { p p \to \tilde { n } } = N _ { \text{IS} } \, \left ( \frac { d n } { d x _ { f } } \right ) _ { p p \to \tilde { p } } \,.$$
Here N IS denotes the isospin factor which parameterizes a possible asymmetry between ¯ and n ¯ production in p pp collisions.
In figure 3, we depict the antiproton multiplicity in pp and np collisions measured by NA49 [6]. In the case N IS = 1 both multiplicity distributions should match which appears inconsistent with the data. Indeed using a ∆ χ 2 -test and taking the isospin factor to be constant, we obtain 1 37 . ± 0 06 at the 90% confidence level. . The multiplicity distribution corresponding to the best fit point is also indicated in the figure.
This argumentation indicates a preference for the production of pn ¯ pairs compared to np ¯ pairs in pp collisions. One should note, however, that the np data of NA49 are still at the preliminary level, systematic uncertainties have not been discussed. There are additional sources of concern: deviations from the independent target and projectile fragmentation would affect the above result. As we shall discuss later, the p C scattering data of NA49 are consistent with isospin effects, but hint at a somewhat smaller N IS ∼ 1 2. . The hadronization models implemented in the Monte Carlo generators PYTHIA, GEANT and DPMJET yield no clear preference for ¯ production at all. n LHC data indicate that at very high energies, the ¯ p/p ratio in the mid-rapidity window approaches unity [26] which seems to speak against a preference for pn ¯ pairs in pp collisions. Finally, it is not guaranteed that the isospin factor is constant in the whole phase space: the data in figure 3 prefer a larger isospin effect at low x f , although not at a statistically significant level.
As the current situation is inconclusive, a conservative treatment of the isospin effect seems appropriate. In the following, we will assume an isospin factor N IS = 1 0 . -1 43. . The lower end of this window corresponds to the standard assumption f pp → ¯ p = f pp → ¯ n , the upper end to the 95% CL upper limit on the isospin factor deduced from np scattering (see above).
An additional contribution to antineutron production arises from hyperon decay, completely analogous as in the antiproton case (see section 3.2.2). Assuming equal production of Σ ¯ -and ¯ Σ + and using the branching fractions from [24], we find that the number of hyperon-induced antineutrons is by a factor of 1.05 higher than the number of hyperon-induced antiprotons.
## 3.3 Corrections to Radial Scaling
Let us now turn to the energy dependence of the invariant cross section. The latter is expected to be independent of the CMS energy for √ s > 10GeV if expressed in terms of the radial scaling variable and the transverse momentum. This can be verified by considering high energy collider data.
In figure 4, we used the invariant cross section extracted from the NA49 data, to predict f pp → p ( x , p f T ) at a higher CMS energy √ s = 53 GeV. 2 Corresponding experimental data from the CERN ISR collider by Capiluppi et al. [27] and Alper et al. [28] are also shown. We have included the antiprotons from hyperon decay in the NA49 prediction as a large fraction of them is expected to be contained in the CERN ISR data [4]. The indicated uncertainty band includes the systematic error of NA49, uncertainties related to our extrapolation of the NA49 data to low x R as well as the uncertainty in the hyperon contribution. Given that there is a considerable scatter in the CERN ISR data, one may take the shown experimental error bars with a grain of salt. But there is an overall good agreement between the measured cross section and the NA49 expectation.
2 Note that in the radial scaling regime f ( x , p f T ) still depends on the CMS energy. This is because x f is a function of x R and √ s . Only in the very high energy limit x f → x R and f ( x , p f T ) becomes independent of √ s . This corresponds to the Feynman scaling regime.
D
GLYPH<144>
@
Figure 4: Invariant antiproton cross section measured at CERN ISR compared with the fit used in this work (see text).
For further comparison, we use a data set from the BRAHMS experiment at Brookhaven [29] which was taken at CMS energy √ s = 200GeV for two different rapidities y = 2 95 and . y = 3 3. . The measured invariant cross section is in very good agreement with the NA49 prediction as can be seen in figure 5. This gives further confidence in the radial scaling hypothesis as well as our extrapolation of the NA49 data towards low x R .
We note that deviations from radial scaling are expected at very high energies. The BRAHMS data indicate that, for p T < 1 5 GeV, radial scaling is valid up to CMS energies . √ s = 200 GeV. Deviations from radial scaling at higher p T can be neglected as the cross section is anyway highly suppressed in this region of the phase-space. The BRAHMS energy corresponds to the scattering of an incoming proton with E p ∼ 20 TeV on the interstellar hydrogen. More energetic protons only marginally contribute to the antiproton flux in the energy window covered by AMS-02. Further, the growing uncertainty in the primary proton flux dilutes any effect caused by the violation of radial scaling at high √ s . Therefore, we can safely work under the hypothesis of radial scaling for √ s > 10 GeV. √
At energies s glyph[lessmuch] 10 GeV, however, radial scaling is heavily broken. In order to model the low energy antiproton flux, we clearly have to quantify the violation of radial scaling. We follow [2] and define the energy-dependent ratio
$$\Phi _ { R } ( \sqrt { s }, x _ { R }, p _ { T } ) = \frac { f _ { p p } ( \sqrt { s }, x _ { R }, p _ { T } ) } { f _ { p p } ^ { R S } ( x _ { R }, p _ { T } ) } \,,$$
where f RS pp ( x R , p T ) denotes the invariant cross section in the radial scaling limit. In [17] a large number of hadronic processes was considered. Independent of the inclusive chan-
D
GLYPH<144>
@
Figure 5: Invariant antiproton cross section measured by the BRAHMS experiment compared with the fit used in this work (see text).
nel, it was found that the radial scaling limit is approached asymptotically from above, i.e. R ( √ s, x R , p T ) ≥ 1. Further, it was noted that the cross section enters the scaling regime faster at low x R , while R only weakly depends on p T [2, 17]. √
Unfortunately, experimental data on inclusive ¯-production at p s < 10GeV are rare. There exist only two data sets [30,31] with reasonable phase-space coverage. Both were recorded at the CERN Proton Synchrotron in the early 1970s. We have attempted to find a parameterization of R with the qualitative behavior described above. For this, we neglected the dependence of R on p T , and required R x ( R , √ s = 10 GeV) = 1. Empirically we arrived at the function
$$R ( x _ { R }, \sqrt { s } ) = \left ( 1 + C _ { 1 } \left ( \frac { 1 0 \, \text{GeV} - \sqrt { s } } { \text{GeV} } \right ) ^ { 5 } \right ) \exp \left [ C _ { 2 } \, \left ( \frac { 1 0 \, \text{GeV} - \sqrt { s } } { \text{GeV} } \right ) \, ( x _ { R } - x _ { R, \min } ) ^ { 2 } \right ] \,, \ \ ( 1 8 )$$
for √ s ≤ 10GeV. The parameters were determined as C 1 = (1 ± 0 4) . × 10 -3 and C 2 = 0 7 . ± 0 04. . Given that error bars are only partially available for the data sets, we have chosen the uncertainties in C 1 2 , such that the error band encloses the data points. Our parameterization as well as the experimental data [30,31] can be seen in figure 6. A slightly more involved function R was suggested in [2]. We have verified that the antiproton source term does virtually not depend on which of the two parameterizations we use. This is not surprising as [2] employed the same limited amount of data available at low √ s .
## 3.4 Analytic Approximation
In this work, we decided to use the fit to the Na49 data for the invariant antiproton cross section. To allow for a quick comparison with our results, we nevertheless want to specify an analytic approximation for f pp . In the NA49 data we observe an almost perfect factorization of f 0 pp → ¯ p with respect to the variables x R and p T . A good fit ( χ 2 = 1 4 . / d.o.f.) can be obtained with the parameterization
$$f _ { p p \to p } ^ { 0 } ( x _ { R }, p _ { T } ) = \left ( 4 0 0 \, \text{mb} \, \text{GeV} ^ { - 2 } \right ) \times ( 1 - x _ { R } ) ^ { 7. 7 6 } \exp \left ( - 5. 9 5 \, m _ { T } \right ) \,,$$
Figure 6: Full invariant cross section divided by the cross section in the radial scaling regime at two different beam energies. Data points stem from the CERN Proton Synchrotron. Our parameterization of R is indicated by the colored bands.
where m T stands for the transverse mass. To obtain f pp one has to include the contributions from antineutron and hyperon decay. Further in the low energy regime the correction function R must be applied. Using central values for the hyperon induced antinucleons, we arrive at
$$f _ { p p } & ( x _ { R }, p _ { T }, \sqrt { s } ) \\ & = ( 4 0 0 \, \text{mb GeV} ^ { - 2 } ) \times R ( x _ { R }, \sqrt { s } ) \times ( 1. 5 1 + N _ { \text{IS} } ) \times ( 1 - x _ { R } ) ^ { 7. 7 6 } \exp { ( - 5. 9 5 \, m _ { T } ) } \. \quad ( 2 0 )$$
where we left the isospin factor unspecified. We have verified that the antiproton source term calculated from the analytic approximation agrees with the source term obtained from the NA49 fit within a few % precision.
## 4 Antiproton Production in p A and AA Scattering
Given the lack of experimental data on p He and HeHe scattering, the corresponding differential cross sections must be determined by means of extrapolation. Duperray et al. [3] have therefore employed a modified version of the Kalinovskii formula [32] which describes generic proton nucleus collisions. The parameters of the formula were fitted by using a large set of data, including scattering of protons on very heavy nuclei like lead. However, one should be aware that the nuclear medium effects in such scatterings depend crucially on the size of the nucleus. While they are almost absent for p He processes, they completely distort the phase space distribution of resulting hadrons in p Pb scattering. It is, therefore, questionable to which extent data involving heavy nuclei are useful for the extrapolation of the p He cross section.
Figure 7: Measured antiproton multiplicity in proton carbon scattering and prediction from the empirical model discussed in the text. The multiplicity in proton proton scattering is shown for comparison.
In this work, we construct the p He and HeHe cross sections from the elementary hadronic processes.
## 4.1 Empirical Description
In [5], it has been realized that the differential antiproton multiplicity in p C scattering is closely related to the multiplicity in pp scattering. Indeed, it was found that both multiplicities virtually match in the projectile hemisphere, i.e. in the forward direction. At the same time, a significant increase of the multiplicity is observed in the target hemisphere. The origin of this excess can be traced back to multiple scatters of the projectile in the nucleus. By measuring the increase of pion yields in the backward hemisphere, the average number of proton interaction was found to be 〈 ν C 〉 = 1 6 in carbon. Using the overlap functions introduced in section 3.2.3, . the multiplicity can be written as
$$\left ( \frac { d n } { d x _ { f } } \right ) _ { p C \rightarrow \bar { p } } = \left ( \frac { 1 + N _ { \text{IS} } } { 2 } \left \langle \nu _ { _ { C } } \right \rangle F _ { \text{tar} } ( x _ { f } ) + F _ { \text{pro} } ( x _ { f } ) \right ) \left ( \frac { d n } { d x _ { f } } \right ) _ { p p \rightarrow \bar { p } }.$$
Note that in addition to multiple scatters, there appears a possible isospin enhancement due to the fact that the carbon nucleus contains half neutrons. In figure 7, we depict the measured antiproton multiplicity in p C collisions and the prediction derived from (21). For the isospin factor we have used the uncertainty band N IS = 1 -1 43 as given in section 3.2.3. . It can be seen that the empirical model yields a very good description of p C scattering within the given uncertainties. Minor deviations arising in the projectile hemisphere can be related to nuclear medium effects like the Cronin effect [33]. They reside, however, at the level of a few per cent for carbon and are expected to be completely negligible for helium.
## 4.2 Cross Section for p He and HeHe Scattering
In order to determine the p He, He p and HeHe cross section we adopt the same method as for p C scattering. We assume that the differential antiproton multiplicity is given as
$$\left ( \frac { d n } { d x _ { f } } \right ) _ { A _ { 1 } A _ { 2 } } = \left ( \langle \nu _ { _ { A _ { 1 } } } \rangle \, F _ { \text{tar} } ( x _ { f } ) + \langle \nu _ { _ { A _ { 2 } } } \rangle \, F _ { \text{pro} } ( x _ { f } ) \right ) \left ( \frac { d n } { d x _ { f } } \right ) _ { p p } \,, \quad \left ( A _ { 1, 2 } = p, \text{He} \right ) \,.$$
Here n A A 1 2 = n A A 1 2 → ¯ p + n A A 1 2 → ¯ n stands for the total antiproton multiplicity including the contributions from hyperon and antineutron decay. Note that all isospin effects are already contained in n pp , there appear no extra isospin factors. This is because isospin effects may induce an asymmetry between antiproton and antineutron production, but they do not affect the total number of antinucleons which are produced.
The average number of interactions is 〈 ν p 〉 = 1 for protons. In the case of helium, we determine 〈 ν He 〉 from the total inelastic cross section as
$$\langle \nu _ { _ { \text{He} } } \rangle = 4 \, \frac { \sigma _ { p p, \text{inel} } } { \sigma _ { p \text{He}, \text{inel} } } \,,$$
where the factor 4 originates from the nucleon number of helium. Extracting the total inelastic proton helium cross section from [34], we find 〈 ν He 〉 = 1 25. . 3 The invariant cross section is related to the multiplicity via (10). Assuming the same transverse momentum distribution of antiprotons in proton and helium scattering, we arrive at the invariant cross section
$$f _ { A _ { 1 } A _ { 2 } } = \frac { \sigma _ { A _ { 1 } A _ { 2 }, \text{inel} } } { \sigma _ { p p, \text{inel} } } \left ( \langle \nu _ { A _ { 1 } } \rangle \, F _ { \text{tar} } ( x _ { f } ) + \langle \nu _ { A _ { 2 } } \rangle \, F _ { \text{pro} } ( x _ { f } ) \right ) \, f _ { p p } \,, \quad ( A _ { 1, 2 } = p, \text{He} ) \,.$$
We will assume that (24) holds independent of the CMS energy. Note that f p He and f HeHe inherit the uncertainties contained in f pp .
## 5 The Secondary Antiproton Flux
## 5.1 The Source Term
Primary cosmic rays create antiprotons by inelastic interactions with the interstellar gas in the galactic disc. In the following, we want to determine the secondary antiproton source term q ¯ p which is defined as the differential ¯ production rate per volume, time and energy. p For this, we make use of the invariant cross sections evaluated in the previous sections. To obtain the uncertainty band for the source term, we perform a conservative linear error propagation which is necessitated by the fact that errors are not normally distributed in our case. 4 We include the previously discussed uncertainties arising from systematic errors in the NA49 experiment, from the extrapolation of cross sections towards low x R , from the hyperon-induced antinucleon production as well as from isospin effects. We also account for the uncertainty in the function R which relates the invariant cross section at low collision energies with the cross section in the radial scaling regime. For our reference source term we use central values and set N IS = 1.
3 In [34] an interpolation formula for inelastic proton nucleus cross sections is presented. It is noted in the text that this formula overestimates the proton helium cross section by 20%. Applying this correction, one arrives at σ p He,inel = 45 mb × 4 0 7 . / . 1 2.
4 In the case of isospin effects, there is e.g. a finite probability that N IS = 1, which is at the edge of the considered uncertainty band.
T GeV @ D Figure 8: Primary cosmic ray fluxes measured by AMS-02 and power law fits employed in this work (see text).
The secondary source term is determined by
$$q _ { \bar { p } } ( T ) = \sum _ { A _ { 1, 2 } = p, \text{He} } \ 4 \pi \overset { \infty } { \underset { E _ { \text{th} } } { \int } } d T ^ { \prime } \left ( \frac { d \sigma } { d T } \right ) _ { \varLambda _ { 1 } \varLambda _ { 2 } } \varrho _ { \text{A} _ { 2 } } \ \Phi _ { \text{A} _ { 1 } } ^ { \text{IS} } ( T ^ { \prime } ) \.$$
Here T ′ denotes the kinetic energy per nucleon of the incoming primary cosmic ray particle (proton or helium), T the kinetic energy of the outgoing antiproton. The differential antiproton production cross section ( dσ/dT ) A 1 A 2 can be obtained from the invariant cross section f A 1 A 2 (cf. (1)). The threshold energy is given as E th = 6 m p , while for the interstellar number densities of hydrogen and helium, we set glyph[rho1] p = 0 9 cm . -3 and glyph[rho1] He = 0 1 cm . -3 [35]. 5 The contribution of heavier elements to q ¯ p can be neglected [10].
The interstellar fluxes of primary cosmic rays Φ IS p, He exhibit a power law behavior in the energy range T > 10 GeV relevant for the secondary antiproton production. They can in principle be taken from previous studies [36, 37]. However, as a new precision measurement of proton and helium fluxes by AMS-02 became available [38], we decided to reevaluate the primary fluxes. Experimental data always refer to the top-of-the-atmosphere fluxes Φ TOA p, He . Therefore, we have to account for the effects of solar modulation. This is done by means of the force field approximation [39]. To obtain the interstellar fluxes, we have fitted power law functions to the AMS-02 data, which we modulated by a force field φ = 700 MV. 6
In figure 8, we depict the envelope of (modulated) power law functions which are consistent with the measured proton and helium fluxes (at the 90% CL using a χ 2 -metric). In the calculation of the source term we include the shown bands as an additional source of uncertainty.
5 For a self-consistent approach, we have to employ the same input values for the hydrogen and helium densities as were used for the determination of the propagation parameters which we adopt later.
6 The force field φ = 700 MV was determined by comparing the low energy proton flux measured by AMS-02 with the interstellar flux of [36].
D
L
@H
10 -1 1 10 1 10 2 10 3 T GeV @ D @ D Figure 9: Secondary source term determined from the NA49 data as described in the text and corresponding uncertainty band (also available in the Ancillary Files). The pp component of the source term, scaled by 0.1, is shown separately. Also depicted are the source terms arising from previous parameterizations of the invariant cross sections (left panel). Source terms obtained with different Monte Carlo tools are shown in the right panel.
For our reference source term we take the best-fit functions
$$\Phi _ { p } ^ { \text{IS} } ( T ) = 1 7 4 0 7 \, \mathrm m ^ { - 2 } \text{sr} ^ { - 1 } \text{s} ^ { - 1 } \text{GeV} ^ { - 1 } \times \left ( \frac { T } { \text{GeV} } \right ) ^ { - 2. 7 7 5 } \,,$$
$$\Phi _ { \text{He} } ^ { \text{IS} } ( T ) = 5 9 7. 2 \, m ^ { - 2 } \text{sr} ^ { - 1 } \text{s} ^ { - 1 } \text{GeV} ^ { - 1 } \times \overset { \cdot } { \left ( \frac { T } { \text{GeV} } \right ) } ^ { \cdot } \.$$
In figure 9 we depict the antiproton source term which results from the invariant cross sections used in this work. The source term includes contributions from pp , p He, He p and HeHe scattering, the pp component is shown separately. The associated error in q ¯ p resides at the level of 20% for intermediate energies T ∼ 10 GeV. In this regime it is dominated by the uncertainty in the isospin factor which affects antineutron production. Towards low T , the width of the error band increases drastically and reaches ∼ 50% at T = 1 GeV. This results from the uncertainty in the hadronic cross sections due to the breakdown of radial scaling and the lack of low-energy collider data. Towards high kinetic energies, the error in q ¯ p also rises which is related to the growing uncertainties in the primary cosmic ray fluxes.
For comparison we have determined q ¯ p by using previous parameterizations of the invariant antiproton cross section by Tan et al. [2] and Duperray et al. [3] (see left panel of figure 9). In the pp channel the Tan source term virtually matches with our reference source term. This is remarkable as our description of pp scattering relied on a disjoined set of experimental data compared to Tan et al.. Duperray et al. have attempted to find a single parameterization of the invariant cross section f pp in- and outside the radial scaling regime. This induces a tendency to underestimate f pp at high energies and to overestimate it at low energies. We believe that this explains the shape of the Duperray source term. We note, however, that the latter still resides within the uncertainty band of our q ¯ p .
In the right panel of figure 9, we show q ¯ p as obtained with the Monte Carlo tools PYTHIA, GEANT and DPMJET. While the Monte Carlo tools qualitatively reproduce the expected shape of the source term, they generically predict too high q ¯ p . At T > 10 GeV the GEANT and DPMJET source terms are only marginally outside the uncertainty band of our data-based source term, while PYTHIA already overestimates q ¯ p by a factor of 2. The level of discrepancy increases continuously towards low T which is most likely caused by the hadronization models, arriving at the edge of their validity. We have traced back the discrepancy between databased and Monte Carlo approach to the fact, that the Monte Carlo generators predict too large multiplicities already in pp collisions. As an example, the Monte Carlos find n pp → ¯ p = 0 006 . , 0 005 . , 0 008 (GEANT, DPMJET, PYTHIA) at a collision energy . √ s = 6 1 GeV which . has to be compared with the experimental value n pp → ¯ p = 0 002 [40]. . Therefore, it seems that the tested Monte Carlo tools can not be used 'out of the box' to determine the secondary antiproton flux.
## 5.2 Propagation in the Diffusion Model
The antiproton flux induced by the source term q ¯ p is determined through the diffusion equation
$$\nabla ( - K \, \nabla N _ { \bar { p } } + V _ { c } \, N _ { \bar { p } } ) + \partial _ { T } ( b _ { \text{tot} } \, N _ { \bar { p } } - K _ { E E } \, \partial _ { T } N _ { \bar { p } } ) + \Gamma _ { \text{ann} } \, N _ { \bar { p } } = q _ { \bar { p } } + q _ { \bar { p } } ^ { \text{ter} } \,, \quad ( 2 8 )$$
where N ¯ p denotes the antiproton space-energy density, K accounts for diffusion and V c is the galactic wind. The function b tot describes energy losses, ionization and reacceleration, it can be taken from [41], K EE is the so-called energy diffusion coefficient [42] and Γ ann the annihilation rate which describes the antiproton annihilation with interstellar matter. The tertiary source term q ter ¯ p arises from inelastic scattering of secondary antiprotons in the galactic disc.
We solve this equation semi-analytically within the two-zone diffusion framework introduced in [8,9]. Further details can be found in our previous publication [43]. 7 In the two-zone diffusion model, propagation depends on five parameters: the height of the diffusion cylinder 8 L , the galactic wind V c , the Alfv´n speed e V a which enters K EE as well as the diffusion parameters K 0 and δ which are related to K as
$$K = K _ { 0 } \beta \left ( \frac { p } { \text{GeV} } \right ) ^ { \delta } \,.$$
Here β and p stand for the antiproton velocity and momentum. We use the propagation parameters shown in table 1 which were taken from a recent boron to carbon analysis [35]. As
$$\frac { \delta \quad K _ { 0 } \left ( \text{kpc} ^ { 2 } \cdot \text{Myr} ^ { - 1 } \right ) \ \ L \left ( \text{kpc} \right ) \ \ V _ { c } \left ( \text{km} \cdot s ^ { - 1 } \right ) \ \ V _ { a } \left ( \text{km} \cdot s ^ { - 1 } \right ) } { 0. 8 6 \quad 0. 0 0 4 2 \quad 4 \quad 1 8. 7 \quad 3 5. 5 }$$
Table 1: Propagation parameters taken from the boron to carbon analysis in [35].
the focus of this study is on the particle physics uncertainties contained in the antiproton flux, we do not attempt to include the uncertainties in the propagation parameters . 9 The latter can be obtained by propagating the antiproton source term with all configurations consistent with the boron to carbon ratio and taking the envelope of the obtained antiproton fluxes (see e.g. [50]).
7 For a full numerical approach to the diffusion equation, see [11,44,45].
8 See [46-48] for recent discussions on L .
9 See e.g. [49] for a recent discussion of astrophysical uncertainties.
-
10
1
D
1
-
2
L
10
-
s sr GeV
-
3
2
m
10
-
4
@H
10
TOA
-
5
F
10
p
f=
500 600 MV
-
-
10 1
10 -1 1 T GeV @ D Figure 10: Interstellar antiproton flux and particle physics uncertainty band obtained in this work (upper panel, also available in the Ancillary Files). In the lower panel, the top-of-theatmosphere flux obtained by modulating Φ IS ¯ p with a force field φ = 500 -600 MV is compared with the experimental data from PAMELA and BESS-Polar II.
10 2
10
6
## 5.3 Prediction for the Antiproton Flux and Comparison with Data
In figure 10 (upper panel), we depict the interstellar antiproton flux obtained from the source term of figure 9 within the two-zone diffusion model. The shown error band comprises the particle physics uncertainties and the uncertainties of the primary cosmic ray fluxes, propagation uncertainties have not been included. The error band is practically inherited from the source term. Only at low T , the uncertainty in Φ IS ¯ p is slightly reduced compared to the uncertainty in q ¯ p due to the tertiary component of the flux.
To compare the obtained flux with existing experimental data, we have to account for
@
D
solar modulation. For this we use the force field approximation. The currently most accurate measurements of the antiproton flux were performed by the PAMELA [51] and the BESS-Polar II [52] collaborations. Both data sets were obtained in a period of low solar activity, where a force field of φ = 500 -600MV can be used to account for modulation [36]. In figure 10 (lower panel) we compare the top-of-the-atmosphere flux from our calculation with the experimental data of PAMELA and BESS-Polar II. It can be seen that a very good agreement within the given uncertainties is obtained. We notice that the measured flux resides more towards the lower end of the uncertainty band which seems to speak against strong isospin effects in the antineutron production. However, this conclusion might be premature as we have not included the propagation uncertainties in the antiproton flux.
## 6 Conclusion
The AMS-02 experiment is about to release a high precision measurement of the cosmic ray antiproton flux. In this light, we have rederived the secondary antiproton source term which arises from cosmic ray spallations in the galactic disc (figure 9). We employed new experimental data on proton proton and proton nucleus scattering recorded by the NA49 experiment at CERN. Important improvements of our study include the careful treatment of antiprotons arising from antineutron and hyperon decay as well as a revised approach to proton helium scattering. For the first time, we were able to assign a realistic particle physics error band to the secondary source term, which is an important input for constraining possible primary antiproton signals in the galaxy. Our source term is publically available and can be used with a generic propagation setup.
We have identified two major sources of uncertainty in the secondary source term: one stems from the lack of experimental data on low energy proton collisions, where hadronic cross sections do not follow a scaling behavior. The other is related to a possible isospin effect which may result in an enhanced ¯-production in n pp collisions. In this regard, the experimental situation, however, remains dubious and further investigation is required.
Independently, we determined the source term by use of the Monte Carlo tools PYTHIA, GEANT and DPMJET. While there appears a qualitative agreement with the data-based evaluation of the source term at antiproton energies T > 10 GeV, the Monte Carlo generators substantially overestimate q ¯ p . We have identified the origin of the discrepancy in the too large antiproton multiplicities predicted by the Monte Carlo generators. At T glyph[lessmuch] 10 GeV the predictions of all three Monte Carlo tools become rather poor which is most likely explained by the breakdown of the underlying hadronization models.
In the last step, we have propagated the source term within the two-zone diffusion framework in order to obtain the secondary antiproton flux. The latter was then confronted with existing experimental data by the PAMELA and BESS-Polar II experiments and a very satisfying agreement was obtained (figure 10). The comparison with the upcoming AMS-02 data is eagerly awaited.
## Note added
In the final stage of this work we became aware of a related study by F. Donato, A. Goudelis, M. Di Mauro and P. Serpico [53]. While it also determines the proton scattering cross sections relevant for cosmic ray physics, the approach is quite complementary to ours. We focus on the
NA49 data sets which we use as a basis to discuss prompt and hyperon-induced antiproton production, antineutron production and proton nucleus scattering. The focus of [53] on the other hand is on a global reanalysis of the existing pp scattering data. In this light [53], provides a very important comparison for our work.
## Acknowledgments
This work has been supported by the German Science Foundation (DFG) within the Collaborative Research Center 676 'Particles, Strings and the Early Universe' and the SFB-Transregio TR33 'The Dark Universe'.
## References
- [1] AMS Collaboration, M. Aguilar et al., Phys.Rev.Lett. 110 (2013), 141102.
- [2] L. Tan and L. Ng, Phys.Rev. D26 (1982), 1179-1182.
- [3] R. Duperray, C.-Y. Huang, K. Protasov, and M. Buenerd, Phys.Rev. D68 (2003), 094017, [astro-ph/0305274].
- [4] NA49 Collaboration, T. Anticic et al., Eur.Phys.J. C65 (2010), 9-63, [0904.2708].
- [5] NA49 Collaboration, B. Baatar et al., Eur.Phys.J. C73 (2013), 2364, [1207.6520].
- [6] NA49 Collaboration, H. Fischer, Heavy Ion Phys. 17 (2003), 369-386.
- [7] NA49 Collaboration, O. Chvala, Eur.Phys.J. C33 (2004), S615-S617, [hep-ex/0405053].
- [8] D. Maurin, F. Donato, R. Taillet, and P. Salati, Astrophys.J. 555 (2001), 585-596, [astroph/0101231].
- [9] F. Donato, D. Maurin, P. Salati, A. Barrau, G. Boudoul, et al., Astrophys.J. 563 (2001), 172-184, [astro-ph/0103150].
- [10] M. Simon, A. Molnar, and S. Roesler, Astrophys.J. 499 (1998), 250.
- [11] A. Strong and I. Moskalenko, Astrophys.J. 509 (1998), 212-228, [astro-ph/9807150].
- [12] T. Sjostrand, S. Mrenna, and P. Z. Skands, Comput.Phys.Commun. 178 (2008), 852-867, [0710.3820].
- [13] S. Roesler, R. Engel, and J. Ranft, (2000), 1033-1038, hep-ph/0012252.
- [14] GEANT4, S. Agostinelli et al., Nucl.Instrum.Meth. A506 (2003), 250-303.
- [15] R. Brun and F. Rademakers, Nucl.Instrum.Meth. A389 (1997), 81-86.
- [16] D. C. Carey, J. Johnson, R. Kammerud, M. Peters, D. Ritchie, et al., Phys.Rev.Lett. 33 (1974), 327-330.
- [17] F. Taylor, D. C. Carey, J. Johnson, R. Kammerud, D. Ritchie, et al., Phys.Rev. D14 (1976), 1217.
- [18] R. P. Feynman, Phys.Rev.Lett. 23 (1969), 1415-1417.
- [19] F. Low, Phys.Rev. D12 (1975), 163-173.
- [20] S. Nussinov, Phys.Rev.Lett. 34 (1975), 1286-1289.
- [21] S. J. Brodsky and J. Gunion, Phys.Rev.Lett. 37 (1976), 402-405.
- [22] S. J. Brodsky and J. Gunion, Phys.Rev. D17 (1978), 848-857.
- [23] NA49 Collaboration, C. Alt et al., Eur.Phys.J. C45 (2006), 343-381, [hep-ex/0510009].
- [24] Particle Data Group, J. Beringer et al., Phys.Rev. D86 (2012), 010001.
- [25] G. Barr, O. Chvala, H. Fischer, M. Kreps, M. Makariev, et al., Eur.Phys.J. C49 (2007), 919-945, [hep-ex/0606029].
- [26] LHCb Collaboration, R. Aaij et al., Eur.Phys.J. C72 (2012), 2168, [1206.5160].
- [27] P. Capiluppi, G. Giacomelli, A. Rossi, G. Vannini, A. Bertin, et al., Nucl.Phys. B79 (1974), 189.
- [28] British-Scandinavian Collaboration, B. Alper et al., Nucl.Phys. B100 (1975), 237.
- [29] BRAHMS Collaboration, I. Arsene et al., Phys.Rev.Lett. 98 (2007), 252001, [hepex/0701041].
- [30] J. Smith, (1972).
- [31] J. Allaby, F. Binon, A. Diddens, P. Duteil, A. Klovning, et al., (1970).
- [32] A. Kalinovskii, M. Mokhov, and Y. Nikitin, American Institute of Physics (1989).
- [33] J. Cronin, H. J. Frisch, M. Shochet, J. Boymond, P. Piroue, et al., Phys.Rev.Lett. 31 (1973), 1426-1429.
- [34] J. Letaw, R. Silberberg, and C. Tsao, Astrophys.J.S. 51 (1983), 271.
- [35] A. Putze, L. Derome, and D. Maurin, Astron.Astrophys. 516 (2010), A66, [1001.0551].
- [36] M. Potgieter, E. Vos, M. Boezio, N. De Simone, V. Di Felice, et al., (2013), 1302.1284.
- [37] Y. Shikaze, S. Haino, K. Abe, H. Fuke, T. Hams, et al., Astropart.Phys. 28 (2007), 154167, [astro-ph/0611388].
- [38] S. Haino, Talk at the 33rd ICRC Conference, 2013.
- [39] L. Gleeson and W. Axford, Astrophys.J. 154 (1968), 1011.
- [40] M. Antinucci, A. Bertin, P. Capiluppi, M. D'Agostino-Bruno, A. Rossi, et al., Lett.Nuovo Cim. 6 (1973), 121-128.
- [41] D. Maurin, R. Taillet, F. Donato, P. Salati, A. Barrau, et al., (2002), astro-ph/0212111.
- [42] D. Maurin, R. Taillet, and F. Donato, Astron.Astrophys. 394 (2002), 1039-1056, [astroph/0206286].
- [43] R. Kappl and M. W. Winkler, Phys.Rev. D85 (2012), 123522, [1110.4376].
- [44] I. V. Moskalenko, A. W. Strong, J. F. Ormes, and M. S. Potgieter, Astrophys.J. 565 (2002), 280-296, [astro-ph/0106567].
- [45] C. Evoli, D. Gaggero, D. Grasso, and L. Maccione, JCAP 0810 (2008), 018, [0807.4730].
- [46] J. Lavalle, D. Maurin, and A. Putze, (2014), 1407.2540.
- [47] T. Bringmann, F. Donato, and R. A. Lineros, JCAP 1201 (2012), 049, [1106.4821].
- [48] J. Lavalle, Phys.Rev. D82 (2010), 081302, [1007.5253].
- [49] C. Evoli, I. Cholis, D. Grasso, L. Maccione, and P. Ullio, Phys.Rev. D85 (2012), 123511, [1108.0664].
- [50] T. Bringmann and P. Salati, Phys.Rev. D75 (2007), 083006, [astro-ph/0612514].
- [51] PAMELA Collaboration, O. Adriani et al., Phys.Rev.Lett. 105 (2010), 121101, [1007.0821].
- [52] K. Abe, H. Fuke, S. Haino, T. Hams, M. Hasegawa, et al., Phys.Rev.Lett. 108 (2012), 051102, [1107.6000].
- [53] F. Donato, A. Goudelis, M. Di Mauro, and P. Serpico, to appear. | 10.1088/1475-7516/2014/09/051 | [
"Rolf Kappl",
"Martin Wolfgang Winkler"
] | 2014-08-01T20:27:07+00:00 | 2014-08-01T20:27:07+00:00 | [
"hep-ph",
"astro-ph.CO",
"astro-ph.HE"
] | The Cosmic Ray Antiproton Background for AMS-02 | The AMS-02 experiment is measuring the cosmic ray antiproton flux with high
precision. The interpretation of the upcoming data requires a thorough
understanding of the secondary antiproton background. In this work, we employ
newly available data of the NA49 experiment at CERN, in order to recalculate
the antiproton source term arising from cosmic ray spallations on the
interstellar matter. We systematically account for the production of
antiprotons via hyperon decay and discuss the possible impact of isospin
effects on antineutron production. A detailed comparison of our calculation
with the existing literature as well as with Monte Carlo based evaluations of
the antiproton source term is provided. Our most important result is an updated
prediction for the secondary antiproton flux which includes a realistic
assessment of the particle physics uncertainties at all energies. |
1408.0300v1 | SI-HEP/14-18, QFET/14-12 01 August 2014
## Non-Leptonic Heavy Meson Decays - Theory Status
Thorsten Feldmann 1
Theoretische Physik 1, Naturwissenschaftlich-Technische Fakult¨t, a Universit¨t Siegen, Walter-Flex-Straße 3, D-57068 Siegen, Germany a
I briefly review the status and recent progress in the theoretical understanding of non-leptonic decays of beauty and charm hadrons. Focusing on a personal selection of topics, this covers perturbative calculations in quantum chromodynamics, analyses using flavour symmetries of strong interactions, and the modelling of the relevant hadronic input functions.
## PRESENTED AT
Flavor Physics and CP Violation (FPCP-2014), Marseille, France, May 26-30 2014
1 Supported in parts by the Bundesministerium f¨r Bildung und Forschung (BMBF), and by the u Deutsche Forschungsgemeinschaft (DFG) within Research Unit FOR 1873 ('Quark Flavour Physics and Effective Field Theories').
## 1 Motivation
Non-leptonic decays of hadrons containing a heavy bottom or charm quark may provide important information on the angles of the Cabibbo-Kobayashi-Maskawa (CKM) matrix in the Standard Model (SM). They also may reveal deviations from the SM, in particular the presence of new CP-violating phases from 'new physics' (NP). The non-trivial hadronic dynamics in such flavour transitions further allows to assess the accuracy of theoretical methods (perturbative or non-perturbative) in Quantum Chromodynamics (QCD), based on the separation ('factorization') of shortand long-distance strong-interaction effects. Finally, comparison of experimental data and theoretical parametrizations may also lead to a better understanding of the hadronic structure of heavy-light bound states.
Experimental studies of non-leptonic B -and D -meson decays, which have been successfully carried out at flavour factories and at hadron colliders in the past, will be continued at present and future experiments, notably at LHCb and Belle-II. With the foreseen increasing experimental precision, theoretical calculations should therefore catch up in accuracy in order to achieve reliable phenomenological conclusions about the validity of the SM or hints for NP (for comprehensive reviews, see e.g. [1-4]).
In this proceedings contribution, I highlight some recent theoretical results and developments which contribute to improving our understanding of exclusive heavymeson decays.
## 2 Perturbative Calculations
Theory predictions for non-leptonic exclusive decays, by definition, depend on hadronic matrix element which cannot be calculated in QCD perturbation theory. Still, the presence of a heavy quark mass (notably for the b -quark) implies that certain dynamical effects are related to short-distance physics (on length scales of the order 1 /m b ) and may thus be accessible in perturbative QCD. The challenge is then to systematically separate short- and long-distance phenomena, where the latter should be described by as few independent hadronic parameters as possible.
## QCD Factorization for B Decays into Light Mesons
B -decays into two (energetic) light mesons ( M ,M 1 2 ) are described by hadronic matrix elements of weak transition operators O i . In the limit of infinitely heavy quark masses, the strong-interaction dynamics factorizes according to [5]
$$\langle M _ { 1 } M _ { 2 } | \mathcal { O } _ { i } | B \rangle \ \simeq \ & F ^ { B \to M _ { 1 } } \, \int d u \, T _ { i } ^ { I } ( u ) \, \phi _ { M _ { 2 } } ( u ) \\ & \quad + \int d \omega \, d u \, d v \, T _ { i } ^ { \text{II} } ( \omega, u, v ) \, \phi _ { B } ( \omega ) \, \phi _ { M _ { 1 } } ( v ) \, \phi _ { M _ { 2 } } ( u ) \,.$$
Here, the hadronic input functions are given by (universal) transition form factors F B → M evaluated at large recoil energy, and light-cone distribution amplitudes (LCDAs) φ B,M ,M 1 2 for heavy and light mesons which depend on the momenta (respectively momentum fractions) of the light quarks. The short-distance kernels T I II , i can be calculated perturbatively, including renormalization-group (RG) improvement in the framework of soft-collinear effective theory (SCET [6, 7]). One generic phenomenological consequence of this 'QCD-improved' factorization (QCDF) is that direct CP violation in these decays - which requires strong rescattering phases - is suppressed by O ( α s ) and/or O (1 /m b ) in the heavy-quark limit.
The result of QCDF calculations can be parametrized in terms of (decay-channel dependent) tree- and penguin-amplitude parameters α i , for instance, ∗
$$\langle \pi ^ { + } \pi ^ { - } | \mathcal { H } _ { \text{eff} } | \overline { B } ^ { 0 } \rangle \quad = \quad A _ { \pi \pi } \{ \lambda _ { u } \left [ \alpha _ { 1 } ( \pi \pi ) + \alpha _ { 4 } ^ { u } ( \pi \pi ) \right ] + \lambda _ { c } \, \alpha _ { 4 } ^ { c } ( \pi \pi ) \} \\ \langle \pi ^ { + } \overline { K } ^ { - } | \mathcal { H } _ { \text{eff} } | \overline { B } ^ { 0 } \rangle \quad = \quad A _ { \pi \overline { K } } \{ \lambda _ { u } ^ { ( s ) } \left [ \alpha _ { 1 } ( \pi K ) + \alpha _ { 4 } ^ { u } ( \pi K ) \right ] + \lambda _ { c } ^ { ( s ) } \, \alpha _ { 4 } ^ { c } ( \pi K ) \} \\ \text{ etc.}$$
The current status of higher-order calculations (NNLO, i.e. second order in the strong coupling α s ) is as follows.
- · 2-loop vertex corrections contributing to T I i for tree-amplitude parameters α 1 2 , have been determined independently in [9,10]. The corresponding 2-loop vertex corrections for penguin amplitudes are currently under study [11], and preliminary results will be shown below.
- · The 1-loop spectator corrections to T II i for tree amplitudes have been analyzed in [13-15]. The 1-loop spectator corrections for penguin amplitudes can be found in [16,17].
Let us briefly discuss the numerical significance of the individual contributions. Considering e.g. the colour-allowed (-suppressed) tree amplitude α 1 ( α 2 ) in B → ππ decays, we have
$$\begin{array} { r c l } \alpha _ { 1 } ( \pi \pi ) & = & [ 1. 0 0 8 ] _ { V _ { 0 } } + [ 0. 0 2 2 + 0. 0 0 9 i ] _ { V _ { 1 } } + [ 0. 0 2 4 + 0. 0 2 6 i ] _ { V _ { 2 } } \\ & & - [ 0. 0 1 4 ] _ { S _ { 1 } } - [ 0. 0 1 6 + 0. 0 1 2 i ] _ { S _ { 2 } } - [ 0. 0 0 8 ] _ { 1 / m _ { b } } \\ & = & 1. 0 1 5 ^ { + 0. 0 2 0 } _ { - 0. 0 2 9 } + \left ( 0. 0 2 3 ^ { + 0. 0 1 5 } _ { - 0. 0 1 5 } \right ) i \,, \end{array}$$
and
$$\begin{array} { r c l } \alpha _ { 2 } ( \pi \pi ) & = & [ 0. 2 2 4 ] _ { V _ { 0 } } - [ 0. 1 7 4 + 0. 0 7 5 i ] _ { V _ { 1 } } - [ 0. 0 2 9 + 0. 0 4 6 i ] _ { V _ { 2 } } \\ & & + [ 0. 0 8 4 ] _ { S _ { 1 } } + [ 0. 0 3 7 + 0. 0 2 2 i ] _ { S _ { 2 } } + [ 0. 0 5 2 ] _ { 1 / m _ { b } } \end{array}$$
∗ Here λ q denote combinations of CKM elements, and the normalization factors A M M 1 2 are given in terms of form factors and decay constants [8].
$$= \quad 0. 1 9 4 ^ { + 0. 1 3 0 } _ { - 0. 0 9 5 } - \left ( 0. 0 9 9 ^ { + 0. 0 5 7 } _ { - 0. 0 5 6 } \right ) i \,,$$
where V 0 1 2 , , stand for (LO,NLO,NNLO) contributions to T I i , while (NLO,NNLO) spectator contributions from T II i are labeled by S 1 2 , . Estimates of 1 /m b power corrections are also quoted. One observes that
- · the perturbative expansion is well behaved, with individual NNLO corrections V 2 and S 2 being significant but tending to cancel in the sum;
- · precise predictions are achieved for the colour-allowed tree amplitude α 1 , while larger hadronic uncertainties remain for the colour-suppressed amplitude α 2 ;
- · the relative phase between α 1 and α 2 stays small at NNLO.
Penguin amplitudes are currently known at NLO, and for the ππ channel one gets
$$\begin{array} { r c l } \alpha _ { 4 } ^ { u } ( \pi \pi ) & = & - 0. 0 2 4 ^ { + 0. 0 0 4 } _ { - 0. 0 0 2 } + \left ( - 0. 0 1 2 ^ { + 0. 0 0 3 } _ { - 0. 0 0 2 } \right ) i \\ \alpha _ { 4 } ^ { c } ( \pi \pi ) & = & - 0. 0 2 8 ^ { + 0. 0 0 5 } _ { - 0. 0 0 3 } + \left ( - 0. 0 0 6 ^ { + 0. 0 0 3 } _ { - 0. 0 0 2 } \right ) i \end{array}$$
The calculation of penguin amplitudes at NNLO, which is currently worked out [11], involves O (70) 2-loop diagrams with up to 3 independent mass scales ( m b , m c , um b ) and 4 external legs and non-trivial charm thresholds at (1 -u m ) 2 b = 4 m 2 c (where u is the momentum fraction of a quark in a light meson). Preliminary results have been shown in recent conference talks [12].
In the past, the QCDF results have been used for comprehensive phenomenological studies:
- · The decays B → ππ, πρ, ρρ have been investigated in [10, 18]. Predictions for colour-suppressed modes are rather uncertain and typically underestimated in QCDF (depending on the hadronic matrix element governing the size of spectator-scattering contributions). The uncertainties from hadronic form factors and the CKM element | V ub | can be reduced by considering ratios with the semi-leptonic rates.
- · Estimates for tree-dominated B s decays can be found in [19]. Here the relevant hadronic parameters (form fators and LCDAs) are less well known than for the previous case. On the other hand, the pattern of annihilation contributions is simpler. It has also been emphasized that the size of charming-penguin effects can be tested from ratios of colour-allowed modes [20].
- · One can also find results for charmless B -meson decays into scalar mesons [21].
Figure 1: Dalitz plot of B → πππ decays, highlighting the 'mercedes-star' configuration where QCDF is applicable (from [24]).
## Other modes / Further Activities
NNLO corrections in QCDF for the decays B → Dπ are currently studied as well. In this case, the charm and bottom quark are usually treated as heavy quarks in HQET, and their hard fluctuations are integrated out at a common matching scale. Compared to charmless decays, this leads to a number of new master integrals which depend on the mass ratio m /m 2 c 2 b . The status of the computation has been recently reported in [22].
The systematics of QCDF in e.g. B → ππ decays can also be independently addressed by phenomenological studies of the related decay B → ππglyph[lscript]ν . In the region where the dipion invariant mass is large, m 2 ππ ∼ O ( m 2 b ), one obtains a factorization theorem [23] which takes a similar form as in B → ππ (with ⊗ denoting the convolution integrals as in (1)),
$$\langle \pi ^ { + } \pi ^ { - } | \overline { u } \, \Gamma \, b | B \rangle \ \simeq \ F ^ { B \to \pi } \cdot \phi _ { \pi } \otimes T ^ { I } _ { \Gamma } ( q ^ { 2 } ) \ + \ \phi _ { \pi } \otimes \phi _ { \pi } \otimes \phi _ { B } ( \omega ) \otimes T ^ { I I } _ { \Gamma } ( q ^ { 2 } ) \,. \quad ( 6 )$$
In contrast to B → ππ , the decay is now induced by semi-leptonic operators, where Γ = γ µ (1 -γ 5 ) in the SM. In the considered kinematic region, at least one hard gluon is required to produce the additional back-to-back quark-antiquark pair in the final state. As a consequence, the kernel T I Γ starts at O ( α s ), while the kernel T II Γ includes additional spectator interactions and thus starts at O ( α 2 s ). Moreover, the variable q 2 representing the invariant mass of the lepton pair, provides a new lever arm to assess systematic uncertainties related to non-factorizable effects.
Finally, let me mention that the QCDF approach can also be applied to 3-body decays B → πππ in the kinematic region where each individual dipion mass m 2 ij = ( p i + p j ) 2 is sufficiently large, i.e. the three momenta forming a 'mercedes-star'-like configuration, see Fig. 1. Asystematic theoretical and phenomenological investigation of this idea will be pursued in [24].
## 3 Flavour Symmetries in QCD
The approximate flavour symmetries (FS) of light quarks in strong interaction dynamics (isospin for u, d quarks, U -spin for d, s quarks, or the full SU (3) F for u, d, s ) have always been a standard tool in understanding hadronic physics. The wealth of experimental data on non-leptonic b -and c -decays nowadays allows to draw conclusions about first-order FS-breaking corrections. In combination with factorization approaches, this can be used to test assumptions about subleading terms in the 1 /m b expansion, and in the long run this may also enhance the sensitivity to finding deviations from the SM in these decay modes.
## Isospin and SU (3) F in B → PP and B → PV
The complete set of isospin, U -spin and SU (3) F relations among the CP asymmetries in B -meson decays to two pseudoscalars (or to one pseudoscalar and one vector meson), together with first-order symmetry-breaking effects, has recently been analyzed in [25] (see also [26,27]). Comparing with experimental data, the amount of SU (3) F breaking turns out to be of reasonable size, e.g.
$$\tilde { \Delta } \ \equiv \ \frac { \delta _ { \text{CP} } [ B _ { d } \to K ^ { + } \pi ^ { - } ] + \delta _ { \text{CP} } [ B _ { s } \to K ^ { - } \pi ^ { + } ] } { \delta _ { \text{CP} } [ B _ { d } \to K ^ { + } \pi ^ { - } ] - \delta _ { \text{CP} } [ B _ { s } \to K ^ { - } \pi ^ { + } ] } = 0. 0 2 6 \pm 0. 1 0 6 \,,$$
if the observables are properly normalized, δ CP [ i → f ] = 8 π m 2 i | glyph[vector] p i → f | ∆ CP [ i → f ]. Additional constraints on SU (3) F breaking can be obtained from certain theory approaches. As an example, the authors of [25] compare two phenomenological approaches based on different treatment of non-factorizable effects. In the so-called 'BBNS' approach [8], the combination of amplitude parameters α α 1 c 4 determines the numerically dominant source of strong phases and direct CP violation. In the 'BPRS' approach [28] the dominant source of non-factorizable effects is expected from charmpenguin contributions. This leads to different correlations between SU (3) F breaking in BRs and CP asymmetries for ( B s → K π -+ , B d → π -K + ) and ( B d → π -π + , B s → K K -+ ) which can be confronted with experimental data. Present data are still consistent with both alternatives, but with improved precision in future experiments more decisive conclusions about non-factorizable SU (3) F -breaking effects in non-leptonic decays will be possible.
Figure 2: Fit of penguin-enhancement parameters δ , δ 3 ′ 3 to experimental data on non-leptonic charm decays (after Charm2013 ). Orange (red) regions indicate the 95% (68%) C.L. regions, see [29] for more details. Fig. taken from [29].
## Non-leptonic Charm Decays
Factorization-based approaches to describe exclusive charm decays suffer from the relatively small charm-quark mass, such that non-factorizable corrections are more important than for the corresponding bottom decays. Again, approximate flavour symmetries of QCD turn out to be helpful. The flavour anatomy of D -meson decays into two pseudoscalars has been systemetically studied in [29]. Taking into account the complete set of first order SU (3) F -breaking effects, a consistent fit of the available experimental data could be achieved, with corrections to the flavour-symmetry limit of natural size, O (30%). At the time of the analysis, the measured value of the difference between CP asymmetries in D 0 → K K + -and D 0 → π + π -decays, ∆ a dir CP ( K K ,π π + -+ -), required a drastic penguin enhancement (in the SM), where also a dir CP ( D 0 → K K S S ), a dir CP ( D s → K π S + ), and a dir CP ( D s → K π + 0 ) contribute. After the Charm2013 conference, the effect is less dramatic, as illustrated in Fig. 2. Alternatively, the SU (3) F analysis of various CP asymmetries in non-leptonic D -meson decays allows to discriminate between different NP scenarios.
A related SU (3) F analysis of non-leptonic D -meson decays to two pseudoscalars or one pseudoscalar and one vector meson can be found in [30]. The focus of that work is to derive sum rules among decay amplitudes or decay rates such that O ( m s )
effects drop out. For instance, from the CKM-weighted amplitude relations
$$| ( D ^ { 0 } | \pi ^ { + } \rho ^ { - } ) | / \lambda + | ( D ^ { 0 } | K ^ { + } K ^ { * - } ) | / \lambda \ = \ | ( D ^ { 0 } | K ^ { + } \rho ^ { - } ) | / \lambda ^ { 2 } + | ( D ^ { 0 } | \pi ^ { + } K ^ { * - } ) \quad ( 8 )$$
one can infer a prediction for one of the yet unmeasured branching ratios (BRs),
$$\Rightarrow \ B r ( D ^ { 0 } \to \rho ^ { - } K ^ { + } ) \simeq ( 1. 7 \pm 0. 4 ) \cdot 1 0 ^ { - 4 } \,.$$
The formalism can also be combined with the '∆ U = 0 rule' for large penguins [31] to predict direct CP asymmetries in D → PV decays.
## Other modes
Recent analyses based on SU (3) F symmetry have also been performed for particular final state configurations in B -meson decays to three light pseudoscalars [32]; and for 2-body B -meson decays into octet or decuplet baryons [33]. In a systematic study of SU (3)-breaking effects in B → J/ψP decays [34] it has been shown how penguin corrections can be extracted from data, constraining the pollution in the extraction of the CKM angle sin 2 β to be very small, | ∆ S | ≤ 0 01. . With more precise data on CP asymmetries and BRs this uncertainty can be further reduced in the future.
## 4 Hadronic Input Functions
Besides the perturbative computation of short-distance kernels in factorization theorems, and factorization-independent constraints from flavour symmetries, an important role for the quantitative prediction of nonleptonic decays is played by universal hadronic input parameters like decay constants, form factors and light-cone distribution amplitudes (LCDAs). As these contain the information about hadronic binding effects, they have to be determined by nonperturbative methods (i.e. lattice or sum rules) or extracted from experimental data.
## Transition Form Factors
If factorization in (1) holds, the form factors for B → M 1 transitions together with the decay constants f M 2 determine the overall magnitude of B → M M 1 2 decay amplitudes and BRs at leading order. The sensitivity to this type of hadronic input can be reduced by considering ratios with semileptonic decays or among different nonleptonic decays. For theoretical estimates of transition form factors between heavy and light mesons, it has become customary to perform combined fits of light-cone sumrule results (valid at large energy transfer, see e.g. [35, 36]) and results from QCD simulations on the lattice (valid at low recoil energy, see the discussion in [37] and references given therein) on the basis of the so-called ' z -expansion'.
## Light-Cone Distribution Amplitudes for the B -Meson
The B -meson LCDA φ B ( ω ) determines the size of the spectator interactions in QCDF. The leading term in the short-distance kernels T II i is proportional to the inverse moment
$$\lambda _ { B } ^ { - 1 } ( \mu ) \ \equiv \ \int _ { 0 } ^ { \infty } \frac { d \omega } { \omega } \, \phi _ { B } ( \omega, \mu ) \,,$$
where ω denotes the light-cone projection of the spectator quark's energy. The size of this parameter is crucial for phenomenological analyses. While comparison with data on B → ππ, πρ, ρρ decays prefers relatively small values around λ B ∼ 200 MeV within the QCDF/BBNS approach, QCD-sum-rule based estimates typically lead to values in the region λ B ∼ (350 -500) MeV [38]. Recent OPE analyses of the large-scale behaviour of φ B , using the concept of 'dual' LCDAs [39] (see also [40]), have shown that - contrary to naive expectation - the parameter λ B is essentially independent of other HQET parameters like Λ = M B -m b [41]. The most promising approach to independently determine λ B is to extract its value from experimental data on B → γglyph[lscript]ν decays, on the basis of QCD factorization theorems and estimates for non-factorizable 1 /m b corrections [42, 43]. Presently, using the BaBar bound for that decay rate from 2009, one finds λ B > 115 MeV.
## Annihilation Parameters in QCDF
Amplitude topologies where the spectator quark in a B -meson annihilates with the b -quark via weak interactions also play a crucial role for the phenomenological analyses of non-leptonic B decays. Notably, these decay topologies cannot be described by heavy-to-light form factors. The 1 /m b power corrections induced by annihilation topologies lead to IR-sensitive convolution integrals in QCDF and can thus only be modelled in a crude and ad-hoc manner. In a recent phenomenological analysis of pure annihilation decays of B d and B s mesons [44] (see also [45]) the flavour dependence of annihilation parameters in QCDF has been studied. Comparing, on the one hand, B d → π -K + and B s → π + K -decays, one expects similar strong rescattering phases because the final states are related by charge conjugation, which turns out to be in line with experimental observation. On the other hand, comparing B s → π + π -and B d → K K + -within that approach, sizeable SU (3) F breaking effects are required.
The size of strong phases in non-leptonic B -meson decays has also been studied within a phenomenological rescattering model [46]. Distinguishing different topological amplitudes ('exchange (E)', 'annihilation (A)', 'penguin-annihilation (PA)'), experimental data reveals a relatively regular pattern, where (E) ∼ (5 -10)% and (PA) ∼ (15 -20)% of the largest amplitude from which they can rescatter. This allows one to estimate several BRs for not yet observed B and B s decays.
## 5 Summary/Outlook
The dynamics of strong interactions in non-leptonic decays of heavy mesons is extremely complex. While one has to admit that on the theory side a conceptual breakthrough for the systematic calculation of non-factorizable hadronic effects is still lacking, the combination of several theoretical methods in many cases still gives a satisfactory phenomenological picture.
- · Short-distance kernels in the QCD factorization approach are now being calculated at NNLO for a variety of decays.
- · Systematic studies of SU (3) F flavour-symmetry breaking effects on the basis of phenomenological data are available.
- · The ongoing improvement of the experimental situation leads to better knowledge on hadronic input parameters and more reliable estimates of systematic theoretical uncertainties.
We are thus looking forward to phenomenological updates that combine the stateof-the-art results for radiative corrections, hadronic input parameters, and SU (3) F -breaking effects.
## ACKNOWLEDGEMENTS
I would like to thank the organizers of FPCP'14 for the invitation and for a stimulating workshop program. I am also grateful to Guido Bell, Tobias Huber and Martin Jung for valuable scientific input.
## References
- [1] M. Artuso, D. M. Asner, P. Ball, E. Baracchini, G. Bell, M. Beneke, J. Berryhill and A. Bevan et al. , Eur. Phys. J. C 57 (2008) 309 [arXiv:0801.1833 [hep-ph]].
- [2] M. Antonelli, D. M. Asner, D. A. Bauer, T. G. Becher, M. Beneke, A. J. Bevan, M. Blanke and C. Bloise et al. , Phys. Rept. 494 (2010) 197 [arXiv:0907.5386 [hep-ph]].
- [3] RAaij et al. [LHCb Collaboration], Eur. Phys. J. C 73 (2013) 2373 [arXiv:1208.3355 [hep-ex]].
- [4] A. J. Bevan, B. Golob, T. Mannel, S. Prell, B. D. Yabsley, K. Abe, H. Aihara and F. Anulli et al. , arXiv:1406.6311 [hep-ex].
- [5] M. Beneke, G. Buchalla, M. Neubert and C. T. Sachrajda, Phys. Rev. Lett. 83 (1999) 1914 [hep-ph/9905312]. Nucl. Phys. B 591 (2000) 313 [hep-ph/0006124]; Nucl. Phys. B 606 (2001) 245 [hep-ph/0104110];
- [6] C. W. Bauer, S. Fleming, D. Pirjol and I. W. Stewart, Phys. Rev. D 63 (2001) 114020 [hep-ph/0011336].
- [7] M. Beneke, A. P. Chapovsky, M. Diehl and T. Feldmann, Nucl. Phys. B 643 (2002) 431 [hep-ph/0206152].
- [8] M. Beneke and M. Neubert, Nucl. Phys. B 675 (2003) 333 [hep-ph/0308039].
- [9] G. Bell, Nucl. Phys. B 795 (2008) 1 [arXiv:0705.3127 [hep-ph]]; Nucl. Phys. B 822 (2009) 172 [arXiv:0902.1915 [hep-ph]].
- [10] M. Beneke, T. Huber and X.-Q. Li, Nucl. Phys. B 832 (2010) 109 [arXiv:0911.3655 [hep-ph]].
- [11] G. Bell, M. Beneke, T. Huber,and X.-Q. Li, in preparation.
- [12] M. Beneke at '3rd KEK Flavor Factory Workshop', Tsukuba, Feb 13-15, 2014; T. Huber at 'Loops and Legs 2014', Weimar, Apr 27 - May 2; G. Bell at 'BEACH 2014', Birmingham, Jul 2126, 2014.
- [13] M. Beneke and S. J¨ger, Nucl. Phys. B a 751 (2006) 160 [hep-ph/0512351].
- [14] N. Kivel, JHEP 0705 (2007) 019 [hep-ph/0608291].
- [15] V. Pilipp, Nucl. Phys. B 794 (2008) 154 [arXiv:0709.3214 [hep-ph]].
- [16] M. Beneke and S. J¨ger, Nucl. Phys. B a 768 (2007) 51 [hep-ph/0610322].
- [17] A. Jain, I. Z. Rothstein and I. W. Stewart, arXiv:0706.3399 [hep-ph].
- [18] G. Bell and V. Pilipp, Phys. Rev. D 80 (2009) 054024 [arXiv:0907.1016 [hep-ph]].
- [19] G. Bell, talk presented at CKM 2010, University of Warwick, Sep 2010.
- [20] G. Zhu, JHEP 1005 (2010) 063 [arXiv:1002.4518 [hep-ph]].
- [21] H.-Y. Cheng, C.-K. Chua, K.-C. Yang and Z.-Q. Zhang, Phys. Rev. D 87 (2013) 11, 114001 [arXiv:1303.4403 [hep-ph]].
- [22] T. Huber and S. Kr¨nkl, arXiv:1405.5911 [hep-ph]. a
- [23] T. Feldmann, D. van Dyk, work in progress.
- [24] T. Huber, S. Kr¨nkl, T. Mannel, D. Rosenthal, J. Virto, work in progress (private a communication).
- [25] Y. Grossman, Z. Ligeti and D. J. Robinson, JHEP 1401 (2014) 066 [arXiv:1308.4143 [hep-ph]].
- [26] X.-G. He, S.-F. Li and H.-H. Lin, JHEP 1308 (2013) 065 [arXiv:1306.2658 [hepph]].
- [27] M. Gronau, Phys. Lett. B 727 (2013) 136 [arXiv:1308.3448 [hep-ph]].
- [28] C. W. Bauer, D. Pirjol, I. Z. Rothstein and I. W. Stewart, Phys. Rev. D 70 (2004) 054015 [hep-ph/0401188].
- [29] G. Hiller, M. Jung and S. Schacht, Phys. Rev. D 87 (2013) 014024 [arXiv:1211.3734 [hep-ph]]; PoS EPS -HEP2013 (2014) 371 [arXiv:1311.3883 [hep-ph]].
- [30] Y. Grossman and D. J. Robinson, JHEP 1304 (2013) 067 [arXiv:1211.3361 [hepph]].
- [31] J. Brod, Y. Grossman, A. L. Kagan and J. Zupan, JHEP 1210 (2012) 161 [arXiv:1203.6659 [hep-ph]]; see also J. Brod, talk at this conference.
- [32] B. Bhattacharya, M. Gronau, M. Imbeault, D. London and J. L. Rosner, arXiv:1402.2909 [hep-ph].
- [33] C.-K. Chua, Phys. Rev. D 89 (2014) 056003 [arXiv:1312.2335 [hep-ph]].
- [34] M. Jung, Phys. Rev. D 86 (2012) 053008 [arXiv:1206.2050 [hep-ph]].
- [35] P. Ball and R. Zwicky, Phys. Rev. D 71 (2005) 014015 [hep-ph/0406232]; Phys. Rev. D 71 (2005) 014029 [hep-ph/0412079].
- [36] G. Duplancic, A. Khodjamirian, T. Mannel, B. Melic and N. Offen, JHEP 0804 (2008) 014 [arXiv:0801.1796 [hep-ph]].
- [37] L. Lellouch, talk at this conference; S. Meinel, talk at this conference.
- [38] V. M. Braun, D. Y. Ivanov and G. P. Korchemsky, Phys. Rev. D 69 (2004) 034014 [hep-ph/0309330].
- [39] G. Bell, T. Feldmann, Y.-M. Wang and M. W. Y. Yip, JHEP 1311 (2013) 191 [arXiv:1308.6114 [hep-ph]].
- [40] V. M. Braun and A. N. Manashov, arXiv:1404.0863 [hep-ph].
- [41] T. Feldmann, B. O. Lange and Y. -M. Wang, Phys. Rev. D 89 (2014) 114001 [arXiv:1404.1343 [hep-ph]].
- [42] M. Beneke and J. Rohrwild, Eur. Phys. J. C 71 (2011) 1818 [arXiv:1110.3228 [hep-ph]].
- [43] V. M. Braun and A. Khodjamirian, Phys. Lett. B 718 (2013) 1014 [arXiv:1210.4453 [hep-ph]].
- [44] K. Wang and G. Zhu, Phys. Rev. D 88 (2013) 014043 [arXiv:1304.7438 [hep-ph]].
- [45] Q. Chang, X.-W. Cui, L. Han and Y.-D. Yang, Phys. Rev. D 86 (2012) 054016 [arXiv:1205.4325 [hep-ph]].
- [46] M. Gronau, D. London and J. L. Rosner, Phys. Rev. D 87 (2013) 3, 036008 [arXiv:1211.5785 [hep-ph]]. | null | [
"Thorsten Feldmann"
] | 2014-08-01T20:28:36+00:00 | 2014-08-01T20:28:36+00:00 | [
"hep-ph"
] | Non-Leptonic Heavy Meson Decays - Theory Status | I briefly review the status and recent progress in the theoretical
understanding of non-leptonic decays of beauty and charm hadrons. Focusing on a
personal selection of topics, this covers perturbative calculations in quantum
chromodynamics, analyses using flavour symmetries of strong interactions, and
the modelling of the relevant hadronic input functions. |
1408.0302v1 | ## Imitative learning as a connector of collective brains
Jos´ e F. Fontanari
Instituto de F´ ısica de S˜o Carlos, a Universidade de S˜o Paulo, a Caixa Postal 369, 13560-970 S˜o Carlos, a S˜ ao Paulo, Brazil
The notion that cooperation can aid a group of agents to solve problems more efficiently than if those agents worked in isolation is prevalent, despite the little quantitative groundwork to support it. Here we consider a primordial form of cooperation - imitative learning - that allows an effective exchange of information between agents, which are viewed as the processing units of a social intelligence system or collective brain. In particular, we use agent-based simulations to study the performance of a group of agents in solving a cryptarithmetic problem. An agent can either perform local random moves to explore the solution space of the problem or imitate a model agent - the best performing agent in its influence network. There is a complex trade-off between the number of agents N and the imitation probability p , and for the optimal balance between these parameters we observe a thirtyfold diminution in the computational cost to find the solution of the cryptarithmetic problem as compared with the independent search. If those parameters are chosen far from the optimal setting, however, then imitative learning can impair greatly the performance of the group. The observed maladaptation of imitative learning for large N offers an alternative explanation for the group size of social animals.
PACS numbers: 87.23.Ge, 89.75.Fb, 05.50.+q
## I. INTRODUCTION
Imitative learning or, more generally, social learning offers a means whereby information can be transferred between biological or artificial agents, being thus a crucial factor for the emergence of social intelligence or collective brains [1]. Its relevance in this context is neatly expressed by Bloom: 'Imitative learning acts like a synapse, allowing information to leap the gap from one creature to another'[2]. Not surprisingly, the advantages of this learning strategy were perceived and exploited by nature well before the advent of the human species as attested by its widespread use in the animal kingdom [3-6]. We note that imitation as a mechanism of social learning in humans was extensively studied by Bandura in the 1960s [7, 8].
Social learning has inspired the design of several optimization techniques, such as the particle swarm optimization algorithm [9, 10] and the adaptive culture heuristic [11, 12]. Despite the success of these heuristics in producing optimal or near optimal solutions to combinatorial optimization problems, we know little about the factors that make cooperation effective, as well as about the quantitative improvements that results from it [13]. The reason is probably that those heuristics and the problems they are set to solve are too complex to yield to a first-principle analysis. In this contribution we address these issues by tackling a simple combinatorial problem and by endowing the agents with straightforward search strategies in which the strength of collaboration is controlled by a single parameter of the model. In particular, we solve a cryptarithmetic problem using a group of N agents which, in addition to the capacity to carry out random local searches, can learn from (or imitate) a model agent - the best performing agent in their influence networks at a given trial. The frequency of the imitative or cooperative behavior is determined by the imitation probability parameter p ≥ 0. Hence our model exhibits two critical ingredients of a collective brain, namely, imitative learning and a dynamic hierarchy among the agents [2].
We find that imitative learning can greatly improve the performance of the group of agents provided that the control parameters are not too distant from their optimal values. For instance, in an optimal setting, say a fully connected system of N = 7 agents with imitation probability p = 0 6, we find a thirtyfold decrease of the . mean number of trials necessary to find the solution of the cryptarithmetic problem as compared with the baseline case p = 0 where the N agents explore the solution space independently. Most significantly, however, we find that, for a fixed value of the imitation probability, increasing the number of agents N beyond a certain value impairs the group performance which can then perform much worse than in the case of the independent search. The following of a bad model is the culprit for the poor performance in this case. This harmful effect can be mitigated somewhat by reducing the connectivity of the agents so as to limit the influence of a model agent to only a fraction of the group. We argue that the maladaptation of imitative learning for large systems may be an alternative explanation for the group size of social animals.
## II. METHODS
First we will describe the particular cryptarithmetic problem the agents must solve, explain how the digitto-letter identifications are encoded in strings and introduce the cost value associated to those strings. We will present also the elementary move that transforms any
valid string into an adjacent valid string, allowing thus the exploration of the solution space. Once these basic elements are introduced we will describe the mechanism of imitation between agents, thus completing the specification of the agent-based model we use to evaluate the efficacy of imitative learning in solving a complex task.
## A. The cryptarithmetic problem
Cryptarithmetic problems such as
$$D O N A L D + G E R A L D = R O B E R T \quad ( 1 ) \quad \text{cryp}$$
are constraint satisfaction problems in which the task is to find unique digit assignments to each of the letters so that the numbers represented by the words add up correctly [14]. In the cryptarithmetic problem of eq. (1), there are 10! different digit-to-letter assignments, of which only one is the solution to the problem, namely, A = 4 , B = 3 , D = 5 , E = 9 , G = 1 , L = 8 , N = 6 , O = 2 , R = 7 , T = 0. This type of cryptarithmetic problem, in which the letters form meaningful words, are also termed alphametics [15] and were popularized in the 1930s by the Sphinx , a Belgian journal of recreational mathematics [14]. Of course, from the perspective of evaluating the performance of search heuristics on solving cryptarithmetic problems, the meaningfulness of the words is inconsequential, but in this contribution we will focus mainly on the alphametic problem (1). Nonetheless we will offer evidence to support the validity of our conclusions by considering a few randomly generated cryptarithmetic problems as well.
A non-random search heuristics to solve cryptarithmetic problems requires the introduction of some arbitrary quality measure or cost value to each possible digitto-letter assignment. For the alphametic problem of eq. (1) we encode a digit-to-letter assignment by the string i = ( i 1 , i 2 , . . . , i 10 ) where i n = 0 , . . . , 9 represent the 10 digits and the subscripts n = 1 , . . . , 10 label the letters according to the convention
$$\text{onvention} \quad \begin{matrix} 1 & \rightarrow & A & & \text{re} \\ 2 & \rightarrow & B & & b ( \\ 3 & \rightarrow & D & & \text{ac} \\ 4 & \rightarrow & E & & \text{in} \\ 5 & \rightarrow & G & & \text{es} \\ 6 & \rightarrow & L & & m \\ 7 & \rightarrow & N & & M \\ 8 & \rightarrow & O & & \text{tc} \\ 9 & \rightarrow & R & & \text{in} \\ 10 & \rightarrow & T. & & ( 2 ) & m \\ \text{c string} & ( 0, 2, 9, 4, 8, 1, 7, 6, 3, 5 ) & \text{core} - \text{if} \end{matrix}$$
For example, the string (0 , 2 9 4 8 1 7 6 3 5) , , , , , , , , corresponds the the digit-to-letter assignment A = 0 , B = 2 , D = 9 , E = 4 , G = 8 , L = 1 , N = 7 , O = 6 , R = 3 , T =
5. A somewhat natural way to associate a cost to a string i is through the expression [16]
$$C \left ( \mathbf i \right ) = \left | R - \left ( F + S \right ) \right | \text{ \quad \ \ } \text{(3)}$$
where R is the result of the operation ( ROBERT ), F is the first operand ( DONALD ) and S is the second operand ( GERALD ). In our example we have R = 362435, F = 967019 and S = 843019 so that the cost associated to string (0 , 2 9 4 8 1 7 6 3 5) is , , , , , , , , C = 1447603. If the cost of a string is C = 0 then the digit-to-letter assignment coded by that string is the solution of the cryptarithmetic problem. We must note that the cost value defined in eq. (3) applies to all strings except those for which i 3 = 0 corresponding to the assignment D = 0, i 5 = 0 corresponding to the assignment G = 0 and i 9 = 0 corresponding to the assignment R = 0. In principle, those are invalid strings because they violate the rule of the cryptarithmetic puzzles that an integer should not have the digit 0 at its leftmost position. For those strings we assign an arbitrary large cost value, namely, C = 10 , 8 so that they can be considered valid strings and hence part of the solution space.
In addition to the assignment of the cost values to all 10! strings that code the possible digit-to-letter mappings for the alphametic problem (1), we introduce also an elementary move that connects two valid digit-toletter mappings. We define the elementary move as follows. Starting from a particular digit-to-letter mapping, say (0 2 9 4 8 1 7 6 3 5), , , , , , , , , , we choose two letter labels at random and then interchange the digits assigned to them. For example, say we pick letter labels 1 and 5, then the resulting mapping after the elementary move is (8 , 2 9 4 0 1 7 6 3 5). Clearly, the repeated application , , , , , , , , of our elementary move is capable of producing all 10! strings starting from any arbitrary valid digit-to-letter mapping.
## B. Imitative learning
The system is composed of N agents or strings which represent valid digit-to-letter assignments as described before. Each agent is connected unidirectionally to exactly M = 1 , . . . , N -1 distinct, randomly chosen agents in the system. We will refer to those agents as the 'influencers' of the target agent. More specifically, for each agent we sample M influencers from the N -1 remaining agents without replacement. The extreme case M = N -1 corresponds to the fully connected network. An agent has a probability p ∈ [0 , 1) of copying a digitto-letter assignment from a model string in its group of influencers, and probability 1 -p of performing the elementary move. We choose the model string as the lowest cost string among the M influencers of the target agent. If the cost associated to the target string is lower than the cost of the model string then the copying process is aborted.
To illustrate the copying process let us assume for the sake of concreteness that the target agent is our already familiar example string (0 2 9 4 8 1 7 6 3 5), , , , , , , , , , whose cost is C = 1447603, and that the model string is (5 , 3 9 4 8 1 6 2 7 0) whose cost is , , , , , , , , C = 1050568. In the copying process the target agent selects at random one of the distinct digit-to-letter assignments in the model string and assimilates it. In our example, the distinct assignments occur at the letter labels n = 1 2 7 8 9 10. , , , , , Say that the letter label n = 1, which corresponds to the assignment A = 5 according to our convention (2), is chosen. To assimilate this assignment the target agent needs to reassign the digit 0 to the letter label which was previously assigned to digit 5 so that the resulting string becomes (5 2 9 4 8 1 7 6 3 0), whose cost is , , , , , , , , , C = 1448608. As expected, a result of the imitative learning process is the increase of the similarity between the target and the model strings. The case p = 0 corresponds to the baseline limit where the N agents perform independent searches. The specific copying procedure proposed here was inspired by the mechanism used to model the influence of an external media [17-19] in Axelrod's model of culture dissemination [20]. It is important to note that in the case the target string is identical to the model string, as well as in the case the cost of the target string is lower than the cost of the model string, the opportunity of update is wasted.
We may interpret the imitation (or copying) process of a model string as a blackboard cooperation system where a central control exhibits hints (i.e., the lowest cost string) in a public space [13, 21], but here we prefer to use the interpretation of learning by imitation as it allows us to view the group of agents as a neural system where the agents play the role of neurons and dynamical synapses connect them to the best performing agents in the neural network [2]. Nevertheless, since the process of imitation results in an effective collaboration among agents, in the sense that there is an exchange of information between them, we refer to this search strategy as collaborative search to contrast with the independent search which occurs when the copying process is turned off, i.e, the imitation probability p is set to zero.
## C. Search dynamics
We begin by generating the N influence networks, i.e., a group of M influencers for each agent. These networks are kept fixed during the entire search. In this initial stage, at trial number t = 0, we also associate a random digit-to-letter assignment (a valid string) to each agent and determine its corresponding model string by evaluating and comparing the cost values of its M influencers.
A new trial begins with the choice of the update order of the N agents, so that at the end of the trial all N agents are updated. The agent to be updated - the target agent - has the possibility to imitate its model string or perform the elementary move with probabilities p and
1 -p , respectively. After update, we must re-evaluate the model string status in all groups of influencers to which the target agent belongs. After all N agents are updated we increment the trial number t by one unit and check whether any string has cost zero, in which case the search is halted. The trial number at which the search ends or, alternatively, the number of trial to success is denoted by t ∗ .
Except for the independent search ( p = 0), the update of the N agents is not strictly a parallel process since the model strings may change several times within a given trial. Nonetheless, since in a single trial all agents are updated, the total number of agent updates at trial t is with given by the product Nt .
## III. RESULTS
The efficiency of a search strategy is measured by the total number of agent updates necessary to find the solution of the cryptarithmetic problem (i.e., Nt ∗ ) and in the following we will refer to this measure as the computational cost of the search. Since we expect that the typical number of trials to success t ∗ scales with the size of the solution space (i.e., 10!), we will present the results in terms of the rescaled variable τ = t ∗ / 10!. For the purpose of comparison we will consider first the baseline case of independent search for which the agents can perform the elementary move only ( p = 0) and then the general case of cooperative search ( p > 0).
## A. Independent search
In this case there is no imitation and so the influence networks have no role in the outcome of the search. The main results of the independent search are summarized in fig. 1, which shows the probability distribution P ( Nτ ) of the rescaled computational cost Nτ of the search for several system sizes. The data is very well fitted by the exponential distribution P ( Nτ ) = a exp( -aNτ ) with a = 1 1 14 which is shown by the solid straight line in / . the figure.
As expected, the mean rescaled computational cost 〈 Nτ 〉 ≈ 1 14 is insensitive to the system size . N , but the finding that it does not equal 1 is somewhat surprising. In fact, if we replace our elementary move by a global move in which the entire string is generated randomly at each update then we find that this mean equals 1, as expected. The reason that our elementary move is slightly less efficient than the global move in exploring the solution space is because it is not too unlikely to reverse a change made by the elementary move. For example, the probability to reverse a change in a subsequent trial is 2 / 10 × 1 / 9 = 2 / 90 for the the elementary move, whereas it is 1 / 10! for the global move.
FIG. 1: Exponential distribution of the rescaled computational cost for the independent search. Probability distribution P ( Nτ ) that a search employing N independent agents finds the solution of the cryptarithmetic problem (1) using a total of Nt ∗ updates for N = 1( ) 5( ◦ , glyph[triangle] ) , 10( glyph[triangleinv] ) and 20( glyph[square] ). Here τ = t ∗ / 10! is the ratio between number of trials to success and the size of the solution space. These distributions were generated using 10 5 independent searches for each N . The solid straight line is the exponential distribution P Nτ ( ) = a exp( -aNτ ) with a = 1 1 14. / .
## B. Cooperative search
As pointed out before, the cooperation among agents stems from the possibility that they copy potentially relevant digit-to-letter assignments from the model strings in their influence networks. We will consider first the fully connected system where M = N -1 and then the partially connected systems where 1 ≤ M < N -1.
## 1. Fully connected system
Figure 2 shows how the mean rescaled computational cost is affected by varying the imitation probability p while the number of agents N is kept at a fixed value. For N = 20 and p = 0 5 we observe a twentyfold de-. crease of the mean cost in comparison with the cost of the independent search, which corresponds to p = 0 and yields 〈 Nτ 〉 ≈ 1 14. . This is a remarkable evidence of the power of imitative learning to speed up the search on the solution space of the cryptarithmetic problem. In the limit p → 1 one expects the computational cost to diverge since the solution space cannot be fully explored as the option for the elementary move is never made in this limit. This harmful effect of learning by imitation becomes more pronounced as the number of agents increases.
In the region where the mean computational cost decreases monotonically with increasing p (e.g., p < 0 5 for . N = 20) we found that the probability distribution of
FIG. 2: The effect of the imitation probability on the computational cost of the fully connected system. The symbols represent the mean rescaled computational cost 〈 Nτ 〉 for cooperative systems of size N = 20( · ) , N = 5( glyph[trianglesolid] ) , N = 3( glyph[triangledownsld] ) and N = 2( glyph[squaresolid] ). The independent variable p is the probability that an agent will copy a digit-to-letter assignment from the model string, chosen as the lowest cost string in the entire system. Each symbol represents an average over 10 5 searches and the lines are guides to the eye. The error bars are smaller than the size of the symbols.
FIG. 3: Deviation from the exponential distribution for a large imitation probability. Probability distribution P ( Nτ ) of the rescaled computational cost for a search employing N = 20 fully connected agents with imitation probability p = 0 6. . The mean of this distribution is 〈 Nτ 〉 ≈ 8 0. . The solid straight line is the fitting function a exp( -bNτ ) with a = 0 03 and . b = 1 15 in the regime of large cost. / The distribution was generated using 10 5 independent searches.
the computational cost is well described by an exponential distribution, in the sense that the ratio between the standard deviation and the mean is always very close to 1. (We recall that this ratio equals 1 for an exponential distribution.) However, in the region where 〈 Nτ 〉 increases
with increasing p we found that in the low cost regime P ( Nτ ) gives values significantly greater than those predicted by an exponential distribution, as illustrated in fig. 3, though those values are not greater than those obtained in the case of the independent search (see fig. 1).
FIG. 4: The effect of the system size on the computational cost of the fully connected system. The symbols represent the mean rescaled computational cost 〈 Nτ 〉 for the imitation probability p = 0( ) · , p = 0 3( . glyph[diamondsolid] ) , p = 0 4( . glyph[squaresolid] ) , p = 0 5( . glyph[triangledownsld] ) and p = 0 6( . glyph[trianglesolid] ) The independent variable N is the number of agents in the system. Each symbol represents an average over 10 5 searches and the lines are guides to the eye. The error bars are smaller than the size of the symbols.
The effect of increasing the number of agents N for a fixed value of the imitation probability p is summarized in fig. 4. The mean computational cost of the cooperative system exhibits a non-monotonic dependence on N , except in the case of the independent search ( p = 0) when it takes on a constant value. The benefit of cooperation is seen in this figure by the initial decrease of the computational cost as the number of agents increases. However, for all p > 0 we find that the presence of too many agents greatly harms the performance of the system and that for a fixed p > 0 there exists an optimum value of N that maximizes the search efficiency of the cooperative system. For instance, although not shown in the scale of fig. 4, the minimum computational cost for p = 0 3 occurs . at N ≈ 270. The efficiency at this optimum, however, is not affected significantly by the choice of the parameters N and p . In other words, the costs corresponding to the minima shown in figs. 2 and 4 are not very sensitive to changes in N and p , respectively. In particular, for the parameter settings we have explored, the best efficiency 〈 Nτ 〉 ≈ 0 041 is achieved for . N = 7 and p = 0 6 . and amounts to a thirtyfold speed up with respect to the independent search.
We conjecture that the reason the efficiency of the cooperative system deteriorates as N increases beyond its optimum value (e.g., in the range N > 6 for p = 0 6 as .
shown in fig. 4) is that for N not too small there is a good chance that the cost of one of the strings is significantly lower than the cost of the other N -1 strings. Provided p is not too small too, this string may remain as the model string for a few trials thus biasing the search to the vicinity of the model string. In the (typical) case that the model string is far from the solution of the cryptarithmetic problem, imitative learning may lead to the observed impairment of the performance of the cooperative system. In sum, the following of a bad leader is likely the culprit of the poor performance of the system.
FIG. 5: Probability distribution of the mean duration of the stases in a search. Probability distribution of the mean number of trials ζ for which a cost value stays as the lowest cost among the N solutions in 10 5 searches for the imitation probability p = 0 6 in a fully connected system. . Panel A : N = 6 (low computational cost regime). Panel B : N = 17 (high computational cost regime). The slope of the straight line shown in the semi-log scale of panel B is 0 003. .
To check the validity of this conjecture we calculate the mean number of consecutive trials for which a cost value stays as the lowest cost among the N strings. The procedure to obtain this quantity, which we denote by ζ , is straightforward. At trial t = 0 we evaluate the cost of
the N strings and record the minimal cost among them. Then at the next trial t = 1, after the N strings are updated, we re-evaluate again their costs and record the minimal cost. If the minimal cost at t = 1 is different, i.e., greater or less, than the minimal cost at t = 0 we say that a change event has occurred. The comparison of the values of the minimal costs at consecutive trials is repeated and the cumulative number of change events is recorded until the solution is found at t = t ∗ . The desired quantity ζ is given simply by the ratio between the total number of change events and the total number of trials t ∗ . Hence for each search we obtain a single value for ζ , which can then be interpreted as the mean number of trials between consecutive change events or as the mean duration of the stases for that search.
In fig. 5 we present the probability distribution Q ζ ( ) using 10 5 searches for the imitation probability p = 0 6 . and two representative values of N . Figure 5 A shows this distribution for N = 6, which corresponds to a regime of low computational cost according to fig. 4. We observe a pronounced maximum at ζ ≈ 3 7 so that in most searches . the model cost remains unaltered for 3 to 5 trials. This is an optimum scenario since no string stays on the top tier long enough to influence the entire system. For N < 6, we find that Q ζ ( ) exhibits a similar shape but the maximum becomes sharper and its location is shifted towards lower values of ζ as N decreases. Figure 5 B , which shows the results for N = 17, reveals a very different scenario: the distribution Q ζ ( ) exhibits a plateau indicating that the model cost remains unchanged for hundreds to a few thousands trials. For very large values of ζ , the distribution Q ζ ( ) seems to exhibit an exponential decay to zero, namely, Q ζ ( ) ∼ exp( -0 003 . ζ ). We stress that for the two cases exhibited in fig. 5 the probability that an agent will imitate the model rather than perform an elementary move is the same, namely p = 0 6, and so the qualitative . differences reported in the figure are due solely to the change on the number of agents.
## 2. Partially connected system
If the poor performance of large collaborative systems based on imitative learning is due to the influence of bad models then a natural way to reduce this harmful effect is to limit the influence of those models. This was the motivation to introduce the influence networks scheme where each agent picks its model among M randomly chosen agents predetermined at the beginning of the search. In fact, fig. 6 shows that the reduction of the connectivity of the agents increases somewhat the range of values of the imitation probability p for which the cooperative system outperforms the system composed of independent agents. More pointedly, for N = 20 this range is extended from p ≈ 0 57 for . M = 19 to p ≈ 0 87 for . M = 1. In addition, the value of the optimal mean computational cost does not seem to vary significantly with M . Figure 7 offers another perspective on the role of the number of influ-
FIG. 6: The effect of the imitation probability on the computational cost of partially connected systems. The symbols represent the mean rescaled computational cost 〈 Nτ 〉 for a system composed of N = 20 agents, each one connected to M = 19( · ) , M = 9( glyph[trianglesolid] ) , M = 4( glyph[triangledownsld] ) and M = 1( glyph[squaresolid] ) influencers. The independent variable p is the imitation probability. Each symbol represents an average over 10 5 searches and the lines are guides to the eye. The error bars are smaller than the size of the symbols.
FIG. 7: The effect of the number of influencers on the mean computational cost. The symbols represent the mean rescaled computational cost 〈 Nτ 〉 for a system composed of N = 40 agents and imitation probability p = 0 5( . · ) , p = 0 45( . glyph[trianglesolid] ) , p = 0 4( . glyph[triangledownsld] ) and p = 0 3( . glyph[squaresolid] ). The independent variable M is the size of the group of influencers of each agent. Each symbol represents an average over 10 5 searches and the lines are guides to the eye. The error bars are smaller than the size of the symbols.
encers M . It shows that for small values of the imitation probability the fully connected system (i.e., M = N -1) exhibits the best performance. However, as p increases (e.g., p > 0 4 for . N = 40), the optimal performance is obtained with partially connected systems. Moreover, we
found that for any fixed value of p > 0 and M the performance of the system is always impaired when the number of agents N is very large. Finally, we note that similarly to our findings for the fully connected system, the probability distribution of the computational cost P ( Nτ ) departs significantly from an exponential distribution only in the regions where the mean computational cost becomes an increasing function of the control parameters of the model.
## 3. Random cryptarithmetic problems
In order to verify the generality of our findings, which were obtained for the specific alphametic problem DONALD + GERALD = ROBERT , we have considered a variety of random cryptarithmetic problems with 10 letters and a unique solution, so that the sizes of their solution spaces are the same as that of the alphametic problem. The comparison between the mean computational costs to solve four such random problems and our alphametic problem is shown in fig. 8 for the fully connected system. The results are qualitatively the same, as expected. The alphametic problem, however, was somewhat easier to solve by the cooperative system than the random problems, perhaps because of the coincidence of the last three letters ('ALD') in the first and second operands. Interestingly, the independent system ( p = 0) cannot distinguish between the problems but the cooperative system ( p > 0) can, and this distinction is most pronounced when the parameters are set so as to achieve the optimal performance. It is as if the cooperative system had adapted to the specific task posed to it. We expect that our conclusions remain valid, in a qualitative sense of course, for any constraint satisfaction problem characterized by a rugged cost landscape.
## IV. DISCUSSION
Rather than offer any novel method to solve cryptarithmetic problems, our aim in this contribution is to assess quantitatively the potential of imitative learning as the underlying mechanism - the critical connector - of collective brains [2]. Here imitative learning is implemented by allowing an agent to copy clues from the best performing agent - the model agent - in its group of influencers. More pointedly, at trial t each agent has the probability p of imitating the model and the probability 1 -p of executing a random rearrangement of the digitto-letter mapping which is its guess to the solution of the cryptarithmetic problem. In an optimal setting, say a fully connected system of N = 7 agents with imitation probability p = 0 6, we find a thirtyfold decrease of the . mean number of trials needed to find the solution of the problem (i.e., of the mean computational cost), as compared with the case p = 0 when the agents search the solution space independently (see fig. 4).
FIG. 8: Computational cost of the alphametic problem and of four random cryptarithmetic problems. The mean rescaled computational cost for the alphametic problem DONALD + GERALD = ROBERT (symbol · ) and for four ten-letter random cryptarithmetic problems with a unique solution (symbols glyph[triangledownsld] , glyph[trianglesolid] , glyph[squaresolid] and glyph[diamondsolid] ). The symbols represent the mean rescaled computational cost 〈 Nτ 〉 for a system composed of N = 20 fully connected agents. The independent variable p is the imitation probability. Each symbol represents an average over 10 5 searches and the lines are guides to the eye. The error bars are smaller than the size of the symbols.
In the optimal setting, as well as in the regions where the computational cost is a decreasing function of the control parameters of the model, the probability distribution of the computational cost is given by an exponential distribution, rather than by a lognormal distribution as predicted by a general theory of cooperative processes [13, 22]. In fact, the reason the cooperative scheme implemented in [13] is so efficient is that all discovered digitto-letter assignments that add up correctly modulo 10 for at least one column are permanently exposed as hints in a blackboard for use by all agents, which can pick a hint at each trial. There is no place for any kind of learning in that scenario since in the case there are no hints in the blackboard or the agent has already used the chosen one, the agent selects a new random digit-to-letter assignment which is completely uncorrelated to its previous assignment.
Most significantly, for fixed values of the imitation probability p and of the number of influencers M , we find that increasing the number of agents N beyond a certain quantity impairs the working of the cooperative system, which then performs much worse than if the agents had executed independent searches. Our analysis indicates that the following of a bad model is the culprit of the poor performance of the system in this case. In that sense, the efficacy of imitative learning could be a factor determinant of group size [23]. In contrast to the cognitive load that constrains the number of individuals with whom it is possible to maintain stable relationships and
leads to Dunbar's number for primates [24, 25], the group size here (i.e., the value of N corresponding to the lowest computational cost) does not stem from a limitation of the neocortical processing capacity of the individuals. Rather, it is a property of the group of agents as a whole, since for any fixed non-vanishing value of the imitation probability, which may be seen as an individual trait, a too large number of agents, which is a group property, will impair the performance of the cooperative system. Of course, if p were allowed to decrease with increasing N then the system could be maintained at the highest level of perform regardless of the group size (see figs. 2 and 4). In other words, in order to perform at the optimal level a system based on imitative learning should decrease the frequency of the interactions among individuals as its size increases.
To conclude, our findings indicate that imitative learning has a great potential to improve the task-solving capability of a group of agents, provided the model param-
- [1] C. L. Nehaniv and K. Dautenhah, in Imitation and Social Learning in Robots, Humans and Animals , edited by C. L. Nehaniv and K. Dautenhah (Cambridge University Press, Cambridge, UK, 2007) pp. 1-18
- [2] H. Bloom, Global Brain: The Evolution of Mass Mind from the Big Bang to the 21st Century (Wiley, New York, 2001)
- [3] B. R. Moore, in Social learning in animals: The roots of culture , edited by C. M. Heyes and B. G. Galef, Jr (Academic Press, San Diego, CA, 1996) pp. 245-265
- [4] G. Fiorito and P. Scotto, Science 256 , 545 (1992)
- [5] K. N. Laland, N. Atton, and M. M. Webster, Phil. Trans. R. Soc. B 366 , 958 (2011)
- [6] C. Heyes, Phil. Trans. R. Soc. B 367 , 2181 (2012)
- [7] A. Bandura, in Nebraska Symposium on Motivation , edited by M. R. Jones (University of Nebraska Press, Lincoln, NE, 1962) pp. 211-274
- [8] A. Bandura, Social learning theory (Prentice Hall, New York, 1977)
- [9] J. Kennedy and R. Eberhart, in Proceedings of the IEEE International Conference on Neural Networks , Vol. 4 (IEEE Computer Society, Washington, DC, 1995) pp. 1942-1948
- [10] E. Bonabeau, M. Dorigo, and G. Theraulaz, Swarm Intelligence: From Natural to Artificial Systems (Oxford University Press, Oxford, UK, 1999)
- [11] J. Kennedy, J. Conflict Res. 42 , 56 (1998)
- [12] J. F. Fontanari, Phys. Rev. E 82 , 056118 (2010)
eters - number of agents ( N ), imitation probability ( p ) and number of influencers ( M ) - are not too far from their optimal values. In the cases that N or p are too large, the imitative learning strategy leads the cooperative system astray, in a sort of maladaptive behavior that has actually been observed in fishes [26]. It would be interesting to find out what new ingredients one should add to our model in order to prevent the catastrophic effect of imitative learning on large populations.
## Acknowledgments
This work was partially supported by grant 2013/17131-0, S˜ ao Paulo Research Foundation (FAPESP) and by grant 303979/2013-5, Conselho Nacional de Desenvolvimento Cient´ ıfico e Tecnol´gico o (CNPq).
- [13] S. H. Clearwater, B. A. Huberman, and T. Hogg, Science 254 , 1181 (1991)
- [14] B. Averbach and O. Chein, Problem Solving Through Recreational Mathematics (Freeman, San Francisco, 1980)
- [15] J. A. Hunter, Mathematical Brain Teasers (Dover, New York, 1976)
- [16] R. Abbasian and M. Mazloom, in Proceedings of the Second International Conference on Computer and Electrical Engineering , Vol. 2 (IEEE Computer Society, Washington, DC, 2009) pp. 308-312
- [17] Y. Shibanai, S. Yasuno, and I. Ishiguro, J. Conflict Res. 45 , 80 (2001)
- [18] J. Gonz´ alez-Avella, M. Cosenza, V. M. Egu´ ıluz, and M. S. Miguel, New J. Phys. 12 , 013010 (2010)
- [19] L. R. Peres and J. F. Fontanari, Europhys. Lett. 96 , 38004 (2011)
- [20] R. Axelrod, J. Conflict Res. 41 , 203 (1997)
- [21] R. Englemore and T. Morgan, Blackboard Systems (Addison-Wesley, New York, 1988)
- [22] B. A. Huberman, Physica D 42 , 38 (1990)
- [23] E. Wilson, Sociobiology (Harvard University Press, Cambridge, MA, 1975)
- [24] R. I. M. Dunbar, J Human Evol 22 , 469 (1992)
- [25] B. Gon¸ calves, N. Perra, and A. Vespignani, PLoS ONE 6 , e22656 (2011)
- [26] K. N. Laland and K. Williams, Behav. Ecol. 9 , 493 (1998) | 10.1371/journal.pone.0110517 | [
"José F. Fontanari"
] | 2014-08-01T20:34:26+00:00 | 2014-08-01T20:34:26+00:00 | [
"nlin.AO",
"physics.soc-ph"
] | Imitative learning as a connector of collective brains | The notion that cooperation can aid a group of agents to solve problems more
efficiently than if those agents worked in isolation is prevalent, despite the
little quantitative groundwork to support it. Here we consider a primordial
form of cooperation -- imitative learning -- that allows an effective exchange
of information between agents, which are viewed as the processing units of a
social intelligence system or collective brain. In particular, we use
agent-based simulations to study the performance of a group of agents in
solving a cryptarithmetic problem. An agent can either perform local random
moves to explore the solution space of the problem or imitate a model agent --
the best performing agent in its influence network. There is a complex
trade-off between the number of agents N and the imitation probability p, and
for the optimal balance between these parameters we observe a thirtyfold
diminution in the computational cost to find the solution of the
cryptarithmetic problem as compared with the independent search. If those
parameters are chosen far from the optimal setting, however, then imitative
learning can impair greatly the performance of the group. The observed
maladaptation of imitative learning for large N offers an alternative
explanation for the group size of social animals. |
1408.0303v4 | ## Heterogeneous, Weakly Coupled Map Lattices
M Dolores Sotelo Herrera , Jes´s San Mart´ ın , Mason A. Porter a a u a b
- a Departamento de Matem´ atica Aplicada, E.T.S.I.D.I., Universidad Polit´cnica de Madrid, e Madrid, Spain
- b Oxford Centre for Industrial and Applied Mathematics, Mathematical Institute, University of Oxford, Oxford, United Kingdom
## Abstract
Coupled map lattices (CMLs) are often used to study emergent phenomena in nature. It is typically assumed (unrealistically) that each component is described by the same map, and it is important to relax this assumption. In this paper, we characterize periodic orbits and the laminar regime of type-I intermittency in heterogeneous weakly coupled map lattices (HWCMLs). We show that the period of a cycle in an HWCML is preserved for arbitrarily small coupling strengths even when an associated uncoupled oscillator would experience a period-doubling cascade. Our results characterize periodic orbits both near and far from saddle-node bifurcations, and we thereby provide a key step for examining the bifurcation structure of heterogeneous CMLs.
Keywords: heterogeneous CML, intermittency, period preservation, synchronization
## 1. Introduction
Numerous phenomena in nature - such as human waves in stadiums [1] and flocks of seagulls [2] - result from the interaction of many individual elements, and they can exhibit fascinating emergent dynamics that cannot arise in individual or even small numbers of components [3]. In practice, however, a key assumption in most such studies is that each component is described by the same dynamical system. However, systems with heterogeneous elements are much more common than homogeneous systems. For example, a set of interacting cars on a highway that treats all cars as the same ignores different types of cars (e.g., their manufacturer, their age, different levels of intoxication among the drivers, etc.), and a dynamical system that governs the behavior of different cars could include different parameter values or even different functional forms entirely for different cars. Additionally, one needs to use different functional forms to address phenomena such as interactions among cars, traffic
Email address: [email protected] (Jes´s San Mart´ ın) u
December 11, 2015
lights, and police officers. Unfortunately, because little is known about heterogeneous interacting systems [4, 5], the assumption of homogeneity is an important simplification that allows scholars to apply a plethora of analytical tools. Nevertheless, it is important to depart from the usual assumption of homogeneity and examine coupled dynamical systems with heterogeneous components.
The study of coupled map lattices (CMLs) [6, 7] is one important way to study the emergent phenomena (e.g., cooperation, synchronization, and more) that can occur in interacting systems. CMLs have been used to model systems in numerous fields that range from physics and chemistry to sociology, economics, and computer science [7, 8, 9, 10, 11]. In a CML, each component is a discrete dynamical system (i.e., a map). There are a wealth of both theoretical and computational studies of homogeneous CMLs [6, 7, 12, 13, 14, 15, 16, 17, 18], in which the interacting elements are each governed by the same map. Such investigations have yielded insights on a wide variety of phenomena. As we mentioned above, the assumption of homogeneity is a major simplification that often is not justifiable. Therefore, we focus on heterogeneous CMLs, in which the interacting elements are governed by different maps or by the same map with different parameter values. The temporal evolution of a heterogeneous coupled map lattice (CML) with p components is given by
/negationslash
$$X _ { i } ( n + 1 ) = f _ { R _ { i } } ( X _ { i } ( n ) ) + \varepsilon \sum _ { \substack { h = \frac { 1 } { h } \\ h \neq i } } ^ { p } f _ { R _ { h } } ( X _ { h } ( n ) ) \,, \quad i \in \{ 1, \dots, p \} \,, \quad ( 1 )$$
where X n i ( ) represents the state of the entity at instant n at position i of a lattice and ε > 0 weights the coupling between these entities. We consider entities in the form of oscillators, where the i th oscillator evolves according to the map
$$X _ { i } ( n + 1 ) = f _ { R _ { i } } ( X _ { i } ( n ) ) \,, \quad i \in \{ 1, \dots, p \} \,, \quad \quad ( 2 )$$
where the f R i are, in general, different functions that depend on a parameter R i (where i ∈ { 1 , . . . , p } ). We assume that each f R i is a C 2 unimodal function that depends continuously on the parameter R i with a critical point C at R i . As usual, f m means that f is composed with itself m times. If an uncoupled oscillator X n i ( ) takes the value x i,n , then the evolution of this value under the map is x i,n +1 = f R i ( x i,n ).
In this paper, we examine heterogeneous, weakly coupled map lattices (HWCMLs). Weakly coupled systems can exhibit phenomena (e.g., phase separation because of additive noise [19]) that do not arise in strongly coupled systems, and one can even use weak coupling along with noise to fully synchronize nonidentical oscillators [20]. Thus, it is important to examine HWCMLs, which are amenable to perturbative approaches. In our paper, we characterize periodic orbits both far away from and near saddle-node (SN) bifurcations. Understanding periodic orbits is interesting by itself and is also crucial for achieving an understanding of more complicated dynamics (such as chaos) [21, 22]. We then characterize the laminar regime of type-I intermittency in our HWCMLs. Finally, we summarize our results and briefly comment on applications.
## 2. Theoretical Results
Before discussing our results, we need to define some notation. Let x i,n R | i denote the points in a periodic orbit of the i th uncoupled oscillator with control parameter R i . The parameter value r i is a bifurcation value of R i for the i th map, so x i,n r | i denotes the points in a periodic orbit at this parameter value.
Suppose that R i = r i + ε α , where ε is the same as in the coupling term of the CML (1) and α ∈ (0 , ∞ ) is a constant. We seek to derive results that are valid at size O ε ( ). We need to consider the following situations:
α < 1 In this case, when we expand to size O ε ( ), the coupling term does not contribute at all. Therefore, the oscillators in (1) behave as if they were uncoupled at this order of the expansion.
- α > 1 In this case, the coupling term controls the ε bifurcation terms. Thus, to size O ε ( ), we cannot study the behavior of the bifurcation.
α = 1 In this case, we are considering a perturbation of the same size as the coupling term, and we can simultaneously study the coupling and the bifurcation analytically.
/negationslash
To study orbits close to bifurcation points, we thus let R i = r i + , where ε ε is the same as in the coupling term of the CML (1). In our numerical simulations (see Section 3), we will also briefly indicate the effects of considering α = 1 (see Section 3.3).
## 2.1. Study of the CML Far from and Close to Saddle-Node Bifurcations
In this section, we examine heterogeneous CMLs in which the uncoupled oscillators have periodic orbits either far from or near SN bifurcations. As periodic orbits exhibit different dynamics from each other depending on whether they are near or far from SN bifurcations [23, 24], it is important to distinguish between these two situations.
A periodm SN orbit is a periodic orbit that is composed of m 'SN points' of the composite map f m r i . Each of these m SN points is a fixed point of f m r i at which f m r i undergoes an SN bifurcation. Periodm SN orbits play an important role in a map's bifurcation structure, because they occur at the beginning of periodic windows in bifurcation diagrams. Studying them is thus an important step towards examining the general bifurcation structure of a map.
When f m r i undergoes an SN bifurcation, the map f r i has two properties that we highlight. Let { x i, 1 | r i , x i, 2 | r i . . . , x i,m r | i } be an periodm SN orbit. It then follows that
1.
Consequently, orbits that are near an SN orbit satisfy
$$\frac { \partial f _ { r _ { i } } ^ { m } } { \partial x } ( x _ { i, j | r _ { i } } ) = 1 = \prod _ { k = j } ^ { j + m - 1 } \frac { \partial f _ { r _ { i } } } { \partial x } ( x _ { i, k | r _ { i } } ) \\ \text{tlv. orbits that are near an SN orbit satisfy}$$
$$1 - \prod _ { k = j } ^ { j + m - 1 } \frac { \partial f _ { R _ { i } } } { \partial x } ( x _ { i, k | R _ { i } } ) = o ( 1 ) \,.$$
By contrast, if
$$\iota, \, \Pi & 1 - \prod _ { k = j } ^ { j + m - 1 } \frac { \partial f _ { R _ { i } } } { \partial x } ( x _ { i, k | R _ { i } } ) = O ( 1 ) \,. & ( 4 ) \\ t \, \text{an orbit is the far from} " a \, \text{SN orbit}$$
we say that an orbit is 'far from' a SN orbit.
2. Because f r i has a critical point at C, so does f m r i . Suppose that x i,n r | i is the point of the SN orbit that is closest to C. As ∂f r i ∂x (C) = 0, for sufficiently large periods, we can find SN orbits with arbitrarily small ∣ ∣ ∣ ∣ ∂f r i ∂x ( x i,n r | i ) ∣ ∣ ∣ ∣ (see Fig. 1), and in particular we can find examples where ∣ ∣ ∣ ∣ ∂f r i ∂x ( x i,n r | i ) ∣ ∣ ∣ ∣ < ε . We use the term small-derivative SN orbits for such orbits. Additionally, a small-derivative SN orbit includes points that are not close to the critical point C, so that ∂f ∂x ( x i ) = O (1) in general, and the associated terms cannot be neglected.
The overall bifurcation pattern in a typical unimodal map of the interval is topologically equivalent to the bifurcation pattern in any other typical unimodal map of the interval [25], so it is sensible to focus on a particular such map. The standard choice for such a map is the logistic map. Orbits of any period occur in the logistic map, which contains infinitely many small-derivative SN orbits. In particular, such orbits include the periodq SN orbits from which supercycles with symbol sequences CRL q -2 originate. 1 Given this fact and the broad applicability of results for the logistic map, we note that our results are relevant in numerous situations.
## Lemma 1
Let | ε | < 1 in the CML (1), and suppose that the map f q i R i has an SN bifurcation at R i = r i , such that the associated SN orbit of f r i is a smallderivative SN orbit. Additionally, suppose that R i = r i + ε for i ∈ { 1 , . . . , s } , but that R i for i ∈ { s +1 , . . . , p } are far away from r i . Consider the following initial conditions:
- · For i ∈ { 1 , . . . , s } , let X n i ( ) = x i,n r | i + εA i,n + O ε ( 2 ), where x i,n r | i is the point of the SN orbit closest to the critical point C of f R i at R i = r i .
- · For i ∈ { s +1 , . . . , p } , let X n i ( ) = x i,n R | i + εA i,n + O ε ( 2 ).
The temporal evolution of the CML (1) is then given by
1 Recall that a 'supercycle' is a periodic orbit that includes C; if its period is q (i.e., if f q R ( C ) = C for some parameter value R ), then ∂f q R ∂x ( x k ) = 0 for all points x k in the supercycle.
## 1. For i ∈ { 1 , . . . , s } ,
/negationslash
$$X _ { i } ( n + 1 ) = x _ { i, n + 1 | r _ { i } } + s \left ( \frac { \partial f _ { r _ { i } } } { \partial r } ( x _ { i, n | r _ { i } } ) + \sum _ { \substack { h = 1 \\ h \neq i } } ^ { s } x _ { h, n + 1 | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, n + 1 | R _ { h } } \right ) + O ( \varepsilon ^ { 2 } ) \,, \quad m = 1 \,,$$
/negationslash
/negationslash
$$X _ { i } ( n + m ) = x _ { i, n + m | r _ { i } } + \left [ \frac { \partial f _ { r _ { i } } } { \partial r } ( x _ { i, n + m - 1 | r _ { i } } ) + \sum _ { k = n } ^ { n + m - 2 } \frac { \partial f _ { r _ { i } } } { \partial r } ( x _ { i, k | r _ { i } } ) \prod _ { l = k + 1 } ^ { n + m - 1 } \frac { \partial f _ { r _ { i } } } { \partial x } ( x _ { i, l | r _ { i } } ) \\ + \sum _ { k = n + 1 } ^ { n + m - 1 } \left ( \left ( \sum _ { \substack { s \\ h = 1 \\ h \neq i } } ^ { s } x _ { h, k | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, k | R _ { h } } \right ) ^ { n + m - 1 } \prod _ { l = k } ^ { \partial f _ { r _ { i } } } \frac { \partial f _ { r _ { i } } } { \partial x } ( x _ { i, l | r _ { i } } ) \right ) \\ + \sum _ { \substack { h = 1 \\ h \neq i } } ^ { s } x _ { h, n + m | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, n + m | R _ { h } } \right ] \varepsilon + O ( \varepsilon ) \,, \quad m \in \{ 2, \dots, q \}.$$
## 2. For i ∈ { s +1 , . . . , p } ,
/negationslash
$$X _ { i } ( n + 1 ) = x _ { i, n + 1 | R _ { i } } + \varepsilon \left ( \frac { \partial f _ { R _ { i } } } { \partial x } ( x _ { i, n | R _ { i } } ) A _ { i, n } + \sum _ { h = 1 } ^ { s } x _ { h, n + 1 | R _ { h } } + \sum _ { \substack { h = s + 1 \\ h \neq i \\ ( 7 ) } } ^ { p } x _ { h, n + 1 | R _ { h } } \right ) + O ( \varepsilon ^ { 2 } ) \,, \quad m = 1 \,, \\ - \quad \dots \quad.$$
/negationslash
$$X _ { i } ( n + m ) = x _ { i, n + m | R _ { i } } + \left [ \prod _ { k = n } ^ { n + m - 1 } \frac { \partial f _ { R _ { i } } } { \partial x } ( x _ { i, k | R _ { i } } ) A _ { i, n } + \sum _ { k = n + 1 } ^ { n + m - 1 } \left ( \left ( \sum _ { h = 1 } ^ { s } x _ { h, k | r _ { h } } \sum _ { h = s + 1 } ^ { p } + x _ { h, k | R _ { h } } \right ) \\ \prod _ { \begin{array} { c } n + m - 1 \\ \end{array} } ^ { n + m - 1 } \frac { \partial f _ { R _ { i } } } { \wedge } ( x _ { i, l | R _ { i } } ) + \stackrel { s } { \lambda } ^ { s } x _ { h \, n + m | r _ { i } } + \end{array} \right ] _ { \begin{array} { c } n } \cdot \\ \end{array} } \right ] _ { \varepsilon } + O ( \varepsilon ^ { 2 } ) \,. \quad m \in \{ 2. \dots. q \}.$$
/negationslash
$$\prod _ { l = k } ^ { n + m - 1 } \frac { \L L } { \partial x } ( x _ { i, l | R _ { i } } ) \Big ) + \sum _ { h = 1 } ^ { s } x _ { h, n + m | r _ { h } } + \sum _ { \substack { h = s + 1 \\ h \neq i } } ^ { p } x _ { h, n + m | R _ { h } } \Big ] \varepsilon + O ( \varepsilon ^ { 2 } ) \,, \quad m \in \{ 2, \dots, q \}.$$
## Proof of Lemma 1
We proceed by induction. Substitute X n i ( ) = x i,n r | i + εA i,n + O ε ( 2 ) and X n i ( ) = x i,n R | i + εA i,n + O ε ( 2 ) into equation (1) and expand in powers of ε . Note that we need to consider i ∈ { 1 , . . . , s } and i ∈ { s +1 , . . . , p } separately.
- 1. We initiate the iteration at the point x i,n r | i of the SN orbit closest to the critical point C. Because we have a small-derivative SN orbit, ∂f r i ∂x ( x i,n r | i ) is arbitrarily small, although this is not true for other points in the SN orbit.
For i ∈ { 1 , . . . , s } , we have
$$X _ { i } ( n + 1 ) = f _ { \underset { \varepsilon } { r _ { i } } } ( x _ { i, n | r _ { i } } ) + \varepsilon \frac { \partial f _ { r _ { i } } } { \partial x } ( x _ { i, n | r _ { i } } ) A _ { i, n } + \varepsilon \frac { \partial f _ { r _ { i } } } { \partial r } ( x _ { i, n | r _ { i } } )$$
/negationslash
$$+ \varepsilon \sum _ { \substack { h = 1 \\ h \neq i } } ^ { s } f _ { r _ { h } } ( x _ { h, n | r _ { h } } + O ( \varepsilon ) ) + \sum _ { h = s + 1 } ^ { p } f _ { R _ { h } } ( x _ { h, n | R _ { h } } + O ( \varepsilon ) ) + O ( \varepsilon ^ { 2 } ) \\ \cdot$$
/negationslash
$$& \quad ^ { n \ne i } \\ = & \, x _ { i, n + 1 | r _ { i } } + \varepsilon \left ( \frac { \partial f _ { r _ { i } } } { \partial r } ( x _ { i, n | r _ { i } } ) + \sum _ { \substack { h = 1 \\ h \ne i } } ^ { s } x _ { h, n + 1 | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, n + 1 | R _ { h } } \right ) + O ( \varepsilon ^ { 2 } ) \,,$$
In the last step, we have neglected terms that contain ε ∂f r i ∂x ( x i,n r | i ) because ∂f r i ∂x ( x i,n r | i ) is arbitrarily small.
- 2. For i ∈ { s +1 , . . . , p } , we have
/negationslash
$$X _ { i } ( n + 1 ) = x _ { i, n + 1 | R _ { i } } + \varepsilon \left ( \frac { \partial f _ { R _ { i } } } { \partial x } ( x _ { i, n | R _ { i } } ) A _ { i, n } + \sum _ { h = 1 } ^ { s } x _ { h, n + 1 | r _ { h } } + \sum _ { \substack { h = s + 1 \\ h \neq i } } ^ { p } x _ { h, n + 1 | R _ { h } } \right ) + O ( \varepsilon ^ { 2 } ) \,.$$
When using the induction hypothesis, we need to distinguish the case i ∈ { 1 , . . . , s } from the case i ∈ { s +1 , . . . , p } . For the CML (1), equations (9,10) yield the following equations.
- 1. For i ∈ { 1 , . . . , s } , we write the induction hypothesis for m ≥ 2 as
/negationslash
/negationslash
$$1. & \text{ For } i \in \{ 1, \dots, s \}, \text{ we write the induction hypothesis for } m \geq 2 \text{ as} \\ X _ { i } ( n + m ) = x _ { i, n + m | r _ { i } } + \left [ \frac { \partial f _ { r _ { i } } } { \partial r } ( x _ { i, n + m - 1 | r _ { i } } ) + \sum _ { k = n } ^ { n + m - 2 } \frac { \partial f _ { r _ { i } } } { \partial r } ( x _ { i, k | r _ { i } } ) \prod _ { l = k + 1 } ^ { n + m - 1 } \frac { \partial f _ { r _ { i } } } { \partial x } ( x _ { i, l | r _ { i } } ) \\ & + \sum _ { k = n + 1 } ^ { n + m - 1 } \left ( \left ( \sum _ { h = 1 } ^ { s } x _ { h, k | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, k | R _ { h } } \right ) \prod _ { l = k } ^ { n + m - 1 } \frac { \partial f _ { r _ { i } } } { \partial x } ( x _ { i, l | r _ { i } } ) \right ) \\ & + \sum _ { h = 1 } ^ { s } x _ { h, n + m | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, n + m | R _ { h } } \right ] \varepsilon + O ( \varepsilon ^ { 2 } ), \quad ( 1 1 )$$
which implies that
/negationslash
/negationslash
$$\text{which implies that} \\ X _ { i } ( n + m + 1 ) & = f _ { R _ { i } } \left [ x _ { i, n + m | r _ { i } } + \left ( \frac { \partial f _ { r _ { i } } } { \partial r } ( x _ { i, n + m - 1 | r _ { i } } ) + \sum _ { k = n } ^ { n + m - 2 } \frac { \partial f _ { r _ { i } } } { \partial r } ( x _ { i, k | r _ { i } } ) \prod _ { l = k + 1 } ^ { n + m - 1 } \frac { \partial f _ { r _ { i } } } { \partial x } ( x _ { i, l | r _ { i } } ) \\ & + \sum _ { k = n + 1 } ^ { n + m - 1 } \left ( \left ( \sum _ { h = \overset { s } { \in i } } x _ { h, k | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, k | R _ { h } } \right ) \prod _ { l = k } ^ { n + m - 1 } \frac { \partial f _ { r _ { i } } } { \partial x } ( x _ { i, l | r _ { i } } ) \right ) \\ & + \sum _ { h = \overset { s } { \in i } } x _ { h, n + m | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, n + m | R _ { h } } \right ) \varepsilon \right ] + \varepsilon \sum _ { h = \overset { s } { \in i } } f _ { r _ { h } } ( x _ { h, n + m | r _ { h } } + O ( \varepsilon ) ) \\ & + \varepsilon \sum _ { h = s + 1 } ^ { p } f _ { R _ { h } } ( x _ { h, n + m | R _ { h } } + O ( \varepsilon ) ) \\ \text{We Taylor expand all occurrences of $f$ and its derivatives to obtain}$$
/negationslash
We Taylor expand all occurrences of f and its derivatives to obtain
/negationslash
/negationslash
$$\text{We Taylor expand all occurrences of $f$ and its derivatives to obtain} \\ X _ { i } ( n + m + 1 ) = x _ { i, n + m + 1 | r _ { i } } + \left [ \frac { \partial f _ { r _ { i } } } { \partial r } ( x _ { i, n + m | r _ { i } } ) + \sum _ { k = n } ^ { n + m - 1 } \frac { \partial f _ { r _ { i } } } { \partial r } ( x _ { i, k | r _ { i } } ) \prod _ { l = k + 1 } ^ { n + m } \frac { \partial f _ { r _ { i } } } { \partial x } ( x _ { i, l | r _ { i } } ) \\ + \sum _ { k = n + 1 } ^ { n + m } \left ( \left ( \sum _ { h = 1 } ^ { s } x _ { h, k | r _ { i } } + \sum _ { h = s + 1 } ^ { p } x _ { h, k | R _ { i } } \right ) \prod _ { l = k } ^ { n + m } \frac { \partial f _ { r _ { i } } } { \partial x } ( x _ { i, l | r _ { i } } ) \right ) \\ + \sum _ { h = 1 } ^ { s } x _ { h, n + m + 1 | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, n + m + 1 | R _ { h } } \right ] \varepsilon + O ( \varepsilon ^ { 2 } ) \,. \\ 2. \text{ For } i \in \{ s + 1, \dots, p \}, \text{ we write the induction hypothesis for $m \geq 2$ as}$$
2. For i ∈ { s +1 , . . . , p } , we write the induction hypothesis for m ≥ 2 as
X n i ( + m ) = x i,n + m R | i
/negationslash
$$X _ { i } ( n + m ) & = x _ { i, n + m | R _ { i } } \\ + & \left [ \prod _ { k = n } ^ { n + m - 1 } \frac { \partial f _ { R _ { i } } } { \partial x } ( x _ { i, k | R _ { i } } ) A _ { i, n } + \sum _ { k = n + 1 } ^ { n + m - 1 } \left ( \left ( \sum _ { h = 1 } ^ { s } x _ { h, k | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, k | R _ { h } } \right ) \prod _ { l = k } ^ { n + m - 1 } \frac { \partial f _ { R _ { i } } } { \partial x } ( x _ { i, l | R _ { i } } ) \right ) \\ & \quad + \sum _ { h = 1 } ^ { s } x _ { h, n + m | r _ { h } } + \sum _ { \substack { h = s + 1 \\ h \neq i } } ^ { p } x _ { h, n + m | R _ { h } } \right ] \varepsilon + O ( \varepsilon ^ { 2 } ) \,, & ( 1 3 )$$
/negationslash
which implies that
X n
$$\text{which implies that} \\ X _ { i } ( n + m + 1 ) = \\ f _ { R _ { i } } \left ( x _ { i, n + m | R _ { i } } + \left [ \prod _ { k = n } ^ { n + m - 1 } \frac { \partial f _ { R _ { i } } } { \partial x } ( x _ { i, k | R _ { i } } ) A _ { i, n } + \sum _ { k = n + 1 } ^ { n + m - 1 } \left ( \left ( \sum _ { h = 1 } ^ { s } x _ { h, k | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, k | R _ { h } } \right ) \\ \prod _ { l = k } ^ { n + m - 1 } \frac { \partial f _ { R _ { i } } } { \partial x } ( x _ { i, l | R _ { i } } ) \right ) + \sum _ { h = 1 } ^ { s } x _ { h, n + m | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, n + m | R _ { h } } \right ] \varepsilon + O ( \varepsilon ^ { 2 } ) \right ) \\ + \varepsilon \sum _ { h = 1 } ^ { s } f _ { r _ { h } } ( x _ { h, n + m | r _ { h } } + O ( \varepsilon ) ) + \sum _ { h = s + 1 } ^ { p } f _ { R _ { h } } ( x _ { h, n + m | R _ { h } } + O ( \varepsilon ) ) \\ \text{We Taylor expand of all occurrences of $f$ and its derivatives to obtain} \\ \nu / \cdot \quad. \quad. \quad.$$
/negationslash
/negationslash
We Taylor expand of all occurrences of f and its derivatives to obtain
X n i ( + m +1) = x i,n + m +1 | R i
/negationslash
$$X _ { i } ( n + m + 1 ) & = x _ { i, n + m + 1 | R _ { i } } \\ & + \left [ \prod _ { k = n } ^ { n + m } \frac { \partial f _ { R _ { i } } } { \partial x } ( x _ { i, k | R _ { i } } ) A _ { i, n } + \sum _ { k = n + 1 } ^ { n + m } \left ( \left ( \sum _ { h = 1 } ^ { s } x _ { h, k | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, k | R _ { h } } \right ) \prod _ { l = k } ^ { m } \frac { \partial f _ { R _ { i } } } { \partial x } ( x _ { i, l | R _ { i } } ) \right ) \\ & + \sum _ { h = 1 } ^ { s } x _ { h, n + m + 1 | r _ { h } } + \sum _ { \substack { h = s + 1 \\ h \neq i } } ^ { p } x _ { h, n + m + 1 | R _ { h } } \right ] \varepsilon + O ( \varepsilon ^ { 2 } ) \\ & & \Box$$
/negationslash
## Theorem 1
Let | ε | < 1 in the CML (1), and suppose that the hypotheses of Lemma 1 are satisfied. That is, we assume that the map f q i R i has an SN bifurcation at R i = r i , such that the associated SN orbit of f r i is a small-derivative SN orbit, that R i = r i + ε for i ∈ { 1 , . . . , s } , and that R i for i ∈ { s + 1 , . . . , p } are far away from r i . Let { x i, 1 | r i , x i, 2 | r i , . . . , x i,q i | r i } be a periodq i orbit for the uncoupled oscillator X i for i ∈ { 1 , . . . , s } , and let { x i, 1 | R i , x i, 2 | R i , . . . x i,q i | R i } be a periodq i orbit for the uncoupled oscillator X i for i ∈ { s +1 , . . . , p } . Consider the following initial conditions:
- · For i ∈ { 1 , . . . , s } ,
$$\., s \}, & \text{let} \\ & X _ { i } ( n ) = x _ { i, n | r _ { i } } + \varepsilon A _ { i, n } + O ( \varepsilon ^ { 2 } ) \,,$$
where x i,n r | i is the point of the SN orbit closest to the critical point of f R i at R i = r i , and A i,n is an arbitrary O (1) value.
/negationslash
- · For i ∈ { s +1 , . . . , p } , let
$$X _ { i } ( n ) = x _ { i, n | R _ { i } } + \varepsilon A _ { i, n } + O ( \varepsilon ^ { 2 } ) \,,$$
where x i,n R | i is a point of the orbit, and
/negationslash
/negationslash
$$& \text{where $x_{i,n|R_{i}}$ is a point of the orbit, and} \\ A _ { i, n } = \left [ \sum _ { k = n + 1 } ^ { n + q - 1 } \left ( \left ( \sum _ { h = 1 } ^ { s } x _ { h, k | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, k | R _ { h } } \right ) \prod _ { l = k } ^ { n + q - 1 } \frac { \partial f _ { R _ { i } } } { \partial x } ( x _ { i, l | R _ { i } } ) \right ) \\ & \quad + \sum _ { h = 1 } ^ { s } x _ { h, n + q | r _ { h } } + \sum _ { \substack { h = s + 1 \\ h \neq i } } ^ { p } x _ { h, n + q | R _ { h } } \right ] \left ( \frac { 1 } { 1 - \prod _ { k = n } ^ { n + q - 1 } \frac { \partial f _ { R _ { i } } } { \partial x } ( x _ { i, k | R _ { i } } ) } \right ). \\ \\ & \text{The CML} \left ( 1 \right ) \text{has the solution}$$
The CML (1) has the solution
$$X _ { i } ( n + m ) = \begin{cases} x _ { i, n + m | r _ { i } } + \varepsilon A _ { i, n + m } + O ( \varepsilon ^ { 2 } ) \,, \quad i \in \{ 1, \dots, s \} \,, m \in \{ 1, \dots, q \} \\ x _ { i, n + m | R _ { i } } + \varepsilon A _ { i, n + m } + O ( \varepsilon ^ { 2 } ) \,, \quad i \in \{ s + 1, \dots, p \} \,, m \in \{ 1, \dots, q \} \,, \end{cases}$$
where the coefficients A i,n + m are periodic with period q = lcm( q 1 , q 2 , . . . , q p ) and satisfy the following formulas:
$$1. \text{ For } i \in \{ 1, \dots, s \},$$
/negationslash
$$A _ { i, n + 1 } = \frac { \partial f _ { r _ { i } } } { \partial r } ( x _ { i, n | r _ { i } } ) + \sum _ { \substack { h = 1 \\ h \neq i } } ^ { s } x _ { h, n + 1 | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, n + 1 | R _ { h } } \,, \quad m = 1 \quad ( 1 7 )$$
(18)
/negationslash
$$A _ { i, n + m } = \frac { \partial f _ { r _ { i } } } { \partial r } ( x _ { i, n + m - 1 | r _ { i } } ) + \sum _ { k = n } ^ { n + m - 2 } \frac { \partial f _ { r _ { i } } } { \partial r } ( x _ { i, k | r _ { i } } ) \prod _ { l = k + 1 } ^ { n + m - 1 } \frac { \partial f _ { r _ { i } } } { \partial x } ( x _ { i, l | r _ { i } } ) \\ + \sum _ { k = n + 1 } ^ { n + m - 1 } \left ( \left ( \sum _ { h \neq i } ^ { s } x _ { h, k | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, k | R _ { h } } \right ) \prod _ { l = k } ^ { n + m - 1 } \frac { \partial f _ { r _ { i } } } { \partial x } ( x _ { i, l | r _ { i } } ) \right ) \\ + \sum _ { h = 1 } ^ { s } x _ { h, n + m | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, n + m | R _ { h } } \,, \quad m \in \{ 2, \dots, q \} \,.$$
/negationslash
2. For i ∈ { s +1 , . . . , p } ,
/negationslash
/negationslash
$$2. \text{ For } i \in \{ s + 1, \dots, p \}, \\ A _ { i, n + m } & = \left [ \sum _ { k = n + m + 1 } ^ { n + m + q - 1 } \left ( \left ( \sum _ { h = 1 } ^ { s } x _ { h, k | r _ { h } } + \sum _ { \substack { h = s + 1 \\ h \neq i } } ^ { p } x _ { h, k | R _ { h } } \right ) ^ { n + m + q - 1 } \prod _ { l = k } ^ { 0 } \frac { 0 f R _ { i } } { \partial x } ( x _ { i, l | R _ { i } } ) \right ) \\ & + \sum _ { h = 1 } ^ { s } x _ { h, n + m + q | r _ { h } } + \sum _ { \substack { h = s + 1 \\ h \neq i } } ^ { p } x _ { h, n + m + q | R _ { h } } \right ] \left ( \frac { 1 } { 1 - \prod _ { k = n + m } ^ { n + m + q - 1 } \frac { 0 f R _ { i } } { \partial x } ( x _ { i, k | R _ { i } } ) } \right ), \\ & \quad m \in \{ 1, \dots, q \}.$$
Remark. Although the initial conditions given in the statement of Theorem 1 may seem restrictive, our numerical computations demonstrate that - independently of the type of the orbit (i.e., either close to or far away from the SN) - it is sufficient to take as an initial condition any point of the unperturbed orbit plus a perturbation of size O ε ( ).
## Proof of Theorem 1.
We need to consider i ∈ { 1 , . . . , s } and i ∈ { s +1 , . . . , p } separately.
- 1. Using Lemma 1, it follows from X n i ( ) = x i,n r | i + εA i,n + O ε ( 2 ) that
/negationslash
$$X _ { i } ( n + 1 ) = x _ { i, n + 1 | r _ { i } } + \varepsilon \left ( \frac { \partial f _ { r _ { i } } } { \partial r } ( x _ { i, n | r _ { i } } ) + \sum _ { h = 1 } ^ { s } x _ { h, n + 1 | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, n + 1 | R _ { h } } \right ) + O ( \varepsilon ^ { 2 } )$$
and
/negationslash
$$\text{and} \\ X _ { i } ( n + q + 1 ) = x _ { i, n + q + 1 | r _ { i } } + \left [ \frac { \partial f _ { r _ { i } } } { \partial r } ( x _ { i, n + q | r _ { i } } ) + \sum _ { k = n } ^ { n + q - 1 } \frac { \partial f _ { r _ { i } } } { \partial r } ( x _ { i, k | r _ { i } } ) \prod _ { l = k + 1 } ^ { n + q } \frac { \partial f _ { r _ { i } } } { \partial x } ( x _ { i, l | r _ { i } } ) \\ & + \sum _ { k = n + 1 } ^ { n + q } \left ( \left ( \sum _ { \hbar { = } i } ^ { s } x _ { h, k | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, k | R _ { h } } \right ) \prod _ { l = k } ^ { n + q } \frac { \partial f _ { r _ { i } } } { \partial x } ( x _ { i, l | r _ { i } } ) \right ) \\ & + \sum _ { \hbar { = } i } ^ { s } x _ { h, n + q + 1 | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, n + q + 1 | R _ { h } } \right ] \varepsilon + O ( \varepsilon ^ { 2 } ) \,.$$
/negationslash
Because n + q ∏ l = k ∂f r i ∂x ( x i,l r | i ) includes the arbitrarily small term ∣ ∣ ∣ ∣ ∂f r i ( x i,n r | i ) ∂x ∣ ∣ ∣ ∣ , it follows from (22) that
/negationslash
With q = lcm( q , q 1 2 , . . . , q p ), we have
$$X _ { i } ( n + q + 1 ) = x _ { i, n + q + 1 | r _ { i } } + \left ( \frac { \partial f _ { r _ { i } } } { \partial r } ( x _ { i, n + q | r _ { i } } ) + \sum _ { \substack { h = 1 \\ h \neq i } } ^ { s } x _ { h, n + q + 1 | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, n + q + 1 | R _ { h } } \right ) \varepsilon + O ( \varepsilon ^ { 2 } ) \,. \\ \text{With } q = \text{lcm} ( q _ { 1 }, q _ { 2 }, \dots, q _ { p } ), \text{ we have}$$
$$\psi$$
$$x _ { i, n + q + 1 | r _ { i } } & = x _ { i, n + 1 | r _ { i } } \,, \\ x _ { i, n + q + 1 | R _ { i } } & = x _ { i, n + 1 | R _ { i } } \,,$$
/negationslash
$$\sum _ { \substack { h = 1 \\ h \neq i } } ^ { s } x _ { h, n + q + 1 | r _ { h } } + \sum _ { \substack { h = s + 1 \\ h \neq i } } ^ { p } x _ { h, n + q + 1 | R _ { h } } = \sum _ { \substack { h = 1 \\ h \neq i } } ^ { s } x _ { h, n + 1 | r _ { h } } + \sum _ { \substack { h = s + 1 \\ h \neq i } } ^ { p } x _ { h, n + 1 | R _ { h } } \,,$$
/negationslash
/negationslash
/negationslash because x i,j is a point of a q i -period orbit. Consequently, equations (21) and (22) become the same equation. From equations (9) and (21), we can write equation (6) in Lemma 1 as X n i ( + m ) = x i,n + m r | i + εA i,n + m + O ε ( 2 ) to obtain
/negationslash
/negationslash
$$\text{to obtain} \quad & \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \ \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \, \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cd. \quad \end{array} \quad \text{to obtain} \quad & \quad \partial f _ { r _ { i } } ( x _ { i, n + m - 1 } ) + \sum _ { k = n } ^ { p } \frac { \partial f _ { r _ { i } } } { \partial x } ( x _ { i, l | r _ { i } } ) \prod _ { l = k + 1 } ^ { n + m - 1 } \frac { \partial f _ { r _ { i } } } { \partial x } ( x _ { i, l | r _ { i } } ) \right ) \\ & + \sum _ { \substack { h = 1 \\ h = 1 \\ h = 1 } } ^ { s } x _ { h, n + m | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, n + m | R _ { h } } \,, \quad \quad m \in \{ 1, \dots, q \} \,. \quad \quad \quad ( 2 3 ) \\ 2. \ \text{With} \ X _ { i } ( n + m ) = x _ { i, n + m | R _ { i } } + \varepsilon A _ { i, n + m } + O ( \varepsilon ^ { 2 } ), \ \text{Lemma} \ 1 \text{ implies that}$$
2. With X n i ( + m ) = x i,n + m R | i + εA i,n + m + O ε ( 2 ), Lemma 1 implies that
$$X _ { i } ( n + m + q ) = x _ { i, n + m + q | R _ { i } } + \prod _ { k = n + m } ^ { n + m + q - 1 } \frac { \partial f _ { R _ { i } } } { \partial x } ( x _ { i, k | R _ { i } } ) A _ { i, n + m } \,.$$
/negationslash
/negationslash
$$X _ { i } ( n + m + q ) & = x _ { i, n + m + q | R _ { i } } + \left [ \prod _ { k = n + m } ^ { n + m + q - 1 } \frac { \partial f _ { R _ { i } } } { \partial x } ( x _ { i, k | R _ { i } } ) A _ { i, n + m } \\ & \quad + \sum _ { k = n + m + 1 } ^ { n + m + q - 1 } \left ( \left ( \sum _ { h = 1 } ^ { s } x _ { h, k | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, k | R _ { h } } \right ) ^ { n + m + q - 1 } \prod _ { l = k } ^ { \partial f _ { R _ { i } } } \frac { \partial f _ { R _ { i } } } { \partial x } ( x _ { i, l | R _ { i } } ) \right ) \\ & \quad + \sum _ { h = 1 } ^ { s } x _ { h, n + m + q | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, n + m + q | R _ { h } } \right ] \varepsilon + O ( \varepsilon ^ { 2 } ) \,. \\ & 1 1$$
By taking q = lcm( q , q 1 2 , . . . , q p ), we obtain x i,n + m r | i = x i,n + m q r + | i and x i,n + m R | i = x i,n + m q R + | i because x i,j is a point of a periodic orbit. Consequently, X n i ( + m ) -X n i ( + m + ) = q O ε ( ) whenever
/negationslash
/negationslash
$$\text{query}, \, \Lambda _ { i } ( n + m ) - \Lambda _ { i } ( n + m + q ) = \mathcal { U } _ { \infty } ( \varepsilon ) \text{ whenever} \\ A _ { i, n + m } & = \prod _ { k = n + m } ^ { n + m + q - 1 } \frac { \partial f _ { R _ { i } } } { \partial x } ( x _ { i, k | R _ { i } } ) A _ { i, n + m } + \sum _ { k = n + m + 1 } ^ { n + m + q - 1 } \left ( \left ( \sum _ { h = 1 } ^ { s } x _ { h, k | r _ { h } } + \sum _ { \substack { h = s + 1 \\ h \neq i } } ^ { p } x _ { h, k | R _ { h } } \right ) \\ & \prod _ { l = k } ^ { n + m + q - 1 } \frac { \partial f _ { R _ { i } } } { \partial x } ( x _ { i, l | R _ { i } } ) \right ) + \sum _ { h = 1 } ^ { s } x _ { h, n + m + q | r _ { h } } + \sum _ { \substack { h = s + 1 \\ h \neq i } } ^ { p } x _ { h, n + m + q | R _ { h } } \cdot \, ( 2 4 ) \\ \text{Furthermore} \quad A \colon \dots \text{ is neriodic}$$
Furthermore, A i,n + m is periodic. Equation (24) now implies that
/negationslash
/negationslash
$$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{ \text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{ $\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{#\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{ $\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text$$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text $\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text
$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text}$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text#\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text$
$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\test{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\test$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text'$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text<$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text.$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text`$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text$$
It follows that A i,n + m has period q because it is given by sums and products of q -periodic functions evaluated at points of a q -periodic orbit.
Observe that the formula for A i,j for i ∈ { 1 , . . . , s } in equation (19) does not contain the term 1 -j + q -1 ∏ k = j ∂f r i ∂x ( x i,k r | i ) in the denominator [see equation
(3)]. Otherwise, A i,j would be of size O (1 /ε ), and the expansion that we used to prove Theorem 1 would not be valid. By contrast, the formula for A i,j for
$$& \text{to prove $1$ theorem $1$ would not be valid. By contrast, the formula for $A_{i,j}$ for} \\ & i \in \{ s + 1, \dots, p \} \text{ in equation (20) includes the term } \left [ 1 - \prod _ { k = j } ^ { j + q - 1 } \frac { \partial f _ { R _ { i } } } { \partial x } ( x _ { i, k | R _ { i } } ) \right ] \\ & \text{in the denominator because the oscillators are far from SN bifurcations for} \\ & i \in \{ s + 1, \dots, p \}. \text{ Therefore.}$$
in the denominator because the oscillators are far from SN bifurcations for i ∈ { s +1 , . . . , p } . Therefore,
$$1 - \prod _ { k = j } ^ { j + q - 1 } \frac { \partial f _ { R _ { i } } } { \partial x } ( x _ { i, k | R _ { i } } ) = O ( 1 ) \,, \\. \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cap \, \cdot$$
and it follows that A i,j also has size O (1).
,
## 2.2. Type-I Intermittency Near Saddle-Node Bifurcations
Theorem 1 concerns the behavior of the CML (1) with a mixture of periodic oscillators that are near the SN bifurcation with others that are far from the SN bifurcation. If an SN orbit takes place at R i = r i , then the oscillators with R i = r i + ε are the ones that are close to the SN orbit.
We now want to study the behavior of the CML (1) when an uncoupled oscillator has type-I intermittency [26] at R i = r i -ε (i.e., just to the left of where it undergoes an SN bifurcation). Type-I intermittency is characterized by the alternation of an apparently periodic regime (a so-called 'laminar phase'), whose mean duration follows the power law 〈 〉 l ∝ ε -1 2 (so the laminar region becomes longer as ε becomes smaller), and chaotic bursts. As R i = r i -ε , we expand f r i -ε in powers of ε to obtain
$$f _ { r _ { i } - \varepsilon } ^ { q _ { i } } ( x _ { j | r _ { i } } ) = f _ { r _ { i } } ^ { q _ { i } } ( x _ { j | r _ { i } } ) - \varepsilon \frac { \partial f _ { r _ { i } } ^ { q _ { i } } } { \partial r } ( x _ { j | r _ { i } } ) + O ( \varepsilon ^ { 2 } ) \,,$$
where x j a point of a periodq i SN orbit. Therefore the laminar phase is driven by the periodq i SN orbit associated with the SN bifurcation. Thus, as ε becomes smaller, the orbit spends more iterations in the laminar regime, and it thus more closely resembles the periodq i SN orbit. In particular, ∣ ∣ ∣ x j r | i -f q i r i -ε ( x j r | i ) ∣ ∣ ∣ = O ε ( ) .
To approximate the temporal evolution of the laminar regime using the periodq i SN orbit, we proceed in the same way as in Theorem 1, except that we replace R i = r i + ε by R i = r i -ε . We thus write
/negationslash
$$X _ { i } ( n + 1 ) = x _ { i, n + 1 | r _ { i } } + \left [ - \frac { \partial f _ { r _ { i } } } { \partial r } ( x _ { i, n | r _ { i } } ) + \sum _ { \substack { h = 1 \\ h \neq i } } ^ { s } x _ { h, n + 1 | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, n + 1 | R _ { h } } \right ] \varepsilon + O ( \varepsilon ^ { 2 } ) \,, \quad m = 1 \,,$$
/negationslash
$$X _ { i } ( n + m ) = x _ { i, n + m | r _ { i } } + \left [ - \frac { \partial f _ { r _ { i } } } { \partial r } ( x _ { i, n + m - 1 | r _ { i } } ) - \sum _ { k = n } ^ { n + m - 2 } \frac { \partial f _ { r _ { i } } } { \partial r } ( x _ { i, k | r _ { i } } ) \prod _ { l = k + 1 } ^ { n + m - 1 } \frac { \partial f _ { r _ { i } } } { \partial x } ( x _ { i, l | r _ { i } } ) \\ + \sum _ { k = n + 1 } ^ { n + m - 1 } \left ( \left ( \sum _ { h = 1 } ^ { s } x _ { h, k | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, k | R _ { h } } \right ) \prod _ { l = k } ^ { n + m - 1 } \frac { \partial f _ { r _ { i } } } { \partial x } ( x _ { i, l | r _ { i } } ) \right ) \\ \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat{ \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat { \ } \hat$$
$$( n + m ) = x _ { i, n + m | r _ { i } } + \left [ - \frac { \partial f _ { r _ { i } } } { \partial r } ( x _ { i, n + m - 1 | r _ { i } } ) - \sum _ { k = n } ^ { n + m - 2 } \frac { \partial f _ { r _ { i } } } { \partial r } ( x _ { i, k | r _ { i } } ) \prod _ { l = k + 1 } ^ { n + m - 1 } \frac { \partial f _ { r _ { i } } } { \partial x } ( x _ { i, l | r _ { i } } ) \\ + \sum _ { k = n + 1 } ^ { n + m - 1 } \left ( \left ( \sum _ { h = 1 } ^ { s } x _ { h, k | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, k | R _ { h } } \right ) ^ { n + m - 1 } \prod _ { l = k } ^ { \partial f _ { r _ { i } } } \frac { \partial f _ { r _ { i } } } { \partial x } ( x _ { i, l | r _ { i } } ) \right ) \\ + \sum _ { h = 1 } ^ { s } x _ { h, n + m | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, n + m | R _ { h } } \right ] \varepsilon + O ( \varepsilon ^ { 2 } ) \,, \quad m \in \{ 2, \dots, q \} \\ \dots \,.$$
/negationslash
$$+ \sum _ { \substack { h = 1 \\ h \neq i } } ^ { s } x _ { h, n + m | r _ { h } } + \sum _ { h = s + 1 } ^ { p } x _ { h, n + m | R _ { h } } \Big { ] } \, \varepsilon + O ( \varepsilon ^ { 2 } ) \,, \quad m \in \{ 2, \dots, q \}$$
which determines the temporal evolution of the CML (1) in the laminar regime.
## 3. Numerical Computations
Theorem 1 proves the existence of an approximately periodic orbit. In principle, one can deduce the existence of a periodic orbit by using the Implicit Function Theorem (IFT). However, the IFT fails at the SN bifurcation (i.e., at R i = r i ) for free oscillators and consequently fails near an SN bifurcation (i.e., for R i = r i + ) of the HWCML (1), because the Jacobian determinant vanishes. ε
Had we expanded all terms in Theorem 1, we would have obtained terms of size O ε ( 2 ) that depend on the coefficients of the terms of size O ε ( ) (i.e., as functions of the A i,n + m terms in Theorem 1), so terms of size O ε ( 2 ) would have the same period as the A i,n + m . We could then obtain terms of size O ε ( 3 ) as a functions of the coefficients of lower-order terms. These terms would also have the same period as A i,n + m , and the same is true for all higher-order terms if we continued the expanding in powers of ε . This reasoning suggests the existence of a periodic orbit of period q = lcm( q , . . . , q 1 p ) (not just an approximate one), and our numerical simulations successfully illustrate the existence of such periodic orbits.
For simplicity, we consider a pair of coupled oscillators,
$$X ( n + 1 ) & = f ( X ( n ) ) + \varepsilon g ( Y ( n ) ) \,, \\ Y ( n + 1 ) & = g ( Y ( n ) ) + \varepsilon f ( X ( n ) ) \,,$$
where f ( x ) = R x 1 (1 -x ) and g y ( ) = cos( R y 2 ). We initially fix the coupling to be ε = 0 0001, though we will later consider 2 . ε , 3 , and so on. ε The uncoupled oscillator Y ( n ) has a fixed period of 4 and is far away from a SN bifurcation for R 2 = 1 9. . We use values of R 1 such that the uncoupled oscillator X n ( ) is near an SN bifurcation, and we consider SN orbits with different periods.
## 3.1. Uncoupled Oscillator X n ( ) with a Period-3 Orbit
For the oscillator X n ( ), we fix R 1 = r 1 +2 , where ε r 1 ≈ 3 828427 is an SN . bifurcation point of f . When there is no coupling, the free oscillator X n ( ) has a period-3 SN orbit, and the free oscillator Y ( n ) has a period-4 orbit. When coupled, both X n ( ) and Y ( n ) have a periodic orbit with period q = lcm(3 4) = , 12 (see Fig. 2).
At R 1 = r 1 + , the HWCML (29) exhibits type-I intermittency associated ε with the SN bifurcation (see Fig. 3). However, for larger R 1 (e.g., r 1 + 2 , ε r 1 +3 , ε . . . , r 1 +7 ), the periods of the uncoupled oscillators ε X n ( ) and Y ( n ) are preserved because we are farther away from the bifurcation point. We observe type-I intermittency for R 1 = r 1 +0 , ε R 1 = r 1 -ε , R 1 = r 1 -2 . ε
Remark. When R 1 = r 1 +2 , we calculate 1 ε -∏ j + m -1 k = j ∂f r i ∂x ( x i,k r | i ) ≈ 0 24 for . ε = 0 0001. (For . R 1 = r 1 + , we obtain a smaller value for the second quantity). ε Recall the quantifications of 'far from' and 'near' in section 2.1. Although ε can be very small, the periodic windows that are born with an SN orbit can be even smaller than ε . Thus, from the dynamical standpoint, a very small
value of the coupling parameter ε can nevertheless be large as a variation on a bifurcation parameter.
In Section 2.2, we determined the temporal evolution of the oscillators in the laminar regime of type-I intermittency up to size O ε ( ). By comparing Fig. 2 (which depicts the dynamics for a parameter value slightly larger than the SN bifurcation point) and Fig. 4 (which depicts the dynamics right before the bifurcation), we observe periodic behavior just after the bifurcation and laminar behavior just before it.
## 3.2. Uncoupled Oscillator X n ( ) with a Period-5 Orbit
We proceed as in Section 3.1 and obtain similar results.
For the oscillator X n ( ), we fix R 1 = r 1 +2 , where ε r 1 ≈ 3 738173 is an SN . bifurcation point of f . When there is no coupling, the free oscillator X n ( ) has a period-5 SN orbit, and the free oscillator Y ( n ) has a period-4 orbit. When coupled, both X n ( ) and Y ( n ) have a periodic orbit with period q = lcm(5 4) = , 20 (see Fig. 5).
At R 1 = r 1 + , the HWCML (29) exhibits type-I intermittency associated ε with the SN bifurcation (see Fig. 6). However, for larger R 1 (e.g., r 1 + 2 , ε r 1 +3 , ε r 1 +4 ε . . . ), the periods of the uncoupled oscillators X n ( ) and Y ( n ) are preserved because we are farther away from the bifurcation point. We observe type-I intermittency for R 1 = r 1 +0 , ε R 1 = r 1 -ε , R 1 = r 1 -2 . ε
## 3.3. Summary of HWCML Dynamics
Our results allow us to deduce the dynamics of the HWCML (29) when R i = r i + ε α . We worked with a coupling strength of ε = 0 0001 and a control . parameter of R i = r i + kε . In our numerical computations, we observed the following behavior:
- (a) intermittency for R i ≤ r i + ; ε
- (b) periodic behavior for R i ≥ r i +2 . ε
Therefore, the following occurs.
- (i) If we choose R i = r i + ε α with α > 1, then R < r i i + , and the HWCML ε exhibits intermittent behavior according to (a).
- (ii) If we choose R i = r i + ε α with 0 < α < 1, then R > r i i +2 ; this holds ε even for α close to 1, as long as ε α > 2 ε (e.g., 0 < α /lessorapproxeql 0 92 for . ε = 0 0001). . Therefore, the HWCML exhibits periodic behavior according to (b).
Based on our numerical computations, we can thus establish the following statement: 'Under the hypotheses of Theorem 1, the oscillators in the CML (1) have periodic orbits that persist with the same period as in Theorem 1 for perturbations of size O ε ( ). That is, higher-order terms do not change the period, as we heuristically stated at the beginning of Section 3.'
We now discuss the consequences of all oscillators in an HWCML having the same period q = lcm( q , q 1 2 , . . . , q p ), where q , . . . , q 1 p are the periods of the free oscillators. One can adjust the parameters to obtain periods q 1 . . . q p so
/negationslash that q = lcm( q 1 , q 2 , . . . , q p ) remains constant. For example, if q 1 = 3 and q 2 = 2 , then k q = lcm(3 2 ) = 3 , k × 2 k (for integers k > 0). If the first oscillator undergoes a period-doubling cascade, then its period is 3, 3 × 2, 3 × 2 , and so 2 on. However, the period m of the HWCMLs is q = lcm(3 2 ) = lcm(3 , k × 2 2 ) = , k · · · = lcm(3 × 2 k , 2 ) = 3 k × 2 , k so it does not change even after an arbitrary number of period-doubling bifurcations. That is, for arbitrarily small ε = 0, the HWCML period remains the same even amidst a period-doubling cascade.
We illustrate the above phenomenon with a simple computation. Consider the HWCML (29) and suppose that R 1 = 3 83 and . R 2 = 1 9. When . ε = 0 (i.e., when there is no coupling), the free oscillator X n ( ) has a period-3 orbit and the free oscillator Y ( n ) has a period-4 orbit. However, when ε = 0 001, both . X n ( ) and Y ( n ) have a periodic orbit with period q = lcm(3 4) = 12. As we show in , Table 1, the free oscillator X n ( ) undergoes period-doubling bifurcations, but the HWCML exhibits synchronization and still has period-12 orbits for ε = 0 001. .
## 4. Conclusions and Discussion
We have examined heterogeneous weakly coupled map lattices (HWCMLs) and have given results to describe periodic orbits both near and far from saddlenode orbits and to describe the temporal evolution of the laminar regime in typeI intermittency. All periodic windows of the bifurcation diagram of unimodal maps originate from SN bifurcations, so it is important to explore the dynamics near such bifurcation points.
An important implication of our results is that HWCMLs of oscillators need not behave approximately like their associated free-oscillator counterparts. In particular, they can have periodic-orbit solutions with completely different periods even for arbitrarily small coupling strengths ε = 0.
/negationslash
Our numerical calculations illustrate an important result about period preservation when oscillator parameters change. Even when one varies the parameters R i of the functions f R i such that the uncoupled oscillator X i undergoes a perioddoubling cascade, the periods of each of the coupled oscillators are preserved as long as the least common multiple of the periods remains constant. That is, the oscillation period is resilient to changes.
Period preservation is a rather generic phenomenon in CMLs. Suppose, for example, that one oscillator has period of q × 2 n , which can originate either from period doubling or from an SN bifurcation [27]. One can then change parameters so that different individual oscillators (if uncoupled) would undergo a period-doubling cascade, whereas the least common multiple of the periods of those oscillators will remain constant until one oscillator (if uncoupled) has period p × 2 n +1 . In a CML, a very large number of oscillators can each undergo a period-doubling cascade, so the period of a CML can be very resilient even in situations when other conditions - in particular, the values of the parameters in the CML - are changing a lot. Moreover, one can adjust the parameters to obtain oscillations of arbitrary periods q 1 . . . q p with q = lcm( q 1 , q 2 , . . . , q p ) = constant. Consequently, period preservation is a very common phenomenon: it
is not limited to the aforementioned period-doubling cascade but rather appears throughout a bifurcation diagram.
Periodic orbits anticipated by Theorem 1 and confirmed in Section 3 correspond to traveling waves in a one-dimensional HWCML and to periodic patterns in a multidimensional HWCML. Such patterns have been studied in homogeneous CMLs [13, 29], and our results can help to describe such dynamics in heterogeneous CMLs both near and far from bifurcations. Our observation about period resilience implies that there will be many different patterns with the same period. Small changes in an HWCML can change the specific pattern, but the period itself is rather robust.
Our results also have implications in applications. A toy macroscopic traffic flow model, governed by the logistic map, was proposed in [30]. The derivation of the model is based on very general assumptions involving speed and density. When these assumptions are satisfied, one can use the model to help examine the evolution of flows of pedestrians, flows in a factory, and so on. When such flows interact weakly, then equations of the form that we discussed in Section 2.1 can be useful for such applications. For example, one could do a simple examination of the temporal evolution of two groups of football fans around a stadium (or of sheep around an obstacle [31]). The two groups have different properties, so suppose that they are governed by an HWCML. From our results, if each group is regularly entering the stadium on its own (i.e., their behavior is periodic), then both groups considered together would continue to enter regularly at the same rate, provided that the interaction between the two groups is weak. This suggests that it would be interesting to explore a security strategy that models erecting a light fence to ensure that the interaction between the two groups remains weak.
The model in Ref. [30] also admits chaotic traffic patterns. One can construe the intermittent traffic flow in a traffic jam as being formed by regular motions (i.e., a laminar regime) and a series of acceleration and braking (i.e., chaotic bursts). Our results give the temporal evolution of such a laminar regime in a chaotic intermittent flow if the interaction between entities is weak (i.e., when the laminar regime is long, as we discussed in Section 2.2). Indeed, as has been demonstrated experimentally for the flow of sheep around an obstacle [31], it is possible to preserve laminar behavior for a longer time through the addition of an obstacle.
## Acknowledgements
We are grateful to the anonymous referee for his/her enlightening and detailed suggestions. We also thank Daniel Rodr´ ıguez P´rez for his help in the e preparation of this manuscript.
## References
## References
- [1] Farkas I, Helbing D, Vicsek T. Human waves in stadiums. Physica A 2003; 330:18-24
- [2] Sumpter DJT. Collective Animal Behavior. Princeton: Princeton University Press; 2010.
- [3] Boccara N. Modelling Complex Systems. Second Edition. Berlin: SpringerVerlag; 2010.
- [4] Coca D, Billings SA. Analysis and reconstruction of stochastic coupled map lattice models. Phys. Lett. A 2003; 315:61-75
- [5] Pavlov EA, Osipov GV, Chan CK, Suykens JAK. Map-based model of the cardiac action potential. Phys. Lett. A 2011; 375:2894-2902.
- [6] Kaneko K. Theory and Applications of Coupled Map Lattices. New York: John Wiley& Sons; 1993.
- [7] Kaneko K. From globally coupled maps to complex-systems biology. Chaos 2015; 25:097608
- [8] Special issue. Chaos 1992; 2(3).
- [9] Special issue. Physica D 1997; 103.
- [10] Tang Y, Wang Z, Fang J. Image encryption using chaotic coupled map lattices with time-varying delays. Commun. Nonlinear Sci. Numer. Simulat. 2010; 15:2456-2468.
- [11] Wang S, Hu G, Zhou H. A one-way coupled chaotic map lattice based selfsynchronizing stream cipher. Commun. Nonlinear Sci. Numer. Simulat. 2014; 19:905-913.
- [12] Kaneko K. Spatiotemporal chaos in one- and two-dimensional coupled map lattices. Physica D 1989; 37:60-82.
- [13] He G. Travelling waves in one-dimensional coupled map lattices. Commun. Nonlinear Sci. Numer. Simulat. 1996; 1:16-20.
- [14] Franceschini V, Giberti C, Vernia C. On quasiperiodic travelling waves in coupled map lattices. Physica D 2002; 164:28-44.
- [15] Cherati ZR, Motlagh MRJ. Control of spatiotemporal chaos in coupled map lattice by discrete-time variable structure control. Phys. Lett. A 2007; 370:302-305.
- [16] Jakobsen A. Symmetry breaking bifurcation in a circular chain of N coupled logistic maps. Physica D 2008; 237:3382-3390.
| [17] | Sotelo Herrera D, San Mart´n ı J. Analytical solutions of weakly coupled map lattices using recurrence relations. Phys. Lett. A 2009; 373:2704-2709. |
|--------|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [18] | Xu L, Zhang G, Han B, Zhang L, Li MF, Han YT. Turing instability for a two-dimensional Logistic coupled map lattice. Phys. Lett. A 2010; 374:3447-3450. |
| [19] | Angelini L, Pellicoro M, Stramaglia S. Phase ordering in chaotic map lattices with additive noise. Phys. Lett. A 2001; 285:293-300. |
| [20] | Lai YM, Porter MA. Noise-induced synchronization, desynchronization, and clustering in globally coupled nonidentical oscillators. Phys. Rev. E 2013; 88:012905 |
| [21] | Moehlis J, Josi´ c K, Shea-Brown ET. Periodic orbit. Scholarpedia 2006; 1(7):1358 |
| [22] | Cvitanovi´ c P, Artuso R, Mainieri R, Tanner G, Vattay G, Whelan N, Wirzba A. Chaos: Classical and Quantum. Version 14. (available at http://chaosbook.org ) (2012). |
| [23] | Sotelo Herrera D, San Mart´n ı J. Travelling waves associated with saddle- node bifurcation in weakly coupled CML. Phys. Lett. A 2010; 374:3292- 3296. |
| [24] | Sotelo Herrera D, San Mart´n ı J, Cerrada L. Saddle-node bifurcation cas- cades and associated travelling waves in weakly coupling CML. Int. J. Bif. Chaos. 2012; 22:1250172. |
| [25] | Guckenheimer J, Holmes P. Nonlinear Oscillations, Dynamical Systems, and Bifurcations of Vector Fields Berlin: Springer-Verlag; 1983. |
| [26] | Pomeau Y, Manneville P. Intermittent Transition to Turbulence in Dissi- pative Dynamical Systems. Commun. Math. Phys. 1980; 74:189-197. |
| [27] | San Mart´n ı J. Intermittency cascade. Chaos Solitons & Fractals 2007; 32:816-831. |
| [28] | Milnor J. Self-similarity and hairiness in the Mandelbrot set. Lect. Notes Pure App. Math. 1989; 114:211-257. |
| [29] | dos S. Silva FA, Viana RL, de L. Prado T, Lopes SR. Characterization of spatial patterns produced by a Turing instability in coupled dynamical systems. Commun. Nonlinear Sci. Numer. Simul. 2014; 19:1055-1071. |
| [30] | Lo S-C, Cho H-J. Chaos and control of discrete dynamic traffic model. J. Franklin Inst. 2005; 342:839-851. |
| [31] | Zurigel I et al. Clogging transition of many-particle systems flowing through bottlenecks. Sci. Rep. 2014; 4:7324. |
## Figure Legends
Figure 1: Left: The maps f 3 and f 4 (where f is the logistic map) and the fixed points at which SN bifurcations occur . Observe that there are SN points far away from the critical point C. Because 4 > 3, the extremum of f 4 near the critical point C is narrower than the extremum of f 3 near C. Right: Magnification of the extrema near the critical point C . We show the distances d between the SN point and the critical point for both f 3 and f 4 . Observe that the distance between this pair of points decreases as the period increases.
Figure 2: Temporal evolution of the HWCML (29) for R 1 = r 1 + 2 ε , where r 1 ≈ 3 828427 . (which is an SN bifurcation point) and R 2 = 1 9. . The uncoupled oscillators have (a) period 3 and (b) period 4. When ε = 0 0001 (i.e., weak coupling), . the oscillators X ε ( n ) and Y ε ( n ) both have period lcm(3 , 4) = 12. In panels (c) and (d), we plot X ε ( n ) -X 0 ( n ) and Y ε ( n ) -Y 0 ( n ) (i.e., the solution in the coupled case minus the solution in the ε = 0 case) to better observe the period-12 dynamics.
Figure 3: Temporal evolution of the HWCML (29) for R 1 = r 1 + ε , where r 1 ≈ 3 828427 . (which is an SN bifurcation point) and R 2 = 1 9 . . The uncoupled oscillators have (a) period 3 and (b) period 4. When ε = 0 0001 (i.e., weak coupling), . the oscillators X ε ( n ) and Y ε ( n ) exhibit type-I intermittency. In panels (c) and (d), we plot X ε ( n ) -X 0 ( n ) and Y ε ( n ) -Y 0 ( n ) (i.e., the solution in the coupled case minus the solution in the ε = 0 case) to better observe the intermittency dynamics.
Figure 4: Magnification of the laminar regime of type-I intermittency from Fig. 3d . We can clearly see the resemblance with the temporal evolution of the oscillator in the periodic regime (see Fig. 2d).
Figure 5: Temporal evolution of the HWCML (29) for R 1 = r 1 + 2 ε , where r 1 ≈ 3 738173 . (which is an SN bifurcation point) and R 2 = 1 9 . . The uncoupled oscillators have (a) period 5 and (b) period 4. When ε = 0 0001 (i.e., weak coupling), . the oscillators X ε ( n ) and Y ε ( n ) both have period lcm(5 , 4) = 20. In panels (c) and (d), we plot X ε ( n ) -X 0 ( n ) and Y ε ( n ) -Y 0 ( n ) (i.e., the solution in the coupled case minus the solution in the ε = 0 case) to better observe the period-20 dynamics.
Figure 6: Temporal evolution of the HWCML (29) for R 1 = r 1 + ε , where r 1 is an SN bifurcation point and R 2 = 1 9. . The uncoupled oscillators have (a) period 5 and (b) period 4. When ε = 0 0001 (i.e., . weak coupling), the oscillators X ε ( n ) and Y ε ( n ) show intermittency. In panels (c) and (d), we plot X ε ( n ) -X 0 ( n ) and Y ε ( n ) -Y 0 ( n ) (i.e., the solution in the coupled case minus the solution in the ε = 0 case) to better observe the intermittent dynamics.
## Tables
Table 1: Period of the CML (29) for r 2 = 1 9 . and ε = 0 001. The parameter . r 1 indicates when the logistic map, which is satisfied by the free oscillator X n ( ), exhibits orbits of various periods during a period-doubling cascade in the window of period-3 orbits in the bifurcation diagram. Although the period of X n ( ) changes, the period of the CML remains the same.
| r 1 | Period of X ( n ) | Period of the CML |
|------------|---------------------|-----------------------|
| 3 . 831874 | 3 | lcm(3 , 4) = 12 |
| 3 . 844568 | 3 × 2 | lcm(3 × 2 , 4) = 12 |
| 3 . 848344 | 3 × 2 2 | lcm(3 × 2 2 , 4) = 12 | | null | [
"Dolores Sotelo Herrera",
"Jesús San Martín",
"Mason A. Porter"
] | 2014-08-01T20:39:43+00:00 | 2015-12-10T17:22:43+00:00 | [
"nlin.PS",
"math.DS"
] | Heterogeneous, Weakly Coupled Map Lattices | Coupled map lattices (CMLs) are often used to study emergent phenomena in
nature. It is typically assumed (unrealistically) that each component is
described by the same map, and it is important to relax this assumption. In
this paper, we characterize periodic orbits and the laminar regime of type-I
intermittency in heterogeneous weakly coupled map lattices (HWCMLs). We show
that the period of a cycle in an HWCML is preserved for arbitrarily small
coupling strengths even when an associated uncoupled oscillator would
experience a period-doubling cascade. Our results characterize periodic orbits
both near and far from saddle--node bifurcations, and we thereby provide a key
step for examining the bifurcation structure of heterogeneous CMLs. |
1408.0304v1 | ## Polariton-based optical switch
Oleg L. Berman, Roman Ya. Kezerashvili, and German V. Kolmakov Physics Department, New York City College of Technology, City University of New York, Brooklyn, NY 11201, USA
Based on the studies of propagation of an exciton-polariton condensate in a patterned optical microcavity with an embedded graphene layer, we propose a design of a Y-shaped electrically-controlled optical switch. The polaritons are driven by a time-independent force due to the microcavity wedge shape and by a time-dependent drag force owing to the interaction of excitons in graphene and the electric current running in a neighboring quantum well. It is demonstrated that one can control the polariton flow direction by changing the direction of the electric current. The simulations also show that an external electric field normal to the microcavity plane can be utilized as an additional parameter that controls the propagation of the signals in the switch. By considering the transient dynamics of the polariton condensate, we estimate the response speed of the switch. Finally, we propose a design of the polariton switch in a flat microcavity based on the geometrically identical Y-shaped quantum well and graphene pattern where the polariton flow is only induced by the drag force.
## I. INTRODUCTION
In the past decades, substantial experimental and theoretical efforts were devoted to find the optimal ways to tune the optical properties of semiconductors and graphene by application of external electric and magnetic fields. The motivation of this research lies in the potential applications for integrated circuits in optical and quantum computers, for secure information transfer, and in new light sources. 1 One of the promising approaches is in the use of polaritons, which are a quantum superposition of cavity photons and excitons in a nanometer-wide semiconductor layer (a quantum well) or graphene. 2-7 Since polaritons are interacting Bose particles, a polariton gas can transit to a superfluid state that, under certain conditions, propagates in a microcavity almost without dissipation. 7-10 Due to a small effective mass, 10 -4 of the free electron mass, the superfluid transition occurs at relatively high temperatures that are comparable with the room temperature. Polaritons can propagate in the sample with a speed up to a few percent of the speed of light that is, much faster than that of electric-field driven electrons and holes in semiconductors. 11 However, one of the main problems to be solved in actual design of polariton-based optical devices is in weak response of polaritons to an external electric field owing to their net zero electric charge.
In this paper, we propose a design of polariton-based, electrically controlled switch, in which the polariton propagation is controlled by means of a drag force. Recently, it was shown that the drag caused by an electric current results in entrainment of polaritons and in formation of a persistent, directed polariton current in the microcavity. 12,13 In this case, this is a drag force that is exerted on the exciton component of polaritons that results in generation of the polariton flux proportional to the driving electric field E ,
$$j = \gamma E,$$
where γ is the temperature-dependent drag coefficient. In addition to the electrically controlled drag force, the polaritons can also be accelerated by a constant, time independent force, which is caused by the wedge-like shape of the microcavity. 14,15 In the latter case, the effect of the constant force is similar to application of a dc voltage in electric circuits. By changing the magnitude and direction of the electric current one can vary the net force exerted on the polaritons and thus, to control the polariton propagation.
In this paper, we propose the polariton switches of two types. In the first design, the polariton flow in the switch is produced by a constant force due to the wedge-like shape of the microcavity whereas the drag force is utilized to switch the direction of the flow. In the second design, the polariton flow in the required direction is only produced by the drag force. The photons propagate in the planar microcavity, and the excitons are located in a quantum well embedded in the microcavity. Below we describe the both types of the switches in details.
In the first design, we take advantage of a patterned microcavity to create a potential-energy landscape for polaritons in the form of one-dimensional channels, or 'polariton wires', which are somewhat similar to the ordinary wires that conduct electrical current. However, in our case, the wires conduct polaritons and help one to deliver photons to a desired location in the semiconductor structure. Recently developed experimental techniques, which enable one to produce such guiding potentials for polaritons, include deposition of a metal pattern on the Bragg reflectors 16 and modulation of the cavity layer thickness in a mesa structure. 17,18 The polariton channels of complex shape and topology can also be engineered directly in an air-filled microcavity by creating quasi-one-dimensional waveguides, or micropillars. 19-23
FIG. 1: (a) Schematic of the wedge-shaped microcavity formed by two distributed Bragg reflectors (DBR) that encompasses an embedded graphene layer. Polaritons in the microcavity are formed as a quantum superposition of excitons in graphene and cavity photons. The force F x is exerted on the polaritons in x -direction owing to a small opening angle α ≈ ( h 2 -h 1 ) /L /lessmuch 1 of the microcavity, where h 1 and h 2 are the microcavity lengths at the opposite edges and L is the horizontal size of the microcavity. A metal layer deposited on the upper DBR creates a Y-shaped potential energy landscape for the polaritons shown in (b). The polaritons are created by an external laser radiation in the stem of the channel and propagate towards the junction due to the force F x . A driving electric current I runs perpendicularly to the stem of the channel in a quantum well QW (green) in the area of the junction. The polariton flux j is formed in one branch of the channel in response to the drag created by the current.
FIG. 2: Schematic of the polariton switch where the polariton flux in a flat microcavity is induced by the drag due to an electric current running in the circuit. Both the quantum well that carries the electron current and the graphene layer containing the excitons have identical geometrical Y shape and are positioned parallel to each other. The diffraction Bragg reflectors positioned parallel to the graphene layer and the quantum well are not shown. The position of switch S controls the direction of the polariton flow.
To design a polariton switch, we utilize a Y-shaped channel shown in Fig. 1. We assume that polaritons are created at a constant rate by the external laser pumping at the stem of the channel and then, propagate towards the junction under the action of a constant force due to the microcavity wedge-like shape. After reaching the junction, the polariton flow splits between the two branches. If no drag force is applied to the polaritons, the flux is distributed equally between the branches. However, if the drag force is exerted on the polaritons in the region of the junction in the direction normal to the stem, the polaritons are 'pushed' towards one branch and hence, the flux in that branch increases. Simultaneously, the flux through another branch decreases due to approximate conservation of the total number of polaritons and in effect, the polariton flux is redistributed between the branches. The drag force in the junction region is produced by the electric current running in a neighboring quantum well, which has a form of a stripe placed in a microcavity perpendicular to the stem of the channel. Thus, the heterostructure encompasses two parallel quantum wells, separated by a nanometer-wide semiconductor of dielectric barrier, placed in an optical microcavity. Heterostructures embedded in a microcavity are widely used in the studies of exciton dynamics 24,25 and in experiments with polaritons. 22,26,27 We demonstrate below that more than 70% of the total polariton flux can be dynamically redistributed between the branches for realistically achievable drag forces.
The second type of the polariton switch is based on a flat microcavity where the mirrors are parallel to each other. In this design, the excitons are located in a Y-shaped semiconductor quantum well, and the neighboring currentcarrying quantum well geometrically repeats the shape of the structure, see Fig. 2. In contrast to the first type of
the switch described above, here we do not use the patterned microcavity. Instead, the polaritons are guided by the Y-shaped semiconductor structure. The electric current in the quantum well is generated by the external voltage. The drag force exerted on the cavity polaritons by the electric current is directed parallel to the channel and induces the polariton flow along one of the branches. The polariton flow is redistributed between the branches in response to switching of the electric voltage applied to the current-carrying quantum well.
In both designs, the quantum wells containing the excitons can be fabricated by using a number of materials, including GaAs, 24,25 CdTe, 28 WSe , 2 29 and gapped graphene. 30,31 While the polariton condensate was experimentally observed when the polaritons are formed by excitons in a semiconductor quantum well , in this paper we focus on 6 microcavity polaritons formed by excitons in graphene. One of the advantages to use graphene is related to the properties of polaritons in gapped graphene that can be tuned by application of a static electric field normal to the microcavity plane. Specifically, the energy gap between the valence and conduction bands can be created and changed by the electric field in a wide range. 32-34 This enables one to tune the polariton effective mass and the sound velocity in the spectrum of the quasiparticles 35 and through that to optimize the performance of the system, as detailed below. Another advantage of considering excitons in a graphene layer compared to a semiconductor quantum well is the absence of the fluctuations of the width due to the fact that graphene is a 2D one atom-thick material. Such fluctuations cause the disorder which results to the decrease of the drag coefficients due to the scattering of the excitons on the random field. Thus, in both designs of the polariton switch described above we consider graphene embedded in microcavity. Our consideration of the switch with polaritons formation in a quantum well currently is in progress.
## II. THE DYNAMICS OF A POLARITON CONDENSATE
The dynamics of the polariton condensate was captured via the non-equilibrium Gross-Pitaevskii equation for the condensate wave function Ψ( r , t ) 36
$$i \hbar { \frac { \partial \Psi ( r, t ) } { \partial t } } & = \, - \, \frac { \hbar { ^ { 2 } } } { 2 m } \Delta \Psi ( r, t ) + U ( r, t ) \Psi ( r, t ) + g \Psi ( r, t ) | \Psi ( r, t ) | ^ { 2 } \\ & \quad - \, \frac { i } { 2 \tau } \Psi ( r, t ) + i P ( r ), \\. \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdots$$
where m is the polariton mass, r = ( x, y ) is a two-dimensional vector in the plane of the microcavity, and time t , g is the polariton-polariton interaction strength, τ is the polariton lifetime, and the source terms P ( r ) describes incoherent laser pumping of the polariton reservoir. In our simulations, we set τ = 100 ps as a representative value. 15,37
The effective potential for the polaritons
$$U ( r, t ) = U _ { c o n f } ( r ) + U _ { w } ( r ) + U _ { d r a g } ( r, t )$$
is the sum of the confining potential owing to microcavity patterning U conf ( r ), a linear potential corresponding to a constant accelerating force in a wedge-shaped microcavity U w ( r ), and a time-dependent drag U drag ( r , t ) caused by the driving electric current. The confining potential U conf ( r ) that forms the Y-shaped channel is shown in Fig. 1b. The potential depth compared to zero value outside the waveguide is taken equal 1 eV that is a representative value for waveguides in the microcavities. 16-18 The average force acting upon a polariton wave packet in a wedge-shaped microcavity is F r ( ) = -∇ E C ( r ), where E C ( r ) is the energy of the polariton band taken at the in-plane wavevector of the polariton k = 0. For the wedge-like microcavity considered in this paper, the energy E C ( r ) is a linear function of the spatial coordinate 14,15 thus, the force is coordinate-independent. The corresponding potential is
$$U _ { w } ( r ) = - F _ { x } x,$$
where F x = | ∂E C ( r ) /∂x | , and we suppose that the force is applied in x -direction, along the stem of the Y-shaped channel.
The potential that described the drag force in y -direction is taken equal
$$U _ { d r a g } ( r, t ) = F _ { d r a g } ( t ) y$$
in a stripe 400 µ m < x < 600 µ m and U drag ( r , t ) = 0 otherwise. The drag force exerted on polaritons by charges moving in a neighboring quantum well is estimated in the τ -approximation as F drag ( ) t = 〈 p /τ 〉 p = mγE t /n τ ( ) n p , where 〈 p 〉 is the average gain of the linear momentum of polaritons owing to the drag, τ p is the polariton momentum relaxation time, E t ( ) is a time-dependent electric field applied in the plane of the quantum well with free electrons, n n = 3 (3) ( ζ s k B T ) 3 / 2 π /planckover2pi1 2 c 4 s m is the density of the normal component in a polariton superfluid, 38 ζ (3) ≈ 1 202 is the .
Riemann zeta function, s = 4 is the spin degeneracy factor, k B is the Boltzmann constant, T is temperature, and c s = √ gn/m is the sound velocity in the polaritonic system. By taking the polariton condensate density n = 10 14 m -2 , the separation between the graphene layer with the excitons and the quantum well with the electrons D = 15 nm, γ = 6 × 10 16 (Vs) -1 and the relaxation time τ p = 6 × 10 -11 s as representative parameters for temperature T = 20 K, 13,39-41 one obtains the maximum drag force F drag = 0 3 . -6 meV/mm for the working range of electric fields E = 10 -2 × 10 2 mV/mm. In the simulations, we varied the maximum drag force F drag from 1 to 4 meV/mm.
Formation of the excitons in graphene requires a gap in the electron and hole excitation spectra, which can be created and dynamically tuned by applying an external electric field in the direction normal to the graphene layer. 32-34 The gap can also be opened by chemical doping of graphene. 42 The polariton mass m and the interaction strength g in gapped graphene depend on the gap energy δ . The polariton effective mass in gapped graphene is 7
$$m = 2 ( m _ { \text{ex} } ^ { - 1 } + c L _ { C } / \sqrt { \varepsilon } \pi \hbar { ) } ^ { - 1 },$$
where m ex is the exciton effective mass in graphene, L C is the length of the microcavity, ε is the dielectric constant of the microcavity, and c is the speed of light in vacuum. In the simulations, we set ε = 13 for a GaAs-based microcavity. 13 We consider the case of zero detuning where the cavity photons and the excitons in graphene are in resonance at k = 0. In this case the length of the microcavity is 7
$$L _ { C } & = \frac { \hbar { \pi } c } { ( 2 \delta - V _ { 0 } + C / \delta ^ { 2 } ) \sqrt { \varepsilon } }, \quad \begin{array} { c } \\ \\ \\ \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \colon \end{array} \quad \begin{array} { c } \\ \\ \\ \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \\ \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \\ \cdot \cdot \cdot \cdot. \end{array}$$
where V 0 = e / 2 4 πεε r 0 ′ , C = ( /planckover2pi1 ev F ) 2 / 8 πεε r 0 ′ 3 , ε 0 is the permittivity of free space, and v F ≈ 10 6 m/s is the Fermivelocity of the electrons in graphene. The parameter r ′ is found from the equation
$$2 \delta ^ { 2 } ( 2 \delta - \hbar { \omega } ) r ^ { \prime 3 } - 2 D \delta ^ { 2 } r ^ { \prime 2 } + D ( \hbar { v } _ { F } ) ^ { 2 } = 0,$$
where D = e / 2 4 πεε 0 . For dipolar excitons in GaAs/AlGaAs coupled quantum wells, the energy of the recombination peak is /planckover2pi1 ω = 1 61 eV . 43 . We expect similar photon energies in graphene. However, its exact value depends on the graphene dielectric environment and substrate properties. The exciton effective mass in Eq. (6) is 7
$$m _ { \text{ex} } = \frac { 2 \delta ^ { 4 } } { C v _ { F } ^ { 2 } }.$$
The polariton-polariton interaction strength is 7,44
$$g = \frac { 3 e ^ { 2 } a _ { B } } { 8 \pi \varepsilon _ { 0 } \varepsilon },$$
where a B = 2 πε ε 0 /planckover2pi1 2 /m e r 2 is the two-dimensional Bohr radius of the exciton, m r = 1 4 m ex is the exciton reduced mass.
## III. RESULTS AND DISCUSSION
## A. Electrically controlled optical switch in a wedge-shaped microcavity
We study propagation of the polariton condensate in a patterned microcavity by numerically integrating the nonequilibrium Gross-Pitaevskii equation (2) for the wave function Ψ( r , t ) of the polariton condensate in the microcavity. In the model that is described in details in Sec. II we we determine the conditions that enable one to control the polariton spreading in a potential landscape.
The layered semiconductor structure considered in this section is shown in Fig. 1a. The excitons are positioned in the graphene layer whereas the free electrons are located in the neighboring, parallel quantum well. The heterostructure is placed between two high-quality multilayer mirrors (distributed Bragg reflectors, or DBR). In what follows, we consider a microcavity decorated with a Y-shaped pattern. Fig. 1b demonstrates the potential-energy landscape for the polaritons due to patterning of the microcavity. Additionally, the reflectors form a wedge opened towards the positive direction of the x -axis (that is, to the right in Fig. 1) that results in a constant force exerted on the polaritons in a microcavity, as explained in Sec. II. In the simulation, we set the constant force equal F x = 13 meV/mm that corresponds to a representative experimental value. 14,15 The polaritons are created by an external laser radiation in the region of the stem. In what follows, an excitation spot has a Gaussian profile with a full width at half maximum (FWHM) of 65 µ m, centered in the stem part of the channel at r = (250 250) , µ m.
FIG. 3: Propagation of the polariton condensate in a wedge-shaped microcavity with embedded graphene (a) with and (b) without the pattern deposited on the Bragg reflectors. The constant force acting on the polaritons in the x -direction owing to the wedge is F x = 13 meV/mm. The color bar shows the polariton density n = Ψ( | r , t ) | 2 . The figure shows the steady state reached after the polariton source is turned on. The polariton flux splits between the upper and lower branches of the channel in Figure (b). Red arrows show the direction of the polariton flow.
The effect of patterning on the polariton condensate propagation is shown in Fig. 3. It is seen that without the pattern (that is, without the confining potential in the ( x, y ) plane), a polariton condensate driven by the constant force F x forms a wide trail in the direction of the force with the length of ∼ 400 µ m. At larger distances from the excitation spot the condensate density rapidly decreases due to finite lifetime of polaritons. Propagation of the condensate under the same conditions in the presence of a Y-shaped pattern is shown in Fig. 3b. As is seen in this Figure, the polariton flow propagating in the patterned microcavity is localized in the channel formed by the pattern. In the region of the junction x ≈ 500 µ m, the polariton condensate flow splits between the 'upper' and 'lower' branches of the Y-channel. The polariton density in the channel at large distances x ∼ 500 µ m from the source is significantly higher than that in the microcavity without the channel (cf. Fig. 3a and b). The increase of the polariton density is caused by the confining potential that prevents the polariton flow from spreading in the ( x, y ) plane.
To probe the effect of the drag force on the polariton flow redistribution between the branches, in addition to the constant force F x we applied the force F drag directed in y -direction in a stripe 400 µ m < x < 600 µ m in the region of the junction. In Fig. 4a the area, in which the drag force F drag is exerted, is bounded by the dashed lines. We found in the simulations that, due to the drag, the polaritons tend to propagate in the channel in the direction of the drag force F drag and therefore, they mostly move in the upper branch of the channel in Fig 4a. To characterize the redistribution of the polaritons between the upper and lower branches we calculated the total polariton flux propagating in the branch,
$$J = \int d s j _ { \nu } ( r, t ),$$
where j ν ( r , t ) ≡ j r ( , t ) · ν is the component of the polariton flux density parallel to the channel, ν is a unit vector along the branch, and
$$j ( r, t ) = \frac { \hbar } { 2 i m } ( \Psi ^ { * } ( r, t ) \nabla \Psi ( r, t ) - \Psi ( r, t ) \nabla \Psi ^ { * } ( r, t ) )$$
is the flux density. 45 We integrate in Eq. (11) over the cross section of the branch. The flux J was calculated at x = x 1 ≡ 700 µ m that is, at a distance from the junction much larger than the width of the channel a ch ∼ 60 µ m.
Fig. 4b shows the dependencies of the total polariton flux trough the upper (red curve) and lower (blue curve) branches as functions of time t . In these simulations, the energy gap in the graphene quasiparticle excitation spectrum is set δ = 0 3 eV, the drag force is . F drag = 4 meV/mm, and the polariton source is turned on at the moment t = 0. According to the results shown in Fig. 4b, the polaritons propagate to the point x 1 at t ∼ 70 ps after the source is
FIG. 4: (a) Redistribution of the polariton flux between the upper and lower branches of the channel in response to the driving drag force F drag in the region of the junction. The boundaries of a stripe 400 µ m < x < 600 µ m in which the drag force is exerted on the polaritons are shown by vertical dashed lines. The cross-section of the branches, at which the polariton flux J in Eq. (11) is determined, is labeled as x 1 . (b) Dependence of the polariton flux in the upper J up and lower J low branches on time after the polariton source is turned on. The transient oscillations last for t ∼ 200 ps after the source is switched on and then, the system comes to a steady state. The energy gap in the graphene layer in the microcavity is δ = 0 3 eV. .
turned on. During the time interval 70 ps < t < 200 ps the polariton flux exhibits transient oscillations and then, tends to a constant value at t ≥ 200 ps. As it follows from Fig. 4b, the flux through the upper branch of the channel in the steady state is J ≈ 2 7 . × 10 14 s -1 , that is ≈ 2 8 . × larger than that through the lower branch. Therefore, the polariton flow is redistributed among the branches in response to the drag force owing to the external, driving electric current. To quantify the redistribution of the polariton flux in the channel, we studied the dependence of the flux J on the drag force F drag . In the simulations, we varied F drag from 1 meV/mm to 4 meV/mm, as explained in Sec. II. It is shown in Fig. 5a that the flux in the upper branch J up gradually increases with the rise of the drag force F drag . In its turn, the flux through the lower channel, J low decreases thus, the total flux J tot = J up + J low remains approximately constant. To more fully characterize the polariton flow in the channel, we determined the performance of the system,
$$Q = ( J _ { u p } / J _ { t o t } ) \times 1 0 0 \%,$$
as a function of the energy gap in the graphene layer. We varied the gap δ from 0.1 to 0.5 eV that corresponds to the experimentally accessible range. 32-34,42 The results for the constant drag force F drag = 4 meV/mm are summarized in Fig. 5b. It is seen that the performance Q can be approximated by a linear function of the gap δ and it reaches ∼ 77% for the maximum gap δ = 0 5 eV. In other words, more than 3 . / 4 of the total polariton flux propagates through the upper channel for F drag = 4 meV/mm, and less than 1 / 4 of the flux only propagates through the lower channel. Summarizing, we demonstrated that one can govern the propagation of a polariton condensate in the Y-shaped channel by means of the external electric current running in a neighboring quantum well. This make it possible to construct an optical, polariton-based switch controlled by an external voltage.
Finally, to estimate the response speed of the system, we studied the transient dynamics of a polariton condensate in the channel when the direction of the driving current is inverted. In these simulations, we turned on the pumping and waited t ∼ 180 ps until the system reached an equilibrium in the presence of a drag force. In this steady state, the polaritons mostly propagated in the upper channel, in accordance with our previous consideration. At t = 180 ps, we changed the direction of the drag force to opposite. Since the drag force now pushes the polaritons down in the junction area, the polaritons tend to propagate via the lower channel, whereas the flux through the upper channel now decreases. After the transient oscillations were damped, the system came to a new equilibrium state where most of the polaritons propagated through the lower channel. The dependencies of the polariton fluxes J up and J low on time are shown in Fig. 6. According to this Figure, the characteristic switching time of the system between the states where
/s40/s97/s41
/s32
/s32
/s32
/s32
FIG. 5: (a) Redistribution of the polariton flux between the upper and lower branches in response to the magnitude of the driving drag force F drag . The polaritons fluxes in the upper and lower channel, J up and J low , are determined at the cross-section labeled as x 1 in Fig. 4a in a steady state after the transient oscillations damped. (b) Performance of the system Q, Eq. (13), as a function of the energy gap in the graphene layer. It is seen that up to 77% of polaritons are propagating through the upper channel if the drag force F drag = 4 meV/mm is applied at the junction.
/s32
/s32
FIG. 6: Response of the system to switching direction of the drag force. The drag is directed towards the upper channel at t < t 0 ≡ 180 ps after the source is turned on; the polaritons mostly propagate via the upper branch. The direction of the drag is changed to opposite at t = t 0 and then, the transient oscillations of the polariton flux occur for ∼ 70 ps. After the oscillations are damped, most of polaritons propagate via the lower branch. The magnitude of the drag force and the energy gap are labeled in the figure.
the polaritons propagate via the upper and lower branches is t sw ≈ 70 ps. This means that the maximum switch frequency for this junction can reach f = t -1 sw ≈ 14 GHz at given linear dimensions of the system. This frequency, however, should rise with decreasing channel length. In this paper we only consider the switch of length L ∼ 600 µ m and defer the size-dependent effects to the future studies.
## B. Polariton switch in a flat microcavity
We now consider another design of the polariton switch based on the drag effect. In this case, the reflectors are parallel to each other. In such 'flat' microcavity the only force acting on the polaritons is the drag force due to the interactions with the current. Such flat microcavities are widely used in the studies of exciton polaritons in semiconductor heterostructures. 24-31
In this design, the microcavity has no decoration. The Y-shaped channel is formed by graphene stripes. 46 While the design of nanojunction of the graphene stripes is technologically challenging, the possibility of fabrication of a junction of graphene stripes with help of the covalent functionalization has recently been proposed. 47 We note that the utility of electric splitters based on graphene structures has recently been discussed in the literature. 48 The current-carrying quantum well is now Y-shaped that is, it repeats the shape of the channel. Similarly to the design above, the excitons in graphene are excited due to the laser pumping at the stem of the channel. The main concept is schematically illustrated in Fig. 2.
The voltage V is applied across the electron QW between the stem and one of the branches. The polaritons formed by excitons in graphene and cavity photons are dragged by the electron current in the electron QW along one of the branches. The direction of the electric current and, hence, of the polariton flux is defined by the electric switch S (see Fig. 2). If the switch S is closed in position A then, the electric current in the electron QW runs in the upper branch. Due to the drag effect caused by the interactions between the excitons in graphene and the electrons in QW, 12,13 the electron current induces the flow of polariton quasiparticles in the cavity along the upper branch. Then, if the switch is closed in position B , the flow of polariton quasiparticles propagates along the lower branch. Thus, the path of the polaritons in the Y-shaped channel can be dynamically changed by changing the position of the switch S .
The flux density j of the polariton superfluid in the channel is given by Eq. (1). For the polaritons velocities smaller than the speed of sound c s in the polariton superfluid, the polariton flux density can be found as follows 13
$$j \equiv n _ { n } v = \gamma E$$
where n n is the density of the normal component in the polariton subsystem formed by the quasiparticles, and v is the average velocity of the dragged polariton quasiparticles. Estimating the magnitude of the electric field as | E | /similarequal V/L , where V is the voltage across the channel and L is the length of the channel, one obtains for the magnitude of the averaged polariton velocity
$$v & = \frac { \gamma V } { n _ { n } L } \.$$
The time t , required for a quasiparticle to propagate across the channel, is estimated as
$$t \simeq \frac { L } { v } = \frac { n _ { n } L ^ { 2 } } { \gamma V }.$$
In the relevant parameter range discussed in Sec. II, Eq. (16) gives the estimate t ≈ 72 ps for the channel of length L = 100 µ m, the voltage V = 20 mV, and energy gap δ = 0 4 eV. The average velocity of the polaritons calculated . from Eq. (15) is v = 1 38 . × 10 6 m/s that is smaller than the sound velocity c s = 1 81 . × 10 6 m/s in the polariton superfluid. Therefore, the breakdown of superfluidity in the system does not occur. 10,13 For polaritons with lifetime τ ∼ 100 ps, 15,37 the time t required for the quasiparticles to propagate across the channel is smaller than τ that validates the proposed design.
## IV. CONCLUSIONS
In the above studies we considered two designs of an electrically-controlled optical switch. In both cases, we focused on propagation of polaritons in a microcavity, which are a quantum superposition of cavity photons and excitons. A significant challenge is designing a system where the direction of the polariton flow is controlled by means of the electrical current since both photons and excitons are electrically neutral. In our studies, to control the polariton propagation we utilize the drag effect that is, the entrainment of polaritons by an electric current running in the neighboring quantum well. The drag effect for excitons in a planar geometry is well studied. Recently, the drag of polaritons by the electric current in an unrestricted planar microcavity has been considered theoretically. In the above studies, we utilize this approach to design nanostructures, in which the polaritons are localized in a circuit made by quasi-one dimensional polariton wires.
In the first setup, the polaritons propagate in the Y-shaped channel owing to a constant force created in a wedgeshaped microcavity. In this case, the controlling drag force is directed perpendicularly to the direction of the stem of
the channel and it pushes the polariton in one of the two branches. Changes in the direction of the electric current results in switching of the path in which the polaritons propagate.
In the second setup, the polaritons in a flat microcavity are considered. In this case, the constant force is absent and the polaritons are only dragged by an electric current running in a Y-shaped quantum well. The channel containing the excitons is formed by graphene stripes embedded in the microcavity. Despite the latter setup is more technologically challenging, mostly due to the requirement of a high-quality junction between the stripes, this scheme provides additional advantages for the polariton flow control. Specifically, the magnitude of the polariton flow can be tuned by setting the voltage across the channel to a given value.
In both cases, we focused on microcavities with embedded gapped graphene. The advantage of graphene, compared to standard semiconductor quantum wells, is in the possibility to dynamically tune the effective mass of the polaritons by application of an electric field normal to the microcavity plane. This provides an additional opportunity to control the polariton propagation in the device.
## Acknowledgments
The authors are gratefully acknowledge support from Army Research Office, grant #64775-PH-REP. G.V.K. is also grateful for support to Professional Staff Congress - City University of New York, award #66140-0044. The authors are grateful to the Center for Theoretical Physics of New York City College of Technology of the City University of New York for providing computational resources.
- 1 H. Gibbs, G. Khitrova, and N. Peyghambarian, eds., Nonlinear Photonics (Springer, London, 2011).
- 2 M. Zamfirescu, A. Kavokin, B. Gil, G. Malpuech, and M. Kaliteevski, Phys. Rev. B 65 , 161205(R) (2002).
- 3 H. Deng, G. Weihs, D. Snoke, J. Bloch, and Y. Yamamoto, Proc. Natl. Acad. Sci. U.S.A. 100 , 15318 (2002).
- 4 T. C. H. Liew, A. V. Kavokin, T. Ostatnicky, M. Kaliteevski, I. A. Shelykh, and R. A. Abram, Phys. Rev. B 82 , 033302 (2010).
- 5 V. M. Menon, L. I. Deych, and A. A. Lisyansky, Nature Photonics 4 , 345 (2010).
- 6 H. Deng, H. Haug, and Y. Yamamoto, Rev. Mod. Phys. 82 , 1489 (2010).
- 7 O. L. Berman, R. Y. Kezerashvili, and K. Ziegler, Phys. Rev. B 86 , 235404 (2012).
- 8 I. Carusotto and C. Ciuti, Phys. Rev. Lett. 93 , 166401 (2004).
- 9 A. Amo, J. Lefrere, S. Pigeon, C. Adrados, C. Ciuti, I. Carusotto, R. Houdre, E. Giacobino, and A. Bramati, Nature Physics 5 , 805 (2009).
- 10 M. Wouters and V. Savona, Phys. Rev. B 81 , 054508 (2010), URL http://link.aps.org/doi/10.1103/PhysRevB.81.054508 .
- 11 M. Wouters and I. Carusotto, Phys. Rev. Lett. 105 , 020602 (2010).
- 12 O. L. Berman, R. Y. Kezerashvili, and Y. E. Lozovik, Phys. Lett. A 374 , 3681 (2010).
- 13 O. L. Berman, R. Y. Kezerashvili, and Y. E. Lozovik, Phys. Rev. B 82 , 125307 (2010).
- 14 B. Sermage, G. Malpuech, A. V. Kavokin, and V. Thierry-Mieg, Phys. Rev. B 64 , 081303(R) (2001).
- 15 B. Nelsen, G. Liu, M. Steger, D. W. Snoke, R. Balili, K. West, and L. Pfeiffer, Phys. Rev. X 3 , 041015 (2013).
- 16 S. Utsunomiya, L. Tian, G. Roumpos, C. W. Lai, N. Kumada, T. Fujisawa, M. Kuwata-Gonokami, A. L¨ffler, S. H¨fling, o o A. Forchel, et al., Nature Physics 4 , 700 (2008).
- 17 O. E. Da¨ ıf, A. Baas, T. Guillet, J.-P. Brantut, R. I. Kaitouni, J. L. Staehli, F. Morier-Genoud, and B. Deveaud, Appl. Phys. Lett. 88 , 061105 (2006).
- 18 R. I. Kaitouni, O. E. Da¨ ıf, A. Baas, M. Richard, T. Paraiso, P. Lugan, T. Guillet, F. Morier-Genoud, J. D. Gani`re, J. L. e Staehli, et al., Phys. Rev. B 74 , 155311 (2006).
- 19 E. Wertz, L. Ferrier, D. D. Solnyshkov, R. Johne, D. Sanvitto, A. Lemaitre, I. Sagnes, R. Grousson, A. V. Kavokin, P. Senellart, et al., Nat. Physics 6 , 860 (2010).
- 20 L. K. Van Vugt, B. Piccione, C.-H. Cho, P. Nukala, and R. Agarwal, Proc. Natl. Acad. Sci. U.S.A. 108 , 10050 (2011).
- 21 H. S. Nguyen, D. Vishnevsky, C. Sturm, D. Tanese, D. Solnyshkov, E. Galopin, A. Lemaˆ ıtre, I. Sagnes, A. Amo, G. Malpuech, et al., Phys. Rev. Lett. 110 , 236601 (2013).
- 22 A. Das, P. Bhattacharya, J. Heo, A. Banerjee, and W. Guo, Proc. Natl. Acad. Sci. U.S.A. 110 , 2735 (2013).
- 23 T. Boulier, M. Bamba, A. Amo, C. Adrados, A. Lemaitre, E. Galopin, I. Sagnes, J. Bloch, C. Ciuti, E. Giacobino, et al., Nat. Commun. 5 , 3260 (2014).
24
R. Balili, B. Nelsen, D. W. Snoke, L. Pfeiffer, and K. West, Phys. Rev. B
79
,
075319 (2009).
- 25 L. Butov, Solid State Communications 127 , 89 (2003).
- 26 C. W. Lai, N. Y. Kim, S. Utsunomiya, G. Roumpos, H. Deng, M. D. Fraser, T. Byrnes, P. Recher, N. Kumada, T. Fujisawa, et al., Nature 450 , 529 (2007).
- 27 M. Aßmann, J.-S. Tempel, F. Veit, M. Bayer, A. Rahimi-Iman, A. L¨ffler, S. H¨fling, S. Reitzenstein, L. Worschech, and o o A. Forchel, Proc. Natl. Acad. Sci. U.S.A. 108 , 1804 (2011).
- 28 J. Kasprzak, M. Richard, S. Kundermann, A. Baas, P. Jeambrun, J. M. J. Keeling, F. M. Marchetti, M. H. Szymanska, R. Andre, J. L. Staehli, et al., Nature 443 , 409 (2006).
- 29 S. Wu, S. Buckley, A. M. Jones, J. S. Ross, N. J. Ghimire, J. Yan, D. G. Mandrus, W. Yao, F. Hatami, J. Vuckovic, et al., 2D Materials 1 , 011001 (2014).
- 30 M. Furchi, A. Urich, A. Pospischil, G. Lilley, K. Unterrainer, H. Detz, P. Klang, A. M. Andrews, W. Schrenk, G. Strasser, et al., Nano Letters 12 , 2773 (2012).
- 31 N. Youngblood, Y. Anugrah, R. Ma, S. J. Koester, and M. Li, Nano Letters 14 , 2741 (2014).
- 32 A. B. Kuzmenko, I. Crassee, D. van der Marel, P. Blake, and K. S. Novoselov, Phys. Rev. B 80 , 165406 (2009).
- 33 K. F. Mak, C. H. Lui, J. Shan, and T. F. Heinz, Phys. Rev. Lett. 102 , 256405 (2009).
- 34 Y. Zhang, T.-T. Tang, C. Girit, Z. Hao, M. C. Martin, A. Zettl, M. F. Crommie, Y. R. Shen, and F. Wang, Nature 459 , 820 (2009).
- 35 O. L. Berman, R. Y. Kezerashvili, Y. E. Lozovik, and D. W. Snoke, Phil. Trans. R. Soc. A 368 , 5459 (2010).
- 36 I. Carusotto and C. Ciuti, Rev. Mod. Phys. 85 , 299 (2013).
- 37 A. Amo, D. Sanvitto, F. P. Laussy, D. Ballarini, E. del Valle, M. D. Martin, A. Lemaitre, J. Bloch, D. N. Krizhanovskii, M. S. Skolnick, et al., Nature 457 , 291 (2009).
- 38 O. L. Berman, R. Y. Kezerashvili, and K. Ziegler, Phys. Rev. B 85 , 035418 (2012).
- 39 N. N. Zinov'ev, L. P. Ivanov, I. G. Leng, S. T. Pavlov, A. V. Prokaznikov, and I. D. Yaroshetskii, Sov. Phys. JETP 57 , 1254 (1983).
- 40 T. Takagahara, Quantum Coherence, Correlation and Decoherence in Semiconductor Nanostructures (Academic Press, London, 2003).
- 41 P. K. Basu and P. Ray, Phys. Rev. B 44 , 1844 (1991).
- 42 D. Haberer, D. V. Vyalikh, S. Taioli, B. Dora, M. Farjam, J. Fink, D. Marchenko, T. Pichler, K. Ziegler, S. Simonucci, et al., Nano Lett. 10 , 3360 (2010).
- 43 V. Negoita, D. W. Snoke, and K. Eberl, Phys. Rev. B 60 , 2661 (1999).
- 44 C. Ciuti, P. Schwendimann, and A. Quattropani, Semicond. Sci. Technol. 18 , S279 (2003).
- 45 A. Messiah, Quantum Mechanics , Series in physics (Interscience, New York, 1961).
- 46 C. Berger, Z. Song, X. Li, X. Wu, N. Brown, C. Naud, D. Mayou, T. Li, J. Hass, A. N. Marchenkov, et al., Science 312 , 1191 (2006).
- 47 C. Cocchi, A. Ruini, D. Prezzi, M. J. Caldas, and E. Molinari, J. Phys. Chem. C 115 , 2969 (2011).
- 48 X. Zhu, W. Yan, N. A. Mortensen, and S. Xiao, Opt. Express 21 , 3486 (2013). | null | [
"Oleg L. Berman",
"Roman Ya. Kezerashvili",
"German V. Kolmakov"
] | 2014-08-01T20:42:47+00:00 | 2014-08-01T20:42:47+00:00 | [
"cond-mat.mes-hall"
] | Polariton-based optical switch | Based on the studies of propagation of an exciton-polariton condensate in a
patterned optical microcavity with an embedded graphene layer, we propose a
design of a Y-shaped electrically-controlled optical switch. The polaritons are
driven by a time-independent force due to the microcavity wedge shape and by a
time-dependent drag force owing to the interaction of excitons in graphene and
the electric current running in a neighboring quantum well. It is demonstrated
that one can control the polariton flow direction by changing the direction of
the electric current. The simulations also show that an external electric field
normal to the microcavity plane can be utilized as an additional parameter that
controls the propagation of the signals in the switch. By considering the
transient dynamics of the polariton condensate, we estimate the response speed
of the switch. Finally, we propose a design of the polariton switch in a flat
microcavity based on the geometrically identical Y-shaped quantum well and
graphene pattern where the polariton flow is only induced by the drag force. |
1408.0308v1 | ## Opinion Dynamics and Price Formation: a Nonlinear Network Model
Draft - please do not quote or cite without permission
First version: November 2013. This version: August 5, 2014
Marco D'Errico 1 Silvana Stefani Giovanni Zambruno Department of Statistics and Quantitative Methods, University of Milano - Bicocca
Gulnur Muradoglu School of Business and Management, Queen Mary University of London
## Abstract
Opinions and beliefs determine the evolution of social systems. This is of particular interest in finance, as the increasing complexity of financial systems is coupled with information overload. Opinion formation, therefore, is not always the result of optimal information processing. On the contrary, agents are boundedly rational and naturally tend to observe and imitate others in order to gain further insights. Hence, a certain degree of interaction, which can be envisioned as a network, occurs within the system. Opinions, the interaction network and prices in financial markets are then heavily intertwined and influence one another. We build on previous contributions on adaptive systems, where agents have hetereogenous beliefs, and introduce a dynamic confidence network that captures the interaction and shapes the opinion patterns. The analytical framework we adopt for modeling the interaction is rooted in the opinion dynamics problem. This will allow us to introduce a nonlinear model where the confidence network, opinion dynamics and price formation coevolve in time. A key aspect of the model is the classification of agents according to their topological role in the network, therefore showing that topology matters in determining how of opinions and prices will coevolve. We illustrate the dynamics via simulations, discussing the stylized facts in finance that the model is able to capture. Last, we propose an empirical validation and calibration scheme that makes use of social network data.
## Keywords
opinion dynamics, network, price formation
1 Corresponding author: [email protected]
## Contents
| 1 Introduction | 1 Introduction | 1 Introduction | 3 |
|------------------|--------------------------------------------|------------------------------------------------------|-----|
| 2 | Opinions and prices: a network model | Opinions and prices: a network model | 8 |
| | 2.1 | Introducing the network interaction . . . . . . . | 9 |
| | | 2.1.1 Bounded confidence model . . . . . . . . | 11 |
| | | 2.1.2 Price adaptive strategy . . . . . . . . . . | 12 |
| | | 2.1.3 Fundamental benchmark . . . . . . . . . . | 12 |
| | 2.2 | Combinations of strategies . . . . . . . . . . . . . | 13 |
| 3 | Simulation results and discussion | Simulation results and discussion | 14 |
| | 3.1 Opinion shifts . . | . . . . . . . . . . . . . . . . . . | 17 |
| 4 | Empirical validation and model calibration | Empirical validation and model calibration | 21 |
| 5 | Conclusions and future research | Conclusions and future research | 23 |
| A | Materials and methods | Materials and methods | 25 |
| | A.1 | Graphs and networks . . . . . . . . . . . . . . . . | 25 |
| | | A.1.1 Determining the Gantmacher form . . . . | 26 |
| | A.2 | Opinion dynamics: the mathematical background | 28 |
| | | A.2.1 The homogeneous opinion dynamics process | 28 |
| | | A.2.2 Inhomogeneous opinion dynamics process | 30 |
| | | A.2.3 Convergence in the inhomogenous case . . | 31 |
| | | A.2.4 Opinion dynamics under bounded confidence | 31 |
## 1 Introduction
If men define situations as real, they are real in their consequences W. I. Thomas and D. S. Thomas (1928, p. 572)
.
Social systems are endogenously driven by how agents perceive, think of and, ultimately, model the system itself. Therefore, the interpretation of a particular situation yielding to a subsequent action often affects the situation. However, the interpretation is almost never objective and carries a certain degree of subjectivity stemming from individual choices and behaviours. This naturally reflects on the aggregate behaviour of the system. Whether accurate or not, individual views, opinions and beliefs have consequences with a real and tangible impact on the system. The quote reported above is also known as Thomas theorem . 2 This 'theorem' refers to a broad situation where opinions determine reality and, in turn, reality shapes opinions.
This problem can be envisioned within the framework of rationality, arguably one the most debated concepts in both social and economic theory. A 'rational' - in any acceptation of the word - agent would never act in such a way that a potential undesired consequence could originate from the wrong definition of the situation. This idea, in the light of the current financial crisis, is even more challenged within the standard rationality paradigm in finance, i.e. that investors make optimal use of all available information.
The traditional financial paradigm is built upon the assumption of rationality of agents and markets that are perfectly efficient (Fama, 1970). In the Efficient Market Hypothesis (EMH), asset prices at any time reflect all the information available to the agents in a financial system. However, recent studies have shown that investors' behaviour often deviates from the rationality implied by the Efficient Market Hypothesis. The idea that investors' psychology plays a fundamental role in determining price dynamics has led to a new field called behavioural finance , where the behaviour of agents in a financial system departs from the assumption of rationality.
Behavioral aspects can explain a wide range of financial phenomena that cannot be interpreted within the EMH framework (e.g. momentum trading, trend extrapolation, noise trading, overreaction, overshooting, contrarian strategies). Some useful references include DeBondt and Thaler (1985), Hong and Stein (1999, 2003), and review papers such as Hirschleifer (2001), and Barberis and Thaler (2003). Behavioural finance is currently expanding into different directions (DeBondt et al. , 2008).
The debate is enriched by the fact that financial systems are witnessing the emergence of very complex structures. Complexity definitely plays a key role here: the crisis has unequivocally shown that - within the complex financial environment - it is unrealistic to assume that, even in presence of full information, agents will have the cognitive capabilities to process and make optimal decisions from the available informative set.
There is, in fact, a strong consensus that financial systems show an increasing degree of complexity on at least two levels. The first level relates to the rapid expansion in the number of financial products traded. The intrinsic complexity embedded in the financial evaluation of such
2 For further details, see Robert K. Merton's 1995 historical account on the Thomas theorem where the author explains why the statement became known as a theorem .
products has remarkable consequences in terms of systemic risk (Caccioli et al . 2009; Brock & Hommes, 2008). The second level pertains to the complex network structure of financial markets (Haldane and May, 2011; Battiston et al. 2013).
On both levels, complexity in financial markets implicates that a huge amount of information needs to be gathered, analyzed and understood before even attempting to make any decision, let alone an optimal one. In this increasingly complex environment, hence, the view that investors are fully rational implies that they must possess unrealistic cognitive skills.
The key aspect here is precisely linked to the limited cognitive capabilities of human beings. Prompted by such limitations, Herbert Simon (1955) introduced the idea that agents have 'limited knowledge and ability'. This approach led to the introduction of the concept of bounded rationality , which entails the viewpoint that agents would rather adopt a 'satisficing' strategy as opposed to a fully optimal decision-making strategy. Simon refers to environmental conditions leading to this particular behaviour. More specifically, Bawden and Robinson (2008) refer to information overload as a core characteristic yielding to boundedly rational behaviour. Agnew and Szykman (2005) have explored this concept in the context of behavioural finance, stressing that informative overload prevents investors from formulating and deciding upon the correct strategy to pursue and might result in more simplistic investing strategies.
The complexity conundrum implies that some investors might look and imitate the behaviour of those investors who are believed to be better informed, sometimes ignoring fundamentals or private information. Devenow and Welch (1996) refer to such situation as rational herding and, interestingly within the context we intend to model, stress the importance of informational (cascade) learning . Crucial for our analysis, these works show that investors might consider 'pooling' other investors' opinions for a number of reasons, including the fact that they are unsure about the information they possess, for strategic purposes or even because they show a particular aversion for large losses (Kahneman & Tversky, 1984).
Sobel (2000) wonders whether imitation can be attributed to a rational choice and emphasises the separation between 'individual' and 'social' learning. In this light, agents are 'informationally linked, so that the actions and payoffs of one agent provide information about the state of the world' to other agents. Sobel also raises the issue of whether individual na¨ ıveness of agents in large groups translates into the system being 'smart' in the aggregate.
This is the object of the study of a seminal paper by Galton (1907) about the so-called 'wisdom of the crowd' (Surowiecki, 2005) enigma. Galton found that the average guess, among those expressed by a group of people upon the butchered weight of an ox, was actually an extremely accurate way to gauge the actual weight.
Are then aggregate predictions always so accurate ? If we observe the aggregate behaviour in certain critical situations - and, in financial system, also in non-critical ones - one should rather speak of 'madness of crowds' (cit., Mackay, 1841).
Golub and Jackson (2010) tackled the problem of na¨ ıve learning in connection with the 'wisdom of crowds' in a social network framework. In particular, they determine conditions when beliefs converge to a true value and what obstacles might hamper the convergence process. Lorenz et al. (2011) propose an empirical validation of this problem.
This work and, later, the contributions of Hegselmann and Krause (2005) and Lorenz (2005a, 2005b, 2006) provide with detailed analytical conditions for the convergence towards the consolidation or the fragmentation of opinion patters. The opinion dynamics modeling framework used
in these works is routed within a stream of literature where the key motivation is the search of a common opinion, or consensus . Perhaps, the most well known formulation is the one proposed by DeGroot (1974). However, several authors had previously tackled this problem (notably French, 1956; Harary, 1959). Interestingly for our problem, the basic idea beneath DeGroot's approach is that agents 'pool' the opinions of neighbours and revise their own opinion by a simple arithmetic average of these opinions.
This modelling scheme constitutes the basis of our work and motivates Golub and Jackson's approach within a social network framework, as it implies a pairwise interaction among agents. The interaction is represented by the confidence weights agents place onto one another.
Later contributions (namely the already mentioned Hegselmann and Krause, 2005; Lorenz, 2007) extend this approach, by introducing the idea that the confidence network depends on the opinions and, in turn, opinions are updated according to the network. Nonlinearity arises precisely in that opinions influence the confidence network and vice versa. This will be the basis of the nonlinear model we propose. The scheme proposed Hegselmann and Krause (2005) is based on an update rule according to which, at each time, every agent select a subset of agents whose opinions are the closest, within a specified threshold.
For this reason, this opinion dynamics modelling approach is referred to as a bounded confidence scheme. We can therefore state that bounded confidence is one of the many degrees of freedom (Barberis and Thaler, 2003, p. 64) in behavioural finance that can be associated to bounded rationality . The conceptual link between the change in confidencewith bounded rationality is then made explicit in this class of models.
The basic argument here is that the bounded rationality of agents in the marketplace naturally leads to a certain level of interaction among agents. The mutual interaction between agents is a key aspect within this context and will be the main subject of our study.
Several authors have investigated this problem under different perspectives and modelling approaches. Lux and Marchesi (1998) show that scaling power law properties and time-dependent volatily may depend on this interaction. In view of this problem Cont and Bouchaud (2000) model herd and imitative behaviour with a random interaction structure among clusters of agents, finding a potential explanation for heavy tails in the distribution of returns.
Analogously to the bounded confidence model, agents' interaction can be the result of a specific selection . As such, a model featuring an adaptive financial system, is more suitable at explaining these behavioural limitations (Hommes and Wagener, 2008). In particular, an emergent stream of literature deals with models where heterogenous agents interact and determine asset price dynamics. Among the several contributions in this field, those of Brock and Hommes (1998), Lux (1998), Gaunersdorfer and Hommes (2007) and Chiarella et al. (2007) are of particular interest for our work.
The introduction of heterogeneous agents finds its root in overcoming the typical limitations embedded in models with representative agents: as a matter of fact, in the real world, agents naturally have heterogeneous beliefs about asset dynamics. In this context, a logical separation between two categories of agents arises: fundamentalists (i.e. agents who believe that prices will eventually converge towards a fundamental value, simply given by the discounted sum of future dividends) and chartists . Under the EMH, the latter category would be irrational as they would simply lose money in the long run. However, there is evidence that these agents might be able to obtain higher-than-average returns.
Consistently with the approach we will follow, one of the basic tenets of this class of models is that agents want to exploit temporary deviations from a fundamental price (based on rational expectations) and this might lead agents to adopt short-period strategies that depart from rationality (see Brock and Hommes, 1998 for a thorough discussion).
These Agent Based Models (ABMs) often feature an adaptive dynamics (adaptive beliefs) based on the success of previous strategies (either in simple terms of price forecast or in terms of accumulated profits) where the interaction in these models stems from the agents' capability of observing other agents' strategies and, by switching strategy, imitating the most successful ones. However, the number of agent types presented in these model is usually limited and the interaction occurs via price realizations.
We propose to overcome these limitations by modelling agents' interaction via a dynamic network approach, associating the network process to the price dynamics. The pricing equation is based on opinions (Section 2) formed upon a certain rule, therefore introducing the nonlinear network opinion dynamics approach within this class of models.
In our model, agents interact along three intertwined dimensions: i) a network, based on the level of confidence that agents place onto one another, ii) a more classical interaction via the prices, and iii) comparing their opinions.
An important caveat should be made clear. In the model we propose, we are not strictly interested in whether and how the system reaches a consensus of opinions, as opinions in our model can vary significantly in time, although this represents surely an interesting case and will be explored in the case of reversion towards a fundamental financial value. We are rather interested in finding a way to characterize agents according to their role in the nonlinear opinion/price/network process, according to their topological role within the network.
The model captures also situations when changes in the topology are coupled with significant shifts in opinions. This is particularly important within the increasing adoption of high frequency trading. This aspect will be further explored in the following motivating example, where we draw a parallel with the aforementioned Thomas theorem.
## A motivating example
On April 23, 2013, at 1:07 PM the US stock market crashed by about 1% in a couple of minutes, recovering soon afterwards in about the same time. Fig. 1 shows a snapshot of the DJ 30 industrial index one minute intraday time series during those minutes (see the sudden and dramatic drop in the index visibile towards the end of the time series snapshot). What happened during those few minutes?
At 1:07 p.m., Associated Press tweeted that the US president had been reported injured in an explosion at the White House (see Fig. 2). The tweet was fake, probably due to a hacker that had temporary access to AP's account. The tweet was immediately retweeted a number of times (about 4000 in less than a minute, and even more in the next minutes).
This two-minute storm, or 'flash-crash' - which led to a ∼ 1% drop and a loss of $130 billion in stock value - cannot be explained by only one tweet: indirect (network) effects have to be taken into account, probably also due - as several observers have pointed out - to the increasing adoption of high-frequency automated trading strategies.
Figure 1: April 23 DJ 30 Industrial crash - 1 minute intraday.
## Apr 28, 2013 1-07 PM
## Tweet appears
Fake tweet appears on the AP Twitter account, which has more than 1.9 million followers: "Breaking: Two Explosions in the White House and Barack Obama is injured:'
Figure 2: AP tweet reporting explosions at the White House.
We believe that, in addition to this, AP's authority as an independent and reliable agency (coupled with the 'check mark' that certifies a verified Twitter account), led investors (indepedently from their strategy) to immediately trust the consequences of such a tweet, without verifying its truthfulness.
Whether or not the fake tweet was intentionally meant to destabilise the stock market is still an open question. Doubtless, it shows that the financial system is intrinsically vulnerable to opinion shifts , regardless of the truthfulness of the underlying fact. The vulnerability of financial markets to such threats represents an open also for regulators as it leaves observers 'speculating once more about their vulnerability to breaking news in the age of social media'. 3
Wehave now come full circle. Referring, once more, to the Thomas theorem, the consequences were real , although the actual 'situation' was not. 4
## 2 Opinions and prices: a network model
Consider a set V of i = 1 , . . . , n agents, one risky asset and one risk - free asset (with interest rate r ). The price dynamics is denoted by ( p t ( )) t ∈ N . Let z i ( ) t be the quantity (number of shares) of the risky asset purchased by agent i at time t . Finally, let ( y t ( )) t ∈ N be the dividend at t (which will be assumed to be i.i.d. throughout this chapter). The wealth dynamical process for each agent can be described by the following equation (expressed in vector form):
$$\begin{array} {$$
In order to capture the opinion dynamics, we will denote the opinion of agent i about the price and the dividend innovation with the following expected value:
$$x _ { i } ( t ) = \mathbb { E } _ { i } ( p ($$
The opinion at time t is conditioned to the information set available at time t -1. Opinions are heterogenous for what regards the first moment of the distribution. The assumption can be extended to higher moments of the distribution but this case will not be explored here. The opinion about the variance of wealth is constant and equal for all agents, in line with Hommes and Wagener (2005, review paper) and Chiarella et al. . (2007). The opinion (or, in the originl context, belief ) that agent i has about her wealth can be then described as follows:
$$\mathbb { E } _ { i } ( w _ { i } ( t + 1 ) )$$
Agents optimize their wealth in mean/variance, with the following utility function:
3 AP Twitter hack causes panic on Wall Street and sends Dow plunging , The Guardian, April 23, 2013. Online version available at: http://www.theguardian.com/business/2013/apr/23/ ap-tweet-hack-wall-street-freefall
4 Many scholars have pointed out that self - fulfilling prophecies , i.e. statements or predictions that cause themselves to be true, represent a conundrum in the theory of social systems. R. Merton's Social Theory and Social Structure (1968) provides an extensive philosophical and historical portrayal of this fascinating concept.
$$u _ { i } ( w ( t ) ) & = - e ^ { - a _ { i$$
where a i is the Constant Absolute Risk Aversion (CARA) coefficient. This approach is consisent with the literature we refer to (Hommes & Wagener, Chiarella et al .). The wealthmaximisation problem at t can be then expressed as follows:
$$\max _ { z _ { i } ( t ) } \ \left \{ \mathbb {$$
thus leading to
$$z _ { i } ( t ) = \frac { 1 } { a \sigma ^ { 2$$
as the optimal demand of the risky asset for agent i at time t . Hence, the optimal demand (number of shares) of agent i at time t proportional to the difference between the opinion on the risky asset price and dividend at t +1 the capitalized price at t , and inversely proportional to the absolute risk aversion coefficient a and, obviously, the variance.
We assume that there is a limited outside supply per agent , that will be denoted by z s (with nz s obviously being the overal outside supply). We have a convenient way to obtain the equilibrium price of the risky asset via the following equilibrium:
$$\frac { 1 } { n } \sum _ { i = 1 } ^ { N } \frac$$
which finally leads to the following pricing equation :
$$( 1 + r ) p ( t ) & = \underbrace { \frac { 1 } { n$$
As previously mentioned, we will assume that the dividend sequence ( y t ( )) t =1 2 , ... is i.i.d., with variance σ 2 and will be the only source of randomness in the model. If σ = 0, then price is deterministic and given only by the discounted average opinion.
## 2.1 Introducing the network interaction
We now describe the opinion dynamics x ( ) t and the price dynamics p t ( ) for the risky asset. Hommes and Wagener (2005) and, in more detail, Gaunersdorfer and Hommes (2007) assume an interaction based on the success of the trading strategy basing it on a fitness function .
Our contribution in this work lies in adopting an explicit interaction strategy, where agents learn from others according to specific rule. The interaction among agents, capture by a confidence network , whose topology will be crucial in our model. Agents are modeled as vertices in the network. The network is time-dependent and evolves according to specific update rules that
capture different situations in the financial system. The update rule modifies the underlying interaction network topology among agents.
Agents are boundedly rational and DeGroot's approach is particularly suitable in our context because, 'incorporate indirect information in a boundedly rational way' (cit. Golub and Jackson [33]).
Hence, we will assume a pairwise interaction with a dynamics of the kind described within the consensus problem of DeGroot (1974). This issue was later tackled -for example- by Berger (1981) and, in a more general opinion dynamics framework, recently developed by Hegselmann and Krause (2005) and Lorenz (2007). The Methods Section of this paper provides with a detailed and rigorous discussion of these models and the mathematical background that will be needed in develop our model in rest of this Section. In the following paragraphs, we will briefly outline the main aspects and conclusions.
As mentioned in the Introduction, the basic problem of DeGroot was to develop a model to find a consensus (i.e. a shared opinion) within a set of agents V who revise their opinions by averaging (or ''pooling') other agents' opinions. In particular, for each pair of agents ( i, j ), agent i assings a weight a ij ∈ [0 , 1] to the opinion of agent j 'to accommodate the information and expertise, the opinions and judgements, of the rest of the group'.
In this original statement of the problem, it is assumed that the linear combination is convex (i.e. ∑ j =1 a ij = 1 and does not vary over time. As such, the model borrows some analytical results from Markov chain theory in order to find conditions for the convergence to a common opinion. A later, more general approach is the one of Hegselmann and Krause (2005), who introduce the bounded confidence opinion dynamics scheme: the weighing structure is timedependent.
A key aspect in our model is the topological classification of agents into essential and inessential classes. Intuitively, the set of agents can be split into subsets of strongly connected components (SCC). Agents within the same SCC communicate , in that it is always possible to find a path from and to any pair of agents. In other words, within a SCC, agents have direct or indirect positive confidence weight in any other agents within the same SCC. When no agent belonging to a specific SCC puts a positive confidence weight on an agent belonging to another SCC, then the first SCC corresponds to an essential class. Vice versa, when one or more agents belonging to a specific SCC put a positive weight onto an agent belonging to another SCC, then the class is said to be inessential . A rigourous formulation of this concept is provided in the Methods Section.
The relationship between essential and inessential classes, as pointed out by Berger (1981), will determine whether a common or fragmented opinion will be reached and how. Opinions of agents in the essential classes are considered by other inessential agents, but not vice versa. Predominant opinions and hence, prices, will then be determined by a linear combination of the opinions of the essential agents. Loosely speaking, essential agents are the opinion leaders whereas inessential agents are opinion followers .
Our setup, however, allows that essential and inessential classes of agents may vary over time, according to the update rule. In other words, an agent can change her topological role in time. Since we are studying the model by means of simulations, we then need an algorithm to classify essential and inessential agents and find the class they belong to at each t .
The opinion revision process is captured by the following equation:
$$\begin{array} {$$
which, coupled with Equation 4 will give the price at each t .
The process that we will assume in this model involves a change in the confidence network at each t , as a convex linear combination of the previous confidence matrix and an update matrix C ( ): t
$$a _ { i j } ( t ) = \alpha _ { i } \ c _ { i j$$
where α ∈ [0 , 1] is an update propensity parameter. This parameter is key in our model because, as it will be shown in the simulations, the behaviour of the system depends nontrivially on it. The element c ij ( ) t of C ( ) t represents the update (revision) of the confidence weight that i has towards j , and is determined according to the following rule:
$$c _ { i j } ( t ) \coloneqq \begin{cases} \ \frac { 1 } { \# I ( i, \mathbf x ( t - 1 ), p ( t - 1 ) ) } & \text{if $j\in I(i,\mathbf x(t-1), p(t-1))$} \\ 0 & \text{otherwise} \end{cases}$$
It is immediate to prove that, according to this rule, C ( ) is row stochastic t ∀ t and, therefore, also A ( ) is also row - stochastic t ∀ t .
The model is non - linear because of the interaction among the opinions, prices and the confidence network. In the context of consensus/opinion dynamics, Hegselmann and Krause (2005) also refer to this kind of model as non-linear and point out that 'the most difficult type of model occurs if the weights depend on opinions itself because then the model turns from a linear one to a non-linear one'.
Differently from the standard approach, we do not focus on specific trader types . Rather, agents are classified according to the role they play in the topological structure, hence allowing to capture their beahviour along a continuum of possible strategies, coupled with the evolving topology of interactions. The choice of the set I ( i, x ( t -1) , p ( t -1)) for each agent i will lead to different types of models and will be now detailed.
## 2.1.1 Bounded confidence model
In the first case, we will adopt a similar approach to the one in Hegselmann and Krause (2005) and reported in Equation 13 in the Methods section. The authors refer to it as bounded confidence in that a threshold exists when determining to whom an agent puts a confidence weight. We also label this model as bounded confidence (BC). The adaptive update matrix C ( ) is: t
$$I ( i, \mathbf x ( t - 1 ), p ( t - 1 ) ) = \{ j \text{ s.t. } | x _ { i } ( t - 1 ) - x _ { j } ( t - 1 ) | \leq \epsilon _ { i } \}$$
In this scenario, there is no need to look at fundamentals in order to understand the price dynamics. Prices will be driven only by the opinions and the opinions will be pooled according the bounded confidence non - linear model. The non - linearity in the model will lead to opinion polarization and fragmentation, thus the average in Eq. 4 can increase or decrease dramatically
according to such level of polarisation. It is important to notice that { } i is included in the set I ( i, x ( t -1) , p ( t -1)) by construction.
## 2.1.2 Price adaptive strategy
The idea in this setting is that those agents whose opinions are the closest to the actual price at time t (price adaptive, PA) will be followed, in the next time step, by other agents. In an 'adaptive' perspective, these agents are thus those capable of understanding the nontrivial impact of both opinions on price and the impact of dividend innovations.
In this adaptive setting, agents revise the confidence matrix as in Equation 6, but the set I ( i, x ( t -1) , p ( t -1)) is now given by:
$$I ( i, \mathbf x ( t - 1 ), p ( t - 1 ) ) & = \{ i \} \cup \left \{ j \text{ s.t.} \ \left | \frac { p ( t - 1 ) - x _ { i } ( t - 1 ) } { p ( t - 1 ) } \right | \leq \epsilon _ { i } \right \} \quad \quad ( 8 ) \\ \mathfrak { r } _ { 1 } \quad, \quad \mathfrak { r } _ { 2 } \cdot \quad \mathfrak { r } _ { 3 } \cdot \quad \mathfrak { r } _ { 4 } \cdot \quad \mathfrak { r } _ { 5 } \cdot \quad \mathfrak { r } _ { 6 } \cdot \quad \mathfrak { r } _ { 7 } \cdot \quad \mathfrak { r } _ { 8 } \cdot \quad \mathfrak { r } _ { 9 } \cdot \quad \mathfrak { r } _ { 9 } \cdot \quad \mathfrak { r } _ { 9 } \cdot \quad \mathfrak { r } _ { 9 } \cdot \quad \mathfrak { r } _ { 9 } \cdot \quad \mathfrak { r } _ { 9 } \cdot \quad \mathfrak { r } _ { 9 } \cdot \quad \mathfrak { r } _ { 9 } \cdot \quad \mathfrak { r } _ { 9 } \cdot \quad \mathfrak { r } _ { 9 } \cdot \quad \mathfrak { r } _ { 9 } \cdot \quad \mathfrak { r } _ { 9 } \cdot \quad \mathfrak { r } _ { 9 } \cdot \quad \cdot \quad \mathfrak { r } _ { 9 } \cdot \quad \mathfrak { r } _ { 9 } \cdot \quad \mathfrak { r } _ { 9 } \cdot \quad \mathfrak { r } _ { 9 } \cdot \quad \mathfrak { r } _ { 9 } \cdot \quad \mathfrak { r } _ { 9 } \cdot \quad \mathfrak { r } _ { 9 } \cdot \quad \mathfrak { r } _ { 9 } \cdot \quad \mathfrak { r } _ { 9 } \cdot \quad \mathfrak { r } _ { 9 } \cdot \quad \mathfrak { r } _ { 9 } \cdot \quad \mathfrak { r } _ { 9 } \cdot \quad \mathiota _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot 0 \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { l } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cdot \quad \mathfrak { r } _ { 0 } \cd.$$
In other words, in this case, c ij ( ) t represents an adaptive strategy where agents tend to follow more agents that had close opinions to the actual realized price. If glyph[epsilon1] i is a nonzero constant ∀ i ∈ V , then C ( ) is a rank one matrix at each t t . Since the price is driven both by the opinion dynamics and an exogenous divided innovation, this model reflects a situation where agents want to follow other agents that have proven to be successful in the past.
The reason why we insert the set { } i in the set I in the set 8 is because we want to have a non-empty set and also allow the matrix A ( ) t to have a positive diagonal (also in the case α = 1).
## 2.1.3 Fundamental benchmark
The reversion of asset prices towards a fundamental value is a long-debated issue in the financial literature. 5 When all agents have identical long - term expectations (opinions) on the price, i.e. x i ( ) = t c, ∀ i ∈ V , we can assume x ( ) to be the analogous of a consensus limit vector where all t agents belong to the same essential class. The pricing equation 4 can be then rewritten as:
$$( 1 + r ) p ( t ) = c - a \sigma ^ { 2 } z ^ { s } \\. \cdot \cdot \cdot \cdot \cdot \cdot \cdot$$
We can also define a fundamental price given by the discounted sum of all expected future dividends:
$$p ^ { * } ( t ) = \sum _ { k = 1 } ^ { \infty } \frac { \mathbb { E } ( y _ { t + k }, t ) - a \sigma ^ { 2 } z ^ { 2 } } { ( 1 + r ) ^ { k } }$$
which can be set in an analogous way of a perpetuity with interest r . An interesting aspect is that, at a specific time t , should the confidence matrices be fixed, one can assume that there are g different asymptotic opinions, each of which can be associated to an essential class of the
5 References include Lo and Mackinlay (1988), who develop a specification test and claim that 'prices do not follow random walks'; Fama and French (1988); and Poterba and Summers (1988) who focus explicity on mean reversion in asset pricess, tackled by Stefani et al. (2010) in commodity trading.
kind described in the Methods section. When ( y t ( )) t =1 2 , ,... , then E [ y t ] = ¯ and the fundamental y price is constant over t :
$$p ^ { * } = \sum _ { k = 1 } ^ { \infty } \frac { \bar { y } - a \sigma ^ { 2 } z ^ { s } } { ( 1 + r ) ^ { k } } = \frac { \bar { y } - a \sigma ^ { 2 } z ^ { s } } { r }$$
We will show, by simulations, that in this model based on fundamentals (FB), opinion shifts in the essential classes will lead to significant changes in prices.
In this case, we assume the set I to be determined as follows:
$$I ( i, \mathbf x ( t - 1 ), p ( t - 1 ) ) = \{ i \} \cup \left \{ j \text{ s.t.} \ \left | \frac { p ^ { * } - x _ { i } ( t - 1 ) } { p ^ { * } } \right | \leq \epsilon _ { i } \right \}$$
This implies that all agents will see deviations from the fundamental as temporary and will tend to trust agents whose opinion does not deviate significantly from the fundamental. In behavioral terms, agents want to 'find a confirmation' from other agents that the fundamental is the correct long-term benchmark. However, small and temporary deviations are perceived as reasonable and agents will still try to maximize their short - term wealth (myopic maximisation) by exploiting them.
There is one general case when the fundamental price is achieved in 'aggregate' at time t , i.e. when the average opinion on the price is equal to the fundamental ( 1 n ∑ n i =1 x i ( ) = t p ∗ ( )). t As a particular case, obviously, when at t , all the agents' opinions are all equal to the fundamental at time t .
The first, more general, case could occur in a number of situations. For example, even in presence of high variability in the opinions, the system can reach the fundamental price anyway, if the average opinion coincides with it.
## 2.2 Combinations of strategies
The flexibility of the model we propose allows to have different strategies (BC, PA, FB) for each agent; different trader types (e.g. fundamentalists, chartists, contrarians, etc.) can be modeled within this framework. Among the many possible combinations, an interesting analysis would be to assume different strategies for essential and inessential agents assuming, for instance, two interesting scenarios that will be the subject of future research and will not be explored here.
In the first scenario, essential (opinion drivers) agents' behaviour would be encapsulated within a Bounded Confidence model with a very low update propensity, whereas inessential agents (opinion followers) would be following a Price Adaptive strategy with a very high update propensity. This would lead to a situation where a few leaders drive the opinion dynamics and many followers adjust very quickly to those agents whose opinion was the most accurate at each time step.
The second scenario is analogous to the first, with a significant difference. The essential agents are split into two categories: opinion leaders who follow a fundamentalist strategy with a very low update propensity and other 'impulsive' essential agents who follow a price adaptive
strategy. In this case, the essential agents in the second category would enhance little fluctuations when price temporarily depart from the fundamentals.
Although not explicitly explored here, the model potentially allows to capture agents who explicitly want to be followed (thus determining the market), acting as opinion leaders. This obviously introduces the problem of the controllability of the confidence network. The framework might then be useful also for regulators.
## 3 Simulation results and discussion
In this section, we show and discuss simulation results on the dynamics described by Equations 4, 6 according to the three models (BC, Equation 7; PA 8 and FB, 9). We start with the analysis of the price dynamics and later provide some descriptive statistics on the distribution of returns. Given the number of parameters in the model, we will make the following simplifying assumptions:
- 1. The variance on the dividend innovations is constant ∀ t and ∀ i (i.e. on the innovation opinions) and equal to σ ;
- 2. E it ( t t +1 ) = ¯, y ∀ i, t ;
- 3. in any case, glyph[epsilon1] i = , glyph[epsilon1] ∀ i .
The first part of the simulation analysis (Figures 3, 5 and 4) reports a representation of the price p t ( ) in time, obtained from the pricing equation 4, by averaging out 1000 realizations of the process in Equation 6 from the same starting opinion profile x (0), that it is distributed as a log - normal with mean 3. The number of agents is set to n = 100.
In particular, figures 3, 5, 4 show simulations for the price dynamics for the models BC, FB and PA respectively. Some descriptive statistics are shown in Tables 2a - 2f for the price returns r t ( ) = p t /p t ( ) ( -1) -1. These statistics and the price dynamics reflect some well-known stylized facts in financial markets (e.g., clustered volatility, fat tails, asymmetric returns). In particular, although the return distributions are slightly asymmetric, we used the excess kurtosis as a way to compare tail distribution with those of a normal distribution one can see that fat tails arise in the BC non - linear model (where polarisation occurs) when the update propensity parameter α is very high. This means that the quicker the fragmentation, the quicker non linear phenomena arise and fat tails occur. The PA model, on the contrary, produces no fat tails, as the result of a strong reversion towards the fundamental.
Figure 3: BC model for different values of α and σ .
Figure 4: PA model for different values of α and σ .
Figure 5: FB model for different values of α and σ . Fundamental value is the horizontal red line.
Table 1: Descriptive statistics for the price returns p t ( ) over 1000 simulations for the three models and different values of the update propensity parameter α and σ .
| | σ = 0 . 1 | σ = 0 . 5 | σ = 1 |
|-----------|-------------|-------------|---------|
| α = 0 . 1 | 0.2 | -0.03 | 0.02 |
| α = 0 . 5 | -0.21 | -0.16 | 0.25 |
| α = 0 . 9 | 0.12 | 0.05 | -0.02 |
| | σ = 0 . 1 | σ = 0 . 5 | σ = 1 |
|-----------|-------------|-------------|---------|
| α = 0 . 1 | -0.38 | -0.45 | -0.18 |
| α = 0 . 5 | -0 | 0.42 | 0 |
| α = 0 . 9 | 0.29 | -0.08 | 0.59 |
(a) BC model - skewness
| | σ = 0 . 1 | σ = 0 . 5 | σ = 1 |
|-----------|-------------|-------------|---------|
| α = 0 . 1 | 0.14 | -0.02 | -0.09 |
| α = 0 . 5 | 0.09 | -0.16 | -0.13 |
| α = 0 . 9 | 0.08 | 0.11 | 0.12 |
(c) PA model - skewness
| | σ = 0 . 1 | σ = 0 . 5 | σ = 1 |
|-----------|-------------|-------------|---------|
| α = 0 . 1 | 0.05 | -0.18 | 0.12 |
| α = 0 . 5 | -0.02 | 0.19 | -0.17 |
| α = 0 . 9 | 0.18 | -0 | 0.08 |
(e) FB model - skewness
(b) BC - (excess) kurtosis
| | σ = 0 . 1 | σ = 0 . 5 | σ = 1 |
|-----------|-------------|-------------|---------|
| α = 0 . 1 | 0.1 | -0.34 | -0.14 |
| α = 0 . 5 | 0.66 | -0.22 | -0.01 |
| α = 0 . 9 | -0.14 | -0.12 | -0.26 |
(d) PA model - (excess) kurtosis
| | σ = 0 . 1 | σ = 0 . 5 | σ = 1 |
|-----------|-------------|-------------|---------|
| α = 0 . 1 | -0.09 | 0.23 | -0.04 |
| α = 0 . 5 | -0.21 | 0.4 | 0.2 |
| α = 0 . 9 | -0.22 | 0.59 | -0.39 |
(f) FB model - (excess) kurtosis
## 3.1 Opinion shifts
This specific simulation setup aims at explaining sudden drops in price followed by quick recoveries in presence of herding behaviour. In particular, we will try to assess whether the drop in the DJ30 following the fake tweet by AP can be explained in terms of network effects. To do so, we assume a 'sudden' exogenous change at t = 50 for either some essential classes or the union set of all inessential classes. The basic idea is to assess whether changes in the opinion profiles of essential classes might have a different impact with respect to change in the opinion profiles of inessential classes, regardless of convergence . In fact, the conditions outlined in the Methods Section show that essential classes play a role in determining the value towards which opinions converge, but very little can be said, from an analytical perspective, about short-term behaviour. In particular, we will see that changes in inessential classes do have an impact on slowing the recovery.
Figure 6 provides a graphical illustration of the dynamics we assume in this simulation setup. The Figure reports a topology with one essential class (the node on the left hand side), and three inessential classes (in the center and on the right hand side). An opinion shift occurs at a specific time t in an essential class composed of only one agent, producing a limited impact on the price. At t +1, one agents for each inessential class starts witnessing the shift and updates her opinion, the impact on the price therefore starts to increase, now involving four agents. At t +2, agents in two inessential classes update their opinions. The shift has now had an impact on eight agents, with a more significant impact on the price. n price but it will persist less. Should a revision of the previous shift occur in the essential agent, this would take two time steps to lead the price to a full recovery.
Figure 7 shows that, in the BC model, when drops occur in the essential classes, they tend to persist for low values of σ : in fact, the bounded nature of the model will provide a polarisation versus a low market scenario rather than a recovery. This is not the case when the drop occurs in the inessential classes (Fig. 8), where we observe a milder impact. Figures 9 and 10 shows analogous findings for the PA model. The most interesting case is the FB model, where it is important to understand whether a return to the fundamental benchmark will occur and the time to recovery.
Figure 11 and 12 outline an important fact: drops happen and persist when opinion shifts occur in both essential and inessential classes, then recovering quite rapidly. However, the drops are much more pronounced in the essential class case and for lower values of σ , thus showing that a higher variability in dividend innovations might lead to a more stable price dynamics w.r.t. sudden opinion shifts. The role of the update propensity parameter α is important for higher values of σ in that lower propensity implies a lower degree of abandoment of previously formed opinions on the deviations from the fundamental, hence making recovery faster.
Figure 6: Illustration of the opinion shift in an essential class and its reverberations on the inessential classes. Agent i represents an essential class (the only one present in the graph) on her own.
Figure 7: BC, opinion shift (drop) in essential classes
Figure 8: BC, opinion shift (drop) drop in inessential classes
Figure 9: PA, opinion shift (drop) in essential classes
Figure 10: PA, opinion shift (drop) in inessential classes
Figure 11: FB, opinion shift (drop) in essential classes. The fundamental value is in red.
Figure 12: FB, opinion shift (drop) in inessential classes. The fundamental value is in red.
In order to understand the DJ30 drop occurred in the early afternoon of April 23, 2013 described in the Introduction of this work (and reported again in Fig. 13a) we focused on a one-run simulation reported in Fig. 13b and, zoomed-in, in 13c. Here, we simulated a situation where drops in opinion occurred in only two essential classes for a total number of 10 agents shifting opinion. The drop is dramatic and shows the importance of the topological classification of agents.
## 4 Empirical validation and model calibration
The empirical validation of such a model is not an easy task. Although some stylized facts like clustered volatility, drops in prices that lead to fat tails in the returns distribution can be explained by simulations, the calibration of the parameters of the model we have is an open problem that we leave for future research. For example, as we have seen, the correct estimation of the update propensity parameter α is crucial.
Time series for prices are easily available. However, determining and calibrating the interaction network and the update propensity parameter seems to be the hardest part of the problem. We propose to use the wide availability of Big Data (i.e. Twitter microblogging) to have a proxy for such interactions. Applying Twitter data (and, in general, social network big data) in the context of financial markets is still a research endeavour (see Bollen, 2010). However, the recent use of Twitter data in giving signals about markets by the famous data service provider Bloomberg, clearly shows a growing interest in this approach. We can then give an outline of the future research that will need to be done to accomplish this task:
14696.69
129.52
0.89%
USD
(a)
48
50
time
(c)
Figure 13: Temporary large deviations from a fundamental can be explained by opinion shifts . Opinion shift (drop) in 2 essential classes leads to drop in price (blue line) and return to fundamental after confidence weights have shifted due to the update to the fundamental (red dotted line). The figure on the right zooms in the time periods where the opinion shift takes place. This figure provides a possible interpretation of the DJ drop on April 13, 2013 after the fake tweet apperared.
p(t)
0.51
0.5
0.49
0.48
0.47
0.46
0.45
0.44
0.43
0.42
Minute Intraday
40
42
14,720
14,700
14,040
14,020
580
44
46
52
54
56
58
- · Twitter and, more generally, social network provide with a huge amount of data, in the order of terabytes per minute. As such, algorithms must be fast and efficient, especially in the classification of agents.
- · At each t , agents needs to be classified according to the their topological importance (this includes, but it is not limited to, the separation into essential and inessential classes).
- · Social network data naturally embed a multiple network structure , i.e. a structure where several interactions might occur. A possible way to cope with this problem is to use a tensor sequence as a means to treat multiple networks (see Kolda, 2005) ( A ( )) t t ∈ N
$$a _ { i j k } ( t ) = \begin{cases} \in ( 0, 1 ] \text{ if } j \text{ retweets } i ^ { \prime } s \text{ tweet about stock $5kat $t$} \\ 0 \text{ otherwise} \end{cases}$$
## 5 Conclusions and future research
In this work, we have proposed an analytical framework to model the role of opinion dynamics in a financial system. We have stressed the intertwined relationship among opinions, the actual price dynamics and the interaction network among agents. Different cases have been explored and we showed, by simulations, that some stylized facts can be reproduced and explained in this way. Moreover, we proposed an empirical validation scheme for the model.
This work stems from the need to interpret the role of opinion dynamics in an increasingly interconnected world, where network effects play an important role in determining the evolution of social systems.
Particular emphasis has been given to the topology of the interaction network and to the classification of agents according to their role in the dynamical process of opinion formation. We have explored the different models proposed in the literature and we have built on a well - known nonlinear model where opinion fragmentation might occur. In this case, we noticed that such dynamics reflects on the price formation process in a feedback loop in the sense that agents will tend to shape their opinion patterns according to the topology and vice - versa, the topology is determined by the opinion patterns.
## Future research directions
The approach proposed in this work provides a framework that can be extended by future research on these topics.
For what concerns the network interaction topology, a detailed analysis on the properties of network dynamics when the network has a specific topology should be done. We refer to, e.g. random, small - world, scale - free or core - periphery (see Newman, 2010 for details on such network topologies) topology structures. Finding analytical results for specific topologies would be useful for empirical comparison and analysis of the dynamical system. In particular, we think of a scheme where the topology is imposed to the essential classes: this would drive different polarisation/fragmentation patterns that would then need to be compared with the actual empirical data.
Moreover, we would like to extend the model on price formation to more than one risky asset (see Chiarella et al. 2007 for an account on this problem) by assuming different confidence networks for each asset. In other words, we want to investigate a multidimensional opinion dynamics problem. We propose a dynamic multiple network approach to this problem, by using a tensor representation (see Kolda and Bader 2009), as follows (see Fig. 14 for a graphical representation):
Figure 14: Tensor multiple network approach for opinion dynamics.
## A Materials and methods
## A.1 Graphs and networks
We will outline the key concepts in graph and network theory used in the paper . 6 A graph G = ( V, E ) is a pair of sets ( V, E ), where V is the set of n vertices and E is the set of m pairs of vertices of V . The pair ( i, j ) belonging to E is called an edge of G and i and j are called adjacent . A directed graph (digraph) is a graph in which each edge is an ordered pair ( i, j ) of vertices. G is simple if there one edge between two adjacent vertices. A weight w ij can be associated to each edge ( i, j ) and, in this case, we will have a weighted or valued graph. A path is a sequence of distinct adjacent vertices and a i -j path is a path from i to j . For a pair of vertices ( i, j ), i and j communicate if it holds i -j and j -i . In case w ii > 0 ∀ i , it always holds i -i , and we call i self communicating . A vertex i is essential if, ∀ j ∈ V , with i -j , it holds j -i and we denote such property with i ∼ j . A vertex i is inessential if it is not essential.
V can be partitioned into self-communicating equivalence classes of vertices. In fact it easy to see that if i is self communicating and ( i, j ) communicate, then i ∼ i , i ∼ j ⇒ j ∼ i and, finally, if i ∼ j and j ∼ k ⇒ i ∼ k , i.e. a path from i to k and vice versa can always be found passing through j . Hence, ∼ is an equivalence relation. A self-communicating inessential vertex constitutes an equivalence class on its own. In an essential class, all vertices communicate to each other and do not communicate with other vertices belonging to other classes. On the other hand, inessential vertices are linked to either essential and inessential vertices. 7 A particular case arises when all vertices belong to one essential class, i.e. ∀ ( i, j ) i ∼ j . In this case, the graph is called strongly connected .
glyph[negationslash]
A nonnegative n - square matrix W representing the adjacency relationships (in general, weighted and directed), between vertices of G is said to be the adjacency matrix of G . A nonnegative adjacency matrix W is said to be diagonal dominant if w ii ≥ ∑ j = i w ij ∀ i . W is said to be irreducible if for some permutation matrix P , the (permuted) matrix PWP T is not block upper triangular. A matrix that is not irreducible is said to be reducible . The adjacency matrix of a strongly connected graph is irreducible. If there exist a t ∈ N such that W t is (strictly) positive, then W is said to be primitive . An n -vector x is said to be a right eigenvector of W , with associated eigenvalue λ if Wx = λ x . For any n × n matrix, there exist n eigenvalues λ , λ 1 1 , . . . , λ n .
A nonnegative n -square matrix W ∈ R n × n is row - stochastic if the elements in each row sum up to 1, i.e. ∑ n j =1 w ij = 1 ∀ i or, in matrix notation, Wu = u , where u is the n -th unit vector. It is easy to show that a row stochastic matrix has dominant eigenvalue λ 1 = 1 with u its associated eigenvector.
Any square matrix W can be brought in its Gantmacher form :
6 The reader may refer to Harary (1969) and Newman (2010) for more details. Two relevant references for matrix theory are Gantmacher (1959) and Horn & Johnson (1990).
7 This terminology is the one used in Lorenz [57]. In Markov chain theory, essential classes would be associated to a recurrent class of communicating states and inessential vertices would be associated to transient states.
where P is a suitable permutation matrix. The diagonal blocks W 1 1 , , . . . , W p,p are square and irreducible (Gantmacher 1959, Lorenz 2007). For the non diagonal Gantmacher blocks W kl with k = g +1 , . . . , p and l = 1 , . . . , k -1. Then, it holds that for every k = g +1 , . . . , p at least one block of W k, 1 , . . . , W k,k -1 contains at least one positive entry. It turns that W kk , with k = 1 , . . . , g are the sub matrices associated to essential classes. Two non - negative matrices A and B are said to be of the same type , A ∼ B if they share the same non negative pattern, i.e. a ij > 0 ⇔ b ij > 0.
An important results of graph theory (see Harary, 1969) is that a non-connected graph G can be uniquely partitioned into separate strongly connected components (SCC). When each one of these strongly connected component is contracted to a single vertex, the resulting graph is a directed acyclic graph (DAG), the so called condensation of G . A DAG is a digraph with no directed cycles, i.e. there exists no path connecting a node i to itself.
Given a graph G = ( V, E ), its condensation digraph C C ( ) = ( V C , E C ) is a graph where the set of nodes V C is represented by the strongly connected components of G . Accordingly, the condensation digraph of G , denoted C G ( ), is defined as follows: the nodes of C G ( ) are the SCCs of G, and there exists a directed edge in C G ( ) from node H 1 to node H 2 if and only if there exists a directed edge in G from a node of H 1 to a node of H 2 .
## A.1.1 Determining the Gantmacher form
We hereby propose a simple 8 in order to solve the problem. The algorithm outline is as follows:
- 1. find the Strongly Connected Components of G , e.g. via Tarjan's algorithm, of Depth First Search based algorithms ( O V ( | | + | E | )), see Knuth (1997);
- 2. find the condensation digraph C G ( ) of G ;
- 3. for each node k of C G ( ) check links for nodes;
- 4. if node k has only in-coming links (i.e. sinks), then it is associated to an essential class;
- 5. if node k has only out - going links, then is is associated to an innessential class.
- 6. order nodes in C G ( ) in order to find permutation matrix for A .
8 This algorithm is inspired by Fig. 2 in Mirtabadei & Bullo (2011). We thank Jan Lorenz (private communication) for providing useful hints.
Table 2: Agents classification (via SCCs)
The algorithm proves to be quite fast for the applications in Section 4. However, better solutions that do not imply the evaluation of a node of G multiple times could be found (via a Depth-First Search approach.).
On the basis of such classification, a further refinement of Lorenz's classification has been proposed by Mirtabadei and Bullo (2011). See Table 2 for a comparison of the two classifications.
We now deal with the mathematical background related to the opinion dynamics scheme. Several concepts are key in this class of models. We refer, in particular, to how accurate are agents' opinions about reality and how opinion pooling plays a role in this context . 9 However, for sake of brevity, will shall not explore these concepts in detail.
Consider a set of interacting agents and a certain dynamical process leading to the formation of an opinion space, represented by a scalar value. Some interesting questions naturally arise. Will all agents reach the same opinion? Will there be a consolidation of different opinions into fewer ones? We will refer to the first case as consensus of opinions and to the latter as fragmentation of opinions (see, for example, Hegselmann & Krause, 2005).
DeGroot (1974) proposed one of the first mathematical formulations for the dynamics of opinions, by introducing a weighted averaging scheme that will remain the baseline model in later works. The convex weighted average scheme clearly stems from the mathematical properties related to the product of stochastic matrices and, hence, Markov chains. 10
In this light, another interesting work is the one of Berger (1981), in which sufficient conditions for consensus are discussed in more details and an implicit relationships, later analyzed in the literature on economic networks (Jackson, 2008), between consensus and eigenvector centrality is provided. Berger's paper shows also how a common opinion might be achieved only within sub-groups of agents, hence implying a certain degree of opinion fragmentation within the opinion structure(i.e. dissent as opposed to consensus ). Other interesting approaches linking network centrality and opinion dynamics are those of Friedkin (1986, 1991) and Friedkin & Johnsen (1990).
The literature so far reviewed is based on models of interaction that are linear in their nature. In the models proposed by DeGroot and Friedkin's, for example, the focus is on a linear dynamical system. As already mentioned, they make use of very well known properties of linear
9 The so-called 'Wisdom of the Crowd' idea, tracing back to Galton (1907).
10 DeGroot clearly refers to an opinion as a subjective probability distribution , which each agent assigns to a certain parameter. Such a priori opinions are then 'pooled', to obtain an a posteriori set of opinions, thus implying a Bayesian framework. Previous formulations of this problem include the ones by J. French (1956), Harary (1959) and later contributions include the ones (among others) by Chatterjee and Seneta (1977) and Cohen et al . (1986). These works focus on reaching a common opinion, a consensus .
systems and, in particular, of homogeneous Markov chain. Berger's work implicitly introduces a graph - theoretical argument based on connectivity, and Friedkin's approach is entirely graph - based.
A limitation lies in that such models do not capture the evolution in time of the interaction network. Concerning this problem, interesting analytical results for time-dependent networks are found in Moreau (2005) in the context of synchronization analysis and by Fazeli & Jadbabaie (2012), who introduce a martingale network process associated to a Polya Urn process.
So far, opinions and network topology are separated. However, in our model of price formation, there is a strong interaction between opinions and the confidence network. In other words: the interaction topology might be influenced by the opinions themselves.
An interesting way to cope with both these problems is the approach proposed by Krause in a series of works (1997, 2000 and, with Hegselmann, in 2005). In this model, the non-linearity arises because the topology of the confidence network changes in time according to the opinion structure and, viceversa, the new opinion structure is derived by the network interaction. This feedback relationship does not allow to easily derive analytical results. A combination of matrix and graph - theoretical arguments can be used in order to partially overcome this problem, but most of the interpretation must rely on computer simulations. In fact, although it is possible to derive some sufficient conditions for the convergence of a system, it is still an open challenge to fully characterise in an analytical way the dynamical behaviour of such a system.
Hegselmann & Krause (2005) propose a bounded confidence non - linear approach: agents put a positive confidence weight to another agent if the absolute value of the difference of the opinions does not exceed a certain threshold. According to such a threshold, this non - linear model can capture effects of opinion fragmentation or polarization as well as convergence to a consensus.
We define a stable opinion pattern in the dynamical system as that situation where agents stabilize their opinion and no further change occurs. Once such a stable pattern is achieved, we can, in fact, have the two following situations:
- 1. consensus , i.e. all agents have the same opinion;
- 2. dissent , i.e. agents will have different opinions that do not change over time. In particular, we can divide dissent into polarization (i.e. agents can have only two opinions) and fragmentation (when the number of possible opinions is higher than two).
Sufficient conditions for convergence in these models are found in the above mentioned works. In addition to this, we recall the work of Lorenz (2005, 2006, 2007). We will observe, in particular, that the setup suggested by Lorenz provides with useful insights on the topological classification of agents. Further insights and some more technical results on the stabilisation patterns in the classification of agents can be found in Mirtabadei and Bullo (2011).
## A.2 Opinion dynamics: the mathematical background
## A.2.1 The homogeneous opinion dynamics process
Consider a set V of n agents, and an opinion vector x ( ) t ∈ R n at time t ∈ [0 , . . . , + ∞ ). The element x i ( ) represents the opinion of agent t i at time t . The n agents can be seen as vertices
on a network, discussing a particular issue or trying to obtain better guesses about the value of a variable at each time step, such as the price of a financial asset.
The general homogeneous consensus dynamic model can be written as (DeGroot, 1974):
$$\mathbf x ( t + 1 ) = \mathbf A \mathbf x ( t )$$
Where A is a row - stochastic matrix, i.e. ∑ j a ij = 1, ∀ i . We will refer to this matrix as the confidence matrix. The process is said to be homogeneus as the confidence matrix does not change in time (time-invariant process). At time t +1, we have:
$$\mathbf x ( t + 1 ) = \mathbf A ^ { t + 1 } \mathbf x ( 0 )$$
The model in Equation 11 reaches a consensus if ∀ x (0) ∈ R n , ∃ c s.t. lim t →∞ x ( ) t = A x ∞ (0) = c , where c is a real - valued vector with all elements equal. The following twopart theorem gives conditions in this case: 11
Theorem 1 (Conditions for convergence to consensus) . A consensus for system 11 is achieved:
- · if ∀ ( i, j ) ∈ E , ∃ k ∈ V such that a ik > 0 and a jk > 0 [see DeGroot (1974) for the proof];
- · if and only if ∃ t 0 ∈ T s.t. the matrix A t 0 contains at least one strictly positive column [see Berger (1981) for the proof].
Theorem 2 (Conditions for convergence) . Let A be in its Gantmacher form with diagonal Gantmacher blocks A k , k = 1 , . . . , s and g = 0 , . . . s , then
- · lim t →∞ x ( ) t exists ∀ x (0) ∈ R n if and only if the Gantmacher blocks are all primitive;
- · the system reaches a consensus if and only if g = 1 (i.e. there exists only one essential class and no inessential classes).
This first general theorem states that a sufficient condition for consensus is found when any two agents have a positive weight on a same third agent. This conditions is related to the primitivity of the matrix. The value c of the consensus depends, clearly, on the initial opinion profile x (0).
The second theorem employs the Gantmacher form. If each of the sub - groups of essential agents has a primitive structure, then a stable opinion configuration is achieved where only the opinions of the essential agents matter in determining the asymptotic opinion vector. More over, we have a necessary and sufficient condition for consensus when the sub - group of essential agents is only one ( g = 1, i.e. only one essential class).
We define a consensus matrix ˜ A the rank - one strictly positive row - stochastic matrix whose rows are all equal. It is immediate to see that ˜ Ay always gives a consensus.
11 For sake of brevity, we do not report the proofs of these theorem. We refer the reader to the already cited works of Gantmacher (1959), DeGroot (1974), Berger (1981) and Hegselmann and Krause (2005).
## A.2.2 Inhomogeneous opinion dynamics process
A more general approach is obtained when A is time dependent:
$$\mathbf x ( t ) & = \mathbf A ( t ) \cdots \mathbf A ( 0 ) \mathbf x ( 0 ) \\ \cdot \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad.$$
We refer to the process described by Eq. 12 as either inhomogenous (using the Markov chains expression) or time - variant (as opposed to the time - invariant process described in Eq. 11). Quite clearly, the 'inhomogenous' or 'time - variant' part of the model stems from changes in the agents' interaction topology and not directly from the opinion patterns in time. However, we will see that, in the case when the weight (and hence the interaction topology) depends on the opinions themselves, the feedback relationships between the opinions and the network topology will show non - trivial behaviours.
The definitions we previously gave in the homogeneous case naturally extend to the inhomogeneous case. The sequence of n × n non - negative matrices A ( ) in our model is said to be t a sequence of confidence matrices if and only if the matrices:
- 1. are row - stochastic, i.e. ∑ j a ij ( ) = 1, t ∀ t ;
- 2. have strictly positive diagonal values, i.e. a ii ( ) t > 0, ∀ i ∈ V and ∀ t ;
The objective of our analysis will be thus the sequence of matrices ( A 0 , . . . A t ) and, in particular, their right - product A ( ) t . . . A (0) and its asymptotic behaviour for t →∞ . 12
The process described by Equation 12 is clearly more difficult to treat than the homogeneous case and analytical results are harder to obtain. As usual, we are interested in determining sufficient and necessary conditions for the convergence of the system. The structure of the sequence of matrices A ( ) t will determine the asymptotic behaviour of the system. We will observe that, analogously to what happens in the homogenous case, consensus or, more generally, convergence to consensus in the sub - groups will be obtained as long as the weights keep being positive to a certain degree and/or sufficient time passes by.
Several interesting results have been proposed in the literature. Among these, we recall Moreau (2005), who describes a model of dynamic network interaction based on synchronization theory, providing results based on both graph theory and system theory; Fazeli and Ali Jadbabaie (2010), who use a Polya urn argument to show certain asymptotic properties and Fazeli & Jadbabaie (2012), who propose some analytical results on convergence where the underlying network process is a martingale. Alternative averaging schemes are not investigated in this work. 13
12 As opposed to the Markov chain process, in which case, we are interested in the left product of the elements of the sequence: A (0) . . . A ( ). t
13 For a discussion on alternative averaging schemes see Krause (2000 and 2008). For a detailed explanation on the speed of convergence of the system, refer to Isaacson & Madsen (1976), Gross & Rothblum (993); Rothblum & Tan (1995). See Dobrushin, (1956) for an analysis of the ergodicity coefficient, later tackled by Ipsen & Selee (2011) in an analysis that also explores concepts in network centrality.
## A.2.3 Convergence in the inhomogenous case
The intution behind the convergence in the inhomogeneous case is that a time - variant model will converge provided that, as previously noted, both the weights keep beeing sufficiently positive and/or that enough time passes. The positive diagonal assumption will also play a significant role in ensuring this. Clearly, should the weights go to zero too fast, convergence (let alone consensus) might not be achieved.
Krause (2000) proposes an interesting theorem for convergence (not limited to arithmetic means) and later, Hegselmann & Krause provide more general approach based on the arithmetic average setting based on the accumulated weights . The intuition behind this approach is to keep track of the weights of the chain of products.
Another way to see this problem is to exploit the positive diagonal. In fact, given a non negative matrix B with positive diagonal and a non - negative matrix C , then the product BC has at least the same positive entries of C . Based on the above reported results, Lorenz (2005 and 2007) reports a convergence theorem based on the Gantmacher form.
Theorem 3. Given A ( ) t , t = 0 1 2 , , , . . . , non - negative, row - stochastic with strictly positive diagonal, and min + A ( t +1) A ( ) t ≥ δ t and ∑ ∞ t =1 δ t = ∞ Gantmacher blocks converge to:
$$\lim _ { t \to \infty } \left ( \mathbf A ( t ) \cdots \mathbf A ( 0 ) \right ) = \begin{bmatrix} G _ { 1 } & & & 0 \\ & \ddots & & \\ 0 & & G _ { g } & 0 \\ \frac { n. c \ \dots \ n. c. } { 0 } \end{bmatrix} \\. \quad. \quad. \cdots \cdots, \quad. \cdots. \cdots. \cdots. \cdots. \cdots.$$
where g is the number of essential classes, n h is the cardinality of essential class h and G h are n h × n h row - stochastic consensus matrices, i.e. strictly positive stochastic matrices with all rows equal:
$$\mathbf G _ { h } = \left [ \begin{array} { c c c c } c _ { 1 } & c _ { 2 } & \dots & c _ { n } \\ \vdots & \vdots & \vdots & \vdots \\ c _ { 1 } & c _ { 2 } & \dots & c _ { n } \end{array} \right ] = \left [ \begin{array} { c c c c } c _ { 1 } & c _ { 2 } & \Big | \dots & c _ { n } \end{array} \right ]$$
## A.2.4 Opinion dynamics under bounded confidence
The inhomogenous case is often referred to as a time - variant model, in the sense that the weights each agent puts onto other agents changes over time. However, a rule for the evolution of the network topology must be established in order to analyze the system dynamics.
The main question in this case is whether the update rule will influence the reaching of a stable opinion pattern or fragmentation (either consensus within subgroups or dissent across different subgroups) will occur or even no convergence at all will be achieved.
Within this framework, 'the most difficult type of model occurs if the weights depends on opinions itself because then the model turns from a linear one to a non-linear one' (cit. Hegselmann & Krause, 2005).
In particular, we hereby examine a system with bounded confidence , in which agents update their network interaction structure by putting weight on agents that have a similar opinion. The system becomes non - linear in the sense that the averaging process can automatically include or discard certain agents, hence underweighting certain opinions or overweighting other opinions if they are similar.
We define a bounded confidence opinion dynamics process, as the system with initial opinion vector x (0) and dynamics:
$$\mathbf x ( t + 1 ) = \mathbf A ( t + 1 ) \mathbf x ( t ) = \mathbf A ( x ( t ), \epsilon ) \mathbf x ( t )$$
$$a _ { i j } ( t + 1 ) \coloneqq \begin{cases} \ \frac { 1 } { \# I ( i, \mathbf x ( t ) ) } & \text{if $j\in I(i,\mathbf x(t))$} \\ 0 & \text{otherwise} \end{cases}$$
$$I ( i, \mathbf x ( t ) ) & = \{ j \text{ s.t.} \ | x _ { i } ( t ) - x _ { j } ( t ) | \leq \epsilon _ { i } \} \\.. \quad. \quad. \quad. \quad. \quad.$$
This structure implies that, at each time t , every agent compares her opinion with that of others, finding a subset of agents whose opinion does not differ too much from her. The agent then assigns equal weight to these agents. The assumption of equal weights can be relaxed, but it will not be analyzed in this work. More importantly, agent i herself lies in the set I , thus leading to a positive diagonal at each time t .
The non - linearity in the model is due to the fact that the update rule enhances the selection of similar opinions and implies discarding distant opinions. This approach naturally brings to fragmentation and polarisation of opinions, meant as a hardening of opinion patterns. Hegselmann & Krause (2005) describe, by simple examples, the possibility of splits between subgroups that become essential and that reach consensus within the agents belonging to the subgroup.
Simulations The models we have so far described are multi - faceted and computer simulations can give better insights. Since the homogeneous case is of little interest in this work, and it has been reported for sake of completeness, we refer to the non - linear bounded confidence inhomogenous model in Equation 13 in the following simulations.
All simulations are for n = 100 agents. The number of simulations, unless otherwise stated, is 1000. In the first part, the initial opinion vector x (0) is drawn from a uniform continous random distribution with support [0 1]. , Figure 15 reports a simple one - run simulation with different values of glyph[epsilon1] (assumed to be equal for all agents). It is immediate to see that the higher the glyph[epsilon1] , the higher the chances to achieve consensus. However, polarized situations (e.g. glyph[epsilon1] = 0 20) . can be obtained, which show 'hardening' in the opinions. When glyph[epsilon1] = 0 40 a consensus is achieved . in a very limited number of time steps.
Figure 16 shows a tri - dimensional plot of the opinion space with respect to different values of glyph[epsilon1] once a stable configuration is achieved , the y axis is the average relative frequence over 1000 simulations. One can see how the higher the glyph[epsilon1] , the less fragmented the opinion pattern is.
Figure 15: Single - run simulations
Figure 16: Average relative frequencies of x ∈ [0 , 1], n = 100 opinions (after stabilisation) for different values of glyph[epsilon1] (1000 simulations).
Distribution of the initial opinion profile Since in the non - linear bounded confidence model the initial opinion profile can determine the degree of fragmentation of the opinion patterns, it is interesting to conclude this Chapter by modifying the distribution of the initial opinion profile x (0). Quite obviously, if the initial opinion profile is too dispersed (with respect to the confidence level glyph[epsilon1] ), the initial fragmentation pattern will never converge to a consensus or to a less fragmented pattern.
While still keeping the opinion range x (0) ∈ [0 , 1], we will use an initial opinion profile from random numbers drawn from a Beta distribution 14 .
## A.3 Software and data
Data processing and simulations have been done in Python and Matlab. Tri - dimensional plots have been exported to LT E Xvia A matlab2tikz . 15
Twitter ( www.twitter.com ) data have been retrieved and pre - processed (string and hashtag recognition) in Python with Python Twitter 16 (a Python wrapper around the Twitter API
14 The pdf of a Beta distribution is given by
$$f ( x ; a, b ) = \frac { 1 } { B ( a, b ) } x ^ { a - 1 } ( 1 - x ) ^ { b - 1 },$$
where B a, b ( ) is the Beta function and a and b are the shape parameters of the distribution. 15 http://www.mathworks.de/matlabcentral/fileexchange/22022-matlab2tikz/all\_files 16 The current analysis, as of May 2014, is done in Twython.
( https://dev.twitter.com/ ) and double-checked via Twitty for Matlab. Network visualizations are done with Gephi ( https://gephi.org/ ).
## References
- 1. Daron Acemoglu, Asuman Ozdaglar, and Ali ParandehGheibi (2009) Spread of Misinformation in Social Networks . Working paper version. http://arxiv.org/abs/0906.5007
- 2. Julie R. Agnew and Lisa R. Szykman (2005). Asset Allocation and Information Overload: The Influence of Information Display, Asset Choice, and Investor Experience , The Journal of Behavioral Finance 2005, Vol. 6, No. 2, 5770.
- 3. Robert J. Aumann (1976). Agreeing to disagree . Annals of Statistics, 4 , 1236 - 1239.
- 4. Caccioli, F., Marsili, M. & Vivo, P. (2009). Eroding market stability by proliferation of financial instruments . Eur. Phys. J. B 71, 467479
- 5. Nicholas Barberis and Richard Thaler (2003), A Survey of Behavioral Finance . Handbook of the Economics of Finance, Edited by G.M. Constantinides, M. Harris and R. Stulz.
- 6. D. Bawden and L. Robinson (2008). The dark side of information: Overload, anxiety and other paradoxes and pathologies . Journal of Information Science, volume 35, number 2, pp. 180191. http://dx.doi.org/10.1177/0165551508095781
- 7. Stefano Battiston, Guido Caldarelli, Co-Pierre Georg, Robert May & Joseph Stiglitz (2013). Complex derivatives . Nature Physics 9 , 123125 (2013) doi:10.1038/nphys2575.
- 8. Roger L. Berger (1981). A Necessary and Sufficient Condition for Reaching a Consensus Using DeGroot's Method . Journal of the American Statistical Association, Vol. 76 , No. 374 (Jun., 1981), pp. 415- 418
- 9. Johan Bollen, Huina Mao, Xiao-Jun Zeng (2010), Twitter mood predicts the stock market , Journal of Computational Science, 2 (1), Pages 1-8
- 10. Phillip Bonacich (1987) Power and Centrality: A Family of Measures American Journal Of Sociology 92 (5): 1170-1182.
- 11. Phillip Bonacich (1991). Simultaneous group and individual centralities . Social Networks 13 , 155168.
- 12. R. Cont and J. P. Bouchaud (2000). Herd behaviour and aggregate fluctuations in financial markets . Macroeconomic Dynamics, 4:170196, 2000.
- 13. William A. Brock, Cars H. Hommes (1997), A Rational Route to Randomness . Econometrica Vol. 65 , no. 5, 1059-1095.
- 14. William A. Brock, Cars H. Hommes (1998) Heterogeneous beliefs and routes to chaos in a simple asset pricing model . Journal of Economic Dynamics and Control 22 (1998) 1235 - 1274.
- 15. Brock, W. A., Hommes, C. H. & Wagner, F. O. O. (2008). More hedging instruments may destabilise markets . J. Econ. Dynam. Cont. 33, 19121928 (2008)
- 16. F. Bullo, J. Cortes, and S. Martinez (2009), Distributed Control of Robotic Networks . Applied Mathematics Series. Princeton University Press, 2009. Available at http://www.coordinationbook.info
- 17. K. C. Chang, K. Pearson, and T. Zhang (2008), Perron-Frobenius theorem for nonnegative tensors . Commun. Math. Sci. Volume 6 , Number 2, 507-520.
- 18. S. Chatterjee and E. Seneta (1977). Towards Consensus: Some Convergence Theorem on Repeated Averaging . Journal of Applied Probability, 14 , 89-97.
- 19. Carl Chiarella, Roberto Dieci and Xue-Zhong He (2007). Heterogeneous expectations and speculative behavior in a dynamic multi-asset framework , Journal of Economic Behavior & Organization, Elsevier, vol. 62 (3), pages 408-427.
- 20. Werner De Bondt, Werner and Richard Thaler (1985). Does the Stock Market Overreact? , Journal of Finance, vol. 40 (3), pages 793-805.
- 21. Werner F.M. DeBondt, Hersh Shefrin, Yaz Gulnur Muradoglu, Sotiris K. Staikouras (2008), Behavioural Finance: Quo Vadis? Journal of Applied Finance, Vol. 19 , pp.7-21.
- 22. Morris H. DeGroot (1974), Reaching a Consensus . Journal of the American Statistical Association, 69 (345): 118-21.
- 23. Andrea Devenow and Ivo Welch (1996). Rational herding in financial economics in European Economic Review Volume 40, Issues 3-5, April 1996, Pages 603-615.
- 24. R. L. Dobrushin (1956). Central limit theorems for non stationary Markov Chains II . Theory of Probability and its applications, 1 , pp. 329 - 383
- 25. David Easley, Jon Kleinberg (2010), Networks, Crowds, and Markets: Reasoning About a Highly Connected World , Cambridge University Press.
- 26. Eugene F. Fama (1970), Efficient Capital Markets: a Review of Theory and Empirical Work , The Journal of Finance, Vol. 25 , No. 2, pp. 383-417.
- 27. E. F. Fama and K.R. French (1988). Permanent and Temporary Components of Stock Prices . Journal of Political Economy 96, 246-273.
- 28. Arastoo Fazeli and Ali Jadbabaie (2011), On Consensus in a Correlated Model of Network Formation Based on a Polya Urn Process . 50th IEEE Conference on Decision and Control and European Control Conference (CDC-ECC).
- 29. Arastoo Fazeli and Ali Jadbabaie (2012). Consensus Over Martingale Graph Processes . Proceedings of the American Control Conference (ACC). Montreal, E-ISBN: 978-1-4673-2102-0
- 30. Noah E. Friedkin (1991) Theoretical Foundations for Centrality Measures . American Journal of Sociology. Vol. 96 , pp. 1478 - 1504.
- 31. F. Galton (1907), Vox Populi . Nature 75 , pp. 450-451.
- 32. Andrea Gaunersdorfer and Cars Hommes (2007), A nonliner structural model for volatility clustering Long Memory in Economics 2007, pp 265-288, Springer.
- 33. Benjamin Golub and Matthew O. Jackson. (2010), Na¨ ıve Learning in Social Networks and the Wisdom of Crowds , American Economic Journal: Microeconomics, 2 (1): 112-49.
- 34. O. Gross and U. G. Rothblum (1993), Approximations of the spectral radius, corresponding eigenvector, and second largest modulus of an eigenvalue for square, nonnegative, irreducible matrices , SIAM J. Matrix Anal. Appl., 14 .
- 35. Andrew G. Haldane & Robert M. May (2011). Systemic risk in banking ecosystems . Nature 469 , 351355, doi:10.1038/nature09659
- 36. F. Harary (1959), A criterion for unanimity in French's theory of social power . In D. Cartwright, editor, Studies in Social Power, pp. 168-182. University of Michigan.
- 37. F. Harary (1969), Graph theory , Addison - Wesley, Reading MA. ISBN-13: 978-0201410334.
- 38. Rainer Hegselmann, Ulrich Krause (2004), Opinion Dynamics Driven by Various Ways of Averaging , Computational Economics vol. 25 , n. 4, p. 381-405.
- 39. Rainer Hegselmann and Ulrich Krause (2002), Opinion Dynamics and Bounded Confidence Models, Analysis and Simulations . Journal of Artificial Societies and Social Simulation (JASSS) vol. 5 , no. 3.
- 40. David A. Hirshleifer (2001), Investor Psychology and Asset Pricing , AFA 2001 New Orleans Meetings.
- 41. Cars Hommes and Florian Wagener (2008), Complex Evolutionary Systems in Behavioral Finance . Tinbergen Institute discussion papers 08-054/1.
- 42. Harrison Hong and Jeremy C. Stein (1999), A Unified Theory of Underreaction, Momentum Trading, and Overreaction in Asset Markets . The Journal of Finance, Vol. 54 , No. 6.
- 43. Roger A. Horn, Charles R. Johnson (1990), Matrix Analysis , Cambridge University Press.
- 44. Ilse C.F. Ipsen and Teresa M. Selee (2011) Ergodicity Coefficients Defined by Vector Norms . SIAM J. Matrix Anal. Appl. Vol. 32 , No. 1, pp. 153 - 200.
- 45. D.L. Isaacson and R. W. Madsen (1976), Markov Chains: Theory and Applications . Wiley, New York.
- 46. Matthew O. Jackson (2010), Social and Economic Networks , Princeton University Press
- 47. Kahneman, D. and Tversky, A. (1984). Choices, Values, and Frames . American Psychologist 39 (4): 341350. doi:10.1037/0003-066x.39.4.341.
- 48. Donald E. Knuth (1997), The Art of Computer Programming (3rd ed.) Addison-Wesley Professional.
- 49. Tamara G. Kolda and Brett W. Bader (2009), Tensor Decompositions and Applications , SIAM Review 51 (3):455-500, September 2009.
- 50. Ulrich Krause (2000), A discrete nonlinear and non-autonomous Model of Consensus Formation , in S. Elyadi, G. Ladas, J. Popenda, J. Rakowski (eds), Communications in Difference Equations, Gordon and Breach Pub., Amsterdam, p. 227-236.
- 51. Shamanth Kumar, Fred Morstatter, Huan Liu (2013), Twitter Data Analytics , Springer.
- 52. A. W. Lo, and A. C. MacKinlay (1988). Stock market prices do not follow random walks: Evidence from a simple specification test . Review of Financial Studies 1, 41-66.
- 53. Jan Lorenz (2005), Continuous Opinion Dynamics: Insights through Interactive Markov Chains , Proceedings of IASTED Conference 'Modelling, Simulation and Optimization MSO 2005'.
- 54. Jan Lorenz (2005), A Stabilization Theorem for Dynamics of Continuous Opinions , Physica A, vol. 355 , n 1, p. 217-223.
- 55. Jan Lorenz (2006), Continuous opinion dynamics of multidimensional allocation problems under bounded confidence: More dimensions lead to better chances for consensus European Journal of Economic and Social Systems, EJESS 19 /2006 Beliefs, Norms, and Markets, pp. 213-227.
- 56. Jan Lorenz (2006), Consensus strikes back in the Hegselmann-Krause model of continuous opinion dynamics under bounded confidence , Journal of Artificial Societies and Social Simulation. http://jasss.soc.surrey.ac.uk/9/1/8.html
- 57. Jan Lorenz (2006), Convergence of products of stochastic matrices with positive diagonals and the opinion dynamics background , in Positive Systems, Springer Lecture Notes in Control and Information Sciences 341/2006.
- 58. J. Lorenz (2010). Heterogenous bounds of confidence: meet, discuss and find consensus! Complexity 4(15):43-52, 2010.
- 59. Jan Lorenz, Heiko Rauhutb, Frank Schweitzer, and Dirk Helbing (2011), How social influence can undermine the wisdom of crowd effect . PNAS 2011; May 16, 2011, doi:10.1073/pnas.1008636108
- 60. Thomas Lux and Michele Marchesi (1999), Scaling and criticality in a stochastic multi-agent model of a financial market . Nature 397 , 498-500.
- 61. Thomas Lux (2009), Rational Forecasts or Social Opinion Dynamics? Identification of Interaction Effects in a Business Climate Survey . Journal of Economic Behavior & Organization Volume 72 , Issue 2, 2009, Pages 638-65.
- 62. Charles Mackay (1841), Extraordinary Popular Delusions and the Madness of Crowds .
- 63. Robert K. Merton (1968). Social Theory and Social Structure , Free Press, ISBN 0-02-921130-1
- 64. Robert K. Merton (1995). The Thomas Theorem and The Matthew Effect. Social Forces, 74 (2):379-424.
- 65. Carl Meyer (2000), Perron - Frobenius Theory of Nonnegative Matrices , in Matrix analysis and applied linear algebra, SIAM, ISBN 0-89871-454-0
- 66. Anahita Mirtabatabaei and Francesco Bullo (2011), Opinion Dynamics in Heterogeneous Networks: Convergence Conjectures and Theorems . http://arxiv.org/abs/1103.2829. Submitted to SIAM Journal on Control and Optimization (SICON)
- 67. Luc Moreau (2005). Stability of Multiagent Systems With Time-Dependent Communication Links . IEEE Transactions on Automatic Control, 50 , Issue: 2.
- 68. Mark Newman (2010), Networks: An Introduction , Oxford University Press, USA.
- 69. J. Poterba and L. Summers (1988). Mean Reversion in Stock Returns: Evidence and Implications , Journal of Financial Economics 22, 27-60.
- 70. U. G. Rothblum and C. P. Tan (1985), Upper bounds on the maximum modulus of subdominant eigenvalues of nonnegative matrices , Linear Algebra Appl., 66 (1985), pp. 45-86.
- 71. E. Seneta (1981). Non - negative matrices and markov chains. Springer, 2nd edition, 1981.
- 72. H.A. Simon (1955). A behavioral model of rational choice , Quarterly Journal of Economics, volume 69 , number 1, pp. 99118.
- 73. H. A. Simon (1956). Rational Choice and the Structure of the Environment . Psychological Review 63 (2): 129138. doi:10.1037/h0042769
- 74. Jean-Jacques E. Slotine and Wei Wang. A Study of Synchronization and Group Cooperation Using Partial Contraction Theory. Cooperative Control. In Lecture Notes in Control and Information Science Volume 309 , 2005, pp. 207-228.
- 75. Joel Sobel (2000). Economists Models of Learning , Journal of Economic Theory 94(1): 241261.
- 76. Stefani S., Falbo P., & Felletti D. (2010). Commodity Trading. In R. Cont (ed.), Encyclopedia of Quantitative Finance (pp. 308-314). Chichester : John Wiley and Sons.
- 77. M. Stone. The opinion pool . The Annals of Mathematical Statistics, 32 (4):1339-1342, 1961.
- 78. James Surowiecki (2005). The Wisdom of Crowds . Anchor Books. pp. xv. ISBN 0-385-72170-6.
- 79. Thomas, W.I., and Dorothy Swaine Thomas. 1928. The Child in America: Behavior, Problems and Programs University of Chicago Press Programs. Knopf.
- 80. Alireza Tahbaz-Salehi and Ali Jadbabaie (2008), A Necessary and Sufficient Condition for Consensus Over Random Networks . IEEE Transactions on Automatic Control. Volume 53 , Issue 3.
- 81. Stanley Wasserman and Katherine Faust (1994), Social Network Analysis: Methods and Applications , Cambridge University Press. | null | [
"Marco D'Errico",
"Gulnur Muradoglu",
"Silvana Stefani",
"Giovanni Zambruno"
] | 2014-08-01T21:27:15+00:00 | 2014-08-01T21:27:15+00:00 | [
"q-fin.GN",
"physics.soc-ph"
] | Opinion Dynamics and Price Formation: a Nonlinear Network Model | Opinions and beliefs determine the evolution of social systems. This is of
particular interest in finance, as the increasing complexity of financial
systems is coupled with information overload. Opinion formation, therefore, is
not always the result of optimal information processing. On the contrary,
agents are boundedly rational and naturally tend to observe and imitate others
in order to gain further insights. Hence, a certain degree of interaction,
which can be envisioned as a network, occurs within the system. Opinions, the
interaction network and prices in financial markets are then heavily
intertwined and influence one another. We build on previous contributions on
adaptive systems, where agents have hetereogenous beliefs, and introduce a
dynamic confidence network that captures the interaction and shapes the opinion
patterns. The analytical framework we adopt for modeling the interaction is
rooted in the opinion dynamics problem. This will allow us to introduce a
nonlinear model where the confidence network, opinion dynamics and price
formation coevolve in time. A key aspect of the model is the classification of
agents according to their topological role in the network, therefore showing
that topology matters in determining how of opinions and prices will coevolve.
We illustrate the dynamics via simulations, discussing the stylized facts in
finance that the model is able to capture. Last, we propose an empirical
validation and calibration scheme that makes use of social network data. |
1408.0309v1 | ## Coarse-Grained Molecular Dynamics Simulations of Depletion-Induced Interactions for Soft Matter Systems
Tyler N. Shendruk, 1, ∗ Martin Bertrand, 2 James L. Harden, 2 Gary W. Slater, 2 and Hendrick W. de Haan 3
1 The Rudolf Peierls Centre for Theoretical Physics,
Department of Physics, Theoretical Physics, University of Oxford,
1 Keble Road, Oxford, OX1 3NP, United Kingdom
2 University of Ottawa, Department of Physics, 150 Louis-Pasteur, Ottawa, ON, K1N 6N5, Canada
3 University of Ontario Institute of Technology, Faculty of Science, 2000 Simcoe St. (Dated: August 5, 2014)
North, Oshawa, ON, Canada
Given the ubiquity of depletion effects in biological and other soft matter systems, it is desirable to have coarse-grained Molecular Dynamics simulation approaches appropriate for the study of complex systems. This paper examines the use of two common truncated Lennard-Jones (WCA) potentials to describe a pair of colloidal particles in a thermal bath of depletants. The shiftedWCA model is the steeper of the two repulsive potentials considered, while the combinatorial-WCA model is the softer. It is found that the depletion-induced well depth for the combinatorial-WCA model is significantly deeper than the shifted-WCA model because the resulting overlap of the colloids yields extra accessible volume for depletants. For both shifted- and combinatorial-WCA simulations, the second virial coefficients and pair potentials between colloids are demonstrated to be well approximated by the Morphometric Thermodynamics (MT) model. This agreement suggests that the presence of depletants can be accurately modelled in MD simulations by implicitly including them through simple, analytical MT forms for depletion-induced interactions. Although both WCA potentials are found to be effective generic coarse-grained simulation approaches for studying depletion effects in complicated soft matter systems, combinatorial-WCA is the more efficient approach as depletion effects are enhanced at lower depletant densities. The findings indicate that for soft matter systems that are better modelled by potentials with some compressibility, predictions from hard-sphere systems could greatly underestimate the magnitude of depletion effects at a given depletant density.
## 1. INTRODUCTION
form lock-and-key complexes [22-24] and to control particle motion and modify curvature in vesicles [25, 26].
Depletion forces of entropic origin are universal in soft matter systems, where they dictate the formation of a variety of soft solid phases [1-3]. In particular, for biological systems, these forces have been shown to play a prominent role in a myriad of phenomena [4, 5]. For example, we know depletion induced attractions promote fiber bundling [6], aggregation of red blood cells [7], and DNA collapse [8]. Reaction kinetics can also be influenced such as in DNA loop formation [5], protein folding [9, 10], and modified ligand-protein binding within nuclei [11]. Furthermore, these short-ranged depletion forces have been used to control biological systems, such as form active bundles [12]) and improved polymerase chain reactions [13].
Additionally, depletion forces occur in countless nonbiological soft matter contexts. They are often used to engineer novel microscopic systems by inducing ordered structures [14], as is done to align tobacco mosaic viruses [15-17]. Depletion forces can be significant enough to cause separation in colloid-polymer mixtures [3] and determine the phase diagrams of hardsphere mixtures [18, 19]. The presence of depletants can also alter the conformations of polymer chains [20, 21],
∗ Electronic address: [email protected]
Yet, despite their experimental significance, simulations of complex situations that include depletioninduced interactions due to a thermal bath of smaller depletant particles are relatively rare. Computational research primarily focuses on the nature of the depletion forces themselves. A rich literature on binary hardsphere fluids has benefited from Density Functional Theories (DFT) (see Refs. [19, 27]). The majority of the work using DFT has been focused on understanding the statistical state of hard-sphere mixtures [18, 28-36]. Likewise, Monte Carlo (MC) simulations have played an essential role in verifying various depletion theories [37-40]. The computational efficiency of MC simulations is continually improving [41, 42] and they are being applied to more exotic mixtures than binary hard-spheres [17, 43, 44] and material properties [45].
Since such numerical calculations are computationally expensive, they have primarily been limited to simple situations. Furthermore, it is not always obvious how they can be used in conjunction with the generic particlebased computational tools that soft matter physicists have found to be useful for studying complex microscopic matter, such as Molecular Dynamics (MD) [46]. In order to study the physical phenomena that the depletants induce in complex soft matter systems, coarse-grained computational models that can seamlessly couple to those used by the soft matter community are needed. Previ-
ous work has used hard-core Yukawa potentials to study how additional interactions between depletants can impact depletion-induced effects [47], while molecular simulations have been applied to better understand gelation in depletant-colloid systems [39] with large size ratios [48] and the effects of using polymers as depletants [49-51].
Kim and Szleifer [52] studied the attraction between two approaching polymer chains in a bath of depletants via MD simulations using a truncated Lennard-Jones potential, which is also called the Week-Chandler Andersen (WCA) potential. The interaction was characterized by calculation of the potential of mean force (PMF), but as the system was quite extensive in terms of the number of particles, only a small number of volume fractions of depletants could be studied. In this work, we focus on an elementary system: two colloids in a bath of depletants (Fig. 1). This allows us to explore depletant interactions across a wide range of volume fractions for two common WCA implementations. In addition, the results are compared to theoretical predictions for the depletion potentials, including the Morphometric Thermodynamics model [36, 53, 54].
Our work demonstrates that the details of the excluded volume interactions used in the MD simulations greatly impacts depletion-induced forces. Indeed, an appropriate choice of parameters can significantly increase the depth of the attractive well, which in turn decreases the second virial coefficient and effective volume of the colloids, in a predictable manner. As a result, simulations using 'soft' yet steep potentials can reveal important entropic effects at lower volume fractions of depletants than hard sphere simulations and thus reduce computational costs for future studies into depletion effects in complex soft matter systems. Further, for soft matter systems in which the most appropriate coarse-grained model allows for some softness in the interaction between two constituents, the magnitude of depletion effects would be significantly underestimated by hard sphere models and better approximated by potentials that are not infinitely steep at the nominal edge of the particle.
## 1.1. Depletion-Induced Attraction
Because they can produce 'order from disorder' [55], at first glance entropic forces can feel counter-intuitive. In fact, because they are purely entropic in nature, depletion interactions are universal in biology and common in soft matter systems. Before discussing the simulation techniques or results found in this work, let us first review depletion-induced pair potentials.
As shown in Fig. 1, consider two large, hard, colloidal spheres of radius R c and volume V c suspended in a thermal bath of small, hard depletant spheres of radius R d and volume V d with thermal energy k B T = β -1 . The large colloids each have a thin sheath of radius R = R c + R d from which the centres of the depletants are excluded. When the two large spheres are brought
FIG. 1: Two relatively large spherical colloidal particles (of radius R c ) in a solution of depletants (radius R d < R c ); each possesses an excluded volume of radius R = R c + R d . When the colloids are near (2 R c < r < 2 R ) an overlap volume V o exists causing the number of states available to the depletants to be greater than when the colloids are separate ( r ≥ 2 R ). This leads to an effective force that favours bringing the colloids together.
together such that their centre-to-centre separation r is less than 2 R , their excluded volumes overlap and the free space available to the depletants increases. This results in a free energy change as a function of available volume for the depletants giving rise to an osmotic pressure pushing the colloids together.
## 1.2. Asakura-Oosawa Pair Potential
Asakura and Oosawa were the first to estimate this depletion-induced pair potential [56]. They discussed the attraction in terms of an entropic, depletant-induced (and therefore statistical in nature), short-ranged pair potential u ρ ( dep , r ). The Asakura-Oosawa (AO) model has also been referred to as the penetrating hard spheres (PHS) model [1] because the depletants are assumed to be hard with respect to the large colloids but an ideal gas amongst themselves, which is only appropriate for small volume fractions of depletants. In the thermodynamic, dilute limit of depletants ( N d → ∞ while φ d glyph[lessmuch] 1), one can approximate the osmotic pressure to be given by a van't Hoff relation Π ≈ -( φ /V d d ) β -1 . Explicitly, the
-
d
FIG. 2: Depletion-induced attractive pair potential between two spherical, hard colloids for a variety of colloid-depletant ratios. The inset shows the entropic pair potentials for various hard-sphere colloid and depletant sizes, while the main figure shifts these curves. The main abscissa is a shifted and scaled by the centre-to-centre separation emphasizing differnces in well depths. The hard-sphere Aakura-Oosawa (A0) model (Eq. (1) and Eq. (2)) is shown as dashed lines, while the Morphometric Thermodynamics (MT) predictions are shown as solid lines. Scaling the pair potentials removes φ d dependence from the AO model but not the MT model. The MT curves shown are for the example volume fraction of φ d = 0 2 .
AO entropic depletion pair potential is
$$\text{AO entropic depletion pair potential is} & & \text{ due to rep} \\ \beta u _ { \text{AO} } = \begin{cases} - \left ( \frac { \mathcal { R } } { R _ { d } } \right ) ^ { 3 } \left [ 1 - \frac { 3 } { 4 } \frac { r } { \mathcal { R } } + \frac { 1 } { 1 6 } \left ( \frac { r } { \mathcal { R } } \right ) ^ { 3 } \right ] \phi _ { d } & 2 R _ { c } \leq r \leq 2 \mathcal { R } \, & \text{Compu!} \\ r > 2 \mathcal { R } & \text{$\text{$\text{$\text{$\text{$\text{$by}$ and}$} } \\ ( 1 ) & \text{between $\cdot$} } \\ \text{Note that the AO model (and MT model, to be dis-} \\ \text{$therely s1$} \\ \text{$therely s2$} \end{cases}$$
Note that the AO model (and MT model, to be discussed below) is generally defined for hard spheres and must be supplemented with a steric-repulsion condition for r < 2 R c . For hard spheres of radius R c ,
$$U _ { \text{HS} } = \begin{cases} \infty & r < 2 R _ { \text{c} } \\ 0 & r \geq 2 R _ { \text{c} }. \end{cases} \quad \begin{pmatrix} \P a \P a \\ ( 2 ) & \P ( \text{M} ) \\ \end{pmatrix}$$
Hence, the total interaction is u AO + U HS in this definition. As will be shown, we are free to relax this constraint when the particles are not strictly hard spheres.
The AO pair potential (Fig. 2; dashed lines) is characterized by four points: (i) the potential is strictly attractive; (ii) the range is set by the depletant size R d ; (iii) the minimum occurs when the large colloids are in contact
$$\beta u _ { A O } ^ { \min } = - \left [ 1 + \frac { 3 } { 2 } \frac { R _ { c } } { R _ { d } } \right ] \phi _ { d }, \quad \ ( 3 ) \quad \ u _ { \ } i s \\ \i _ { \dots \dots \dots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots$$
which deepens as the size ratio R /R c d is increased (though is not zero in the limit R /R c d → 1 [57]); (iv)
via osmotic pressure, the volume fraction of depletants and temperature both scale the well depth.
The AO model is beautifully simple; however, it has two major limitations:
- · Depletants must be small R d glyph[lessmuch] R c .
- · The volume fraction must be small φ d glyph[lessmuch] 1.
The essential conclusion is that in order to consider the many interesting situations that do not meet these conditions, we must move beyond the AO limit.
## 1.3. Morphometric Pair Potential
To move beyond the φ d glyph[lessmuch] 1 limit, the free energy cost of inserting the large colloids must be greater than it is in the PHS assumption. Simply replacing the PHS osmotic pressure with a form that includes higher virial coefficients for the hard sphere gas of depletants allows the AO model to predict colloid pressures at higher φ d but is not sufficient to predict the form of the depletion pair potential [58, 59]. The missing ingredient is the realization that colloids that are close to one another mutually perturb the statistical arrangement of their surrounding shells of depletants and cause an inhomogeneous probability distribution of depletants. This results in the pair potential oscillating between attractive and repulsive as a function of separation [59]. The well effectively becomes deeper as it is now the difference between the maximum due to repulsion and the minimum due to attraction, but the minimum itself remains mostly unaffected.
Computationally intensive numerical techniques such as DFT can determine the local depletant particle density and so find the resulting entropic pair potential between colloids. Such techniques have proven extremely successful in exploring non-additive effects [30] but are not straight-forward analytically simple models. One conceptually simple model that can reproduce the first repulsive/anti-correlation component of the pair potential is called Morphometric Thermodynamics (MT) [36, 53, 54].
The MT model discusses free energy in terms of the geometric shapes involved and their thermodynamic conjugates. In the AO model, only the overlap volume V o and osmotic pressure Π associated with it are considered. Just as the osmotic pressure is a thermodynamic force that arises because of restrictions on the placement of depletants within the overlap volume, an entropic surface tension arises due to the cost of disallowing depletants to be at certain points on the shell of depletants surrounding each colloid due to the presence of the other colloid (Fig. 1). The entropic interaction pair potential u is then modelled as arising from changes to the accessible volume V o , to the surface area A restriction with resulting entropic surface tension γ and to the Gaussian curvatures of the overlap surface C 1 and C 2 with corre-
sponding entropic bending rigidities κ 1 and κ 2 :
$$u _ { \text{MT} } = \Pi V _ { o } + \gamma A + \kappa _ { 1 } C _ { 1 } + \kappa _ { 2 } C _ { 2 } \quad \ \ ( 4 ) \quad \ \widehat { \xi } ^ { \dagger }$$
where the geometric terms are functions of the colloidcolloid separation while the prefactors depend on the volume fraction of depletants (listed in Appendix B). As shown in Fig. 2 (solid lines), the MT framework is able to produce the attractive well and also the primary repulsive barrier, while the AO model only predicts the attractive well. Correlations and anti-correlations beyond the primary contributions cannot be reproduced within the MT model.
## 2. SIMULATION METHODS
Simplified analytical models such as the AO or MT potentials are appealing for straightforward systems but coarse-grained simulations offer a path to studying more complicated soft matter systems. In order to computationally study depletion effects via Molecular Dynamics simulations, a generic steep repulsive potential to produce hard steric effects between colloids and depletants is needed. To coarse-grain the system, we neglect all hydrodynamic interactions that may propagate through the fluid medium [60, 61] and any possible non-equilibrium effects [62]. We consider the solvent to be simply a dissipative thermal bath with Langevin Dynamics. Both the colloidal particles and the depletants are modelled as quasi-'hard spheres' in the framework of coarse-grained MD.
Because depletion forces arise due to excluded volumes between particles that are inaccessible to the smaller depletant particles, any repulsive potential can technically be used to represent the steric repulsion (though as we shall see, the details can have a significant impact on the magnitude of the effective interactions). In typical coarse-grained MD simulations, this is done using the Lennard-Jones (LJ) pairwise potential, which includes two terms, a steric repulsion that goes as r -12 and a dispersion-type interaction that goes as r -6 [46]. In order to neglect attractive dispersion-type attraction, the LJ potential is commonly truncated at its minimum and shifted to positive values, as will be explained in § 2.1 and § 2.2. This truncated LJ potential is referred to as the Weeks-Chandler-Andersen (WCA) potential. This is a simple and commonly employed potential; however, using a potential that is not infinitely steep alters the attractive entropic interactions and they must be carefully characterized. It will be seen in this work that an appropriate choice of parameters can significantly increase the depth of the depletion induced potential well, which in turn decreases the second virial coefficient and the effective volume of the colloids. We consider two versions of the WCA potential as coarse-grained generic simulation models for depletion effects. For both approaches, the characteristic energy for all interactions is glyph[epsilon1] and the
6 0
FIG. 3: Comparison of the shifted-WCA (blue) and combinatorial-WCA (red) potentials for R c = 2 5 . σ , ε = 1 k B T and ∆ ij = 4 σ for the shifted-WCA case. The combinatorialWCA extends further and diverges at a small separation and is thus the 'softer' of the two. However, both potentials are hard in that they rise as 1 /r 12 .
characteristic length scale is σ - which is equal to the diameter of the depletants.
## 2.1. Shifted WCA
The shifted-WCA model (sWCA) associates the same repulsive potential to each spherical particle but shifts the barrier depending on their radii. For a particle of nominal radius R i and one of R j , it is given by
$$\text{typical} \\ \text{ng the} \\ \text{icludes} \\ \text{and a} \\ \text{In or} - \\ \text{in, the} \\ \text{in and} \\ \mathfrak { I } \, \S \, 2. 1 \\ \mathfrak { I } \, \text{ to as} \quad. \quad. \quad.$$
where r ij is the centre-to-centre separation between particles i and j , ε is the depth of the potential well, σ is the nominal lengthscale of the potential, ∆ ij = R i + R j -σ is a shift of the potential and r cut ij = 2 1 / 6 σ +∆ ij is the cutoff distance. The repulsive potential thus rises in the same rapid manner, over the same distance of 2 2 / 6 σ , for all pairs regardless of the size of the particles. Here we have chosen σ and ε in Eq. (5) to explicitly be the characteristic values but one could conceivably chose some other distance and/or energy. As shown in Fig. 3, the shiftedWCA potential U sWCA rises from zero at r cut ij , crosses U sWCA = 1 ε at r ij = R i + R j , and diverges as r → ∆ . ij
## 2.2. Combinatorial WCA
On the other hand, the combinatorial-WCA model (cWCA) does not shift the WCA repulsion but rather sets an effective size between two MD particles. It is defined by where σ ij is no longer constant but instead is given by R i + R j and r cut ij = 2 1 / 6 σ . The resulting repulsion between particles U cWCA ( r ij ) is still hard - the potential still steeply rises as ∼ r -12 ij . However, the combinatorialWCA repulsion between two colloids is slightly softer than its shifted-WCA counterpart. As shown in Fig. 3, the repulsive potential extends further and does not rise as sharply. The same is true for colloid-depletant interactions. Since σ is equal to the size of the depletants, 2 R d , the depletant-depletant interactions are equivalent in the sWCA and cWCA potentials.
$$& \text{defined by} \\ & U = \begin{cases} 4 \varepsilon \left [ \left ( \frac { \sigma _ { i j } } { r _ { i j } } \right ) ^ { 1 2 } - \left ( \frac { \sigma _ { i j } } { r _ { i j } } \right ) ^ { 6 } \end{cases} \right ] + \varepsilon \quad r _ { i j } < r _ { i j } ^ { c u t } \quad & \text{$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$\quad$
& \text{where $\sigma_{ij}$ is no longer constant but instead is given by} \quad & \text{and} \quad r _ { i j } ^ { c u t } = 2 ^ { 1 / 6 } \sigma. \text{ The resulting repulsion be-} \quad & \text{and} \quad r _ { i j } ^ { c u t } = 2 ^ { 1 / 6 } \sigma. \text{ The resulting repulsion be-} \quad & \text{and} \quad r _ { i j } ^ { c u t } = 2 ^ { 1 / 6 } \sigma. \text{ The resulting repulsion be-} \quad & \text{and} \quad r _ { i j } ^ { c u t } \end{cases}$$
## 2.3. Effective Colloid Size
In using continuous potentials instead of perfectly hard spheres, the question arises as to what the effective size of the colloids are. Not only are the potentials not infinite at the nominal radius, but the colloid-colloid interaction extends beyond the 2 R c (Fig. 3). The second virial coefficient for a pairwise-interacting solution of colloids can be used to define an effective volume (and hence radius) for the colloids defined by either sWCA and cWCA:
Here U r ( ) is the potential that defines the colloids, which for hard spheres is given by Eq. (2). Hence for hardspheres, the second virial coefficient is 4 times the volume of the hard sphere B HS c 2 , = 4 ( 4 3 πR 3 C ) = 4 V HS .
$$r & \text{ the colloids defined by either swUCA and } c W U A \colon \\ B _ { c, 2 } = - 2 \pi \int _ { 0 } ^ { \infty } \left ( e ^ { - \beta U ( r ) } - 1 \right ) r ^ { 2 } d r. \quad ( 7 ) \quad T _ { c } \\ \text{ere} \, U \left ( r \right ) & \text{ is the potential that defines the colloids, which} \quad \text{coll} _ { c }$$
For the sWCA potential, we use Eq. (5) for colloids of nominal size 2 R c in Eq. (7) to solve for B c 2 , . Defining the effective volume of the sWCA colloids as V sWCA = B c 2 , / 4, we can calculate V sWCA /V HS , which depends on the nominal radius R c . For the case of R c = 2 5 . σ as studied in this work, we find that
$$\frac { V _ { s W C A } } { V _ { H S } } = 1. 0 1 0. \quad \ \ \ ( 8 ) \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
Hence, using the sWCA potential to define a colloid with a nominal diameter of 5 σ results in a colloid that has a 1% larger volume than the hard sphere equivalent. This corresponds to an effective radius of R sWCA = 2 508. .
Following a similar procedure for the cWCA potential but using Eq. (6), we obtain
$$\frac { V _ { s W C A } } { V _ { H S } } = 1. 0 5 3. \quad \ \ ( 9 ) \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
The cWCA potential thus results in a colloid that has a 5.3% larger volume than the hard sphere result, corresponding to an effective radius of R cWCA = 2 543. . As opposed to the sWCA result, the length of the cWCA potential scales with the nominal size and thus this result is independent of R c .
The R sWCA and R cWCA values given above are used in theoretical calculations in the rest of this work. Although the corrections to the nominal radii appear to be small, they do have a significant effect on the results and yield much better agreement between the simulations and the theory for the pair potentials, well depths, and B c 2 , values. It is important to keep in mind that both of the WCA potentials are 'hard' in that they are both short-ranged, rapidly rising, repulsive potentials.
## 2.4. Net potential between two colloids
The net pair potential between two colloids is simply the sum of the excluded volume potentials U (Eq. (5) or (6)) and the entropic component u :
$$W = U \left ( r _ { i j }, R _ { c }, \sigma \right ) + u \left ( \phi _ { d }, r _ { i j }, R _ { c }, R _ { d } \right ). \quad ( 1 0 )$$
In this notation, W indicates the net potential, U denotes the repulsive, excluded volume potentials (WCA), and u specifies the entropic, depletion potentials. When the WCA potentials are subtracted off the net interaction potential then only the depletion-induced component remains.
## 3. SIMULATIONS OF DEPLETANT-INDUCED PAIR-POTENTIAL
To measure the entropic pair potential between two colloids, the mean force on each colloidal particle held at a fixed separation is measured. Integrating over the entire range of simulated separations produces the potential of mean force W sim between two colloidal particles.
## 3.1. Shifted-WCA
The net pair potential W sim sWCA ( r ), as measured by force integration, is shown in Fig. 4. The strong, shortranged, shifted-WCA repulsion U sWCA is clearly visible at small separations and the potential diverges at r = ∆ ij = 4 σ for these simulations. For clarity, The main figure shows low to moderately volume fractions ( φ d ≤ 0 25) and the inset shows the highest volume frac-. tions considered here ( φ d = 0 3 and 0 35). . . Subtracting U sWCA off leaves only the depletion-induced component u sWCA (Fig. 5). Persistently negative potentials for large centre-to-centre separations suggests that the force integrated results from simulations underpredict the net pair potentials at the highest volume fractions of depletants considered ( φ d > . 0 25).
FIG. 4: Simulated net pair potentials W sim sWCA from integrated force measurements of shifted-WCA beads of size R c = 2 5 . σ in baths of various volume fractions of shifted-WCA depletants of size R d = 0 5 . σ . While the main figure shows low to moderate values, the inset shows the high volume fraction cases. Solid lines are the sum of the entropic pair potential predicted by the Morphometric Thermodynamics (MT) model and the shifted-WCA repulsion, u MT + U sWCA .
At low volume fractions, the simulated pair potentials are well approximated by both the AO and the MT models but as φ d glyph[greaterorsimilar] 0 1 the AO model begins to deviate from . the measured curves (Fig. 5; inset). On the other hand, the MT model continues to capture the behaviour of the primary attractive well and repulsive barrier (the region within approximately two depletants from the colloid's surface: 2 R c ≤ r glyph[lessorsimilar] 2 [ R c +2 R d ]). Of course, the MT model does not capture the secondary features in Fig. 5 because current functionals are not applicable beyond 2 R = 2( R c + R d ) [53].
Although, the AO model does not well represent the pair potential as a function of separation, Fig. 5 shows that it does estimate the contact energy well, i.e. u min sWCA ≈ u min AO . This can be seen quantitatively in Fig. 6. In fact, Fig. 6 shows that the linear AO minimum (Eq. (3)) agrees with the simulated min[ W sim sWCA ].
To better understand these predictions, we theoretically combine the MT model for the depletion potential and the sWCA excluded volume potential: u MT + U sWCA . This approach successfully replicates the net pair potential as compared to simulation for φ d glyph[lessorsimilar] 0 25 . (Fig. 4; main). This is because the depletion potential is well approximated by the MT model for all but the highest depletant volume fractions (Fig. 5). For the highest volume fractions considered (Fig. 4; inset), the MT model appears to overpredict the primary barrier height and underpredict the primary well depth for sWCA. The minima of the u MT + U sWCA data are shown in Fig. 6 as solid lines. While good agreement is found up to
FIG. 5: Simulated depletant-induced entropic pair potentials u sWCA from integrated force measurements of shifted-WCA beads of size R /R c d = 5. Solid lines show the induced pair potentials predicted by the Morphometric Thermodynamics (MT) model, while dashed lines in the inset show the AakuraOosawa (AO) model. These theoretical curves are zero for r > 2 R = 6 σ and infinite for r < 2 R c = 5 σ .
φ = 0 25, . min[ u MT + U sWCA ] actually begins to curve upwards thereafter. This trend is an artifact of the MT model where at high volume fractions, the first repulsive peak is exaggerated - which drags the curve upwards at r ≈ 2 R c and thus increases the minimum [54]. Due to this artifact, even the linear AO model does a better job than the MT model at predicting the minimum in this regime.
To correct this, the difference between the minimum of the contact attraction and the maximum of the anticorrelation barrier can be calculated. This calculation can be done for both the simulations (Fig. 6; red squares) and the u MT + U sWCA potential (this is not possible for the AO potential as no repulsive barrier is predicted). Excellent agreement is now found between the simulation results and the theoretical u MT + U sWCA potential across all studied volume fractions. This suggests that MT's overprediction of the primary barrier and underprediction of the primary well depth cancel.
The well depth competes with thermal energy in determining if colloids prefer to agglomorate in clusters or explore the entire available volume. Qualitatively speaking since the well depth W min sWCA in Fig. 6 is only a few k B T even as φ d → 0 4, . we expect that extremely high number densities of depletants would be required to overcome thermal jostling. This will be further quantified when we consider the second virial coefficients of colloids in thermal baths of depletants ( § 4.1).
d
FIG. 6: Measures of the well depth of the pair potential as a function of depletant volume fraction for shifted-WCA particles of size R c = 2 5 . σ and depletants of size R d = 0 5 . σ . The minimum from simulations ( W sWCA ; black circles) is compared to the Aakura-Oosawa (A0) value ( u min AO from Eq. (3); dashed line) and the theoretical minima of the Morphometric Thermodynamics (MT) model plus the shifted-WCA potential (min[ u MT + U sWCA ]; solid black line). The effective well depths (primary minimum minus primary maximum) from simulations (∆ W sWCA ; red squares) and the theoretical net value (∆[ u MT + U sWCA ]; solid red line) are also shown.
## 3.2. Combinatorial-WCA
The depletant-induced pair potential is measured in the same manner as in § 3.1 for cWCA interactions; however, separations smaller than 2 R c are allowed by the cWCA repulsion and must now be considered as seen in Fig. 7. The competition between U cWCA and the entropic pair potential u cWCA results in a minimum in the net interaction W cWCA for all volume fractions. A comparison between Fig. 4 and Fig. 7 shows that the well is much deeper in the combinatorial-WCA model than for the shifted-WCA. This is because of the additional overlap that can occur. Since the U cWCA repulsion rises less dramatically and does not have an infinitely-hard repulsive core, the combinatorial-WCA wells are broader, extending to smaller separations. For high volume fractions, the combinatorial-WCA repulsion U cWCA does not dominate over the depletion-induced pair potential u cWCA until much smaller separations than in the shifted-WCA model. This can be seen in Fig. 8, which displays the location of the minimum of the net potentials W cWCA and W sWCA as measured in simulations as a function of volume fraction. Excellent agreement is again found between theory (MT) and simulation.
At low volume fraction, colloids are actually more likely to be found further apart for cWCA than sWCA since the cut-off for the former extends further. However,
FIG. 7: Simulated net pair potentials W cWCA from integrated force measurements of combinatorial-WCA beads of size R c = 2 5 . σ in baths of various volume fractions of combinatorial-WCA depletants of size R d = 0 5 . σ . The inset shows the high volume fraction cases and the main figure shows the low to moderate values. Solid lines are the sum of the MT model for the depletion potential and the combinatorial-WCA excluded volume potential, u MT + U cWCA .
due to the relative softness of u cWCA , the minimum in the potential decreases at a much faster rate and, at high volume fractions, colloids defined by cWCA are most likely found to be significantly closer together than for sWCA. It is interesting to note that the dependence of r min on φ d is approximately linear for both WCA potentials, with a cross-over around φ d ≈ 0 14 and . r min ≈ 5 1. .
Once again the WCA potential can be subtracted from the total potential W in order to produce the depletantinduced pair potential u cWCA (Fig. 9). The depletioninduced pair potential continues into separations r < 2 R c . As discussed, the definitions of the AO and MT implicitly have infinitely hard cores at 2 R c . However, this is an imposed restriction and we can remove it without modifying the formalism.
As seen in Fig. 9 (inset), the AO potential crosses the simulated u potential quite near r ≈ 2 R c but when significant overlap occurs the AO potential u AO cannot reproduce the measured depletion-induced pair potential. On the other hand, allowing the MT pair potential u MT to extend to smaller separations does a remarkable job of reproducing the depletant-induced component of the pair potential except at the highest volume fractions of depletants considered (Fig. 7; inset) at which point the primary repulsive barrier is overpredicted. For this reason, summing u MT and U cWCA at low to moderate volume fractions once again reproduces the net interaction potentials.
Because the potentials are much softer, greater overlap
d
FIG. 8: Position of the minimum in the net potential measured from simulations for both sWCA (black diamonds) and cWCA (red triangles). The prediction from the theoretical potentials u MT + U sWCA (solid black line) and u MT + U cWCA (solid red line) are also included.
between the colloids is allowed. This in turn yields a significant boost in the volume accessible to the depletants and thus a deep attractive well. The resulting minima (Fig. 10) are much deeper at high volume fractions than they were for the sWCA potentials (Fig. 6), though not at smaller volume fractions of φ d glyph[lessorsimilar] 0 2. . In fact, because the cWCA potential extends further (Fig. 3), the AO model is not even accurate at low volume fractions. At both high and low depletant volume fractions, the AO model for hard-spheres is not sufficient for these cWCA simulations. However, since summing the MT model and the cWCA potential reproduces the pair potentials so well for φ d glyph[lessorsimilar] 0 25 (Fig. 7), it also predicts the behaviour of . the minima and the well depths as functions of φ d quite well (Fig. 10). At the highest volume fractions considered ( φ d glyph[greaterorsimilar] 0 30), . the MT model once again begins to fail. While in the sWCA case, the primary repulsion was overpredicted but the primary well depth was underpredicted (Fig. 5 and Fig. 6), in the cWCA case both the repulsive barrier height and the well depth are overpredicted (Fig. 9 and Fig. 10).
## 4. VIRIAL COEFFICIENTS: IMPLICIT DEPLETANTS
Considering a reduced system of only two colloids in a bath of depletants has allowed us to characterize the depletant pair potential in detail. However, systems of interest generally consist of many large particles in a bath of depletants - a much more computationally expensive system to study. To circumvent this, the success of the MT potential in replicating the depletant pair potential
FIG. 9: Simulated depletant-induced entropic pair potentials u cWCA as a function of centre-to-centre separation for R /R c d = 5. Solid lines show the depletion-induced pair potentials predicted by the Morphometric Thermodynamics (MT) model, while dashed lines in the inset show the AakuraOosawa (AO) model. The AO model fails to accurately predict the entropic attraction for separations r < 2 R c but the MT model remains highly accurate.
indicates that it could be used to include the effects of the depletants implicitly . In this approach, explicit depletants would not be included in the system but their effects would be included by imposing the MT potential between all pairs of particles. We use the second virial coefficient as a quantitative measure to predict the efficacy of such an approach.
## 4.1. Second Virial Coefficient
Recall that the second virial coefficient was used to calculate the effective size of the colloids according to the potentials that define them. The effective size as modified by the presence of the depletants can be calculated by a similar method where the potential u r ( ) in Eq. (7) can now be: i) the simulated pair potentials W sim sWCA and W sim cWCA , ii) the sum of u MT and U (either U sWCA or U cWCA ), iii) the net AO potential u AO + U HS .
Although virial coefficient expansions of the equation of state breakdown in non-thermalized systems, predictions of B c 2 , < 0 suggests that attraction dominates over thermal jostling and so demixing of colloidal and depletant particles will occur [54, 63]. In other words, the colloids are expected to agglomorate for negative enough B c 2 , values. The depletion-induced interactions demand a correction such that the virial coefficient is reduced from the hard-sphere value and so we shall discuss the reduced second virial coefficient B ∗ c 2 , = B c 2 , /V c .
FIG. 10: The same as Fig. 6 but for the combinatorial-WCA simulation model. Unlike the shifted-WCA case (Fig. 6), the minimum of the Aakura-Oosawa (AO; dashed line) model does not well-approximate the value from simulations at any volume fraction. The theoretical minimum from the MT model plus the combinatorial-WCA potential (solid black line) and the effective well depth are both found to be in good agreement with the corresponding simulation result.
## 4.1.1. Simulations
Consider first the harder, shifted-WCA simulations (Fig. 11; black diamonds). The simulated shifted-WCA second virial coefficient is linear with volume fraction of depletants until approximately φ d ≈ 0 2 after . which point the curvature increases and B ∗ c 2 , begins to drop more rapidly. The second virial coefficient becomes negative somewhere in the range of φ d between 0 25 and 0 3. . .
The results for the combinatorial-WCA simulations (Fig. 11; red triangles) are similar to the sWCA results at low volume fractions. However, the combinatorialWCA B ∗ c 2 , falls quite rapidly. It becomes negative around φ d = 0 2 and drops to dramatically more negative values . at high volume fraction.
## 4.1.2. AO Approximation: φ d glyph[lessmuch] 1
Assuming that the colloids are hard and φ d glyph[lessmuch] 1 simplifies matters immensely. This is the AO model and, as shown in Appendix A, the depletant-induced attractions appear as a correction on the hard-sphere result, specifically taking the form
$$\frac { B _ { c, 2 } ^ { * } } { 4 } \approx 1 - g \phi _ { d }, \text{ \quad \ \ } ( 1 1 ) \text{ \ \ } \text{$\stackrel{----} { \text{bo} } \\ \text{At} \\ \text{um} \text{ }$$
where g ( R /R d c ) is a third order polynomial given by equation Eq. (A2). The AO approximation (Eq. (11))
FIG. 11: Reduced second virial coefficient B ∗ c 2 , = B c 2 , /V c for colloidal particles with R /R c d = 5 as a function of φ d . Results for the linear Aakura-Oosawa (AO) model is shown as a dashed line. The MT based predictions are also included: u MT + U sWCA (solid black line) and u MT + U cWCA (solid red line)
shows that, in this limit, the reduced second virial coefficient decreases as the volume fraction of depletants is increased, reflecting the deepening of the implicit depletion attraction. As is expected for the AO model, the correction from the hard-sphere value is linear in φ d . The linear AO prediction compares well to both the shiftedand the combinatorial-WCA simulations for R /R c d = 5 when the volume fraction of depletants is low (Fig. 11). The AO approximation remains relatively accurate for larger values of φ d for sWCA. For φ d glyph[lessorsimilar] 0 15, . the approximation predicts the second virial coefficient of the shifted-WCA simulations quite well. However, as it must, the AO approximation for B ∗ c 2 , fails at high volume fractions and does not cross to zero until φ d = 0 525 and . 0 527 for sWCA and cWCA, respectively. .
## 4.1.3. MT plus WCA Approximation
While discussing the simulation models, it was shown that summing the MT depletant potential and the WCA excluded volume interactions successfully predicts the total pair potentials. These theoretical pair potentials can be integrated over to predict the second virial coefficient as a function of volume fraction of depletants.
The result for u MT + U sWCA is shown as a solid black line in Fig. 11. The agreement between the AO curve and both simulation results is reasonable until φ d ≈ 0 20. . At this point, B ∗ c 2 , for u MT + U sWCA begins to curve upwards. This again is an artifact of the exaggeration of the repulsive barrier in the MT model at high volume fractions. On the other hand, the B ∗ c 2 , for the cWCA
simulations behaves physically, beginning to drop more rapidly at this point.
At low volume fractions, the result for u MT + U cWCA (solid red line) agrees with the data and in fact all other curves. However, as φ d increases, the B ∗ c 2 , for u MT + U cWCA begins to drop quite dramatically. The results are in remarkable agreement with the values derived from the simulations. Note that since the cWCA potential allows for the position of the minimum to be significantly less than 2 R c and for correspondingly deeper wells, the exaggeration of the repulsive barrier at higher volume fractions has a much smaller impact on the calculation of B ∗ c 2 , .
## 4.1.4. Escape Time
The second virial coefficient represents two-body interactions and is often conveniently interpreted as an effective size. To give an alternative physical interpretation that is appropriate when the second virial coefficient is a large, negative number, a set of simulations was performed in which two colloids are initially in contact. Setting R /R c d = 5, the volume fraction of depletants was varied between 0 and 0.35. Since the depletion force increases with the volume fraction, the particles are held together more tightly with increasing φ d . To quantify this, the average time it takes for the particles to separate was measured as an escape time, τ esc . The colloids are considered to have escaped the depletion potential well when their centre-to-centre distance is larger than 2( R c + R d ).
As shown in Fig. 12, the τ esc curve mirrors the B ∗ c 2 , results. When the volume fraction of depletants is large the primary attractive well is deep and the primary repulsive barrier is high (Fig. 10) such that colloids that are in contact must diffusively overcome the effective barrier of height ∆ W in order to escape contact. In this way, reaching r = 2 R is a Kramers' process and τ esc ∝ e β ∆ W [64, 65]. As the volume fraction of depletants is increased, the effective barrier height becomes greater (Fig. 10), the escape time rises rapidly as seen in Fig. 12 and the colloids effectively become stuck together. Concurrently, the second virial coefficient B ∗ c 2 , is plummeting to large negative values reflecting the strong attractive interactions as the volume fraction increases. The pair potentials (Fig. 7) demonstrate that the primary attractive well and repulsive barrier dominate over the secondary correlations and are indeed the significant contribution to the second virial coefficient. The inset to Fig. 12 shows that B ∗ c 2 , ∼ e β ∆ W - as for τ esc - when the effective barrier is greater than the thermal energy β ∆ W > 1. The relationship τ esc ∼ B ∗ c 2 , suggests a simple physical picture: when primary barrier is large the magnitude of B ∗ c 2 , reflects the relative amount of time that colloids are held together by depletant forces τ esc .
FIG. 12: Overlay of the escape time τ esc onto the reduced second virial coefficient B ∗ c 2 , = B c 2 , /V c (triangles denoted cWCA simulations and the solid red line presents the corresponding MT-based predictions) as a function of φ d . The right-hand vertical axis corresponds to the escape time in simulation units. The inset shows the reduced second virial coefficient as an exponential function of the effective well depth, ∆ W .
## 5. CONCLUDING REMARKS
In order to explore coarse-grained methodologies for modelling depletion interactions in soft matter systems, we charaterized the use of two types of truncated Lennard-Jones (WCA) potentials in Molecular Dynamics simulations of large colloids immersed in a bath of smaller depletants.
The shifted-WCA model is the 'harder' of the two models while the combinatorial-WCA potential both extends further and rise less steeply - and is thus 'softer' (while still being a hard potential). We find that this seemingly small difference between the excluded volume potentials has a dramatic impact on the depletion effects that arise in the system; the well depths are deeper, the minimum separations are closer, and the second virial coefficients are much lower for cWCA than sWCA. Hence, modelling the system with a cWCA approach yields significantly enhanced depletion effects for a fixed volume fraction of depletants.
Comparison was also made between two simple theories for the depletion potentials. The simplest approach, the Asakura-Oosawa model for penetrating hardsphere depletants, was found to give good results in the limit of low volume fractions. However, as φ d increased, this agreement soon broke down for both WCA potentials. The geometric-based Morphometric Thermodynamics model not only reproduces the minimum in the depletion potential, but also the first repulsive barrier. Adding u MT to either U sWCA or U cWCA yielded excellent agreement for the net pair potentials, the depletion potentials, and the effective well depths at low and mod-
erate volume fractions ( φ d glyph[lessorsimilar] 0 25) but begins to fail for . the highest volume fractions considered here. However, the second virial coefficient calculated from u MT + U cWCA closely matched the cWCA simulation results across all volume fractions, although the predictions for sWCA exhibited known artifacts at these size ratios. This result suggests that using the MT depletion potential along side the cWCA excluded volume interaction may be an effective model for implicitly incorporating depletion effects into future coarse-grained MD simulations.
The relative softness of cWCA significantly enhances depletion effects and thus presents a computationally efficient simulation approach since accessible volume fractions generate large primary attractive wells and repulsive barriers. Moreover, not only does the use of WCA potentials allow for MD-based exploration of depletant effects but also is likely to yield more physical results for a host of soft matter systems in which interactions are softer than the hard-sphere model. In many soft matter systems the effective excluded volume interaction has a softness associated with it; examples include polymer coated colloids, solutes that are compressible (e.g. micro or nanogels, dendrimers), charged objects in an electrolyte solution where the interaction is governed by overlapping Debye layers or easily deformed materials that can fit together more effectively (e.g., protein folding, chromosome collapse [66]). The results from hard
- [1] H. Lekkerkerker and R. Tuinier. Colloids and the Depletion Interaction . Springer, 2011.
- [2] J. Bibette, F. Leal Calderon, and P. Poulin. Emulsions: basic principles. Reports on Progress in Physics , 62(6):969, 1999.
- [3] R. Tuinier, J. Rieger, and C.G. de Kruif. Depletioninduced phase separation in colloidpolymer mixtures. Advances in Colloid and Interface Science , 103(1):1 - 31, 2003.
- [4] D. Marenduzzo, K. Finan, and P. R. Cook. The depletion attraction: an underappreciated force driving cellular organization. The Journal of Cell Biology , 175(5):681-686, 2006.
- [5] D. Marenduzzo, C. Micheletti, and P. R. Cook. Entropydriven genome organization. Biophysical Journal , 90(10):3712 - 3721, 2006.
- [6] M. Hosek and J. X. Tang. Polymer-induced bundling of f actin and the depletion force. Physical Review E , 69:051907, May 2004.
- [7] P. Steffen, C. Verdier, and C. Wagner. Quantification of depletion-induced adhesion of red blood cells. Physical Review Letters , 110:018102, Jan 2013.
- [8] J. Pelletier, K. Halvorsen, B.-Y. Ha, R. Paparcone, S. J. Sandler, C. L. Woldringh, W. P. Wong, and S. Jun. Physical manipulation of the escherichia coli chromosome reveals its soft nature. Proceedings of the National Academy of Sciences , 2012.
- [9] Y. Zhou and C. K. Hall. Solute excluded-volume effects on the stability of globular proteins: A statistical thermodynamic theory. Biopolymers , 38(2):273-284, 1996.
sphere calculations thus represent the lower bound for estimating depletion effects at a given volume fraction and may significantly underestimate their impact for such soft matter systems. While the correct modelling will be system specific, this possibility of underestimating depletant effects is important when considering predictions from hard sphere models. Our results have shown that even a small amount of softness leading to a marginal amount of overlap can result in a dramatic increase in the depletion potential, resulting in 'soft' colloids bound together by depletion-induced attraction that would be insufficient for 'harder' colloids.
## Acknowledgements
We gratefully acknowledge support through NSERC Discovery Grants to G.W.S. and J.L.H. and EMBO funding to T.N.S (ALTF181-2013). High performance computational resources were graciously provided by Prof. Laura Ramuno at the University of Ottawa and Sharcnet.
## References
- [10] Y. Snir and R. D. Kamien. Entropically driven helix formation. Science , 307(5712):1067, 2005.
- [11] K. Richter, M. Nessling, and P. Lichter. Macromolecular crowding and its potential impact on nuclear function. Biochimica et Biophysica Acta (BBA) - Molecular Cell Research , 1783(11):2100 - 2107, 2008.
- [12] T. Sanchez, D. T. N. Chen, S. J. DeCamp, M. Heymann, and Z. Dogic. Spontaneous motion in hierarchically assembled active matter. Nature , 491(7424):431-434, 2012.
- [13] R. R. Lareu, K. S. Harve, and M. Raghunath. Emulating a crowded intracellular environment in vitro dramatically improves rt-pcr performance. Biochemical and Biophysical Research Communications , 363(1):171 - 177, 2007.
- [14] A. G. Yodh, K. Lin, J. C. Crocker, A. D. Dinsmore, R. Verma, and P. D. Kaplan. Entropically driven selfassembly and interaction in suspension. 359(1782):921937, 2001.
- [15] M. Adams, Z. Dogic, S. L. Keller, and S. Fraden. Entropically driven microphase transitions in mixtures of colloidal rods and spheres. Nature , 393(6683):349-352, 1998.
- [16] S. C. Glotzer and M. J. Solomon. Anisotropy of building blocks and their assembly into complex structures. Nature materials , 6(8):557-562, 2007.
- [17] W. Li, T. Yang, and H.-r. Ma. Depletion potentials in colloidal mixtures of hard spheres and rods. The Journal of Chemical Physics , 128(4), 2008.
- [18] R. Roth, R. Evans, and S. Dietrich. Depletion potential in hard-sphere mixtures:theory and applications. Physical Review E , 62:5360-5377, Oct 2000.
- [19] R. Roth. Fundamental measure theory for hard-sphere mixtures: a review. Journal of Physics: Condensed Matter , 22(6):063102, 2010.
- [20] R. P. Sear. Coil-globule transition of a semiflexible polymer driven by the addition of spherical particles. Physical Review E , 58:724-728, Jul 1998.
- [21] B. Martnez-Haya and M. C. Gordillo. Effect of macromolecular crowding on the conformation of confined chain polymers. Macromolecular Theory and Simulations , 14(7):421-427, 2005.
- [22] S. Sacanna, W. T. M. Irvine, P. M. Chaikin, and D. J. Pine. Lock and key colloids. Nature , 464(7288):575-578, 2010.
- [23] T. Wang, R. Sha, R. Dreyfus, M. E. Leunissen, C. Maass, D. J. Pine, P. M. Chaikin, and N. C. Seeman. Selfreplication of information-bearing nanoscale patterns. Nature , 478(7368):225-228, 2011.
- [24] D. J. Ashton, Robert L. Jack, and N. B. Wilding. Selfassembly of colloidal polymers via depletion-mediated lock and key binding. Soft Matter , 9:9661-9666, 2013.
- [25] A. D. Dinsmore, A. G. Yodh, and D. J. Pine. Entropic control of particle motion using passive surface microstructures. Nature , 383(6597):239-242, 1996.
- [26] A. D. Dinsmore, D. T. Wong, P. Nelson, and A. G. Yodh. Hard spheres in vesicles: Curvature-induced forces and particle-induced curvature. Physical Review Letters , 80:409-412, Jan 1998.
- [27] M. Schmidt. Geometry-based density functional theory: an overview. Journal of Physics: Condensed Matter , 15(1):S101, 2003.
- [28] R. Roth, B. G¨ otzelmann, and S. Dietrich. Depletion forces near curved surfaces. Physical Review Letters , 83:448-451, Jul 1999.
- [29] M. Schmidt. Density functional for additive mixtures. Physical Review E , 62:3799-3802, 2000.
- [30] R. Roth and R. Evans. The depletion potential in non-additive hard-sphere mixtures. Europhysics Letters , 53(2):271, 2001.
- [31] M. Oettel. Depletion force between two large spheres suspended in a bath of small spheres: Onset of the derjaguin limit. Physical Review E , 69:041404, 2004.
- [32] S. A. Egorov. Effect of repulsive and attractive interactions on depletion forces in colloidal suspensions: A density functional theory treatment. Physical Review E , 70:031402, Sep 2004.
- [33] R. Roth and M. Kinoshita. Depletion potential between large spheres immersed in a multicomponent mixture of small spheres. The Journal of Chemical Physics , 125(8), 2006.
- [34] P.-M. K¨ onig, R. Roth, and S. Dietrich. Depletion forces between nonspherical objects. Physical Review E , 74:041404, 2006.
- [35] A. R. Herring and J. R. Henderson. Hard-sphere fluid adsorbed in an annular wedge: The depletion force of hardbody colloidal physics. Physical Review E , 75:011402, 2007.
- [36] M. Oettel, H. Hansen-Goos, P. Bryk, and R. Roth. Depletion interaction of two spheres full density functional theory vs. morphometric results. Europhysics Letters , 85(3):36003, 2009.
- [37] T. Biben, P. Bladon, and D. Frenkel. Depletion effects in binary hard-sphere fluids. Journal of Physics: Condensed Matter , 8(50):10799, 1996.
- [38] R. Dickman, P. Attard, and V. Simonian. Entropic forces
- in binary hard sphere mixtures: Theory and simulation. The Journal of Chemical Physics , 107(1), 1997.
- [39] L. Lue and L. V. Woodcock. Depletion effects and gelation in a binary hard-sphere fluid. Molecular Physics , 96(9):1435-1443, 1999.
- [40] A. Santos, S. B. Yuste, and M. L´pez de Haro. o Contact values of the particle-particle and wall-particle correlation functions in a hard-sphere polydisperse fluid. The Journal of Chemical Physics , 123(23), 2005.
- [41] J. Liu and E. Luijten. Rejection-free geometric cluster algorithm for complex fluids. Physical Review Letters , 92:035504, 2004.
- [42] D. J. Ashton, V. S´nchez-Gil, and N. B. Wilding. Monte a carlo methods for estimating depletion potentials in highly size-asymmetrical hard sphere mixtures. The Journal of Chemical Physics , 139(14), 2013.
- [43] G. Odriozola, F. Jim´ enez- ´ ngeles, A and M. LozadaCassou. Entropy driven key-lock assembly. The Journal of Chemical Physics , 129(11), 2008.
- [44] D. A. Triplett and K. A. Fichthorn. Entropic forces and directed alignment of hard squares in suspensions of rods and disks. The Journal of Chemical Physics , 133(14), 2010.
- [45] J. M. Rickman and H. D. Ou-Yang. Elastic response of binary hard-sphere fluids. Physical Review E , 84:012401, 2011.
- [46] G. W. Slater, C. Holm, M. V. Chubynsky, H. W. de Haan, A. Dub´ e, K. Grass, O. A. Hickey, C. Kingsburry, D. Sean, T. N. Shendruk, and L. Zhan. Modeling the separation of macromolecules: A review of current computer simulation methods. Electrophoresis , 30(5):792-818, 2009.
- [47] A. A. Louis, E. Allahyarov, H. L¨wen, and R. Roth. Efo fective forces in colloidal mixtures: From depletion attraction to accumulation repulsion. Physical Review E , 65:061407, 2002.
- [48] D. Henderson, A. Trokhymchuk, L. V. Woodcock, and K.-Y. Chan. Simulation and approximate formulae for the radial distribution functions of highly asymmetric hard sphere mixtures. Molecular Physics , 103(5):667674, 2005.
- [49] J. S. Smith, D. Bedrov, and G. D. Smith. A molecular dynamics simulation study of nanoparticle interactions in a model polymer-nanoparticle composite. Composites Science and Technology , 63(11):1599 - 1605, 2003. Modeling and Characterization of Nanostructured Materials.
- [50] S.-C. Kim, B.-S. Seong, and S.-H. Suh. Depletion interactions in two-dimensional colloid-polymer mixtures: Molecular dynamics simulations. Journal of Physics: Condensed Matter , 21(3):035101, 2009.
- [51] S. Kim. Molecular dynamics study of soft matter: Potential dependence of depletion forces. Journal of the Korean Physical Society , 61(7):1102-1107, 2012.
- [52] J. S. Kim and I. Szleifer. Depletion effect on polymers induced by small depleting spheres. The Journal of Physical Chemistry C , 114(48):20864-20869, 2010.
- [53] V. Bot ¸an, F. Pesth, T. Schilling, and M. Oettel. Hardsphere fluids in annular wedges: Density distributions and depletion potentials. Physical Review E , 79:061402, Jun 2009.
- [54] D. J. Ashton, N. B. Wilding, R. Roth, and R. Evans. Depletion potentials in highly size-asymmetric binary hardsphere mixtures: Comparison of simulation results with theory. Physical Review E , 84:061136, Dec 2011.
- [55] R. B. Phillips, J. Kondev, J. Theriot, N. Orme, and H. Garcia. Physical biology of the cell . Garland Science New York, 2009.
- [56] S. Asakura and F. Oosawa. On interaction between two bodies immersed in a solution of macromolecules. The Journal of Chemical Physics , 22(7):1255-1256, 1954.
- [57] A. Zaccone and E. M. Terentjev. Theory of molecular crowding in brownian hard-sphere liquids. Physical Review E , 85:061202, Jun 2012.
- [58] H. N. W. Lekkerkerker. Osmotic equilibrium treatment of the phase separation in colloidal dispersions containing non-adsorbing polymer molecules. Colloids and Surfaces , 51(0):419 - 426, 1990.
- [59] B. G¨ otzelmann, R. Evans, and S. Dietrich. Depletion forces in fluids. Phys. Rev. E , 57:6785-6800, Jun 1998.
- [60] J. Bafaluy, B. Senger, J.-C. Voegel, and P. Schaaf. Effect of hydrodynamic interactions on the distribution of adhering brownian particles. Physical Review Letters , 70:623-626, 1993.
- [61] J. T. Padding and A. A. Louis. Hydrodynamic interactions and brownian forces in colloidal suspensions: Coarse-graining over time and length scales. Physical Review E , 74:031402, Sep 2006.
- [62] J. Dzubiella, H. L¨wen, and C. N. Likos. Depletion forces o in nonequilibrium. Physical Review Letters , 91:248301, 2003.
- [63] M. G. Noro and D. Frenkel. Extended correspondingstates behavior for particles with variable range attractions. The Journal of Chemical Physics , 113(8), 2000.
- [64] H. A. Kramers. Brownian motion in a field of force and the diffusion model of chemical reactions. Physica , 7(4):284 - 304, 1940.
- [65] P. H¨ anggi, P. Talkner, and M. Borkovec. Reaction-rate theory: fifty years after Kramers. Reviews of Modern Physics , 62:251-341, Apr 1990.
- [66] T. N. Shendruk, M. Bertrand, H. W. de Haan, J. L. Harden, and G. W. Slater. Simulating the entropic collapse of coarse-grained chromosomes. http://arxiv.org/abs/1407.2850, 2014.
- [67] Y. Rosenfeld. Free-energy model for the inhomogeneous hard-sphere fluid mixture and density-functional theory of freezing. Physical Review Letters , 63:980-983, Aug 1989.
- [68] R. Roth, R. Evans, A. Lang, and G. Kahl. Fundamental measure theory for hard-sphere mixtures revisited: the white bear version. Journal of Physics: Condensed Matter , 14(46):12063, 2002.
- [69] Y.-X. Yu and J. Wu. Structures of hard-sphere fluids from a modified fundamental-measure theory. The Journal of Chemical Physics , 117(22):10156-10164, 2002.
- [70] H. Hansen-Goos and R. Roth. Density functional theory for hard-sphere mixtures: the white bear version mark ii. Journal of Physics: Condensed Matter , 18(37):8413, 2006.
## Appendix A: Second Virial Coefficient in the φ d glyph[lessmuch] 1 Limit:
## Depletant/Colloid van der Waals Gas
In the φ d glyph[lessmuch] 1 (or equivalently β u ∣ ∣ min AO ∣ ∣ glyph[lessmuch] 1) AO limit, the minimum value of the pair potential is well approximated by the AO model of hard-sphere colloids (Eq. (3))
and furthermore the well depth is less than the thermal energy ( β u ∣ ∣ min AO ∣ ∣ glyph[lessmuch] 1). This limit allows us to approximate the Mayer function f ( r ) = 1 -e -βu r ( ) ≈ βu r ( ), which can then be used to find an approximate, analytical form of the second virial coefficient:
$$\begin{array} { c } \text{can form on the second year} \text{elements} \\ \text{r} \\ B _ { c, 2 } = 2 \pi \left [ \frac { 2 } { \pi } \frac { 4 \pi R _ { c } ^ { 3 } } { 3 } - \int _ { 2 R _ { c } } ^ { 2 \mathcal { R } } \left ( e ^ { - \beta u } - 1 \right ) r ^ { 2 } d r + 0 \right ] \\ g \\ s, & \approx 2 \pi \left [ \frac { 2 } { \pi } \frac { 4 \pi R _ { c } ^ { 3 } } { 3 } + \int _ { 2 R _ { c } } ^ { 2 \mathcal { R } } \beta u _ { \mathrm A O } r ^ { 2 } d r \right ] \\ n & = 4 V _ { c } \left [ 1 - \phi _ { d } g \right ], & \text{(A1)} \\ \text{f-} \\ \text{f-} \end{array}$$
where we have defined
$$\underset { i c a l } { \text{in-} } \underset { g } { \text{s:} } \quad & g \left ( \frac { R _ { d } } { R _ { c } } \right ) \equiv \frac { 3 } { 2 } + \frac { 1 5 } { 8 } \left ( \frac { R _ { d } } { R _ { c } } \right ) + \frac { 3 } { 4 } \left ( \frac { R _ { d } } { R _ { c } } \right ) ^ { 2 } + \frac { 1 } { 8 } \left ( \frac { R _ { d } } { R _ { c } } \right ) ^ { 3 } \\ \underset { \text{c-es} } { \text{s:} }$$
for convenience. Notice that setting φ d = 0 automatically reproduces the hard-sphere value B c 2 , = 4 V c . The number density of depletants scales the correction, which otherwise is a function of R /R d c alone (encapsulated in g ). Less intuitively, the effective volume is predicted to reduce when the size of the depletants relative to the colloids is increased. This is a manifestation of the increased range of the attractive well but one must keep in mind the discussion in that the depletants must be significantly smaller than the colloidal particles. This condition remains.
The AO solution can be described as a van der Waals gas but, unlike a standard van der Waals gas in which only the first term has thermal energy dependence, in an AO solution the pressure scales with β -1 , reflecting the entropic basis of the attractive forces.
## Appendix B: Morphometric Pair Potential Details
For the MT potential defined in Eqn. 4, the geometric variables are simple functions of colloid-colloid separation r [36]:
$$V _ { 0 } & = \frac { 4 \pi \mathcal { R } ^ { 3 } } { 3 } \left [ 1 - \frac { 3 } { 2 } \left ( \frac { r } { 2 \mathcal { R } } \right ) + \frac { 1 } { 2 } \left ( \frac { r } { 2 \mathcal { R } } \right ) ^ { 3 } \right ], \quad \text{(B1a)} \\ A & = 4 \pi \mathcal { R } ^ { 2 } \left ( 1 - \frac { r } { \sim } \right ).$$
$$A & = 4 \pi \, \kappa \, \left ( 1 - \overline { 2 \mathcal { R } } \right ), & \text{(D1D)} \\, & C _ { 1 } & = \frac { A } { \mathcal { R } } + \pi \mathcal { R } \sqrt { 1 - \left ( \frac { r } { 2 \mathcal { R } } \right ) ^ { 2 } } \sin ^ { - 1 } \left ( \frac { r } { 2 \mathcal { R } } \right ), & \text{(B1c)} \\ | & C _ { 2 } & = 4 \pi. & \text{(B1d)}$$
$$\L ^ { \cdot \nu } & \quad 3 \quad \lfloor \, ^ { \cdot } \, \quad 2 \, \Lambda \, \Lambda \, \Lambda \, \quad 2 \, \Lambda \, \Lambda \, \quad \rfloor \,, \quad \langle \, \Lambda \, \Lambda \, \cdots \, \Lambda \, \rangle \\ A & = 4 \pi \mathcal { R } ^ { 2 } \left ( 1 - \frac { r } { 2 \mathcal { R } } \right ), \quad \ \ ( B 1 b ) \\ C. & = \frac { A } { 2 \, \perp \, \pi \mathcal { P }. } \sqrt { 1 \, - \, \left ( \frac { r } { 2 \, \perp \, \Lambda } \, \right ) ^ { 2 } } _ { \text{cin} ^ { - 1 } \left ( \frac { r } { 2 \, \perp \, \Lambda } \, \right ) } \quad \ \ ( \mathcal { R } 1 \in \Lambda )$$
$$C _ { 2 } & = 4 \pi.$$
However, the thermodynamic quantities are less straightforward. We utilize the Rosenfeld functionals [53, 67]:
$$\beta \Pi V _ { d } & = - \phi _ { d } \frac { 1 + \phi _ { d } + \phi _ { d } ^ { 2 } } { \left ( 1 - \phi _ { d } \right ) ^ { 3 } }, & \left ( B 2 a \right ) \left ( \begin{matrix} \beta \\ \phi _ { d }, \, w \\ \text{terms} \end{matrix} \right ) \\ \varepsilon \text{, } \text{, } \text{, } \text{, } \text{, } \text{, } \text{, } \text{, } \text{, } \text{, } \text{, } \text{, } \text{, } \text{, } \text{, } \text{, } \text{, } \text{, } \text{, } \text{, } \text{, } \text{, } \text{, } \text{ } \text{, } \text{ } \text{, } \text{, } \text{ } \text{ } \text{, } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{\ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text$$
$$\beta \kappa _ { 1 } V _ { \text{d} } = - \phi _ { \text{d} } R _ { \text{d} } ^ { 2 } & \frac { \lambda ^ { \prime \prime } _ { \text{d} } } { \left ( 1 - \phi _ { \text{d} } \right ) ^ { 3 } }, \\ & \quad \, \Big / \, - \varepsilon + 7 \theta \, - \, 1 1 \theta ^ { 2 } \quad \ln \, ( 1 - \theta \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot$$
$$\beta \gamma V _ { \text{d} } = \phi _ { \text{d} } R _ { \text{d} } \frac { \stackrel { \ell } { 3 \phi _ { \text{d} } } \stackrel { \ell \mu } { ( 1 + \phi _ { \text{d} } ) } } { 2 \left ( 1 - \phi _ { \text{d} } \right ) ^ { 3 } }, \quad \quad \quad ( \text{B2b} ) \\ \wedge \ \. \ \. \ \. \ \. \ \. \ \.$$
$$& \stackrel { r \cdots \iota } { r \cdots \iota } \quad \stackrel { r \cdots \iota } { \left ( 1 - \phi _ { d } \right ) ^ { 3 } } \,, \quad \stackrel { \vee \, \, \,$$
but other accurate forms are available in the binary hard sphere fluid literature such as the White Bear functionals [68-70]. The leading term in the osmotic pressure is φ d , while in both the surface tension and κ 2 , the leading terms are φ 2 d . In κ 1 the leading term is φ 3 d . | 10.1063/1.4903992 | [
"Tyler N. Shendruk",
"Martin Bertrand",
"James L. Harden",
"Gary W. Slater",
"Hendrick W. de Haan"
] | 2014-08-01T21:29:58+00:00 | 2014-08-01T21:29:58+00:00 | [
"cond-mat.soft"
] | Coarse-Grained Molecular Dynamics Simulations of Depletion-Induced Interactions for Soft Matter Systems | Given the ubiquity of depletion effects in biological and other soft matter
systems, it is desirable to have coarse-grained Molecular Dynamics simulation
approaches appropriate for the study of complex systems. This paper examines
the use of two common truncated Lennard-Jones (WCA) potentials to describe a
pair of colloidal particles in a thermal bath of depletants. The shifted-WCA
model is the steeper of the two repulsive potentials considered, while the
combinatorial-WCA model is the softer. It is found that the depletion-induced
well depth for the combinatorial-WCA model is significantly deeper than the
shifted-WCA model because the resulting overlap of the colloids yields extra
accessible volume for depletants. For both shifted- and combinatorial-WCA
simulations, the second virial coefficients and pair potentials between
colloids are demonstrated to be well approximated by the Morphometric
Thermodynamics (MT) model. This agreement suggests that the presence of
depletants can be accurately modelled in MD simulations by implicitly including
them through simple, analytical MT forms for depletion-induced interactions.
Although both WCA potentials are found to be effective generic coarse-grained
simulation approaches for studying depletion effects in complicated soft matter
systems, combinatorial-WCA is the more efficient approach as depletion effects
are enhanced at lower depletant densities. The findings indicate that for soft
matter systems that are better modelled by potentials with some
compressibility, predictions from hard-sphere systems could greatly
underestimate the magnitude of depletion effects at a given depletant density. |
1408.0310v3 | ## Improved Theory of Neutrino Oscillations in Matter
## Leonard S. Kisslinger
Department of Physics, Carnegie Mellon University, Pittsburgh PA 15213 USA.
November 20, 2014
## Abstract
This is revision of the S-Matrix theory of neutrino oscillations used for many years. We evaluate the transition probability of a µ to e neutrino without an approximation used for many theoretical studies, and find important differences which could improve the extraction of neutrino parameters from experimental data in the future.
.PACS Indices:11.30.Er,14.60.Lm,13.15.+g
Keywords: neutrino, neutrino oscillations, S-Matrix theory
## 1 Introduction
In previous research on neutrino oscillations we used S-Matrix theory with three active neutrinos to evaluate time reversal violation[1], CP violation[2], and recently[3] the transition probability P ( ν µ → ν e ) . One of the main objectives of Ref.[3] was to calculate neutrino oscillations and anti-electron neutrino disappearance to help in the measurement of θ 13 from Daya Bay, Double Chooze, and RENO data. For this it is essential to have an accurate theory of neutrino transition probabilities.
Our previous research using S-Matrix theory with a 3 × 3 mixing matrix for three active neutrinos was based on earlier publications[4, 5, 6, 7]. In carrying out the evaluation of P ( ν µ → ν e ) we used an approximation for an important function, I α ∗ , found in this earlier work, such as Ref[6], whose formulation we used. In the present work we evalute I α ∗ exactly and find corrections to P ( ν µ → ν e ). Our objective is to improve P ( ν µ → ν e ), and thereby the extraction of important parameters from neutrino oscilation data.
## 2 P ( νµ → νe ) Derived Using S-Matrix Theory
Neutrinos are produced as ν f , with f =flavor= e, µ, τ . Neutrinos with definite mass are ν m , m=1,2,3. The flavor neutrinos are related to the neutrinos with definite mass by the 3 × 3 unitary matrix, U ,
$$\nu _ { f } \ = \ U \nu _ { m } \,,$$
where ν , ν f m are 3 × 1 column vectors and U is ( sinθ ij ≡ s ij , etc).
$$U = \left ( \begin{array} { c c c c } c _ { 1 2 } c _ { 1 3 } & s _ { 1 2 } c _ { 1 3 } & s _ { 1 3 } e ^ { - i \delta _ { C P } } \\ - s _ { 1 2 } c _ { 2 3 } - c _ { 1 2 } s _ { 2 3 } s _ { 1 3 } e ^ { i \delta _ { C P } } & c _ { 1 2 } c _ { 2 3 } - s _ { 1 2 } s _ { 2 3 } s _ { 1 3 } e ^ { i \delta _ { C P } } & s _ { 2 3 } c _ { 1 3 } \\ s _ { 1 2 } s _ { 2 3 } - c _ { 1 2 } s _ { 2 3 } s _ { 1 3 } e ^ { i \delta _ { C P } } & - c _ { 1 2 } s _ { 2 3 } - s _ { 1 2 } c _ { 2 3 } s _ { 1 3 } e ^ { i \delta _ { C P } } & c _ { 2 3 } c _ { 1 3 } \end{array} \right ) \\ \mathbb { M } _ { \infty } \, \mathbb { m } _ { \infty } \, \varkappa \, \varepsilon \, \varpi \, \varpi \, = \, \mathbb { M } _ { \infty } \, \varkappa \, \varpi \, = \, \varpi \, \varpi \, = \, \varpi \, \varpi \, \varpi \, = \, \mathbb { M } _ { \infty } \, \mathbb { M } _ { \infty } \, \mathbb { M } _ { \infty } \, \mathbb { M } _ { \infty } \, \mathbb { M } _ { \infty } \, \mathbb { M } _ { \infty } \, \mathbb { M } _ { \infty } \, \mathbb { M } _ { \infty } \, \mathbb { M } _ { \infty } \, \mathbb { M } _ { \infty } \, \mathbb { M } _ { \infty } \, \mathbb { M } _ { \infty } \, \mathbb { M } _ { \infty } \, \mathbb { M } _ { \infty } \, \mathbb { M } _ { \infty } \, \cdots = \, \mathbb { M } _ { \infty }$$
We use c 12 = 83 . , s 12 = 56 . , s 23 = c 23 = 7071, . s 13 = 0.19, and δ CP =0.
The transition probability P ( ν µ → ν e ) is obtained from the S-Matrix element S 12 , where i d dt S t, t ( 0 ) = H t S t, t ( ) ( 0 ), with H t ( ) the Hamiltonian:
$$\mathcal { P } ( \nu _ { \mu } \to \nu _ { e } ) \ = \ ( R e [ S _ { 1 2 } ] ) ^ { 2 } + ( I m [ S _ { 1 2 } ] ) ^ { 2 } \.$$
From Ref[3]
with
$$\begin{array} { r c l } \bar { \Delta } & = & \Delta - ( V + \delta ) / 2 \\ \Delta & = & \delta m _ { 1 3 } ^ { 2 } / ( 2 E ) \\ \delta & = & \delta m _ { 1 2 } ^ { 2 } / ( 2 E ) \,, \end{array}$$
where the neutrino mass differences are δm 2 12 = 7 6 . x 10 -5 ( eV ) 2 , δm 2 13 = 2 4 . x 10 -3 ( eV ) 2 , sin 2 θ = s 12 c 12 δ ω , a = s 13 (∆ -s 2 12 δ ), and E is the neutrino energy. Note that t → L , where L is the baseline, for v ν /similarequal c . The neutrino-matter potential V = 1 13 . × 10 -13 eV.
The main quantity of interest in our present work, I α ∗ , is
$$I _ { \alpha ^ { * } } \ = \ \int _ { 0 } ^ { t } d t ^ { \prime } \alpha ^ { * } ( t ^ { \prime } ) e ^ { - i \bar { \Delta } t ^ { \prime } }$$
with α t ( ) = cos ωt ( ) -icos 2 θsin ωt ( ), ω = δ 2 + V 2 -2 δV cos (2 θ 12 ) /
In Ref.[3], as in Ref.[6], one used δ, ω /lessmuch
√ 2. ∆ to obtain
$$R e [ I _ { \alpha ^ { * } } ] \ & \simeq \ s i n \bar { \Delta } L / \bar { \Delta } \\ I m [ I _ { \alpha ^ { * } } ] \ & \simeq \ ( 1 - c o s \bar { \Delta } L ) / \bar { \Delta } \.$$
On the other hand if one uses Eq(5) to evaluate I α ∗ , one obtains
$$\begin{array} { c c c } \quad \cdot \cdot \\ R e [ I _ { \alpha ^ { * } } ] & = & [ ( \omega - \bar { \Delta } \cos 2 \theta ) \cos \bar { \Delta } L s i n \omega L \\ & & - ( \bar { \Delta } - \omega \cos 2 \theta ) s i n \bar { \Delta } L \cos \omega L ] / ( \omega ^ { 2 } - \bar { \Delta } ^ { 2 } ) \\ I m [ I _ { \alpha ^ { * } } ] & = & [ \bar { \Delta } + \omega \cos 2 \theta - ( \bar { \Delta } + \omega \cos 2 \theta ) \cos \bar { \Delta } L \cos \omega L \\ & & - ( \omega + \bar { \Delta } \cos 2 \theta ) s i n \bar { \Delta } L s i n \omega L ] / ( \omega ^ { 2 } - \bar { \Delta } ^ { 2 } ). \end{array}$$
From this we obtain an improved P ( ν µ → ν e )= (Re[ S 12 ]) 2 +(Im[ S 12 ]) 2 .
$$R e [ S 1 2 ] \ & = \ s _ { 2 3 } a [ \cos ( \bar { \Delta } L ) I m [ I _ { \alpha ^ { * } } ] - s i n ( \bar { \Delta } L ) R e [ I _ { \alpha ^ { * } } ] ] \\ I m [ S 1 2 ] \ & = \ - c _ { 2 3 } s i n 2 \theta s i n \omega L - s _ { 2 3 } a [ \cos ( \bar { \Delta } L ) R e [ I _ { \alpha ^ { * } } ] \\ & \quad + s i n ( \bar { \Delta } L ) I m [ I _ { \alpha ^ { * } } ]$$
Figure 1: P ( ν µ → ν e ) for MINOS(L=735 km), MiniBooNE(L=500m), JHF-Kamioka(L=295 km), and CHOOZ(L=1.03 km) using the improved 3 × 3 mixing matrix. Solid curve for precise I α ∗ ( E,L ) and dashed curve for approximate I α ∗ ( E,L ). s 13 =0.19
## 3 Conclusions
As one can see from the figure, for a large baseline (MINOS) or energy (MiniBooNE) the corrections are small. For some baselines and energies, including JHF-Kamioka and CHOOZ, the corrections to the neutrino transition probability P ( ν µ → ν e ) are large. This will modify the predictions of CP and T violation, and could have a large effect on the extraction of parameters via neutrino oscillations, such as s 13 from P (¯ ν e → ¯ ) ν e by Daya Bay[8], Double Chooz[9], and, RENO[10]. This is an important result in neutrino physics, and the precise expression for I α ∗ we have derived should be used for all future studies of neutrino oscillations.
## Acknowledgements
This work was carried out while LSK was a visitor at Los Alamos National Laboratory, Group P25. The author thanks Dr. William Louis for information about recent and future neutrino oscillation experiments, and Dr. Mikkel B. Johnson for many helpful discussions.
## References
- [1] E.M. Henley, M.B. Johnson, and L.S. Kisslinger, arXiv:1102.5106/ hep-ph; Int. J. Mod. Phys. E20 , 2453 (2011)
- [2] E.M. Henley, M.B. Johnson, and L.S. Kisslinger, arXiv:1105.2741/ hep-ph (2011)
- [3] Leonard S. Kisslinger, Ernest M. Henley, and Mikkel B. Johnson, arXiv:1203.6613/hepph; Int. J. Mod. Phys. E 21 , 1250065 (2012)
- [4] J. Arafune and J. Sato, Phys. Rev. D 55 , 1653 (1997)
- [5] M. Freund, Phys. Rev. D 64 , 053003 (2001)
- [6] E. Akhmedov, P. Huber, M. Linder, and T. Ohlsson, Nucl. Phys B 608 , 394 (2001)
- [7] M. Jacobson and T. Ohlsson, Phys. Rev. D 69 , 013003 (2004)
- [8] F.P. An et.al., Daya Bay Collaboration, arXiv:1210.6327; Chin. Phys. C37 , 01101 (2013)
- [9] M. Kuze (Double Chooz Collaboration), arXiv:1109.0074/hep-ex (2011)
- [10] J., K. Ahn, et.al. (RENO Collaboration), arXiv:1003.1391/hep-ex (2010) | 10.1142/S0217732315500145 | [
"Leonard S. Kisslinger"
] | 2014-08-01T21:38:50+00:00 | 2014-11-19T15:23:28+00:00 | [
"hep-ph"
] | Improved Theory of Neutrino Oscillations in Matter | This is revision of the S-Matrix theory of neutrino oscillations used for
many years. We evaluate the transition probability of a $\mu$ to $e$ neutrino
without an approximation used for many theoretical studies, and find important
differences which could improve the extraction of neutrino parameters from
experimental data in the future. |
1408.0311v1 | ## The Scaling of Chaos vs Periodicity: How Certain is it that an Attractor is Chaotic?
Madhura Joglekar, 1 Edward Ott, 1 and James A. Yorke 1
1
University of Maryland, College Park, Maryland 20742, USA
(Dated: 5th August 2014)
The character of the time-asymptotic evolution of physical systems can have complex, singular behavior with variation of a system parameter, particularly when chaos is involved. A perturbation of the parameter by a small amount glyph[epsilon1] can convert an attractor from chaotic to non-chaotic or viceversa. We call a parameter value where this can happen glyph[epsilon1] -uncertain. The probability that a random choice of the parameter is glyph[epsilon1] -uncertain commonly scales like a power law in glyph[epsilon1] . Surprisingly, two seemingly similar ways of defining this scaling, both of physical interest, yield different numerical values for the scaling exponent. We show why this happens and present a quantitative analysis of this phenomenon.
While low-dimensional chaotic attractors are common and fundamental in a vast range of physical phenomenon, it is predominantly the case that chaotic motions in such systems are 'structurally unstable'. This typically has the consequence that an arbitrarily small change of a system parameter can always be found that results in periodic behavior [1]. Physical models displaying chaotic attractors that are structurally unstable arise very often, e.g., in studies of plasma dynamics [2], Josephson junctions [3], chemical reactions [4] and many others. We also note that even the simple example of the one-dimensional quadratic map, observed uncertainty, applicable when all chaotic attractors of any size are considered [6], and this observation is at the heart of the resolution of the above-mentioned puzzle.
$$x _ { n + 1 } = C - x _ { n } ^ { \ 2 }, \quad \ \ \ ( 1 ) \quad \epsilon _ { - 1 }$$
displays this phenomenon, and, in this paper we will study this example as a convenient paradigm for such situations in general.
One reason for concern with this type of behavior is that physical systems have uncertainties in the values of their parameters, and one might therefore ask how confident one can be about the prediction of chaotic behavior from a model calculation (even when the model and its parameter dependence are precisely known and there is no noise). Despite the fundamental importance of this question, very little study has been done to quantitatively address it [5-7]. It is the purpose of this paper to readdress this general issue, and, in particular, to resolve a long-standing puzzle. This puzzle has to do with the scaling characterization of the fractal-like chaotic/periodic interweaving structure of parameter dependence associated with structural instability. In particular, studies on the quadratic map Eq.(1) have addressed scaling in two slightly different ways and obtained significantly different estimates of the scaling exponent [5-7]. The reason for this surprising discrepancy has remained unresolved. In one of the two ways of addressing scaling, attention was restricted to what might seem to be the most obvious source of uncertainty, namely, when a large (to be defined subsequently) chaotic attractor suddenly turns into a periodic attractor, as a parameter is varied. However, we find such transitions far too rare [5, 7] to account for the
Consider a dynamical system depending on a parameter C , and an attractor A C ( ) that, as C varies continuously in some range, can be uniquely associated with C . We say that a particular value of C is glyph[epsilon1] -uncertain with respect to chaos if A C ( ) is chaotic while either A C ( + ) glyph[epsilon1] or A C ( -glyph[epsilon1] ) or both are not chaotic. For example, for the case of the quadratic map, to which we henceforth restrict our considerations, it has been found that a random choice of C with uniform probability density yields glyph[epsilon1] -uncertainty with respect to chaos with a probability ¯ ( F glyph[epsilon1] ) that scales like a power of glyph[epsilon1] for small glyph[epsilon1] [6], ¯ ( F glyph[epsilon1] ) ∼ glyph[epsilon1] β . As a result of the fine-scaled interweaving of C values for which A C ( ) is chaotic and intervals of C values for which A C ( ) is periodic, the power law exponent β turns out to be less than 1, and the set of C values with A C ( ) chaotic has been called a 'fat-fractal' [5, 6] (more precisely, it is a Cantor set with positive Lebesgue measure [8]). We have repeated the numerical determination of ¯ ( F glyph[epsilon1] ) (Fig. 1) and find that for small glyph[epsilon1] ,
$$\bar { F } ( \epsilon ) \sim \epsilon ^ { \beta } \text{ with } \beta = 0. 3 9 2 \pm 0. 0 3 7. \quad \text{ (2)}$$
In performing this calculation we estimate ¯ ( F glyph[epsilon1] ) by first randomly choosing many C values with uniform probability in the range where the quadratic map has a unique bounded attractor, -1 / 4 ≤ C ≤ 2 . For each such C value we then compute the Lyapunov exponents for C , C + glyph[epsilon1] and C -glyph[epsilon1] , judging the corresponding attractors to be chaotic or not depending on whether the computed Lyapunov exponent is positive. We then estimate ¯ ( F glyph[epsilon1] ) as the fraction of those randomly chosen C -values that are computed to be glyph[epsilon1] -uncertain with respect to chaos. Reference [6] obtains a slightly larger β value of β ≈ 0 41 using . an glyph[epsilon1] -range with larger glyph[epsilon1] -values. We agree with their result in the range they tested, but, by pushing to small glyph[epsilon1] , obtain the result in Eq. (2) (See Fig. 1.)
We comment that Eq. (2) can be interpreted as implying a type of 'probabilistic stability' for chaos. That is,
FIG. 1: ¯ ( F glyph[epsilon1] ) vs glyph[epsilon1] .
while chaos occuring at some parameter value C may be structurally unstable in the sense that a chaos-destroying perturbation C → C + δC , can always be found with | δC | ≤ ˜ for any given ˜; chaos is still stable in the sense glyph[epsilon1] glyph[epsilon1] that, for ˜ small, such a chaos-destroying glyph[epsilon1] δC may have to be very carefully chosen, and the probability that a random choice of δC in | δC | ≤ ˜ destroys chaos approaches glyph[epsilon1] zero as ˜ is made smaller and smaller [9] (Eq. glyph[epsilon1] (2)). (Incidently, we note that quasiperiodicity appears to have this same type of structural-instability / probabilistic stability [6].)
We now recall the concept of a periodic window. A p -periodic window is an interval of the parameter, C ∗ ( p ) ≤ C ≤ C x ( p ) , such that at the beginning of the window, as C increases through C ∗ ( p ) , there is a bifurcation from a chaotic attractor to a periodic orbit attractor of period p , followed by a period-doubling cascade to chaos, followed by a sequence of band-mergings in each of which 2 m p separate pieces ( x -intervals) of the chaotic attractor pairwise merge into a 2 m -1 p -piece chaotic attractor, eventually forming a 'small' p -piece chaotic attractor, which subsequently terminates ('explodes') at the end of the window ( C = C x ( p ) ) through a crisis transition [10] to a larger chaotic attractor that is similar in size to the larger chaotic attractor just before the beginning of the window at C = C ∗ ( p ) . Thus, as is evident from viewing a bifurcation diagram for the quadratic map, 'small' chaotic attractors occur within windows only, and we call a chaotic attractor that is not contained in any window a 'large chaotic attractor'.
Now, instead of considering glyph[epsilon1] -uncertainty with respect to chaos, we consider glyph[epsilon1] -uncertainty with respect to the occurence of large chaotic attractors. That is, we consider C to be glyph[epsilon1] -uncertain if A C ( ) is a large chaotic attractor while either A C ( + ) or glyph[epsilon1] A C ( -glyph[epsilon1] ) or both are not large chaotic attractors. Reference [7] gives a detailed consideration and analysis of the scaling of glyph[epsilon1] uncertainty with respect to the occurence of large chaotic attractors for Eq.
(1) obtaining for the probability F 0 ( glyph[epsilon1] ) of glyph[epsilon1] -uncertainty
$$F _ { 0 } ( \epsilon ) \sim \epsilon ^ { \alpha } \text{ with } \alpha = 0. 5 1 \pm 0. 0 3. \quad \text{ (3)}$$
Thus, there are evidently two distinct scaling exponents, the β and α of Eqs. (2) and (3). We conjecture that the values of these exponents are universal for onedimensional maps with a quadratic maximum. For example, they would apply to physical situations like those in the plasma example in Ref. [2] and the chemical example in Ref. [4], where strong phase-space attraction leads to dynamics closely approximated by a one-dimensional map.
It is important to note that the chaos within windows has its own windows, which, in turn, have their own windows, and so on. Thus, it is useful to distinguish the 'order' of a window: We say a window is of order one (which we also call a 'primary' window), if it is not contained within another window, and we say a window is of order r > 1, if it is contained within a window of order ( r -1), but it is not contained within a window of order ( r + 1). Thus, if a C value is not contained within a primary window, then it is not contained in any window (as is the case for a large chaotic attractor). For later reference, we define N r (∆) to be the number of windows of order r whose widths are greater than ∆ and we write this quantity for r = 1 (primary windows) as
$$N _ { 1 } ( \Delta ) = \sum _ { i = 1 } ^ { \infty } U ( \Delta _ { i } - \Delta ),$$
where ∆ 1 ≥ ∆ 2 ≥ ∆ 3 ≥ . . . denote the widths of the primary windows, and U denotes the unit step function ( U z ( ) = 1 for z > 0, U z ( ) = 0 for z < 0). If we consider all the windows of any order, then the number of these windows whose widths exceed ∆ is
$$\bar { N } ( \Delta ) = \sum _ { r = 1 } ^ { \infty } N _ { r } ( \Delta ).$$
Since glyph[epsilon1] 0 39 . >> glyph[epsilon1] 0 51 . for small glyph[epsilon1] , the results (2) and (3) imply that for small glyph[epsilon1] most of the parameter values that are glyph[epsilon1] -uncertain with respect to chaos lie in windows, and, in fact, as we will demonstrate elsewhere, for a randomly chosen parameter value ˜ C that is glyph[epsilon1] -uncertain with respect to chaos, the expectation value of the order of the lowest order window containing ˜ C approaches ∞ as glyph[epsilon1] → 0.
Preliminary to our analysis of the relationship between the exponents α and β , we need the following two results:
- (i) Self-similarity of windows: Reference [11] shows that the bifurcation structure and dynamics in windows of various orders and periods are self-similar. That is, considering x near 0, by use of uniform linear stretchings (magnification) in x and C , the p -times iterated map with the parameter ranging through the interval corresponding to a periodp
window, C ∗ ( p ) ≤ C ≤ C x ( p ) , very closely quantitatively replicates the behavior of the map in its full range, -1 / 4 ≤ C ≤ 2 . Furthermore, this selfsimilarity approximation is already extremely good even for the period-3 primary window, and becomes better and better for most r -order windows as r is increased. This is illustrated in Fig. 2, where we see that the bifurcation diagram in the full range -1 / 4 ≤ C ≤ 2 in Fig. 2a is virtually identical to a properly magnified, inverted ( x →-x ) version of the bifurcation diagram for the period-3 window, C ∗ (3) ≤ C ≤ C x (3) , blown up in the region near x = 0 (Fig. 2b).
- (ii) glyph[epsilon1] -uncertainty / window-width scaling equivalence: It is shown in the supplementary material that, for small window widths ∆, the scalings of ¯ (∆) and N N 1 (∆) are related to the small glyph[epsilon1] scalings of ¯ ( F glyph[epsilon1] ) and F 0 ( glyph[epsilon1] ) :
$$\bar { F } ( \epsilon ) \sim \epsilon \bar { N } ( \epsilon ) \text{ and } F _ { 0 } ( \epsilon ) \sim \epsilon N _ { 1 } ( \epsilon ). \quad \ \ ( 6 ) \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
Thus, from Eqs. (2) and (3), Eq. (6) yields
$$\bar { N } ( \Delta ) \sim \frac { 1 } { \Delta ^ { 1 - \beta } } \text{ and } N _ { 1 } ( \Delta ) \sim \frac { 1 } { \Delta ^ { 1 - \alpha } }. \quad \text{ (7)}$$
FIG. 2: (a) shows a bifurcation diagram for the quadratic map for -0 25 . ≤ C ≤ 2. (b) shows a blow-up of the bifurcation diagram in the period-3 window in the region near x = 0.
By using Eq. (7), we reduce our analysis of the glyph[epsilon1] -uncertainty exponents α and β to an analysis of the scaling of the distribution of window-widths. Now, using the self-similarity of windows, we can express N r +1 (∆) in terms of N r (∆),
$$N _ { r + 1 } ( \Delta ) = \int _ { \Delta } ^ { 9 / 4 } ( \frac { - d N _ { r } ( \Delta ^ { \prime } ) } { d \Delta ^ { \prime } } ) \ N _ { 1 } ( \frac { 9 \Delta / 4 } { \Delta ^ { \prime } } ) \ d \Delta ^ { \prime }, \quad ( 8 )$$
where 9 4 / = 2 - -( 1 / 4) is the width of the C -range for the quadratic map. In Eq. (8), we have used the window self-similarity result to write the number of r +1 order windows of width ∆ that are contained in windows of width ∆ ′ as N 1 ( 9∆ 4 / ∆ ′ ). Letting u = ln(9 / (4∆)) and M u r ( ) = N r (∆) , Eq. (8) becomes,
$$M _ { r + 1 } ( u ) = \int _ { 0 } ^ { u } M _ { 1 } ( u - u ^ { \prime } ) \ \frac { d M _ { r } ( u ^ { \prime } ) } { d u ^ { \prime } } \ d u ^ { \prime }. \quad ( 9 )$$
Equation (9) is a convolution. It is therefore convenient to introduce the Laplace transform, ˆ M s r ( ) = ∫ ∞ 0 M u e r ( ) -su du , in terms of which Eq. (5) becomes ˆ M r +1 ( s ) = sM ˆ 1 ( s M ) ˆ r ( s ) (where we have made use of M r (0) = N r (9 / 4) = 0). Iterating this result, we obtain ˆ M s r ( ) in terms of ˆ M s 1 ( ),
$$\hat { M } _ { r } ( s ) = \frac { ( s \hat { M } _ { 1 } ( s ) ) ^ { r } } { s }. \quad \quad \quad ( 1 0 )$$
Introducing the Laplace transform ˆ ( ¯ M s ) of ¯ ( M u ) = ¯ (∆), we have from Eq. N (5) that ˆ ( ¯ M s ) = ∑ ∞ r =1 ˆ M s r ( ), which using Eq. (10) gives, upon summing the resultant geometric series,
$$\hat { \hat { M } } ( s ) = \frac { \hat { M } _ { 1 } ( s ) } { 1 - s \hat { M } _ { 1 } ( s ) }. \quad \quad ( 1 1 )$$
This expression is singular at values of s for which sM ˆ 1 ( s ) = 1 . From Eq. (4), M u 1 ( ) = ∑ ∞ i =1 U u ( -u i ) , where u i = ln(9 / (4∆ )) i . Thus, ˆ M s 1 ( ) = s -1 ∑ ∞ i =1 e -su i = ∑ ∞ i =1 (4∆ i / 9) s and we have that singularities of ˆ ( ¯ M s ) occur at s values satisfying 1 = ∑ ∞ i =1 δ i s , where δ i = (4∆ i / 9) are the normalized widths of the primary windows.
Now consider the inverse Laplace transform [12] of ˆ ( ¯ M s ) for large u (i.e., small ∆); we see that ¯ ( M u ) ∼ e γu where γ is the solution for s of 1 = ∑ i δ i s with the largest real part. It can be shown that this solution is real. Thus we obtain for ¯ (∆) N at small ∆, ¯ (∆) N ∼ ∆ -γ , which when compared with the first part of our result Eq. (7) shows that γ = 1 -β. We conclude that we can obtain the exponent β as the real positive root of
$$\sum _ { i = 1 } ^ { \infty } \delta _ { i } ^ { \gamma } = 1, \ \beta = 1 - \gamma.$$
We now use Eq. (12) to investigate the relationship between the exponents α and β . Writing Eq. (12) as ∑ I -1 i =1 δ i γ + ∑ ∞ i = I δ i γ = 1, for sufficiently large I , we can approximate the second summation by
$$\sum _ { i = I } ^ { \infty } \delta _ { i } ^ { \gamma } \approx \int _ { 0 } ^ { \delta _ { I } } ( - d N _ { 1 } ( \delta ) / d \delta ) \delta ^ { \gamma } d \delta. \quad \ ( 1 3 )$$
Since N 1 ( δ ) ∼ δ -(1 -α ) , the integrand is proportional to δ γ + α -2 , and the integral diverges unless γ + α > 1 . Thus, the summation in (12) is infinity unless γ > (1 -α ), and,
since each term in the sum decreases monotonically with increasing γ , we conclude that as γ increases past (1 -α ), the sum in (12) decreases monotonically.
For γ = 1, the sum is the normalized total length of all windows, which, by definition, is less than 1. We conclude that Eq. (12) has a single root for γ and that this root satisfies γ < (1 -α . ) Thus, for γ = 1 -β , we must have that α > β, in agreement with the numerical results α ≈ 0 51 . , β ≈ 0 39 . . Taking N 1 ( δ I ) = I ( δ I /δ ) 1 -α and performing the integration in (13), we obtain
$$\sum _ { i = 1 } ^ { I - 1 } \delta _ { i } ^ { 1 - \beta } + \frac { 1 - \alpha } { \alpha - \beta } I \delta _ { I } ^ { 1 - \beta } = 1. \quad \quad ( 1 4 )$$
For example, assuming availability of an estimate of α, one can use (14) to estimate β given numerical determinations of ( δ 1 , δ 2 , . . . , δ I ). (Note that the second term in (14) becomes smaller and smaller as I is increased and can be omitted for very large I .)
Figure 3 shows a plot of the estimated value of β obtained from (14) as a function of I for α = 0 51. . The result from Fig. 3 is β = 0 39 in good agreement with . the estimate from Fig. 1.
FIG. 3: Estimated value of β vs I . As I is increased, the estimated value of β converges to ≈ 0 39 . .
In conclusion, we have derived an analytic estimate Eq. (12) for β which yields good agreement with the numerical result in Eq. (2) and shows why α > β. More generally, letting Q 0 and ¯ Q respectively denote the set of C values yielding large chaotic attractors and the set of C values yielding chaotic attractors of any size, one can view our work as using the self-similarity of windows to establish a quantitative link between the structure of these two sets. In particular, Eq. (12) relates the primary window widths { δ 1 , δ 2 , δ 3 , . . . } (a characterization of Q 0 ) to the exponent β (a characterization of ¯ ) Q. Although our considerations have focused on the quadratic map, we believe that the numerical results for the exponents α and β are universal for one-dimensional maps with a single quadratic maximum, and thus apply for situations as in [2], [4] where there is a strong phase-space contraction.
The work of E. O. was supported by ARO grant W911NF1210101.
- [1] That is, given a system parameter value C for which there is a chaotic attractor and a small value of ˜ glyph[epsilon1] > 0, one can find a parameter perturbation C → C + δC with | δC | < glyph[epsilon1] ˜ such that the attractor is no longer chaotic, and this can always be done no matter how small ˜ is. glyph[epsilon1]
- [2] J. M. Wersinger, J. M. Finn, and E. Ott, Phys. Fluids 23 , 1142 (1980).
- [3] B. A. Huberman, J. P. Crutchfield, and N. H. Packard, Appl. Phys. Lett. 37 , 750 (1980).
- [4] F. Argoul, A. Arneodo, P. Richetti, J. C. Roux, and H. L. Swinney, Acc. Chem. Res. 20 , 436 (1987).
- [5] J. D. Farmer, Phys. Rev. Lett. 55 , 351 (1985).
- [6] C. Grebogi, S. W. McDonald, E. Ott, and J. A. Yorke, Phys. Lett. A 110 , 1 (1985).
- [7] B. R. Hunt and E. Ott, J. Phys. A 30 , 7067 (1997).
- [8] M. V. Jacobson, Comm. Math. Phys. 81 , 39 (1981), has shown that C values for which A C ( ) is chaotic form a positive Lebesgue measure set, while J. Graczyk and G. Swiatek [Ann. of Math. 146, 1 (1997)] demonstrate that intervals of C values with periodic attractors are dense in the region, C F < C < 2, where chaos can occur (here C F denotes the 'Feigenbaum point' at which period doublings of periods 2 m accumulate).
- [9] We comment that this is a consequence of Lebesgue's density theorem which states that, for a Lebesgue measurable set Q (in this case the set of C values that yield chaotic attractors), for almost every point x in Q , the limit of glyph[epsilon1] → 0 of µ Q ( ∩ [ x -glyph[epsilon1], x + ]) (2 ) is one, where glyph[epsilon1] / glyph[epsilon1] µ denotes the Lebesgue measure.
- [10] C. Grebogi, E. Ott, and J. A. Yorke, Physica D 7(1) , 181 (1983).
- [11] J. A. Yorke, C. Grebogi, E. Ott, and L. TedeschiniLalli, Phys. Rev. Lett. 54 , 1095 (1985), see Sec. B of Supplementary Material.
- [12] In taking the Laplace transform ∫ ∞ 0 M u ( ) exp( -su du ) and the inverse Laplace transform (2 πi ) -1 ∫ p ˆ ( M s ) exp( su ds ) , where the contour P runs from - ∞ i + σ to + i ∞ + σ , we take Re s ( ) > γ and σ > γ , respectively.
- [13] There is some small error attached to our use of the approximate self-similarity of windows in our derivation of Eq. (12). As shown in Table 1 of Ref. [11] (reproduced in Sec. B of the supplementary material), however, the self-similarity assumption is quite good, and, as demonstrated in Sec. C of the supplementary material, we estimate that this small deviation from self-similarity is not significant compared to the error bounds of the numerical determination of β indicated in Eq. (2). | 10.1103/PhysRevLett.113.084101 | [
"Madhura Joglekar",
"Edward Ott",
"James A. Yorke"
] | 2014-08-01T21:51:45+00:00 | 2014-08-01T21:51:45+00:00 | [
"nlin.CD"
] | The Scaling of Chaos vs Periodicity: How Certain is it that an Attractor is Chaotic? | The character of the time-asymptotic evolution of physical systems can have
complex, singular behavior with variation of a system parameter, particularly
when chaos is involved. A perturbation of the parameter by a small amount
$\epsilon$ can convert an attractor from chaotic to non-chaotic or vice-versa.
We call a parameter value where this can happen $\epsilon$-uncertain. The
probability that a random choice of the parameter is $\epsilon$-uncertain
commonly scales like a power law in $\epsilon$. Surprisingly, two seemingly
similar ways of defining this scaling, both of physical interest, yield
different numerical values for the scaling exponent. We show why this happens
and present a quantitative analysis of this phenomenon. |
1408.0312v1 | ## Interpretation of the coherency matrix for three-dimensional polarization states
## José J. Gil
Facultad de Educación. Universidad de Zaragoza. Pedro Cerbuna 12, 50009 Zaragoza, Spain. [email protected]
## Abstract
From an appropriate parameterization of the three-dimensional (3D) coherency matrix R , that characterizes the second-order, classical states of polarization, the coherency matrices are classified and interpreted in terms of incoherent decompositions. The relevant physical quantities derived from R , as the intensity, the degree of polarimetric purity, the indices of polarimetric purity, the angular momentum, the degree of directionality and the degree of linear polarization are identified and interpreted on the light of the case study performed. The information provided by R about the direction of propagation is clarified and it is found that coherency matrices with rank 2, does not always represent states with a well-defined R direction of propagation. Moreover, it is demonstrated the existence of 3D mixed states that cannot be decomposed into a superposition of a pure state, a 2D unpolarized state, and a 3D unpolarized state. Appropriate representation and interpretation for all the different types of 3D coherency matrices is provided through physical consistent criteria. Under the approach proposed, the conventional two-dimensional model arises naturally.
OCIS . 260.0260 Physical optics, 260.2130 Ellipsometry and polarimetry, 260.5430 Polarization, 290.5855 Scattering, polarization
PACS. 42.25.Ja Polarization - 42.68.Mj Scattering, polarization
## 1. Introduction
A proper description of the polarization properties of electromagnetic waves relies on the concept of the coherency matrix, which is applicable regardless the particular band of the electromagnetic spectrum considered.
The study and characterization of three-dimensional (3D) states of polarization is a subject of high interest from both theoretical and experimental points of view. Several relevant contributions have been published in the recent years concerning aspects as the interpretation of physical quantities derived from the eigenvalues of the coherency matrix R [1-12]; geometric interpretation of 3D states [13-19]; coherent composition of pure (or totally polarized) states [20]; incoherent composition and decomposition of pure and mixed states [4,7,9,17,21]; 3D polarimetry [22,23]; statistical and coherence properties of 3D states [2428], generalized Stokes parameters [16,1,9,13,29], etc.
However, additional effort is needed in order to get answers to questions such as: how to represent 3D states of polarization as combinations of states with a simple physical interpretation? Generally, unitary transformations of the coherency matrix do not correspond to rotations of the 3D laboratory reference frame, therefore, which unitary transformations are physically realizable? What kind of information about the propagation direction can be obtained from the coherency matrix? How to interpret 3D polarization states in terms of
meaningful physical quantities? How many different physical situations can be distinguished and how to classify them?
The aim of this work is to provide appropriate and consistent responses to the previous questions by means of 1) an adequate parameterization of the 3D coherency matrix R , as is the one proposed by Dennis [13] in terms of nine physical parameters; 2) the case study of the different physical situations, identified through specific descriptors and analyzed by means of the arbitrary and characteristic decompositions of R , and 3) the identification, definition and interpretation of parameters that provide complete and meaningful physical information.
## 2. 3D states of polarization
In the most general case, the three components of the electric field vector E of the electromagnetic wave should be considered in order to describe the evolution of E , which determines the polarization state. Let us consider a quasi-monochromatic wave of arbitrary form, propagating in an isotropic medium, and let 1 2 3 , e ,e e be a reference basis of orthonormal vectors along the respective axes XYZ . Given a point r in space, the analytic signals of the three components of E can be arranged as the 3D instantaneous Jones vector [9]
$$\tau \, \stackrel { \dots } { s } ( t ) & = \begin{pmatrix} \varepsilon _ { 1 } ( t ) \\ \varepsilon _ { 2 } ( t ) \\ \varepsilon _ { 3 } ( t ) \end{pmatrix} = \begin{pmatrix} A _ { x } ( t ) \\ A _ { y } ( t ) e ^ { i \delta _ { y } ( t ) } \\ A _ { z } ( t ) e ^ { i \delta _ { z } ( t ) } \end{pmatrix} & & ( 1 ) \\. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad.$$
In general, ( ) has slow time dependence with respect to the coherence time, so that, for t ε time intervals shorter than the coherence time, the polarization ellipse can be considered constant. For time intervals higher than the coherence time, the instantaneous Jones vector can vary, resulting in partially polarized states.
Note that totally-polarized (or pure ) states are characterized by the fact that the quantities ( ) ( ), ( ) ( ), ( ) and y x z x y A t A t A t A t t ( ) are constant, and the 3D Jones vector (applicable z t only for pure states) is expressed without time dependence. It is well known that, in the case of pure states, the electric field describes an ellipse perpendicular to the direction of propagation (at point r ). The 2D model is then easily reproduced by taking the direction of propagation as the transformed Z O reference axis, so that the third component of the 3D Jones vector vanishes.
The coherency matrix or polarization matrix , which contains all measurable second-order information about the state of polarization (including intensity) of an electromagnetic wave, is defined as the 3 3 Hermitian matrix ( ) ( ) t t R ε ε whose elements ij r are the secondorder moments * ( ) ( ) ij i j r t t , 1,2,3 i j of the zero-mean analytic signals ( ) , ( i t * i represents the complex conjugate of i and the brackets indicate time averaging over the measurement time). Thus, R is a covariance matrix and, therefore, its three eigenvalues 1, 2, 3 are non-negative. Moreover, R can be expressed as
$$\begin{pmatrix} \cdot & \cdot & \cdot & \cdot & \\ & \cdot & \cdot & \cdot & \\ & \mathbf R = \begin{pmatrix} \langle A _ { x } ^ { 2 } ( t ) \rangle & \langle A _ { x } ( t ) A _ { y } ( t ) e ^ { - i \delta _ { y } ( t ) } \rangle & \langle A _ { x } ( t ) A _ { z } ( t ) e ^ { - i \delta _ { z } ( t ) } \rangle \\ \langle A _ { x } ( t ) A _ { y } ( t ) e ^ { i \delta _ { y } ( t ) } \rangle & \langle A _ { y } ^ { 2 } ( t ) \rangle & \langle A _ { y } ( t ) A _ { z } ( t ) e ^ { i ( \delta _ { y } ( t ) - \delta _ { z } ( t ) ) } \rangle \\ \langle A _ { x } ( t ) A _ { z } ( t ) e ^ { i \delta _ { z } ( t ) } \rangle & \langle A _ { y } ( t ) A _ { z } ( t ) e ^ { - i ( \delta _ { y } ( t ) - \delta _ { z } ( t ) ) } \rangle & \langle A _ { z } ^ { 2 } ( t ) \rangle \end{pmatrix}$$
where the diagonal elements can be interpreted as the intensities associated with the respective XYZ components of the electric field and the total intensity (irradiance or power flux density of the wave, averaged over the measurement time) is given by
$$I = \text{tr} \, \mathbf R = \langle A _ { x } ^ { 2 } ( t ) \rangle + \langle A _ { y } ^ { 2 } ( t ) \rangle + \langle A _ { z } ^ { 2 } ( t ) \rangle \,.$$
As Fano pointed out [30], an appropriate basis for x n n coherency matrices is a set of Hermitian trace-orthogonal operators, that in the case of R is that constituted by the 3 3 identity matrix together with the eight Gell-Mann matrices (generators of the SU(3) group)
$$\text{3.3.3.3} \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \, \dashrightarrow \colon \\ \omega _ { 1 1 } \equiv \left ( \begin{matrix} 1 & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 1 \end{matrix} \right ), \omega _ { 1 2 } \equiv \sqrt { \frac { 3 } { 2 } } \left ( \begin{matrix} 0 & 1 & 0 \\ 1 & 0 & 0 \\ 0 & 0 & 0 \end{matrix} \right ), \omega _ { 1 3 } \equiv \sqrt { \frac { 3 } { 2 } } \left ( \begin{matrix} 0 & 0 & 1 \\ 1 & 0 & 0 \\ 0 & 0 & 0 \end{matrix} \right ), \\ \omega _ { 2 1 } \equiv \sqrt { \frac { 3 } { 2 } } \left ( \begin{matrix} 0 & - i & 0 \\ 0 & 0 & 0 \end{matrix} \right ), \omega _ { 2 2 } \equiv \sqrt { \frac { 3 } { 2 } } \left ( \begin{matrix} 1 & 0 & 0 \\ 0 & - 1 & 0 \\ 0 & 0 & 0 \end{matrix} \right ), \\ \omega _ { 3 1 } \equiv \sqrt { \frac { 3 } { 2 } } \left ( \begin{matrix} 0 & - i \\ 0 & 0 & 0 \\ i & 0 & 0 \end{matrix} \right ), \omega _ { 3 2 } \equiv \sqrt { \frac { 3 } { 2 } } \left ( \begin{matrix} 0 & 0 & 0 \\ 0 & 0 & - i \\ 0 & i & 0 \end{matrix} \right ), \omega _ { 3 3 } \equiv \sqrt { \frac { 3 } { 2 } } \left ( \begin{matrix} 0 & 0 & 0 \\ 0 & 0 & - i \\ 0 & i & 0 \end{matrix} \right ). \\ \text{e Gell-Mann matrices have been normalized in order to ensure that their Euclidean norms} \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \,. \end{matrix}$$
The Gell-Mann matrices have been normalized in order to ensure that their Euclidean norms coincide with that of the 3 3 identity matrix (in analogy to the 2D case, where the Euclidean norms of the Pauli matrices coincides with that of the 2 2 identity matrix). The notation used for these matrices is justified for the sake of simplicity as well as to emphasize the symmetry in some mathematical expressions [9]. Note that, leaving aside the normalization factor, the Pauli matrices are nested in the upper 2 2 submatrices of 12 21 22 , , ω ω ω .
Thus, the coherency matrix R can be expanded as [3]
$$R = \frac { 1 } { 3 } \sum _ { i, j = 1 } ^ { 3 } q _ { i j } \omega _ { i j } \,, \, q _ { i j } = \text{tr} \left ( R \omega _ { i j } \right ),$$
where the nine real coefficients ij q are the so-called the generalized Stokes parameters [16].
A study of the 3D spectral density tensor devoted to plane waves was presented in Ref. [16], where it is shown that the real vector 32 23 31 , , q q q v determines the plane of the polarization ellipse. Note that 00 tr q I R , so that some expressions are simplified by using the normalized form ˆ tr R R R of the coherency matrix.
Let us now consider the Euclidean norm and the trace norm of R , defined respectively as [9]
$$\left \| R \right \| _ { 2 } \equiv \sqrt { \text{tr} \left ( R ^ { 2 } \right ) } ; \ \left \| R \right \| _ { \text{tr} } \equiv \text{tr} R = 1$$
The 3D degree of polarimetric purity (3) P [9] can be defined as [31,32,1]
$$P _ { ( 3 ) } = \sqrt { \frac { 1 } { 2 } \left ( \frac { 3 \left \| { \mathbf R } \right \| _ { 2 } ^ { 2 } } { \left \| { \mathbf R } \right \| _ { t r } ^ { 2 } } - 1 \right ) } \,. \quad \quad \quad$$
This invariant non-dimensional quantity is limited to the interval 0 P (3) 1. The upper limit corresponds to the case that R has only one nonzero eigenvalue (pure state). Conversely, P (3) = 0 is reached when the three eigenvalues of R are equal (equiprobable mixture of states and zero correlation between the electric field components). For reasons
explained in Ref. [9], we consider preferable to refer (3) P to as the degree of polarimetric purity rather than the degree of polarization.
Setälä et al. pointed out [1,2] that (3) P takes into account not only the purity of the mean polarization ellipse, but also the stability of the plane that contains the instantaneous components of the electric field of the wave. Nevertheless, (3) P does not provide complete information about the polarimetric purity of a 3D state of polarization R , and two quantities like the two indices of polarimetric purity [3,9,10,17] are required
$$P _ { 1 } = \frac { \lambda _ { 1 } - \lambda _ { 2 } } { \text{tr} R } \,, \ P _ { 2 } = \frac { \lambda _ { 1 } + \lambda _ { 2 } - 2 \lambda _ { 3 } } { \text{tr} R } \,,$$
where 1 2 3 1 2 3 , , ( ) are the eigenvalues of R . These invariant non-dimensional parameters are restricted by the nested inequalities 0 1 2 1. P P
Thus, while the pair 1 2 , P P provides detailed information of the structure of the polarimetric purity of R , P (3) represents an overall measure of polarimetric purity of R , which can be calculated from 1 P and 2 P by the weighted quadratic average [3,17]
$$P _ { ( 3 ) } = \sqrt { 3 P _ { 1 } ^ { 2 } + P _ { 2 } ^ { 2 } } \, / 2 \,.$$
By taking advantage of the analytic expressions for 1, 2, 3 obtained by Sheppard [18,19] from the del Ferro-Cardano-Tartaglia-Vieta solution for the cubic equation with real roots, the indices of polarimetric purity can be written as
$$\text{or poaimeq} \, & \text{many can use when as} \\ & P _ { 1 } = \frac { 2 } { \sqrt { 3 } } P _ { ( 3 ) } \sin \psi, \ P _ { 2 } = 2 P _ { ( 3 ) } \cos \psi ; \\ & \psi \equiv \frac { 1 } { 3 }$$
excluded the case (3) 0, where P 1 1 0. Note that the value of P P is limited by 0 3. The physically feasible region in the purity space 1 2 , P P has been studied and interpreted by us in previous papers [9,17], while interesting geometric representations for 1, 2, 3 and other derived quantities (including 1 P and 2 P ) have been presented by Sheppard [18,19].
The physical interpretation of 1 P , 2 P and other parameters is considered in later sections on the light of the case study preformed.
## 3. Composition and decomposition of 3D states of polarization
## 3.1. Coherent composition
As a previous step to the study of the incoherent superposition of states of polarization, which must be performed through the additive composition of the respective coherency matrices, it is worth to consider the coherent composition of pure states, which must be realized through the additive composition of the respective Jones vectors.
Let us consider a point r in a linear, homogeneous and isotropic medium where two mutually coherent states, characterized by respective 3D Jones vectors 1 ε and 2 ε are superposed. The resultant state of polarization (at point r ) is a pure state given by the Jones
vector 1 2 ε ε ε . The electric field of the combined state describes a well-defined ellipse, which, at point r , lies in the plane tangent to the wavefront [20].
The polarization ellipse of a pure state with nonzero ellipticity lies in a well-defined plane , and thus, the direction of propagation of the state (at point r ) is well defined because it is necessarily perpendicular to . On the contrary, a 3D Jones vector representing a linearly polarized state is compatible with any direction of propagation perpendicular to the axis where the electric field of the electromagnetic wave lies. Consequently, a 3D Jones vector (and hence the corresponding coherency matrix) does not contain intrinsic information about the direction of propagation, but that information can be deduced from , either as a fixed direction, for states with nonzero ellipticity, or as an arbitrary direction perpendicular to the polarization axis, for linearly polarized states (zero ellipticity). It will be worth keeping in mind this obvious result when we deal later with the incoherent composition and decomposition of linearly polarized states with coincident polarization axes.
Moreover, since the sum of Jones vectors is a Jones vector, given a 3D Jones vector , it can always be expressed, in an infinite number of ways, as the sum of a number of 3D Jones vectors.
## 3.2 Arbitrary decomposition of the coherency matrix
Iterative [9] and constructive [21,32,34] general procedures for the decomposition of a coherency matrix into a convex sum of coherency matrices have been presented by us in previous works and are summarized below for the case of 3D polarization matrices.
Let us now consider the diagonalization † 1 2 3 diag , , R U U , where U is the unitary matrix whose columns are the eigenvectors of R , and diag 1 2 3 , , represents the diagonal matrix composed of the ordered non-negative eigenvalues 3 2 1 0 .
As follows from Ref. [21] (where the case of four-dimensional covariance matrices representing material media is considered), R can be expressed as the following convex sum in terms of a set of r (with rank r R ) arbitrary independent 3D complex unit vectors i w belonging to Range R (i.e., belonging to the subspace generated by the eigenvectors of R with nonzero eigenvalues)
$$R = \sum _ { i = 1 } ^ { r } p _ { i } R _ { i } ; \ R _ { i } \equiv \text{tr} R \left ( w _ { i } \otimes w _ { i } ^ { \dagger } \right ) ; \ \ p _ { i } = \frac { 1 } { \text{tr} H \sum _ { j = 1 } ^ { r } \frac { 1 } { \lambda _ { j } } \left | \left ( U ^ { \dagger } w _ { i } \right ) _ { j } \right | ^ { 2 } } ; \ \sum _ { i = 1 } ^ { r } p _ { i } = 1, \quad ( 1 1 ) \\. \quad + +. \ d \. \ d \. \ d \. \ d \ - \ d \. \ d \ - \ d \. \ d \ - \ d \$$
where indicates the modulus (or Euclidean norm).
Expansion (11)) provides the way for generating arbitrary complete sets of the coherency matrices tr † i i R w w of the pure components as well as their corresponding coefficients i p . Note that the minimum number of pure components of the arbitrary decomposition is equal to r . When 3, r any arbitrary three-dimensional generalized complex basis 1,2,3 i i w can be chosen. When 2 , r any arbitrary two-dimensional generalized complex basis 1 2 , w w belonging to Range R can be chosen. When 1, r R represents a pure state (fixed polarization ellipse) and consequently the arbitrary decomposition has no physical interest, because it becomes a tautology. Hereafter, when appropriate to note that a state R is pure, we will denote its coherence matrix as p R .
By taking as i w the eigenvectors 1,2,3 i i u of R , we get the following spectral decomposition of R as a particular case of the arbitrary decomposition
$$R = \frac { \lambda _ { + } } { \ t r R } U \text{diag} ( \text{tr}, 0, 0 ) U ^ { + } + \frac { \lambda _ { 2 } } { \ t r R } U \text{diag} ( 0, \text{tr}, 0 ) U ^ { + } + \frac { \lambda _ { 3 } } { \ t r R } U \text{diag} ( 0, 0, \text{tr} ) U ^ { + }$$
where each term in the sum is synthesized from the corresponding eigenvector i u of R and its weight in the convex sum is proportional to the corresponding eigenvalue i
$$R = \sum _ { i = 1 } ^ { 3 } \frac { \lambda _ { i } } { \text{tr} R } \left [ ( \text{tr} R ) \left ( \text{u} _ { i } \otimes \text{u} _ { i } ^ { + } \right ) \right ].$$
It should be noted that, when one of the eigenvalues has a multiplicity higher than one, then the eigenvectors of the corresponding invariant subspace are not unique, and consequently the spectral decomposition is not unique.
By considering the possible values of the indices of purity, we observe that they have a direct link with the purity structure of R and provide more detailed information than the very value of rank r R . As a general overview, the following cases can be distinguished [17]: 1) when 1 2 0 1 P P 3 , r R can be considered as composed of three pure states 1 1 2 2 3 3 p p p p p p R R R R ; if, in particular, 2 0 (and hence, P 1 0 ), P R is proportional to the identity matrix, 3 diag 1,1,1 u D I R R , so that it represents a 3D unpolarized state (completely random polarization ellipse and completely random direction of propagation); 2) when 1 2 0 1 P P 2 r , R can be considered as composed of two pure states 1 1 2 2 p p p p R R R ; if, in particular 1 0 , P R corresponds to a 2D unpolarized state propagating along a well-defined direction of propagation. Finally, when 1 1 (and hence, P 2 1 and P 1), r R corresponds to a pure state.
## 3.3. Characteristic decomposition of the coherency matrix
While all the components of the arbitrary decomposition of R are pure states, it is also possible to decompose R by means of the following characteristic (or trivial ) decomposition [9]
$$\mathbf R = \text{Udiag} ( \lambda _ { 1 }, \lambda _ { 2 }, \lambda _ { 3 } ) \, \mathbf U ^ { + } = P _ { 1 } / \hat { \mathbf R } _ { 1 } + ( P _ { 2 } - P _ { 1 } ) \, \hat { \mathbf R } _ { 2 } + ( 1 - P _ { 2 } ) \, \hat { \mathbf I } _ { 3 } \\ \hat { \mathbf R } _ { 1 } \equiv \text{Udiag} ( 1, 0, 0 ) \, \mathbf U ^ { + }, \, \hat { \mathbf R } _ { 2 } \equiv \frac { 1 } { 2 } \, \mathbf U \text{diag} ( 1, 1, 0 ) \, \mathbf U ^ { + }, \, \hat { \mathbf R } _ { 3 } \equiv \frac { 1 } { 3 } \, \mathbf U \text{diag} ( 1, 1, 1 ) \, \mathbf U ^ { + } \\ \cdot \quad \dots \, \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \end{quad \quad ( 1 4 )$$
where all the components have been chosen to have the same intensity tr I R and the respective coefficients of the components are expressed in terms of the indices of polarimetric purity 1 2 , P P . Note that rank 1 ˆ 1, rank R 2 ˆ 2 and rank R 3 ˆ 3. R
Recall that the trivial decomposition of R cannot be always performed in the form of a sum of a pure state and a 3D unpolarized state. This is because, in general, pure n n coherency matrices contain 2 1 independent parameters whereas n m D-unpolarized n n coherency matrices contain 2 1 independent parameters. In the case of a n m 2 2 coherency matrix, the well-known decomposition of a mixed state into a pure state and a 2D unpolarized state is retrieved.
The characteristic decomposition leads to a physical interpretation of 1 P as the ratio of power of the completely polarized part (or pure component) to the total power of the electromagnetic wave. Furthermore, 2 P is the relative portion of power once the completely random component has been subtracted.
## 4. The intrinsic coherency matrix
The physically realizable rotations of the laboratory reference frame XYZ are represented by 3 3 orthogonal matrices Q (with det 1), so that the transformed coherency matrix Q R representing the same state as R but referred to the new reference frame, is given by T R Q RQ . Moreover, let us consider the decomposition of R into its real and imaginary parts R I i R R R , where the real matrix Re R R R is symmetric and positivesemidefinite, whereas the imaginary matrix Im I R R is skew-symmetric T I I R R . As Dennis pointed out [13], R R can always be diagonalized through a particular rotation Q of the reference frame:
$$Q ^ { T } \mathbf R _ { R } \mathbf Q = \text{diag} \left ( a _ { 1 }, a _ { 2 }, a _ { 3 } \right ) ; \, \left ( 0 \leq a _ { 3 } \leq a _ { 2 } \leq a _ { 1 } \right ).$$
Thus, R R defines an ellipsoid (called by Dennis the inertia ellipsoid), whose semiaxes 1 2 3 , , a a a are aligned along the respective transformed axes O O O X Y Z (Fig.1). Thus, 1 2 3 diag , , a a a can be interpreted as the coherency matrix of a state composed of the incoherent superposition of three linearly polarized pure states
$$\text{p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.p.} \\ \text{diag} \left ( a _ { 1 }, a _ { 2 }, a _ { 3 } \right ) = \text{R} _ { p _ { 1 } } + \text{R} _ { p _ { 2 } } + \text{R} _ { p _ { 3 } } ; \\ \text{R} _ { p _ { 1 } } \equiv a _ { 1 } \text{diag} \left ( 1, 0, 0 \right ), \text{ R} _ { p _ { 2 } } \equiv a _ { 2 } \text{diag} \left ( 0, 1, 0 \right ), \text{ R} _ { p _ { 3 } } \equiv a _ { 3 } \text{diag} \left ( 0, 0, 1 \right ),$$
with respective intensities 1 a , 2 a and 3 a . The previous decomposition is compatible with a variety of directions of propagation for each component (the only condition is that the direction of propagation is orthogonal to the respective polarization axis). For the sake of clarity in further physical interpretations, it results convenient the choice of the axis O Z as the common direction of propagation k for the pure states 1 p R and 2 p R , while any axis orthogonal to k can be considered as the direction of propagation of the third pure component 3 p R (Fig.1).
Fig. 1 (color online). Intensity ellipsoid, with semiaxes 1 2 3 a a a , representing a mixed state constituted by the incoherent superposition of three pure linearly polarized states 1 p R , 2 p R and 3 p R . The direction k along the reference axis ZO can be chosen
as the common direction of propagation of the pure components 1 p R and 2 p R (with respective intensities 1 2 , a a ), whereas the third pure component 3 p R (with intensity a ), propagates along any direction orthogonal to k .
3
By applying the rotation Q to the whole matrix R , we observe that the real and imaginary parts transform separately, so that we get the transformed coherency matrix [13]
$$\underset { \alpha } { m o i m } \sec p a a c e y, \, \text{so u a l w e g e u n e u a m i n e u c o n e r y m a u l a } \lfloor \, \mathcal { U } \rfloor \\ \mathbf R _ { o } \equiv Q ^ { T } \mathbf R Q = Q ^ { T } \mathbf R _ { R } Q + i Q ^ { T } \mathbf R _ { I } Q = \text{diag} ( a _ { 1 }, a _ { 2 }, a _ { 3 } ) + i \mathbf R _ { I } ^ { \prime }, \\ \mathbf R _ { o } \equiv \left [ \begin{smallmatrix} a _ { 1 } & - i n _ { 3 } & i n _ { 2 } \\ i n _ { 3 } & a _ { 2 } & - i n _ { 1 } \\ - i n _ { 2 } & i n _ { 1 } & a _ { 3 } \end{smallmatrix} \right ],$$
$$\Phi _ { 1 } \dots \Phi _ { 2 } \Phi _ { 3 } \Phi _ { 4 } \Phi _ { 5 } \Phi _ { 6 } \Phi _ { 7 } \Phi _ { 8 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phicon _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9} \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi_ _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9. } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ { 9 } \Phi _ _ { 9 } \Phi _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi_ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi _ _ { 9 } \Phi }$$
where, as occurs for I R , I R is real and skew-symmetric. The orthogonal matrix Q can be expressed as a product of rotations around the respective axes ZXY
$$\mathbf Q = \mathbf Q _ { z } ( \varphi ) \mathbf Q _ { x } ( \alpha ) \mathbf Q _ { y } ( \beta ) = \begin{bmatrix} \cos \varphi & - \sin \varphi & 0 \\ \sin \varphi & \cos \varphi & 0 \\ 0 & 0 & 1 \end{bmatrix} \begin{bmatrix} 1 & 0 & 0 \\ 0 & \cos \alpha & - \sin \alpha \\ 0 & \sin \alpha & \cos \alpha \end{bmatrix} \begin{bmatrix} \cos \beta & 0 & - \sin \beta \\ 0 & 1 & 0 \\ \sin \beta & 0 & \cos \beta \end{bmatrix}.$$
Thus, R can be expressed as T O R QR Q , and it can be parameterized in the form proposed by Dennis [13] through the following nine independent parameters: the three orientation angles , , ; the semiaxes of the intensity ellipsoid (or inertia ellipsoid [13]) given by the principal intensities 1 2 3 , , a a a , and the three components 1 2 3 , , n n n , along the respective axes O O O X Y Z , of the angular momentum n of the wave. In general, the orientation 2 2 2 1 2 3 ˆ n n n n n of n differs from k [13]. The intensity of the state I O R is given by
$$I = \text{tr} \, \text{R} = \lambda _ { 1 } + \lambda _ { 2 } + \lambda _ { 3 } = \text{tr} \, \text{R} _ { O } = \text{tr} \, \text{R} _ { p 1 } + \text{tr} \, \text{R} _ { p 2 } + \text{tr} \, \text{R} _ { p 3 } = a _ { 1 } + a _ { 2 } + a _ { 3 } \,.$$
Since the orthogonal similarity transformation (17) preserves the non-negativity (or convexity), the intrinsic coherency matrix O R is positive semidefinite and, thus, the quantities 1 2 3 1 2 3 , , ; , , a a a n n n must satisfy the following set of constraining inequalities [13] derived from the non-negativity of the leading principal minors of O R
$$a _ { 1 } \geq a _ { 2 } \geq a _ { 3 } \geq 0 ; \ \ a _ { 1 } a _ { 2 } \geq n _ { 3 } ^ { 2 }, a _ { 1 } a _ { 3 } \geq n _ { 2 } ^ { 2 }, a _ { 2 } a _ { 3 } \geq n _ { 1 } ^ { 2 } ; \ \ a _ { 1 } a _ { 2 } a _ { 3 } \geq a _ { 1 } n _ { 1 } ^ { 2 } + a _ { 2 } n _ { 2 } ^ { 2 } + a _ { 3 } n _ { 3 } ^ { 2 }. \quad ( 2 0 )$$
The smaller is the third principal intensity 3 a , the smaller is the solid angle around the axis O Z that limits the range of compatible orientations of n . When 3 0 , a n is forced to lie along the axis O Z , which, in turn, in this case is precisely the well-defined direction of propagation of the state O R .
Up to the limits set by the five restrictive inequalities (20), the quantities 1 2 3 1 2 3 , , ; , , a a a n n n are independent and are intrinsic of a given coherency matrix R . It should be stressed that the fact that the only physically realizable unitary transformations of R are those that are orthogonal, leads to the indicated set of six physical parameters instead of the only three physical invariants derivable from the eigenvalues of R . In other words, not all the unitary transformations of R are physically realizable in the laboratory, and consequently a proper interpretation of the physical quantities involved in R , as well as an appropriate
analysis of the arbitrary and characteristic decompositions of R must be performed through orthogonal transformations (and hence excluding the unitary transformations that are not orthogonal).
## 5. Case analysis
Once the arbitrary and characteristic decompositions, the structure of purity and the orthogonal transformations of a generic 3 3 coherency matrix R have been considered, we are ready to undertake the study of the possible decompositions of a three-dimensional state of polarization represented by a given coherency matrix R . To perform a proper case analysis, it is necessary to pay attention to the values of the integer parameters rank rank O r R R
$$\text{and } t \equiv \text{rank} \left [ \tilde { \ } R e \left ( \tilde { \mathbf R } \right ) \right ] = \text{rank} \left [ \tilde { \ } R e \left ( \tilde { \mathbf R } _ { o } \right ) \right ].$$
## 5.1. rank 1 R .
In this case, the intrinsic coherency matrix O R takes the form
$$\text{mome conversion} & = \begin{bmatrix} a _ { 1 } & - i \sqrt { a _ { 1 } a _ { 2 } } & 0 \\ i \sqrt { a _ { 1 } a _ { 2 } } & a _ { 2 } & 0 \\ 0 & 0 & 0 \end{bmatrix},$$
so that, the electric field lies in the transformed plane O O X Y . Now, by introducing the parameters 0 1 2 s a a and 3 1 2 2 s a a , together with the pair of parameters 1 1 2 cos2 s a a , 2 1 2 sin 2 s a a , determined by the arbitrary choice of the angle , we see that O R can be expressed as [13]
$$\Gamma \text{ can use a } \Gamma \text{sessu as } \Gamma \text{sessu as } \Gamma \text{sessu as } \Gamma \text{sessu as } \Gamma \text{sessu as } \Gamma \text{sessu as } \Gamma \text{sessu as } \Gamma \text{sessu as } \Gamma \text{sessu as } \Gamma \text{sessu as } \Gamma \text{sessu as } \Gamma \text{sessu as } \Gamma \text{sessu as } \Gamma \text{sessu as } \Gamma \text{sessu as } \Gamma \text{sessu as } \Gamma \text{sessu as } \Gamma \text{sessu as } \Gamma \text{sessu as }$$
which corresponds to a 2D pure state propagating along the O Z axis. Furthermore, the polarization ellipse is oriented in such a manner that the major and minor semiaxes lie respectively in the new reference axes O X and O Y [note that this is a result of the effect of the matrix Z Q in the orthogonal transformation (17)].
Thus, an additional rotation of an arbitrary angle around the axis O Z provides the general expression of the coherency matrix O pZ R of a pure state propagating along the direction O Z in terms of the Stokes parameters 0 1 2 3 , , , s s s s (Fig.2)
$$\text{tion } Z _ { o } \text{ in terms of the Stokes parameters } \left ( s _ { 0 }, s _ { 1 }, s _ { 2 }, s _ { 3 } \right ) \text{(Fig.2)} \\ \mathbf R _ { \rho Z _ { o } } \equiv \mathbf Q _ { z } \left ( \theta \right ) \mathbf R _ { o } \mathbf Q _ { z } ^ { r } \left ( \theta \right ) \equiv \frac { 1 } { 2 } \left [ \begin{matrix} s _ { 0 } + s _ { 1 } & s _ { 2 } - i s _ { 3 } & 0 \\ s _ { 2 } + i s _ { 3 } & s _ { 0 } - s _ { 1 } & 0 \\ 0 & 0 & 0 \end{matrix} \right ], \\ \left ( \cos 2 \theta \equiv \frac { s _ { 1 } } { \sqrt { s _ { 1 } ^ { 2 } + s _ { 2 } ^ { 2 } } }, s _ { 0 } = \sqrt { s _ { 1 } ^ { 2 } + s _ { 2 } ^ { 2 } + s _ { 3 } ^ { 2 } } \right ).$$
Fig. 2 (color online). Coherency matrix O pZ R of a generic pure state propagating along the axis ZO , obtained from its intrinsic coherency matrix O R through an arbitrary rotation around ZO .
By considering now the two possible values of t that are compatible with 1, r the following two sub-cases can be distinguished.
$$5. 1. 1. \ r = 1, \ t = 1, \, \left ( a _ { _ { 1 } } > 0, \, a _ { _ { 2 } } = a _ { _ { 3 } } = 0, \, n _ { _ { 1 } } = n _ { _ { 2 } } = n _ { _ { 3 } } = 0 \right )$$
O R takes the simple form 1 diag ,0,0 O a R , where 1 a can be interpreted as the Stokes parameter 1 s of a linearly polarized state whose polarization ellipse degenerates into a segment along the axis O X . It should be noted that, in this case, O R (and hence R ) is compatible with any direction of propagation perpendicular to the axis O X . Obviously, under experimental conditions, it is common to have specific complementary information about the direction of propagation, but we stress that, in the case of a pure linearly polarized state, the very knowledge of the coherency matrix R does not determine the direction of propagation of the wave (Fig.3). Furthermore, we also note that, at the point in the space where R (and hence O R ) is being considered, the incoherent superposition of a variety of pure states with linear polarizations along the O X axis, but with different directions of propagation, produces a pure state of linear polarization. Thus, without information additional to R , this last case is polarimetrically indistinguishable from a pure state of linear polarization and fixed direction of propagation. The geometric nature of this peculiarity provides a method for superposing incoherently, at a fixed point in the space, a number of linearly polarized pure states whose arbitrary directions of propagation lie in a plane perpendicular to the common polarization axis O X and produce a pure state whose intensity is the sum of the intensities of the combined states.
Fig. 3 (color online). Representation of a linearly polarized pure state as the incoherent superposition of an arbitrary number of linearly polarized states with the same polarization axis along the axis XO , but with different directions of propagation
$$5. 1. 2. \ r = 1 \,, \ t = 2 \,, \, \left ( a _ { 1 } \geq a _ { 2 } > 0 \,, \, a _ { 3 } = 0 \,, \, a _ { 1 } a _ { 2 } = n _ { 3 } ^ { 2 } \,, \, n _ { 2 } = n _ { 3 } = 0 \right )$$
In this case, the polarization ellipse determines a plane O O X Y , so that the propagation direction O Z is orthogonal to the said plane and therefore it is well defined. Once performed the orthogonal transformation of the reference axes, the case is reduced to a 2D pure state characterized by the coherency matrix
$$\text{terized by the coherency matrix} \\ \text{R} _ { p z _ { o } } = \mathbf Q _ { z } \left ( \theta \right ) \mathbf R _ { o } \mathbf Q _ { z } ^ { r } \left ( \theta \right ) = \frac { 1 } { 2 } \left [ \begin{matrix} s _ { 0 } + s _ { 1 } & s _ { 2 } - i s _ { 3 } & 0 \\ s _ { 2 } + i s _ { 3 } & s _ { 0 } - s _ { 1 } & 0 \\ 0 & 0 & 0 \end{matrix} \right ], \\ \left ( \cos 2 \theta \equiv \frac { s _ { 1 } } { \sqrt { s _ { 1 } ^ { 2 } + s _ { 2 } ^ { 2 } } }, s _ { 0 } = \sqrt { s _ { 1 } ^ { 2 } + s _ { 2 } ^ { 2 } + s _ { 3 } ^ { 2 } } \right ).$$
## 5.2. rank 2 R .
The interpretation of the coherency matrix R with rank 2 depends substantially on the R value of . Since the value t 1 is not compatible with t 2 , we distinguish the two possible r cases 2, 2 r t and 2, 3 . r t
$$5. 2. 1. \ r = 2, \, t = 2, \, \left ( a _ { 1 } \geq a _ { 2 } > 0, \, a _ { 3 } = 0, \, a _ { 1 } a _ { 2 } > n _ { 3 } ^ { 2 }, \, n _ { 2 } = n _ { 3 } = 0 \right )$$
It is straightforward to show that, similarly as in the case 1, 2 r t the reference axes can be chosen in such a manner that the elements of the third row and of the third column of the coherency matrix are zero. The orthogonally transformed coherency matrix O Z R represents a partially polarized state with a well-defined direction of propagation O Z , that is to say, a 2D partially polarized (or mixed ) state
$$\text{a 2D partially polarized (or mixed)} state \\ \text{R} _ { z _ { o } } = \text{Q} _ { z } ( \theta ) \text{R} _ { o } \text{Q} _ { z } ^ { T } ( \theta ) = \frac { 1 } { 2 } \left [ \begin{matrix} s _ { 0 } + s _ { 1 } & s _ { 2 } - i s _ { 3 } & 0 \\ s _ { 2 } + i s _ { 3 } & s _ { 0 } - s _ { 1 } & 0 \\ 0 & 0 & 0 \end{matrix} \right ], \\ \left ( \cos 2 \theta \equiv \frac { s _ { 1 } } { \sqrt { s _ { 1 } ^ { 2 } + s _ { 2 } ^ { 2 } } }, \ \ s _ { 0 } > \sqrt { s _ { 1 } ^ { 2 } + s _ { 2 } ^ { 2 } + s _ { 3 } ^ { 2 } } \right ]. \\ \text{a the direction of non-organization } 7 \text{ has been determined the state of normalization can be}$$
Once the direction of propagation O Z has been determined, the state of polarization can be considered two-dimensional and described by a generic 2 2 coherency matrix
$$\Phi ( I, P _ { 1 }, \mathbf u ) = I \frac { 1 } { 2 } \begin{bmatrix} 1 + P _ { i } u _ { 1 } & P _ { 1 } ( u _ { 2 } - i u _ { 3 } ) \\ P _ { 1 } ( u _ { 2 } + i u _ { 3 } ) & 1 - P _ { 1 } u _ { 1 } \end{bmatrix},$$
where the intensity tr I Φ is the time averaged power density flux of the wave; 1 P (the first index of polarimetric purity) is the degree of polarization , and i u are the components of the unit vector u (Pauli axis) that summarizes the information about the azimuth 0 and ellipticity tan 4 4 of the average polarization ellipse
$$\mathbf u \equiv \left ( u _ { 1 }, u _ { 2 }, u _ { 3 } \right ) ^ { T } = \left ( \cos 2 \chi \cos 2 \varphi, \cos 2 \chi \sin 2 \varphi, \sin 2 \chi \right ) ^ { T }.$$
The state 1 , , I P Φ u , can also be represented through the corresponding Stokes vector
$$s \equiv \left ( s _ { 0 }, s _ { 1 }, s _ { 2 }, s _ { 3 } \right ) ^ { T } = I \left [ \begin{smallmatrix} 1 \\ P _ { 1 } \end{smallmatrix} \right ].$$
## 5.2.1.1. Arbitrary decomposition.
As it has been pointed out in previous works [7,9] a mixed 2D state can always be considered as an incoherent composition of two totally polarized (or pure) states 1 p Φ and 2 p Φ . One of them can arbitrarily be chosen and then the second one is totally determined by that choice, so that there exist infinite possibilities for decomposing as a combination of two (or more) pure states
$$\text{two (or more)} \,pure states & \\ \Phi ( I, P _ { 1 }, \mathfrak { u } ) = p \, \Phi _ { p _ { 1 } } ( I, \mathfrak { v } ) + ( 1 - p ) \Phi _ { p _ { 2 } } ( I, \mathfrak { w } ) ; & \\ \Phi _ { p _ { 1 } } ( I, \mathfrak { v } ) \equiv I \frac { 1 } { 2 } \left [ \begin{smallmatrix} 1 + v _ { 1 } & v _ { 2 } - i v _ { 3 } \\ v _ { 2 } + i v _ { 3 } & 1 - v _ { 1 } \end{smallmatrix} \right ], & \Phi _ { p _ { 2 } } ( I, \mathfrak { w } ) \equiv I \frac { 1 } { 2 } \left [ \begin{smallmatrix} 1 + w _ { 1 } & w _ { 2 } - i w _ { 3 } \\ w _ { 2 } + i w _ { 3 } & 1 - w _ { 1 } \end{smallmatrix} \right ], & \\ p = \frac { 1 - P _ { 1 } ^ { 2 } } { 2 \left ( 1 - P _ { 1 } \mathfrak { u } ^ { \tau } \mathfrak { v } \right ) }, & \mathfrak { w } = \frac { P _ { 1 } \mathfrak { u } - p \mathfrak { v } } { 1 - p }. & &$$
A geometric view for the arbitrary decomposition of 3D states can be found in Ref. [7, p. 341]
Obviously, this arbitrary decomposition of can be considered either: a) a convex combination of two pure states with equal intensities , or I b) an additive combination of two pure states 1 p p Φ and 2 1 p p Φ with respective intensities pI and I pI . Despite of the fact that both the said interpretations are equivalent and respectively physically realizable, it is particularly convenient to use the interpretation (a) because it is formulated in terms of representative states taken with the same trace norm tr I Φ [21].
As a particular case of the above arbitrary decomposition, the choice v u leads to the well-known spectral decomposition of into two pure states represented by antipodal points on the Poincaré sphere
$$\Phi ( I, P _ { _ { 1 } }, \mathbf u ) = \frac { 1 + P _ { _ { 1 } } } { 2 } \Phi _ { _ { p } } ( I, \mathbf u ) + \frac { 1 - P _ { _ { 1 } } } { 2 } \Phi _ { _ { p ^ { 2 } } } ( I, - \mathbf u ) \,,$$
Since the 2D states of polarization are usually represented by means of Stokes vectors, it is worth to bring up the corresponding expressions for the arbitrary and spectral decompositions
$$\ s = \stackrel { \mu } { I } \stackrel { \mu } { \left [ \begin{smallmatrix} 1 \\ P _ { 1 } \mathfrak { u } \end{smallmatrix} \right ] } = p I \stackrel { \mu } { I } \stackrel { \mu } { \left [ \begin{smallmatrix} 1 \\ \mathfrak { v } \end{smallmatrix} \right ] }, + ( 1 - p ) I \stackrel { \mu } { \left [ \begin{smallmatrix} 1 \\ \mathfrak { w } \end{smallmatrix} \right ] }, \ s = I \stackrel { 1 } { \left [ \begin{smallmatrix} 1 \\ P _ { 1 } \mathfrak { u } \end{smallmatrix} \right ] } = \frac { 1 + \stackrel { \mu } { P _ { 1 } } } { 2 } I \stackrel { 1 } { \left [ \begin{smallmatrix} 1 \\ \mathfrak { u } \end{smallmatrix} \right ] } + \frac { 1 - P _ { 1 } } { 2 } I \stackrel { \mu } { \left [ \begin{smallmatrix} 1 \\ \mathfrak { u } \end{smallmatrix} \right ] }.$$
We emphasize the potential applications of Eq. (29) in Stokes polarimetry (which has particular importance in several fields as, for instance, astronomic and atmospheric measurements [35]) because it provides a simple procedure for the polarimetric subtraction of a reference Stokes vector from that obtained by experimental measurements [21].
In summary, the state O Z R is expressed as an incoherent superposition of two pure states with the same direction of propagation O Z (Fig. 4)
$$\mathbf R _ { z _ { o } } = p I \hat { \mathbf R } _ { p \mathbf 1 Z _ { o } } + ( 1 - p ) I \hat { \mathbf R } _ { p \mathbf 2 Z _ { o } } \,,$$
and can be interpreted in terms of the following six independent parameters:
- two orientation angles that determine the common direction of propagation of the two components;
- three stokes parameters of the pure state ˆ O pZ I R v ,
- and the coefficient p of ˆ O pZ I R v in the convex sum.
Fig. 4 (color online). Arbitrary representation of a 2D mixed state (i.e., 2, 2 ) as r t the incoherent superposition of two pure states with the same propagation direction ZO .
## 5.2.1.2. Characteristic decomposition .
An alternative interpretation of a 2D mixed state is achieved through the corresponding characteristic decomposition
$$\text{succompositor} \Phi ( I, P _ { 1 }, \mathbf u ) & = P _ { 1 } \Phi _ { p } ( I, \mathbf u ) + ( 1 - P _ { 1 } ) \Phi _ { u }, \ \Phi _ { u } \equiv \begin{bmatrix} 1 & 0 \\ 0 & 1 \end{bmatrix} ; \\ s & = I \begin{bmatrix} 1 \\ P _ { 1 } \mathbf u \end{bmatrix} = P _ { 1 } I \begin{bmatrix} 1 \\ \mathbf u \end{bmatrix} + ( 1 - P _ { 1 } ) I \begin{bmatrix} 1 \\ \mathbf 0 \end{bmatrix},$$
where p Φ represents a 2D pure state and u Φ represents a 2D unpolarized state, both propagating along the same direction O Z . This is the well-known two-dimensional form of the characteristic decomposition.
Returning to the 3D representation, the characteristic decomposition is expressed as
$$\text{eturning to the 3D representation, the characteristic decomposition is expressed as} \\ \mathbf R _ { z _ { o } } = P _ { 1 } \mathbf R _ { p \mathcal { Z } _ { o } } + \left ( 1 - P _ { 1 } \right ) \mathbf R _ { u - 2 D } ; \\ \mathbf R _ { p \mathcal { Z } _ { o } } = I \frac { 1 } { 2 } \left [ \begin{matrix} 1 + u _ { 1 } & u _ { 2 } - i u _ { 3 } & 0 \\ u _ { 2 } + i u _ { 3 } & 1 - u _ { 1 } & 0 \\ 0 & 0 & 0 \end{matrix} \right ], \, \mathbf R _ { u - 2 D } \equiv I \frac { 1 } { 2 } \left [ \begin{matrix} 1 & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 0 \end{matrix} \right ]. \\ \mathbb { m } \quad \cdot \quad \cdot \ \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot.$$
where O pZ R represents a pure state and 2 u D R represents a 2D unpolarized state, both propagating along the same direction O Z ; that is to say, once the laboratory reference axes have been appropriately rotated, the characteristic decomposition of a state R with 2, 2 r t is expressed as that of a generic 2D state [Eq. (33)].
Thus, in the case of 2, 2 , the characteristic decomposition leads to the following r t interpretation of the coherency matrix in terms of six independent parameters (Fig.5):
- two orientation angles that determine the direction of propagation of the pure component O pZ R ;
- three stokes parameters of the pure component O pZ R ,
- and the degree of polarization 1 P of O Z R , which is the coefficient of O pZ R in the convex sum.
Fig. 5 (color online). Characteristic representation of a 2D mixed state (i.e., 2, 2 ) as the incoherent superposition of a pure state and a 2D unpolarized state r t with the same propagation direction ZO .
$$5. 2. 2. \ r = 2 \,, \ t = 3 \,, \, \left ( a _ { 1 } \geq a _ { 2 } \geq a _ { 3 } > 0 ; \ a _ { 1 } a _ { 2 } a _ { 3 } = a _ { 1 } n _ { 1 } ^ { 2 } + a _ { 2 } n _ { 2 } ^ { 2 } + a _ { 3 } n _ { 3 } ^ { 2 } \right )$$
When 3, the electric field t E of the electromagnetic wave has necessarily three nonzero orthogonal components, so that E does not evolve inside a fixed plane, and thus the direction of propagation is not well defined. The analogy with the 2D representation is no longer applicable for the state represented by R .
## 5.2.2.1. Arbitrary decomposition .
Any pure state belonging to Range R (note that Range Range O R R ) can be considered as a component, and the second pure component as well as the respective coefficients, are easily determined. Thus, the arbitrary decomposition 1 2 ˆ ˆ 1 p p pI p I R R R (Fig. 6) can be performed either as indicated at the beginning of the present section, or through the following procedure [9]:
- 1. take an arbitrary 3D complex unit vector 1 w belonging to Range R , and synthesize the normalized coherency matrix † 1 1 1 ˆ p R w w of the first component;
- 2. calculate the coefficient 1 p of 1 ˆ p R : 1 1 2 1 1 2 ˆ tr diag , ,0 ; p p I I R ,
- 3. and calculate the second pure component: 2 1 ˆ ˆ ˆ 1 p p p p R R R .
Obviously, the previous procedure can be applied either to R or to O R depending on the reference frame considered.
Fig. 6 (color online). Arbitrary representation of a mixed state R with rank 2 and R rank Re 3, as the incoherent superposition of two pure states with different R directions of propagation. In spite of rank 2, R R corresponds to a 3D mixed state.
Since any superposition of pure states with the same directions of propagation is represented by a coherency matrix O R with 3 0 , we conclude that, when a 2, 3, the r t state R can be considered as the superposition of two pure states whose propagation directions are different. In other words, the previous analysis demonstrates the surprising result that a coherency matrix R with rank 2 can be synthesized through the superposition of two R pure states propagating along different directions (provided rank 3). O R
Consequently, in the case of 2, 3, the coherency matrix can be interpreted in terms r t of the following set of eight independent parameters:
- two orientation angles that determine the direction of propagation of the first component ˆ 1 p R of the arbitrary decomposition;
- three stokes parameters of the pure state 1 ˆ p R ;
- two orientation angles that determine the direction of propagation of the second pure component ˆ 2 p R of the arbitrary decomposition,
- and the coefficient 1 p of the convex sum in the arbitrary decomposition of R .
## 5.2.2.2. Characteristic decomposition .
In this case, the characteristic decomposition of R is formulated as follows
$$\text{ase, the characteristic decomposition of $R$ is formulated as follows} \\ \text{R} & = U \begin{bmatrix} \lambda _ { 1 } & 0 & 0 \\ 0 & \lambda _ { 2 } & 0 \\ 0 & 0 & 0 \end{bmatrix} \text{U} ^ { \dagger } = P _ { 1 } \text{R} _ { p } + ( 1 - P _ { 1 } ) \text{R} _ { m } ; \\ \text{R} _ { p } & \equiv I \text{U} \begin{bmatrix} 1 & 0 & 0 \\ 0 & 0 & 0 \\ 0 & 0 & 0 \end{bmatrix} \text{U} ^ { \dagger }, \text{ R} _ { m } = \frac { I } { 2 } \text{U} \begin{bmatrix} 1 & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 0 \end{bmatrix} \text{U} ^ { \dagger }, \\ \text{f the
} \text{$R$ \text{$R$
the
\text{$\text{$the $\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{ $\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text $\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\test{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\test$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text
$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text<$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text`$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text#$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text'$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\text{$\$$$
in terms of the pure component p R and the non-pure component m R . While p R has an immediate physical interpretation as a completely polarized state, the physical interpretation of m R requires the consideration of the value of the integer parameter rank Re m m r R :
a) when 2, m r m R represents a 2D unpolarized state 2 u D R propagating along a direction different from that of p R , and therefore R can be interpreted by means of the following eight parameters provided by the characteristic decomposition of R (Fig. 7):
- two orientation angles that determine the direction of propagation of the pure component p R ;
- three stokes parameters of the pure component p R ;
- two orientation angles that determine the direction of propagation of the 2D unpolarized component 2 m u D R R ,
- and the first index of polarimetric purity (or degree of polarization) 1 P of R , which is the coefficient of p R in the convex sum (we recall that, when 3, the second index r of polarimetric purity 2 P is equal to one, and thus the coefficient 2 1 P P of m R becomes 1 1 P ).
b) when 3, m r m R represents a 3D partially polarized (but not totally random) state without a well-defined direction of propagation. The state m R corresponds to the case (2b) studied previously. From the characteristic decomposition of R , it can be interpreted by means of the following eight parameters (Fig. 8):
- two orientation angles that determine the direction of propagation of the pure component p R ;
- three stokes parameters of the pure component p R ,
- and the three principal intensities of the electric field of m R , i.e., the eigenvalues of Re m R . Note that, with this choice, the value of 1 P is obtainable from the indicated set of parameters.
Fig. 7 (color online). Characteristic representation of a mixed state R with rank 2, R rank Re 3 and rank R Re 2, as the incoherent superposition of a m R pure state and a 2D unpolarized state, with different directions of propagation. In spite of rank 2, R R corresponds to a 3D mixed state.
Fig. 8 (color online). Characteristic representation of a mixed state R with rank 2, R rank Re 3 and rank R Re 3, as the incoherent superposition of a m R pure state and a 3D mixed state. In spite of rank 2, R R corresponds to a 3D mixed state.
$$5. 3. \ \text{rank} \, \mathbf R = 3 \,, \ \left ( a _ { 1 } \geq a _ { 2 } \geq a _ { 3 } > 0, \ \ a _ { 1 } a _ { 2 } a _ { 3 } > a _ { 1 } n _ { 1 } ^ { 2 } + a _ { 2 } n _ { 2 } ^ { 2 } + a _ { 3 } n _ { 3 } ^ { 2 } \right ).$$
In this case, the only achievable value of rank Re O t R is 3, so that the electric field t E of the electromagnetic wave has necessarily three nonzero orthogonal components, and thus the direction of propagation is not well defined. As in the previous case, in order to get appropriate physical interpretation, let us consider separately the arbitrary and characteristic decompositions.
## 5.3.1. Arbitrary decomposition
When rank 3, the arbitrary decomposition of R R can be performed either as indicated at the beginning of the present section, or through the following procedure [9]:
- 1. take an arbitrary 3D complex unit vector 1 w (note that 1 w necessarily belongs to Range R , because Range R covers the complete 3D complex space) and synthesize the normalized coherency matrix † 1 1 1 ˆ p R w w of the first component;
- 2. calculate the coefficient 1 p of 1 ˆ p R :
$$\text{$r$.} \text{ caname in the $\Gamma_{1}$} \circ \mathbb { m } _ { p 1 }. \\ p _ { 1 } = \left \{ \text{tr} \left [ \text{diag} ( \lambda _ { 1 }, \lambda _ { 2 }, \lambda _ { 3 } ) \hat { \mathbf R } _ { p 1 } \right ] \right \} & \Big / I \, ; \, \left ( I \equiv \lambda _ { 1 } + \lambda _ { 2 } + \lambda _ { 3 } \right ) ;$$
- 3. calculate the remainder coherency matrix † 1 1 1 1 ˆ ˆ 1 r p p R R w w ;
- 4. take an arbitrary 3D complex unit vector 2 w belonging to Range ˆ r R and synthesize the normalized coherency matrix † 2 2 2 ˆ p R w w of the second component;
- 5. calculate the coefficient 2 p of 2 ˆ p R in the convex sum
$$p _ { 2 } = \frac { 1 } { I } \text{tr} \left \{ \text{diag} \left ( \lambda _ { 1 }, \lambda _ { 2 }, \lambda _ { 3 } \right ) \left [ \text{w} _ { 2 } \otimes \text{w} _ { 2 } ^ { \dagger } \right ] \right \},$$
- 6. and calculate the third pure component and its coefficient through the expressions:
$$\begin{array} { c } \text{$\text{and can cause one una pure component and is convenient among use expressions.} \\ \hat { \mathbf R } _ { p 3 } = \left [ \hat { \mathbf R } - p _ { 1 } \left ( \mathbf w _ { 1 } \times \mathbf w _ { 1 } ^ { \dagger } \right ) - p _ { 2 } \left ( \mathbf w _ { 2 } \times \mathbf w _ { 2 } ^ { \dagger } \right ) \right ] \Big / \left ( 1 - p _ { 1 } - p _ { 2 } \right ), \quad p _ { 3 } = 1 - p _ { 1 } - p _ { 2 }. \end{array}$$
With respect to the directions of propagation of the three pure components of the arbitrary decomposition, there are two possibilities: 1) two components have the same direction of propagation, but different to that of the remainder component, and 2) the three pure components have different directions of propagation.
The arbitrary decomposition of R leads to its interpretation in terms of the following nine independent parameters (Fig. 9):
- two orientation angles that determine the direction of propagation of the first component ˆ 1 p R ;
- three stokes parameters of the pure state 1 ˆ p R ;
- two orientation angles that determine the direction of propagation of the pure component ˆ 2 p R (or, if the said direction coincides with the direction of propagation of
- 1 ˆ p R , the orientation angles that determine the direction of propagation of 3 ˆ p R ),
- and the coefficients 1 p and 2 p (recall that 3 1 2 1 p p p ).
Fig. 9 (color online). Arbitrary representation of a mixed state R with rank 3 as R the incoherent superposition of three pure states with different directions of propagation.
## 5.3.2. Characteristic decomposition
When rank 3 , the characteristic decomposition of R R has the general form
$$R = & \, U \begin{bmatrix} \lambda _ { 1 } & 0 & 0 \\ 0 & \lambda _ { 2 } & 0 \\ 0 & 0 & \lambda _ { 3 } \end{bmatrix} U ^ { \dagger } = P _ { 1 } R _ { p } + ( P _ { 2 } - P _ { 1 } ) R _ { m } + ( 1 - P _ { 2 } ) R _ { u - 3 D } ; \\ R _ { p } = & \, I U \begin{bmatrix} 1 & 0 & 0 \\ 0 & 0 & 0 \\ 0 & 0 & 0 \end{bmatrix} U ^ { \dagger }, \, R _ { m } = \frac { I } { 2 } U \begin{bmatrix} 1 & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 0 \end{bmatrix} U ^ { \dagger }, \, R _ { u - 3 D } = \frac { I } { 3 } \begin{bmatrix} 1 & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 1 \end{bmatrix}.$$
p R represents a pure state and 3 u D R represents a 3D unpolarized state ( 2 0 ). The P characteristic decomposition is completed with the second component m R , whose interpretation can be performed as follows in terms of the value of the auxiliary parameter rank Re m m r R .
a) when 2, m r m R represents a 2D unpolarized state 2 u D R propagating along a direction different than that of p R . Thus, R can be expressed as
$$\mathbf R = P _ { 1 } \mathbf R _ { p } + \left ( P _ { 2 } - P _ { 1 } \right ) \mathbf R _ { u - 2 D } + \left ( 1 - P _ { 2 } \right ) \mathbf R _ { u - 3 D },$$
and therefore, R can be interpreted by means of the following eight independent parameters (Fig. 10):
- two orientation angles that determine the direction of propagation of the pure component p R ;
- three stokes parameters of the pure component p R ;
- two orientation angles that determine the direction of propagation of the 2D unpolarized component 2 m u D R R ,
- and the two indices of polarimetric purity 1 2 , P P of R .
Fig. 10 (color online). Characteristic representation of a mixed state R with rank 3, and rank R Re 2, as the incoherent superposition of a pure m R state, a 2D unpolarized state (with different direction of propagation) and a 3D unpolarized state.
b) when 3, the components m r p R and m R can be recombined and re-decomposed in such a manner that the characteristic decomposition of R can be rewritten in the following appropriate form
$$\text{appropriate form} \\ \hat { R } _ { p 1 } = U \left [ \begin{matrix} 1 & 0 & 0 \\ 0 & 0 & 0 \\ 0 & 0 & 0 \end{matrix} \right ] U ^ { \dagger }, \hat { R } _ { p 2 } = \frac { 1 } { 2 } U \left [ \begin{matrix} 0 & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 0 \end{matrix} \right ] U ^ { \dagger }, \hat { R } _ { u - 3 D } = \frac { 1 } { 3 } \left [ \begin{matrix} 1 & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 1 \end{matrix} \right ], \\ \text{vhaar the two first commonate a runa tactic with different direction of non-union and}$$
where the two first components are pure states with different directions of propagation and 3 ˆ u D R is a 3D unpolarized state. Thus, R can be interpreted by means of the following nine independent parameters (Fig. 11):
- two orientation angles that determine the direction of propagation of the first pure component 1 p R ;
- three stokes parameters of 1 p R ;
- two orientation angles that determine the direction of propagation of the second pure component 2 p R ,
- and the two indices of purity 1 2 , P P of R .
Fig. 11 (color online). Characteristic representation of a mixed state R with rank 3, and R rank Re 3 as the incoherent superposition of two pure states (with different directions of m R propagation) and a 3D unpolarized state.
## 6. The degree of directionality
As it has been pointed out above, the 3D coherency matrix R is defined for a given fixed point r in the space, and does not contain direct information about the direction of
propagation of the electromagnetic wave. Nevertheless, in the previous section we have found that, when Re 2, the evolution of the electric field defines a fixed plane O t rank R , and consequently, the direction k perpendicular to that plane, and hence perpendicular to the wavefront, can properly be considered as the direction of propagation at point r . That is to say, the 3D state of polarization given by R is polarimetrically indistinguishable from a plane wave with the same coherency matrix R and propagating along the direction k .
Moreover, when 3, the evolution of the electric field of the wave defines the intensity t ellipsoid, so that, except for the case of 3D unpolarized states, the direction k perpendicular to the plane of symmetry containing the maximum ellipse (Fig. 1), can be considered as a the average direction of propagation. Thus, it is highly desirable to define a measure of the directional purity of the state R , i.e., a measure of the distance from the directionality of R to the random directionality of a 3D unpolarized state. As a previous attempt to get a definition of such a degree of directionality , we proposed the use of the second index of polarimetric purity 2 P , but the analysis performed in the previous sections shows that, when rank 2 and r R 3, then t 2 1, despite the fact that the direction of propagation is not P well defined. On the light of the interpretation of the intrinsic coherency matrix O R R and inspired by the definition of 2 P in terms of the eigenvalues of R , we find that the desired definition of the degree of directionality d P is given by the nondimensional quantity
$$P _ { d } \equiv \frac { a _ { 1 } + a _ { 2 } - 2 a _ { 3 } } { a _ { 1 } + a _ { 2 } + a _ { 3 } } \,,$$
where, as indicated in Eq. (15), the real, non-negative, quantities 1 2 3 , , a a a , are defined from the diagonalization 1 2 3 diag , , T R a a a Q R Q of the real part R R of R , with the choice 3 2 1 0 a a a . The degree of directionality d P reaches its maximum value 1 for d P 3 0 . a Moreover 0 corresponds exclusively to a 3D unpolarized state d P 3 u D R , whose associated direction of propagation is completely random (i.e., the intensity ellipsoid is a sphere). In agreement with the case study performed above, intermediate cases have appropriate and consistent values in the interval 0 1. Thus, in our opinion, d P d P is a proper measure of the degree of directionality.
The particular case of a pure linearly polarized state ( 2 3 0) is out of the range of a a application of d P . In fact, in this limiting case 1, while the corresponding coherency d P matrix can be obtained trough the incoherent superposition of an arbitrary set of linearly polarized pure states propagating along different directions, provided that such directions lie in the plane perpendicular to the polarization axis.
By taking advantage of the analogies between the expressions derived from the eigenvalues 1 2 3 , , of R and that derived from the eigenvalues 1 2 3 , , a a a of Re R , we define the degree of linear polarization l P as the nondimensional parameter
$$P _ { l } \equiv \frac { a _ { 1 } - a _ { 2 } } { a _ { 1 } + a _ { 2 } + a _ { 3 } } \,,$$
whose possible values are limited by 0 l d P P 1 . The minimum value d P 0 l P 1 2 a a corresponds to a state with an intensity ellipsoid of revolution with semiaxes 1 1 3 , , a a a , so that its intersection with the plane perpendicular to the average direction of
propagation k is a circumference of radio 1 a . Regardless the value of d P , 0 corresponds l P to states with equal principal intensities in the plane perpendicular to k . When, in particular, 0 , the intensity ellipsoid is a sphere and corresponds to a 3D unpolarized state. The d P maximum value 1, entails l P 1 and corresponds to a pure state of linear polarization. In d P the case of 2D states 1 , d P 2 2 1 2 0 l P s s s , as corresponds to the natural and commonly used definition of the degree of linear polarization. Thus, the inspection of the values of l P for all the possible 3D states of polarization shows that l P gives an appropriate and consistent measure of the degree of linear polarization.
Regardless the intrinsic physical meaning of 1 2 3 , , a a a as the semiaxes of the intensity ellipsoid, equivalent physical information can be represented through the set of quantities , , l d I P P , whose physical meaning is particularly appropriate for the study and analysis of three-dimensional states of polarization.
## 7. Conclusions
At a given point in the space, the second-order state of polarization of an arbitrary electromagnetic wave is characterized by means of the corresponding coherency matrix R , which can be interpreted in terms of a set of nine well-defined parameters, namely [13], the principal intensities 1 2 3 , , a a a ; the angular momentum 1 2 3 , , n n n n , and the set of three orientation angles , , . This parameterization constitutes an adequate framework for the analysis and interpretation of any three-dimensional state of polarization through the corresponding arbitrary and characteristic decompositions.
Together with the integer parameter rank r R , which determines the minimum number of pure components in the arbitrary decomposition, we have found that the integer parameter rank Re t R plays a fundamental role in the physical interpretation of the possible threedimensional states of polarization. 1 corresponds to pure states with linear polarization. In t this case, as indicated in the previous section, the direction of propagation is not determined by R , even though it is true that the experimentalists have usually complementary information enough to determine it. When 2 , the state t R is reduced to a conventional two-dimensional state of polarization, with a well-defined direction of propagation; thus, the arbitrary and the characteristic decompositions of R translate into that of two-dimensional states and, in particular, the arbitrary decomposition is necessarily composed of two pure states with the same directions of propagation; moreover, the characteristic decomposition has the wellknown form of a superposition of a pure state and a unpolarized two-dimensional state. In the case that 3, even if t 2 , it is no longer possible to assign a well-defined direction of r propagation to the state R , so that at least two of the three pure arbitrary components have different propagation directions; moreover, the characteristic decomposition adopts particular forms depending on the value of an additional auxiliary integer parameter. The case study performed clarifies the interpretation and the role played by the set of two indices of polarimetric purity 1 2 , P P as physically invariant quantities that give nondimensional appropriate measures of the structure of purity of a state R , beyond the overall information provided by the degree of polarimetric purity 2 2 (3) 1 2 3 2. P P P
Inspired by the expression of the second index of polarimetric purity 2 P in terms of the eigenvalues of R , the degree of directionality d P has been defined, which gives an appropriate
and consistent measure of the stability of the direction of propagation of the state of polarization represented by R .
The transformation of the coherency matrix performed through the appropriate rotation of the reference frame, provides the intrinsic coherency matrix O R characterized by the set of six parameters 1 2 3 1 2 3 , , ; , , a a a n n n . The physical information contained in O R can also be represented through the following alternative set of meaningful parameters: intensity , degree I of polarization 1 P , second index of polarimetric purity 2 P (i.e., the relative portion of the wave obtained once the 3D unpolarized portion has been subtracted); the degree of directionality d P , the degree of linear polarization l P and the magnitude n of the angular momentum n .
In summary, the approach presented, we think, constitutes a useful tool for the study, representation and interpretation of the complete variety of three-dimensional states of polarization in terms of appropriate and well-defined physical parameters.
## Acknowledgement
This research was supported by the Spanish Ministry of Economía y Competitividad, Grant FIS2011-22496, and by the Gobierno de Aragón, group E99.
## References
- 1. T. Setälä, A. Shevchenko, M. Kaivola and A. T. Friberg, 'Degree of polarization for optical near fields' Phys. Rev. E 66 , 016615-7 (2002).
- 2. T. Setälä, M. Kaivola and A. T. Friberg, 'Degree of polarization on near fields of thermal sources: Effects of surface waves' Phys. Rev. Lett. 88 , 1239021-4 (2002)
- 3. J. J. Gil, J. M. Correas, P. A. Melero and C. Ferreira, 'Generalized polarization algebra' Monografías del Seminario Matemático García de Galdeano 31 , 161-167 (2004). Available from <http://www.unizar.es/galdeano/actas\_pau/PDFVIII/pp161-167.pdf>
- 4. J. Ellis and A. Dogariu, 'On the degree of polarization of random electromagnetic fields' Opt. Commun. 253 , 257-265 (2005).
- 5. A. Luis, 'Degree of polarization for three-dimensional fields as a distance between correlation matrices' Opt. Commun. 253 , 10-14 (2005)
- 6. A. Luis, 'Polarization distribution and degree of polarization for three-dimensional quantum light fields' Phys. Rev. A 71 , 063815-8 (2005)
- 7. C. Brosseau and A. Dogariu, 'Symmetry properties and polarization descriptors for an arbitrary electromagnetic wavefield,' in Vol. 49 of Progress in Optics , E. Wolf, ed. (Elsevier, 2006), Chap. 4.
- 8. M. R. Dennis, 'A three-dimensional degree of polarization based on Rayleigh scattering' J. Opt. Soc. Am. A 24 , 2065-2069 (2007).
- 9. J. J. Gil, "Polarimetric characterization of light and media," Eur. Phys. J. Appl. Phys. 40 , 1-47 (2007).
- 10. I. San José and J. J. Gil, "Invariant indices of polarimetric purity. Generalized indices of purity for n×n covariance matrices" Opt. Commun. 284 , 38-47 (2011).
- 11. J. M. Auñón and Manuel Nieto-Vesperinas, "On two definitions of the threedimensional degree of polarization in the near field of statistically homogeneous partially coherent sources," Opt. Lett. 38 , 58-60 (2013)
- 12. O. Gamel and D. F. V. James, "Majorization and measures of classical polarization in three dimensions," J. Opt. Soc. Am. A 31 , 1620-1626 (2014).
- 13. M. R. Dennis, 'Geometric interpretation of the three-dimensional coherence matrix for nonparaxial polarization' J. Opt. A: Pure Appl. Opt. 6 , S26-S31 (2004).
- 14. J. H. Hannay, 'The Majorana representation of polarization, and the Berry phase of light' J. Mod Opt. 45 , 1001-1008 (1998).
- 15. T. Saastamoinnen, J. Tervo, 'Geometric approach to the degree of polarization for arbitrary fields' J. Mod. Opt. 51 , 2039-2045 (2004).
- 16. T. Carozzi, R. Karlsson and J. Bergman, 'Parameters characterizing electromagnetic wave polarization,' Phys. Rev. E 61 , 2024-2028 (2000).
- 17. J. J. Gil, I. San José, "3D polarimetric purity," Opt. Commun. 283 , 4430-4434 (2010).
- 18. C. J. R. Sheppard, "Partial polarization in three dimensions," J. Opt. Soc. Am. A 28 , 2655-2659 (2011).
- 19. C. J. R. Sheppard, "Geometric representation for partial polarization in three dimensions" Opt. Lett. 37 , 2772-2774 (2012).
- 20. R. M. A. Azzam, "Three-dimensional polarization states of monochromatic light fields," J. Opt. Soc. Am. A 28 , 2279-2283 (2011)
- 21. J. J. Gil and I. San José, "Polarimetric subtraction of Mueller matrices," J. Opt. Soc. Am. A 30 , 1078-88 (2013).
- 22. R. Dändliker, P. Tortora, L. Vaccaro and A. Nesci, 'Measuring three-dimensional polarization with scanning optical probes' J. Opt. A Pure Appl. Opt. 6 , S18-S23 (2004)
- 23. J. Ellis and A. Dogariu, 'Optical Polarimetry of Random Fields' Phys. Rev. Lett. 95 , 203905-4 (2005)
- 24. J. Ellis, A. Dogariu, S. Ponomarenko and E. Wolf, 'Correlation matrix of a completely polarized, statistically stationary electromagnetic field' Opt. Lett. 29 1536-1538 (2004)
- 25. J. Ellis and A. Dogariu, S. Ponomarenko and E. Wolf, 'Degree of polarization of statistically stationary electromagnetic fields' Opt. Commun. 248 , 333-337 (2005).
- 26. F. T. Hioe, 'Isotropy and polarization in a statistically stationary electromagnetic field in three-dimensions' J. Mod. Opt. 53 , 1715-1725 (2006)
- 27. P. Réfrégier, M. Roche and F. Goudail, 'Invariant polarimetric contrast parameters of light with Gaussian fluctuations in three dimensions' J. Opt. Soc. Am. A 23 , 124-133 (2006)
- 28. P. Réfrégier, V. Wasik, K. Vynck and R. Carminati, "Analysis of coherence properties of partially polarized light in 3D and application to disordered media," Opt. Lett. 39 , 2362-2365 (2014).
- 29. A. Luis, 'Quantum polarization for three-dimensional fields via Stokes operators' Phys. Rev. A. 71 , 1023810-7 (2005)
- 30. U. Fano, 'Description of states in quantum mechanics by density matrix and operator techniques,' Rev. Mod. Phys. 29 , 74-93 (1957).
- 31. J.C. Samson, 'Description of the polarization states of vector processes: Applications to ULF magnetic fields' Geophys. J. R. Astron. Soc. 34 , 403-419 (1973).
- 32. R. Barakat, ' n -Fold polarization measures and associated thermodynamic entropy of N partially coherent pencils of radiation' Opt. Acta 30 , 1171-1182 (1983).
- 33. J. J. Gil and I. San José, R. Ossikovski, 'Serial-parallel decompositions of Mueller matrices," J. Opt. Soc. Am. A 30 , 32-50 (2013).
- 34. J. J. Gil, " Review on Mueller matrix algebra for the analysis of polarimetric measurements," J. Appl. Remote Sens. 8 , 081599 (2014).
- 35. F. Snik, J. Craven-Jones, M. Escuti, S. Fineschi, D. Harrington, A. de Martino, D. Mawet, J, Riedi and J. S. Tyo, "An overview of polarimetric sensing techniques and technology with applications to different research fields," Proc. SPIE 9099 , Polarization: Measurement, Analysis, and Remote Sensing XI, 90990B (May 21, 2014) | 10.1103/PhysRevA.90.043858 | [
"Jose J. Gil"
] | 2014-08-01T21:54:34+00:00 | 2014-08-01T21:54:34+00:00 | [
"physics.optics",
"quant-ph"
] | Interpretation of the coherency matrix for three-dimensional polarization states | From an appropriate parameterization of the three-dimensional (3D) coherency
matrix R, that characterizes the second-order, classical states of
polarization, the coherency matrices are classified and interpreted in terms of
incoherent decompositions. The relevant physical quantities derived from R, as
the intensity, the degree of polarimetric purity, the indices of polarimetric
purity, the angular momentum, the degree of directionality and the degree of
linear polarization are identified and interpreted on the light of the case
study performed. The information provided by R about the direction of
propagation is clarified and it is found that coherency matrices with, does not
always represent states with a well-defined direction of propagation. Moreover,
it is demonstrated the existence of 3D mixed states that cannot be decomposed
into a superposition of a pure state, a 2D unpolarized state, and a 3D
unpolarized state. Appropriate representation and interpretation for all the
different types of 3D coherency matrices is provided through physical
consistent criteria. Under the approach proposed, the conventional
two-dimensional model arises naturally. |
1408.0313v1 | ## Tropical Optimization Problems ∗
## Nikolai Krivulin † ‡
## Abstract
We consider optimization problems that are formulated and solved in the framework of tropical mathematics. The problems consist in minimizing or maximizing functionals defined on vectors of finite-dimensional semimodules over idempotent semifields, and may involve constraints in the form of linear equations and inequalities. The objective function can be either a linear function or nonlinear function calculated by means of multiplicative conjugate transposition of vectors. We start with an overview of known tropical optimization problems and solution methods. Then, we formulate certain new problems and present direct solutions to the problems in a closed compact vector form suitable for further analysis and applications. For many problems, the results obtained are complete solutions.
Key-Words: idempotent semifield, tropical optimization, nonlinear objective function, linear constraints, Chebyshev approximation.
MSC (2010): 65K10, 15A80, 65K05, 90C48, 41A50
## 1 Introduction
Tropical (idempotent) mathematics presents an area of applied mathematics that focuses on the theory and applications of semirings with idempotent addition. Among the first works, which laid the foundation for the area, were the pioneering papers by Pandit [1], Cuninghame-Green [2], Giffler [3] and Hoffman [4], dating back to the early 1960s.
The classical works published at the same time or somewhat earlier by Kleene [5], Bellman [6], and Bellman and Karush [7, 8] also served as an essential prerequisite. Specifically, a significant effect was exerted on the
∗ Advances in Economics and Optimization: Collected Scientific Studies Dedicated to the Memory of L. V. Kantorovich (L. A. Petrosyan, J. V. Romanovsky, D. W. K. Yeung, eds), pp. 195-214, Nova Science Publishers, New York, 2014.
† Faculty of Mathematics and Mechanics, Saint Petersburg State University, 28 Universitetsky Ave., Saint Petersburg, 198504, Russia, [email protected].
‡ The work was supported in part by the Russian Foundation for Humanities under Grant #13-02-00338.
area by the Bellman-Karush maximum transform. This transform, which was initially introduced in [7] in a multiplicative form, has taken an additive form in [8] and in many later publications. The additive form of the transform is in direct connection with the theory of support functions in convex optimization [9]. Note that this additive form was also proposed by Romanovsky [10].
A notable contribution to the development of the area was made by the Leningrad (St. Petersburg) researchers N. N. Vorob'ev, J. V. Romanovsky and A. A. Korbut. In the paper [11] published in 1963, Vorob'ev provided a groundwork for the theory of linear operators on extremal spaces, which were defined as vector spaces over idempotent semifields with the operations of maximum and minimum in the role of addition.
Then came the works [12, 13, 14, 15], in which Romanovsky studied asymptotic properties of solutions in dynamic programming problems and obtained an idempotent analogue for the Gelfand spectral radius theorem for bounded linear operators in Banach spaces. These results have proved to be conceptually close to the theory of the extremal spaces, which was further developed by Vorob'ev [16, 17] and Korbut [18, 19].
An extensive advancement in the theory and applications of tropical mathematics was achieved by V. P. Maslov and his colleagues, who have introduced idempotent functional analysis as a new important field within the area. The results obtained by the Maslov scientific school and published in the monographs by Maslov and Samborski˘ ı [20], Maslov and Kolokoltsov [21], Kolokoltsov and Maslov [22], as well as in many research papers, have provided a strong analytical framework for the contemporary tropical mathematics.
In the last few decades, tropical mathematics has received considerable attention as both a powerful mathematics apparatus and a flexible application tool. Advances in the theory and applications have resulted in a great many publications, including the monographs by Cuninghame-Green [23], Carr´ e [24], U. Zimmermann [25], Baccelli et al. [26], Golan [27], Heidergott, Olsder and van der Woude [28], Gondran and Minoux [29], Krivulin [30], Butkoviˇ c [31], McEneaney [32], and in a large amount of contributed papers.
Optimization problems, which can be formulated and solved in terms of tropical mathematics, form an important research domain within the area. These problems arise in many real-world applications in job scheduling, location analysis, transportation networks, decision making, discrete event dynamic systems, and in other fields.
Optimization problems have been examined even in the early publications [2, 4]. Since that time a number of studies that investigate optimization problems in the context of tropical mathematics has been published, which includes the monographs [23, 25, 30, 31].
Tropical optimization problems are commonly formulated to minimize or maximize functionals defined on vectors of a finite-dimensional semimodule
over an idempotent semifield, and may include additional constraints in the form of equations and inequalities. Some problems have both the objective function and constraints that are linear in the sense of tropical mathematics. These problems, which can be considered as direct analogues to linear programming problems, were examined in [4, 33, 25].
In another class of problems [34, 23, 25], the constraints are linear, but the objective functions are not. In these problems, the objective function can usually be written by means of a multiplicative conjugate transposition operator, which does not preserve linearity.
To solve the problems considered, a variety of methods were proposed. Some problems in a rather general setting can be completely solved in an explicit form [4, 34, 33, 25]. For other problems, only algorithmic solutions are known in the form of iterative numerical procedures that produce a particular solution or indicate that no solution exists [35, 36].
In this chapter, we offer a brief overview of known tropical optimization problems and solution methods. We formulate certain new problems and present direct solutions to the problems in a closed compact vector form suitable for further analysis and applications. For many problems, the results obtained are complete solutions.
The chapter is organized as follows. In Section 2, we briefly outline the definitions, notation, and key facts of tropical mathematics to be used in the subsequent sections. Section 3 is devoted to known optimization problems. We present new tropical optimization problems and describe obtained solutions in Section 4. The chapter concludes with an extensive list of references on the topic considered.
## 2 Preliminary Definitions, Notation and Results
We start with a brief overview of the basic elements of tropical (idempotent) mathematics to provide a formal unifying framework for formulating tropical optimization problems and representing the solutions to the problems in the subsequent sections. The overview mainly follows the notation system and results developed by Krivulin [37, 38, 39, 30], which underlie the solutions presented.
The literature on the topic offers a wide range of publications written in varied styles and available on both the introductory and advanced levels. Specifically, the recent monographs: Kolokoltsov and Maslov [22], Golan [27], Heidergott, Olsder and van der Woude [28], Gondran and Minoux [29, 31], Butkoviˇ [31]; and overview papers: Kolokoltsov [40], Litvinov [41], c Akian, Bapat and Gaubert [42], provide a broad source for further reading.
## 2.1 Idempotent Semifield
An idempotent semifield is an algebraic system ( X , ⊕ ⊗ , , 0 1 , ), where X is a nonempty set closed under addition ⊕ and multiplication ⊗ , and equipped with zero 0 and one 1 , such that ( X , ⊕ , 0 ) is a commutative idempotent monoid, ( X , ⊗ , 1 ) is an abelian group, multiplication distributes over addition, and 0 is absorbing for multiplication.
In addition, we assume the semifield to admit a linear order, which is consistent with the partial order induced by idempotent addition to define x ≤ y if and only if x ⊕ y = y . Moreover, the semifield is assumed to be algebraically complete (radicable), which implies that the equation x p = a has a root for all a ∈ X and every natural p .
Examples of the linearly ordered and algebraically complete idempotent semifield include R max + , = ( R ∪ {-∞} , max + , , -∞ , 0), R min + , = ( R ∪{ + ∞} , min + + , , ∞ , 0), R max , × = ( R + ∪{ } 0 , max , × , 0 1) and , R min , × = ( R + ∪{ + ∞} , min , × , + ∞ , 1), where R is the set of real numbers and R + = { x ∈ R | x > 0 } .
Below, we drop the multiplication sign ⊗ for the brevity sake. The relation symbols and the optimization problems are thought of in terms of the linear order on the semifield.
## 2.2 Algebra of Matrices and Vectors
The set of matrices with m rows and n columns over X is denoted by X m n × . A matrix with all entries equal to 0 is the zero matrix. A matrix is column-regular (row-regular) if it has no zero columns (rows). A matrix that has no zero rows nor zero columns is called regular.
Addition and multiplication of conforming matrices and scalar multiplication are performed component-wise in the usual way, except that the ordinary scalar addition is replaced by ⊕ and the multiplication by ⊗ .
/negationslash
The multiplicative conjugate transpose of any nonzero matrix A = ( a ij ) ∈ X m n × is the matrix A -= ( a -ij ) ∈ X n × m , where a -ij = a -1 ji if a ji = 0 , and a -ij = 0 otherwise.
A matrix with only one column (row) is a column (row) vector. The set of column vectors of size n forms an idempotent semimodule over X and is denoted X n . A vector with all elements equal to 0 is the zero vector. A vector that has no zero elements is regular.
/negationslash
The multiplicative conjugate transpose of any nonzero vector x = ( x i ) ∈ X n is the row vector x -= ( x -i ), where x -i = x -1 i if x i = 0 , and x -i = 0 otherwise.
For any vector x ∈ X n and matrix A ∈ X m n × , idempotent analogues of the vector and matrix norms are given by
$$\| x \| = \bigoplus _ { i = 1 } ^ { n } x _ { i }, \quad \| \mathcal { A } \| = \bigoplus _ { i = 1 } ^ { m } \bigoplus _ { j = 1 } ^ { n } a _ { i j }.$$
With the vector 1 = ( 1 , . . . , 1 ) T , which has all components equal to 1 , we write
$$\| x \| = \mathbb { 1 } ^ { T } x, \quad \| A \| = \mathbb { 1 } ^ { T } A \mathbb { 1 }.$$
Consider square matrices of order n in X n × n . A matrix with 1 's on the diagonal and 0 's elsewhere is the identity matrix denoted I .
The trace of any matrix A = ( a ij ) is given by tr A = a 11 ⊕··· ⊕ a nn . To represent solutions to optimization problems below, we use a function that takes any matrix A to the scalar
$$\text{Tr} ( A ) = \text{tr} \, A \oplus \cdots \oplus \text{tr} \, A ^ { n }.$$
If the condition Tr( A ) ≤ 1 holds, we apply the asterate operator, which produces the matrix (sometimes called the Kleene star)
$$A ^ { * } = I \oplus A \oplus \cdots \oplus A ^ { n - 1 }.$$
If Tr( A ) = 1 , we form the matrix A + , which holds those columns of the matrix A × = AA ∗ that have 1 on the diagonal.
## 2.3 Linear Operators and Their Spectral Properties
Every matrix A ∈ X n × n defines a linear operator on X n , which is endowed with certain spectral properties. A scalar λ ∈ X and a nonzero vector x ∈ X n are corresponding eigenvalue and eigenvector of A if they satisfy the equation
$$A x = \lambda x.$$
The matrix A always possesses an eigenvalue that is calculated by the formula
$$\lambda = \bigoplus _ { m = 1 } ^ { n } \text{tr} ^ { 1 / m } ( A ^ { m } ).$$
Any other eigenvalue of A , if it exists, is not greater (in the sense of the order on X ) than λ , which is called the spectral radius of A .
All eigenvectors of the matrix A , which correspond to the spectral radius λ , are given by
$$x = ( \lambda ^ { - 1 } A ) ^ { + } u,$$
where u is any regular vector of appropriate size.
## 3 An Overview of Known Problems and Solution Methods
Multidimensional optimization problems form an important research domain in tropical mathematics, dating back to the early works by CuninghameGreen [2] and Hoffman [4]. These problems appear in various application
areas, including job scheduling [2, 34, 23, 25, 36, 43, 44, 45, 46, 47], location analysis [48, 49, 50], transportation networks [25, 43], decision making [51, 52, 53], and discrete event dynamic systems [54, 55, 56, 37].
In this section, we offer a brief overview of known tropical optimization problems and discuss existing solution methods. We consider problems of minimizing or maximizing both linear and nonlinear functionals defined on vectors in finite-dimensional semimodules over idempotent semifields, subject to constraints expressed as linear equations and inequalities. The overview covers the problems with nonlinear objective functions that are, or can be, written by means of multiplicative conjugate transposition of vectors. Our analysis of the literature shows that many problems, which are relevant to tropical optimization, have objective functions that admit this form.
Note that the problems with nonlinear objective functions are usually considered in a different way; these problems are formulated and solved either in the usual setting within the framework of ordinary mathematics [57, 36, 58, 59, 43], or in terms of two dual semifields R max + , and R min + , at once [34, 23, 45, 46].
In the description of the problems below, we use the capital letters A , B and C for known matrices and the lower case letters b , d , p and q for known vectors. The symbol x represents the unknown vector. The matrix and vector operations are thought of as defined in terms of an idempotent semifield. The minus sign in the exponent indicates multiplicative conjugate transposition of vectors and matrices.
Specifically, in terms of the real semifield R max + , , vector addition, matrixvector multiplication and scalar multiplication follow the standard rules, where the operations max and + play the respective roles of addition and multiplication. Multiplicative conjugate transposition of a vector implies replacing each element of the vector by its inverse (which coincides with its opposite in ordinary arithmetic), and transposition.
## 3.1 Problems with Linear Objective Functions
One of the first optimization problems defined and solved in terms of tropical mathematics is a formal analogue of linear programming problems, which takes the form
$$\minimize & \quad p ^ { T } x, \\ \text{subject to } & \ A x \geq d.$$
Exact general solutions to the problem are obtained under various assumptions concerning the idempotent semiring, in which context the problem is examined. Specifically, Hoffman [4] considered a rather general idempotent semiring and proposed a solution based on abstract extension of the duality principle in linear programming. Another general solution was established by U. Zimmermann [25] by means of a residual-based solution tech-
nique. The common approach in [4] was further developed by Superville [33] to examine the problem in the context of the semifield R max + , . The problem was also solved by Gavalec and K. Zimmermann [60] for the semifields R max + , and R max , × in the framework of max-separable functions.
The results of the theory of max-separable functions were applied by K. Zimmermann [57, 36, 58, 59, 43] to solve the problem with more constraints
$$\minimize & \quad p ^ { T } x, \\ & \text{subject to } \quad A x \leq d, \quad C x \geq b, \\ & \quad g \leq x \leq h. \\. &.$$
Under general conditions, an exact solution was obtained, which was, however, given in ordinary terms rather than in terms of tropical mathematics.
Butkoviˇ c [35], Butkoviˇ and Aminu [44, 47] studied a problem that has c constraints in the form of two-sided equation (with the unknown vector on both left and right sides), and can be written as
$$& \minimize \quad p ^ { T } x, \\ & \text{subject to } \quad A x \oplus b = C x \oplus d.$$
Specifically, a pseudo-polynomial algorithm was proposed in [44], which produces a solution if solutions exist or indicates that the problem has no solutions. The algorithm applies an alternating method proposed in [61] to replace the equality constraints by two opposite inequalities, which are alternately solved to achieve more and more accurate estimates for the solution. An heuristic approach that combines a search scheme providing approximate solutions with iterative procedures of solving low-dimensional problems with one or two unknown variables was developed in [47].
## 3.2 Problems with Nonlinear Objective Functions
The tropical optimization problems with nonlinear objective functions, in which nonlinearity results from the application of multiplicative conjugate transposition, form a rich class of various problems that arise in many applications. We divide the problems into a few groups according to the particular form of the objective function and the principal interpretation of the problems.
## 3.2.1 Chebyshev Approximation
Among the first problems with nonlinear objective functions was the problem, which is formulated in the form
$$\minimize & \quad ( A x ) ^ { - } p, \\ \text{subject to } & \ A x \leq p.$$
The problem was considered by Cuninghame-Green [34] in the context of approximation in the semifield R max + , with the Chebyshev metric. The problem required finding a vector x that gives the best underestimating approximation to a vector p by means of vectors Ax . A general solution in an explicit form was obtained based on the theory of linear operators on vectors over the semifield R max + , . A similar solution was provided by U. Zimmermann [25].
An unconstrained approximation problem formulated in terms of a general idempotent semifield was examined by Krivulin [30] in the form
$$\minimize \ ( A x ) ^ { - } p \oplus p ^ { - } A x.$$
To solve the problem, an approach was proposed, which involves the derivation of a sharp lower bound for the objective function. The form of this bound is exploited to construct vectors that produce the bound. The results obtained were applied to establish existence conditions and to describe all solutions in a closed vector form for the equation Ax = p . In addition, the following problems of both underestimating and overestimating approximation were solved as consequences of the results:
$$\minimize & \quad ( A x ) ^ { - } p, & \quad \minimize & \quad p ^ { - } A x, \\ \text{subject to } & \ A x \leq p ; & \quad \text{subject to } & \ A x \geq p.$$
Certain of the problems, which were originally formulated and solved in a different setting, can be represented in terms of tropical mathematics as problems with nonlinear objective functions and linear constraints and thus are also included in the overview. In particular, a constrained problem of the Chebyshev approximation in the semifield R max + , was examined in [36], where the problem was solved with a polynomial-time threshold-type algorithm. In terms of tropical mathematics, this problem can be written in the form
$$\minimize \quad & ( A x ) ^ { - } p \oplus p ^ { - } A x, \\ \text{subject to} \quad & g \leq x \leq h.$$
## 3.2.2 Problems with Span Seminorm
The problems examined by Butkoviˇ c and Tam [45, 46] in the context of the semifield R max + , had the objective function defined as the span seminorm, which calculates the maximum deviation between elements of a vector. Solutions to both minimization and maximization problems were obtained using a combined technique based on the representation in terms of both the semifield R max + , and its dual R min + , . Note that these problems can be described only in terms of the semifield R max + , as follows
$$\minimize \ 1 ^ { T } A x ( A x ) ^ { - } 1, \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
where 1 = ( 1 , . . . , 1 ) T is a vector of ones in the sense of R max + , .
## 3.2.3 A Problem of 'Linear-fractional' Programming
A minimization problem that has constraints in the form of a two-sided inequality, and that is written in terms of the semifield R max + , as
$$\minimize & \quad p ^ { T } x ( q ^ { T } x ) ^ { - 1 }, \\ \text{subject to } & \ A x \oplus b \leq C x \oplus d,$$
was investigated by Gaubert, Katz and Sergeev [62]. The problem, called a tropical linear-fractional programming problem, was solved by means of an iterative computational scheme that exploits the relationship established by Akian, Gaubert and Guterman [63] between the solutions to two-sided vector equations in the sense of R max + , and mean payoff games.
## 3.2.4 Extremal Property of the Spectral Radius
A problem examined by Cuninghame-Green [2] to minimize a functional defined on vectors over the semifield R max + , was apparently the first optimization problem, which arose in the context of tropical mathematics. With the use of multiplicative conjugate transposition, the problem can be represented in the form
$$\minimize \ x ^ { - } A x.$$
As one of the main result in [2], which gave impetus to further research on the spectral theory of linear operators in tropical mathematics, it was shown that the minimum in the problem is equal to the spectral radius λ of the matrix A and is attained at any eigenvector that corresponds to the radius. Moreover, explicit expressions were obtained to calculate the spectral radius and the eigenvector in terms of standard arithmetic operations. Similar results were provided by Engel and Schneider [64] and by Superville [33].
Furthermore, the above solution was extended and represented in more general terms of tropical mathematics by Cuninghame-Green [23]. Analogues results in a compact vector form using multiplicative conjugate transposition were proposed by Krivulin [37, 38, 30].
To find all vectors that yield the minimum in the problem, a computational approach was proposed in [23]. It consists in solving a linear programming problem and thus does not guarantee a direct solution in an explicit form.
Finally, Elsner and van den Driessche [51, 52] indicated that the solution set includes, in addition to the eigenvectors, all vectors x that satisfy the inequality Ax ≤ λ x . An iterative computational procedure was offered to obtain vectors from this solution set.
## 4 Some New Problems and Recent Results
In this part, we present several new problems [65, 66, 67, 68, 69, 70], which are formulated and solved in terms of a general algebraically complete linearly ordered idempotent semifield. If a complete solution to a problem is obtained, the solution is given below by an expression that represents a general vector describing all vectors in the solution set, and that is referred to as the general solution.
## 4.1 Chebyshev-like Approximation Problems
We start with constrained optimization problems with nonlinear objective functions that have the form similar to that in the Chebyshev approximation problems (1) and (2). One of the application areas for these problems is the single facility location problems in multidimensional spaces with Chebyshev distance under various constraints in the form of linear equalities and inequalities (see, e. g., [50, 68, 67]).
To solve the next two problems with simple boundary constraints, we apply a general approach which was proposed and developed in [37, 38, 30]. The approach involves the derivation of a sharp lower bound on the objective function and the use of this bound to obtain solution vectors.
First, consider the following problem: given vectors p q g h , , , ∈ X n , find regular vectors x ∈ X n such that
$$\minimize & \quad q ^ { - } x \oplus x ^ { - } p, \\ \text{subject to } & \quad g \leq x \leq h.$$
Theorem 1 ([68]) . Let x be the general regular solution to problem (5) , where p and q are regular vectors, g and h are vectors such that g ≤ h , and let ∆ = √ q -p .
Then, the minimum in (5) is equal to
$$\mu = \Delta \oplus q ^ { - } g \oplus h ^ { - } p,$$
and the general solution is given by
$$\mu ^ { - 1 } p \oplus g \leq x \leq ( \mu ^ { - 1 } q ^ { - } \oplus h ^ { - } ) ^ { - }.$$
Suppose that, given a matrix A ∈ X m n × and vectors p q , ∈ X m , g ∈ X n , the problem is to find regular vectors x ∈ X n that
$$\minimize \quad & q ^ { - } A x \oplus ( A x ) ^ { - } p, \\ \text{subject to} \quad & x \geq g.$$
Theorem 2 ([68]) . Let x be a regular solution to problem (6) , where A is a regular matrix, p and q are regular vectors, and let ∆ = √ ( A q ( -A ) - -) p .
Then, the minimum in (6) is equal to
$$\mu = \Delta \oplus q ^ { - } A g,$$
and attained at
$$x = \mu ( q ^ { - } A ) ^ { - }.$$
We now consider another problem with conditions that include a linear inequality with a matrix. Given vectors p q g h , , , ∈ X n and a matrix B ∈ X n × n , find regular vectors x ∈ X n that solve the problem
$$\minimize & \quad x ^ { - } p \oplus q ^ { - } x, \\ \text{subject to } & \ B x \oplus g \leq x, \\ & \quad x \leq h.$$
Note that the conditions in the problem can be represented in the equivalent form
$$B x & \leq x, \\ g & \leq x \leq h.$$
The solution of the problem follows an approach that is based on the general solution of linear inequalities obtained in [39, 30] and refined in [71]. We introduce an additional parameter, which indicates the minimum value of the objective function. The problem is then reduced to the solving of a linear inequality with a matrix that depends on the parameter. We exploit the existence condition for solutions of the inequality to evaluate the parameter, and then take the general solution to the inequality as the solution to the initial problem.
Theorem 3 ([67]) . Let x be the general regular solution to problem (7) , where B is a matrix such that Tr( B ) ≤ 1 , p is a nonzero vector, q and h are regular vectors, and g is a vector such that h B g -∗ ≤ 1 . Then, the minimum in (7) is equal to
$$\theta = ( q ^ { - } B ^ { * } p ) ^ { 1 / 2 } \oplus h ^ { - } B ^ { * } p \oplus q ^ { - } B ^ { * } g,$$
and the general solution is given by
$$x = B ^ { * } u, \quad g \oplus \theta ^ { - 1 } p \leq u \leq ( ( h ^ { - } \oplus \theta ^ { - 1 } q ^ { - } ) B ^ { * } ) ^ { - }.$$
As a consequence, we obtain a complete solution to the problem, which was partially solved in [50]. Consider a variant of problem (7) without the boundary constraints,
$$\minimize & \quad x ^ { - } p \oplus q ^ { - } x, \\ \text{subject to } \ B x \leq x.$$
Corollary 4 ([67]) . Let x be the general regular solution to problem (8) , where B is a matrix such that Tr( B ) ≤ 1 , p is a nonzero vector and q is a regular vector.
Then, the minimum in (8) is equal to
$$\theta = ( q ^ { - } B ^ { * } p ) ^ { 1 / 2 },$$
and the general solution is given by
$$x = B ^ { * } u, \quad \theta ^ { - 1 } p \leq u \leq \theta ( q ^ { - } B ^ { * } ) ^ { - }.$$
## 4.2 Problems with Span Seminorm
The problems, in which the objective function is defined through the span seminorm, appeared in the context of job scheduling [45, 46]. Minimization of the span seminorm solves scheduling problems in the just-in-time manufacturing. Maximization problems arise when the optimal schedule aims to spread the initiation or completion time of the jobs over the maximum possible time interval.
The solution to the problem without constraints uses the evaluation of lower or upper bounds on the objective function. To solve the problems with linear equation and inequality constraints, we first obtain a general solution to the equation or inequality, and then substitute it into the objective function to get an unconstrained problem with known solution.
## 4.2.1 Minimization Problems
We start with an unconstrained problem that presents a rather extended version of minimization problem (3). Given matrices A B , ∈ X m n × and vectors p q , ∈ X m , the problem is to find regular vectors x ∈ X n that
$$\minimize \ q ^ { - } B x ( A x ) ^ { - } p.$$
Theorem 5 ([69]) . Let x be a regular solution to problem (9) , where A is a row-regular matrix, B is a column-regular matrix, p is nonzero vector and q is regular vector.
Then, the minimum in (9) is equal to
$$\Delta = ( A ( q ^ { - } B ) ^ { - } ) ^ { - } p,$$
and attained at
$$x = \alpha ( q ^ { - } B ) ^ { - }, \quad \alpha > 0.$$
We now examine particular cases of problem (9). Suppose that B = A = I and p = q = 1 , where 1 denotes the vector having all elements equal to 1 , and consider the problem
$$\minimize \ 1 ^ { T } x x ^ { - 1 }.$$
Application of Theorem 5 shows that the problem has the minimum ∆ = 1 , which is attained at any vector x = α 1 , α > 0 .
If we assume that A = B and p = q = 1 , then we arrive at problem (3). Theorem 5 allows us to obtain the solution to the last problem in a new compact vector form.
Corollary 6. Let x be a regular solution to problem (3) , where A is a regular matrix.
Then, the minimum in (3) is equal to
$$\Delta = ( A ( \mathbb { 1 } ^ { T } A ) ^ { - } ) ^ { - } \mathbb { 1 },$$
and attained at
$$x = \alpha ( 1 ^ { T } A ) ^ { - }, \quad \alpha > 0.$$
Consider a constrained problem: given matrices B C , ∈ X n × n , find regular vectors x ∈ X n such that
$$\minimize & \quad \mathbb { 1 } ^ { T } y y ^ { - 1 }, \\ \text{subject to} & \quad C x = y, \\ & \quad D x \leq x.$$
Theorem 7 ([69]) . Let x be a regular solution to problem (10) , where C is a regular matrix and D is a matrix such that Tr( D ) ≤ 1 . Then, the minimum in (10) is equal to
$$\Delta = ( C D ^ { * } ( 1 ^ { T } C D ^ { * } ) ^ { - } ) ^ { - } 1,$$
and attained at
$$x = \alpha D ^ { * } ( \mathbb { 1 } ^ { T } C D ^ { * } ) ^ { - }, \quad \alpha > 0.$$
## 4.2.2 Maximization Problems
We begin with an unconstrained problem. Given matrices A ∈ X m n × , B ∈ X l × n , and vectors p ∈ X m , q ∈ X l , we need to find regular vectors x ∈ X n that solve the problem
$$\maximize \ q ^ { - } B x ( A x ) ^ { - } p.$$
The following result offers a complete solution to (11) under fairly general conditions.
Theorem 8 ([66]) . Let x be the general regular solution to problem (11) , where A is a matrix with regular columns, B is a column-regular matrix, p and q are regular vectors. Then, the minimum in (11) is equal to
$$\Delta = q ^ { - } B A ^ { - } p,$$
and the general solution x = ( x i ) has components that are given by
/negationslash
$$x _ { k } & = \alpha a _ { k } ^ { - } p, \\ x _ { j } & \leq \alpha a _ { s j } ^ { - 1 } p _ { s }, \quad j \neq k,$$
where α > 0 , and the indices k and s are defined by the conditions
$$k = \arg \max _ { 1 \leq i \leq n } q ^ { - } b _ { i } a _ { i } ^ { - } p, \quad s = \arg \max _ { 1 \leq i \leq m } a _ { i k } ^ { - 1 } p _ { i }.$$
Consider a special case of the problem. Assume that p = q = 1 . Then, we have
$$\mathbb { 1 } ^ { T } B x ( A x ) ^ { - } \mathbb { 1 } = \| B x \| \| ( A x ) ^ { - } \|, \quad \mathbb { 1 } ^ { T } B A ^ { - } \mathbb { 1 } = \| B A ^ { - } \|.$$
Under this assumption, problem (11) becomes
$$\maximize \ \| B x \| \| ( A x ) ^ { - } \|.$$
Application of Theorem 8 leads to the following result.
Corollary 9 ([66]) . Let x be the general regular solution to problem (12) , where A is a matrix with regular columns and B is a column-regular matrix. Then, the minimum in (12) is equal to
$$\Delta = \| B A ^ { - } \|,$$
and the general solution x = ( x i ) has components that are given by
/negationslash
$$x _ { k } & = \alpha \| a _ { k } ^ { - } \|, \\ x _ { j } & \leq \alpha a _ { s j } ^ { - 1 }, \quad j \neq k,$$
where α > 0 , and the indices k and s are defined by the conditions
$$k = \arg \max _ { 1 \leq i \leq n } \| b _ { i } \| \| a _ { i } ^ { - } \|, \quad s = \arg \max _ { 1 \leq i \leq m } a _ { i k } ^ { - 1 }.$$
We now discuss the solution to constrained problems. Suppose that, given a matrix C ∈ X n × n , we need to solve the problem
$$\maximize & \quad q ^ { - } B x ( A x ) ^ { - } p, \\ \text{subject to } & \ C x \leq x.$$
It is known (see, e.g., [30]) that the inequality constraints have regular solutions if and only if Tr( C ) ≤ 1 . Under this condition, the general solution takes the form x = C u ∗ , where u is any regular vector. Problem (13) then reduces to the unconstrained problem
$$\maximize \ q ^ { - } B C ^ { * } u ( A C ^ { * } u ) ^ { - } p.$$
The last problem takes the form of (11), and thus is completely solved by Theorem 8.
In a similar way, we obtain the solution to the problem with equality constraints
$$\maximize \quad & q ^ { - } B x ( A x ) ^ { - } p, \\ \text{subject to } \quad C x = x.$$
Indeed, if the condition Tr( C ) = 1 holds, then the equation Cx = x has the general solution x = C u + , where u is any regular vector [30]. Substitution of the solution into the objective function again leads to a problem in the form of (11),
$$\maximize \ q ^ { - } B C ^ { + } u ( A C ^ { + } u ) ^ { - } p.$$
## 4.3 Problems with Evaluation of Spectral Radius
First, we return to problem (4), which admits a solution based on the extremal property of the spectral radius. A complete solution to the problem was given in [71] as a consequence of solution to a more general optimization problem. A direct complete solution to problem (4) is obtained in [70] in the following form.
Lemma 10. Let x be the general regular solution to problem (4) , where A is a matrix with spectral radius λ > 0 .
Then, the minimum in (4) is equal to λ , and the general solution is given by
$$x = ( \lambda ^ { - 1 } A ) ^ { * } u, \quad u \in \mathcal { K } ^ { n }.$$
The proof of the statement is based on the derivation of a sharp lower bound for the objective function, accompanied by the development of the equation, which specifies the solution set for the problem. We then reduce the equation to an inequality and take the general solution to the inequality as a complete solution to the optimization problem.
Extensions of problem (4), which differ from (4) by more general form of the objective function and constraints, were examined in [72, 65], where general solutions were obtained under various assumptions.
Below, we offer complete direct solutions to certain new problems, which present further extensions of problem (4).
## 4.3.1 An Unconstrained Problem
Suppose that a matrix A ∈ X n × n , vectors p q , ∈ X n , and a scalar c ∈ X are given. Consider the problem of finding regular vectors x ∈ X n that
$$\minimize \ x ^ { - } A x \oplus x ^ { - } p \oplus q ^ { - } x \oplus c.$$
The problem is completely solved as follows.
Theorem 11 ([70]) . Let x be the general regular solution to problem (15) , where A is a matrix with spectral radius λ > 0 and q is a regular vector. Denote
$$\mu = \lambda \oplus \bigoplus _ { m = 1 } ^ { n } ( q ^ { - } A ^ { m - 1 } p ) ^ { 1 / ( m + 1 ) } \oplus c.$$
Then, the minimum in (15) is equal to µ and the general solution is given by
$$x = ( \mu ^ { - 1 } A ) ^ { * } u, \quad \mu ^ { - 1 } p \leq u \leq \mu ( q ^ { - } ( \mu ^ { - 1 } A ) ^ { * } ) ^ { - }.$$
## 4.3.2 Problems with Constraints
Given matrices A B , ∈ X n × n , C ∈ X m n × , and vectors g ∈ X n , h ∈ X m , consider the problem to find regular vectors x ∈ X n that
$$\minimize & \quad x ^ { - } A x, \\ \text{subject to } & \ B x \oplus g \leq x, \\ & C x \leq h.$$
Theorem 12 ([65]) . Let x be the general regular solution to problem (16) , where A is a matrix with spectral radius λ > 0 , B is a matrix such that Tr( B ) ≤ 1 . Suppose that C is a column-regular matrix and h is a regular vector such that h CB g -∗ ≤ 1 .
Then, the minimum in (16) is equal to
$$\theta = \bigoplus _ { k = 1 } ^ { n } \bigoplus _ { 0 \leq i _ { 0 } + i _ { 1 } + \cdots + i _ { k } \leq n - k } \text{tr} ^ { 1 / k } ( B ^ { i _ { 0 } } ( A B ^ { i _ { 1 } } \cdots A B ^ { i _ { k } } ) ( I \oplus g h ^ { - } C ) ),$$
and the general solution is given by
$$x = ( \theta ^ { - 1 } A \oplus B ) ^ { * } u, \quad g \leq u \leq ( h ^ { - } C ( \theta ^ { - 1 } A \oplus B ) ^ { * } ) ^ { - }.$$
In the case when C = 0 , problem (16) takes the form
$$\text{minimize} \quad & x ^ { - } A x, \\ \text{subject to} \quad & B x \oplus g \leq x.$$
Corollary 13 ([71, 65]) . Let x be the general regular solution to problem (17) , where A is a matrix with spectral radius λ > 0 . Then, the minimum in (17) is equal to
$$\theta = \lambda \oplus \bigoplus _ { k = 1 } ^ { n - 1 } \bigoplus _ { 1 \leq i _ { 1 } + \dots + i _ { k } \leq n - k } \text{tr} ^ { 1 / k } ( A B ^ { i _ { 1 } } \cdots A B ^ { i _ { k } } ),$$
and the general solution is given by
$$x = ( \theta ^ { - 1 } A \oplus B ) ^ { * } u, \quad u \geq g.$$
Suppose that B = 0 and C = I . In this case, problem (16) becomes
$$\minimize & \quad x ^ { - } A x, \\ \text{subject to } & \quad g \leq x \leq h.$$
Corollary 14 ([65]) . Let x be the general regular solution to problem (18) , where A is a matrix with spectral radius λ > 0 , g is a vector and h is a regular vector such that h g -≤ 1 . Then, the minimum in (18) is equal to
$$\theta = \lambda \oplus \bigoplus _ { k = 1 } ^ { n } ( h ^ { - } A ^ { k } g ) ^ { 1 / k },$$
and the general solution is given by
$$x = ( \theta ^ { - 1 } A ) ^ { * } u, \quad g \leq u \leq ( h ^ { - } ( \theta ^ { - 1 } A ) ^ { * } ) ^ { - }.$$
The next problem combines a special case of the objective function of problem (15) with the constraints of problem (17). Given matrices A B , ∈ X n × n and vectors p g , ∈ X n , we find regular vectors x ∈ X n that provide solutions to the problem
$$\minimize & \quad x ^ { - } A x \oplus x ^ { - } p, \\ \text{subject to } & \ B x \oplus g \leq x.$$
Theorem 15 ([70]) . Let x be the general regular solution to problem (19) , where A is a matrix with spectral radius λ > 0 and B is a matrix such that Tr( B ) ≤ 1 .
Then, the minimum in (19) is equal to
$$\theta = \lambda \oplus \bigoplus _ { k = 1 } ^ { n - 1 } \bigoplus _ { 1 \leq i _ { 1 } + \dots + i _ { k } \leq n - k } \text{tr} ^ { 1 / k } ( A B ^ { i _ { 1 } } \cdots A B ^ { i _ { k } } ),$$
and the general solution is given by
$$x = ( \theta ^ { - 1 } A \oplus B ) ^ { * } u, \quad u \geq \theta ^ { - 1 } p \oplus g.$$
## Acknowledgments
The author thanks an editor for helpful comments and suggestions on earlier drafts of the manuscript.
## References
$$\begin{smallmatrix} [ 1 ] & S. \, N. \, N. \, P a n d i t, \, \text{``A new matrix calculus}, \end{smallmatrix} \\ & J. \, S I A M \, 9 \, \text{no. } 4, \, ( 1 9 6 1 ) \, 6 3 2 - 6 3 9.$$
- [2] R. A. Cuninghame-Green, 'Describing industrial processes with interference and approximating their steady-state behaviour,' Oper. Res. Quart. 13 no. 1, (1962) 95-100.
- [3] B. Giffler, 'Scheduling general production systems using schedule algebra,' Naval Res. Logist. Quart. 10 no. 1, (1963) 237-255.
- [4] A. J. Hoffman, 'On abstract dual linear programs,' Naval Res. Logist. Quart. 10 no. 1, (1963) 369-373.
- [5] S. C. Kleene, 'Representation of events in nerve nets and finite automata,' in Automata Studies , C. E. Shannon and J. McCarthy, eds., vol. 34 of Annals of Mathematics Studies , pp. 3-41. Princeton University Press, 1956.
- [6] R. Bellman, 'On a quasi-linear equation,' Canad. J. Math. 8 (1956) 198-202.
- [7] R. Bellman and W. Karush, 'On a new functional transform in analysis: The maximum transform,' Bull. Amer. Math. Soc. 67 (1961) 501-503.
- [8] R. Bellman and W. Karush, 'Mathematical programming and the maximum transform,' Journal of the Society for Industrial and Applied Mathematics 10 no. 3, (1962) 550-567. http://www.jstor.org/stable/2098751 .
- [9] R. T. Rockafellar, Convex Analysis . Princeton University Press, Princeton, NJ, 1970.
- [10] I. V. Romanovsky, 'Some remarks on the functional transformation of Bellman-Karush,' Vestnik Leningrad Univ. 17 no. 18, (1962) 148-150. (in Russian).
- [11] N. N. Vorob'ev, 'The extremal matrix algebra,' Soviet Math. Dokl. 4 no. 5, (1963) 1220-1223.
- [12] I. V. Romanovski˘ ı, 'Asymptotic behavior of dynamic programming processes with a continuous set of states,' Soviet Math. Dokl. 5 no. 6, (1964) 1684-1687.
- [13] I. V. Romanovskii, 'Asymptotic representation of recurrence relations in dynamic programming,' Soviet Phys. Dokl. 9 no. 8, (1965) 633-635.
- [14] I. V. Romanovskii, 'Optimization of stationary control of a discrete deterministic process,' Cybernetics 3 (April, 1967) 52-62.
- [15] I. V. Romanovskiy, 'Deterministic processes of dynamic programming with additional constraints,' J. Cybernet. 2 no. 1, (1972) 79-84.
| [16] | N. N. Vorob'ev, 'Extremal algebra of positive matrices,' Elektron. Informationsverarbeit. Kybernetik 3 no. 1, (1967) 39-71. (in Russian). |
|--------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [17] | N. N. Vorob'ev, 'Extremal algebra of non-negative matrices,' Elektron. Informationsverarbeit. Kybernetik 6 no. 4/5, (1970) 303-311. (in Russian). |
| [18] | A. A. Korbut, 'Extremal spaces,' Soviet Math. Dokl. 6 no. 5, (1965) 1358-1361. |
| [19] | A. A. Korbut, 'Extremal vector spaces and their properties,' Elektron. Informationsverarbeit. Kybernetik 8 no. 8/9, (1972) 525-536. (in Russian). |
| [20] | V. P. Maslov and S. N. Samborski˘, ı vol. 13 of Advances in Soviet Mathematics . AMS, Providence, RI, 1992. |
| [21] | V. Maslov and V. Kolokoltsov, Idempotent Analysis and Its Applications to Optimal Control Theory . Nauka, Moscow, 1994. (in Russian). |
| [22] | V. N. Kolokoltsov and V. P. Maslov, Idempotent Analysis and Its Applications , vol. 401 of Mathematics and Its Applications . Kluwer Academic Publishers, Dordrecht, 1997. |
| [23] | R. Cuninghame-Green, Minimax Algebra , vol. 166 of Lecture Notes in Economics and Mathematical Systems . Springer, Berlin, 1979. |
| [24] | B. Carr´, e Graphs and Networks . Oxford Applied Mathematics and Computing Science Series. Clarendon Press, Oxford, 1979. |
| [25] | U. Zimmermann, Linear and Combinatorial Optimization in Ordered Algebraic Structures , vol. 10 of Annals of Discrete Mathematics . Elsevier, Amsterdam, 1981. |
| [26] | F. L. Baccelli, G. Cohen, G. J. Olsder, and J.-P. Quadrat, Synchronization and Linearity: An Algebra for Discrete Event Systems . Wiley Series in Probability and Statistics. Wiley, Chichester, 1993. |
| [27] | J. S. Golan, Semirings and Affine Equations Over Them: Theory and Applications , vol. 556 of Mathematics and Its Applications . Kluwer Academic Publishers, Dordrecht, 2003. |
| [28] | B. Heidergott, G. J. Olsder, and J. van der Woude, Max-plus at Work: Modeling and Analysis of Synchronized Systems . Princeton Series in Applied Mathematics. Princeton University Press, Princeton, NJ, 2006. |
| [29] | M. Gondran and M. Minoux, Graphs, Dioids and Semirings: New Models and Algorithms , vol. 41 of Operations Research / Computer Science Interfaces . Springer, New York, 2008. |
|--------|---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [30] | N. K. Krivulin, Methods of Idempotent Algebra for Problems in Modeling and Analysis of Complex Systems . Saint Petersburg University Press, Saint Petersburg, 2009. (in Russian). |
| [31] | P. Butkoviˇ, c Max-linear Systems: Theory and Algorithms . Springer Monographs in Mathematics. Springer, London, 2010. |
| [32] | W. M. McEneaney, Max-Plus Methods for Nonlinear Control and Estimation . Systems and Control: Foundations and Applications. Birkh¨user, a Boston, 2006. |
| [33] | L. Superville, Various Aspects of Max-Algebra . PhD thesis, The City University of New York, New York, 1978. |
| [34] | R. A. Cuninghame-Green, 'Projections in minimax algebra,' Math. Program. 10 (1976) 111-123. |
| [35] | P. Butkoviˇ, c 'On properties of solution sets of extremal linear programs,' in Algebraic and Combinatorial Methods in Operations Research , R. E. Burkard, R. A. Cuninghame-Green, and U. Zimmermann, eds., vol. 95 of North-Holland Mathematics Studies , pp. 41-54. Elsevier, Amsterdam, 1984. |
| [36] | K. Zimmermann, 'Some optimization problems with extremal operations,' in Mathematical Programming at Oberwolfach II , B. Korte and K. Ritter, eds., vol. 22 of Mathematical Programming Studies , pp. 237-251. Springer, Berlin, 1984. |
| [37] | N. K. Krivulin, 'Evaluation of bounds on the mean rate of growth of the state vector of a linear dynamical stochastic system in idempotent algebra,' Vestnik St. Petersburg Univ. Math. 38 no. 2, (2005) 42-51. |
| [38] | N. K. Krivulin, 'Eigenvalues and eigenvectors of matrices in idempotent algebra,' Vestnik St. Petersburg Univ. Math. 39 no. 2, (2006) 72-83. |
| [39] | N. K. Krivulin, 'Solution of generalized linear vector equations in idempotent algebra,' Vestnik St. Petersburg Univ. Math. 39 no. 1, (2006) 16-26. |
| [40] | V. N. Kolokol'tsov, 'Idempotent structures in optimization,' J. Math. Sci. (NY) 104 no. 1, (2001) 847-880. |
|--------|-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [41] | G. Litvinov, 'Maslov dequantization, idempotent and tropical mathematics: A brief introduction,' J. Math. Sci. (NY) 140 no. 3, (2007) 426-444, arXiv:math/0507014 [math.GM] . |
| [42] | M. Akian, R. Bapat, and S. Gaubert, 'Max-plus algebra,' in Handbook of Linear Algebra , L. Hogben, ed., Discrete Mathematics and Its Applications, pp. 25-1-25-17. Taylor and Francis, Boca Raton, FL, 2007. |
| [43] | K. Zimmermann, 'Interval linear systems and optimization problems over max-algebras,' in Linear Optimization Problems with Inexact Data , pp. 165-193. Springer, New York, 2006. |
| [44] | P. Butkoviˇ c and A. Aminu, 'Introduction to max-linear programming,' IMA J. Manag. Math. 20 no. 3, (2009) 233-249. |
| [45] | P. Butkoviˇ c and K. P. Tam, 'On some properties of the image set of a max-linear mapping,' in Tropical and Idempotent Mathematics , G. L. Litvinov and S. N. Sergeev, eds., vol. 495 of Contemp. Math. , pp. 115-126. American Mathematical Society, 2009. |
| [46] | K. P. Tam, Optimizing and Approximating Eigenvectors in Max-Algebra . PhD thesis, The University of Birmingham, Birmingham, 2010. |
| [47] | A. Aminu and P. Butkoviˇ, c 'Non-linear programs with max-linear constraints: A heuristic approach,' IMA J. Manag. Math. 23 no. 1, (2012) 41-66. |
| [48] | R. A. Cuninghame-Green, 'Minimax algebra and applications,' Fuzzy Sets and Systems 41 no. 3, (1991) 251-267. |
| [49] | N. K. Krivulin, 'An extremal property of the eigenvalue for irreducible matrices in idempotent algebra and an algebraic solution to a Rawls location problem,' Vestnik St. Petersburg Univ. Math. 44 no. 4, (2011) 272-281. |
| [50] | N. Krivulin, 'A new algebraic solution to multidimensional minimax location problems with Chebyshev distance,' WSEAS Trans. Math. 11 no. 7, (2012) 605-614, arXiv:1210.4770 [math.OC] . |
- [51] L. Elsner and P. van den Driessche, 'Max-algebra and pairwise comparison matrices,' Linear Algebra Appl. 385 no. 1, (2004) 47-62.
- [52] L. Elsner and P. van den Driessche, 'Max-algebra and pairwise comparison matrices, II,' Linear Algebra Appl. 432 no. 4, (2010) 927-935.
- [53] B. B. Gursoy, O. Mason, and S. Sergeev, 'The analytic hierarchy process, max algebra and multi-objective optimisation,' Linear Algebra Appl. 438 no. 7, (2013) 2911-2928.
- [54] S. Gaubert, 'Resource optimization and (min,+) spectral theory,' IEEE Trans. Automat. Control 40 no. 11, (1995) 1931-1934.
- [55] B. De Schutter, Max-Algebraic System Theory for Discrete Event Systems . PhD thesis, Katholieke Universiteit Leuven, Leuven, 1996.
- [56] B. De Schutter and T. van den Boom, 'Model predictive control for max-plus-linear discrete event systems,' Automatica 37 no. 7, (2001) 1049-1056.
- [57] K. Zimmermann, 'On max-separable optimization problems,' in Algebraic and Combinatorial Methods in Operations Research , R. E. Burkard, R. A. Cuninghame-Green, and U. Zimmermann, eds., vol. 95 of North-Holland Mathematics Studies , pp. 357-362. Elsevier, Amsterdam, 1984.
- [58] K. Zimmermann, 'Optimization problems with unimodal functions in max-separable constraints,' Optimization 24 no. 1-2, (1992) 31-41.
- [59] K. Zimmermann, 'Disjunctive optimization, max-separable problems and extremal algebras,'
- Theoret. Comput. Sci. 293 no. 1, (2003) 45-54.
- [60] M. Gavalec and K. Zimmermann, 'Duality for max-separable problems,' Cent. Eur. J. Oper. Res. 20 (2012) 409-419.
- [61] R. A. Cuninghame-Green and P. Butkoviˇ c, 'The equation a ⊕ x = b ⊕ y over (max +),' , Theoret. Comput. Sci. 293 no. 1, (2003) 3-12.
- [62] S. Gaubert, R. D. Katz, and S. Sergeev, 'Tropical linear-fractional programming and parametric mean payoff games,' J. Symbolic Comput. 47 no. 12, (2012) 1447-1478, arXiv:1101.3431 [math.MG] .
- [63] M. Akian, S. Gaubert, and A. Guterman, 'Tropical polyhedra are equivalent to mean payoff games,'
| | Internat. J. Algebra Comput. 22 no. 1, (2012) 1250001-1-1250001-43, arXiv:0912.2462 [math.CO] . |
|------|---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [64] | G. M. Engel and H. Schneider, 'Diagonal similarity and equivalence for matrices over groups with 0,' Czechoslovak Math. J. 25 no. 3, (1975) 389-403. |
| [65] | N. Krivulin, 'A constrained tropical optimization problem: Complete solution and application example,' arXiv:1305.1454 [math.OC] . |
| [66] | N. Krivulin, 'A maximization problem in tropical mathematics: A complete solution and application examples,' arXiv:1304.7461 [math.OC] . |
| [67] | N. Krivulin, 'Complete solution of a constrained tropical optimization problem with application to location in Relational and Algebraic Methods in Computer Science , P. H¨fner, o P. Jipsen, W. Kahl, and M. E. M¨ller, u eds., vol. 8428 of Lecture Notes in Computer Science , pp. 362-378. Springer, Switzerland, 2014. arXiv:1311.2795 [math.OC] . |
| [68] | N. Krivulin and K. Zimmermann, 'Direct solutions to tropical optimization problems with nonlinear objective functions and boundary constraints,' in Mathematical Methods and Optimization Techniques in Engineering , D. Biolek, H. Walter, I. Utu, and C. von Lucken, eds., pp. 86-91. WSEAS Press, 2013. arXiv:1311.2311 [math.OC] . |
| [69] | N. Krivulin, 'Explicit solution of a tropical optimization problem with application to project scheduling,' in Mathematical Methods and Optimization Techniques in Engineering , D. Biolek, H. Walter, I. Utu, and C. von Lucken, eds., pp. 39-45. WSEAS Press, 2013. arXiv:1303.5457 [math.OC] . |
| [70] | N. Krivulin, 'Extremal properties of tropical eigenvalues and solutions to tropical optimization problems,' arXiv:1311.0442 [math.OC] . |
| [71] | N. Krivulin, 'A multidimensional tropical optimization problem with nonlinear objective function and linear constraints,' Optimization (2013) , arXiv:1303.0542 [math.OC] . |
| [72] | N. Krivulin, 'A complete closed-form solution to a tropical extremal problem,' in Advances in Computer Science , S. Yenuri, ed., vol. 5 of Recent Advances in Computer Engineering Series , pp. 146-151. WSEAS Press, 2012. arXiv:1212.6106 [math.OC] . | | null | [
"Nikolai Krivulin"
] | 2014-08-01T22:03:11+00:00 | 2014-08-01T22:03:11+00:00 | [
"math.OC",
"cs.SY",
"65K10 (Primary), 15A80, 65K05, 90C48, 41A50 (Secondary)"
] | Tropical optimization problems | We consider optimization problems that are formulated and solved in the
framework of tropical mathematics. The problems consist in minimizing or
maximizing functionals defined on vectors of finite-dimensional semimodules
over idempotent semifields, and may involve constraints in the form of linear
equations and inequalities. The objective function can be either a linear
function or nonlinear function calculated by means of multiplicative conjugate
transposition of vectors. We start with an overview of known tropical
optimization problems and solution methods. Then, we formulate certain new
problems and present direct solutions to the problems in a closed compact
vector form suitable for further analysis and applications. For many problems,
the results obtained are complete solutions. |
1408.0314v1 | ## Normal variation for adaptive feature size
Nina Amenta and Tamal K. Dey
∗ †
August 5, 2014
## Background
Let Σ be a closed, smooth surface in R 3 . For any two sets X,Y ⊂ R 3 , let d X,Y ( ) denote the Euclidean distance between X and Y . The local feature size f ( x ) at a point x ∈ Σ is defined to be the distance d x, M ( ) where M is the medial axis of Σ . Let n p denote the unit normal (inward) to Σ at point p . Amenta and Bern in their paper [1] claimed the following:
Claim 1 Let q and q ′ be any two points in Σ so that d q, q ( ′ ) ≤ ε min { f ( q , f ) ( q ′ ) } for ε ≤ 1 3 . Then, ∠ n , n q q ′ ≤ ε 1 -3 ε .
Unfortunately, the proof of this claim as given in Amenta and Bern [1] is wrong; it also appears in the book by Dey [2]. In this short note, we provide a correct proof with an improved bound of ε 1 -ε .
Theorem 2 Let q and q ′ be two points in Σ with d q, q ( ′ ) ≤ εf ( q ) where ε ≤ 1 3 . Then, ∠ n , n q q ′ ≤ ε .
1 -ε
## Definitions and Preliminaries
For any point p ∈ R 3 , let ˜ p denote the closest point of p in Σ . When p is a point in Σ , the normal to Σ at p is well defined. We extend this definition to any point p ∈ R 3 . Define the normal n p at p ∈ R 3 \ M as the normal to Σ at ˜ p . Similarly, we extend the definition of local feature size f to R 3 . For any point p ∈ R 3 , let f ( p ) be the distance of p to the medial axis of Σ . Notice that f is 1 -Lipschitz. If two points x and y lie on a surface F ⊂ R 3 , let d F ( x, y ) denote the geodesic distance between x and y . The following facts are well known in differential geometry.
∗ Dept. of CS, U. of California, Davis, CA 95616, [email protected]
† Dept. of CSE, Ohio State U., Columbus, OH 43210, [email protected]
Proposition 3 Let F be a smooth surface in R 3 . Let q and q ′ be two points in F . Then,
$$\lim _ { d \to 0 } \frac { d _ { F } ( q, q ^ { \prime } ) } { d ( q, q ^ { \prime } ) } = 1.$$
Proposition 4 Consider the geodesic path between q, q ′ on a smooth surface F in R 3 . Let κ m be the maximum curvature on this geodesic path. Then ∠ n , n q q ′ ≤ κ m F d ( q, q ′ ) .
## The Proof
Weare to measure ∠ n , n q q ′ for two points q and q ′ in Σ . One approach would be to use the propositions above to bound the length of a path from p to q on Σ and then use that length to bound the change in normal direction, but we can get a better bound by considering the direct path from p to q .
Let Σ ω denote an offset of Σ , that is, each point in Σ ω has distance ω from Σ . Formally, consider the distance function
$$h \colon \mathbb { R } ^ { 3 } \rightarrow \mathbb { R }, \ h ( x ) \mapsto d ( x, \Sigma ).$$
↦
$$\text{Then, } \Sigma _ { \omega } = h ^ { - 1 } ( \omega ).$$
Claim 5 For ω ≥ 0 let p be a point in Σ ω where ω < f p (˜) . There is an open set U ⊂ R 3 so that σ p = Σ ω ∩ U is a smooth 2 -manifold which can be oriented so that n x is the normal to σ p at any x ∈ σ p .
PROOF. Since ω < f p (˜) , p is not a point on the medial axis. Therefore, the distance function h is smooth at p . One can apply the implicit function theorem to claim that there exists an open set U ⊂ R 3 where
$$\sigma _ { p } = h ^ { - 1 } ( \omega ) \cap U$$
is a smooth 2 -manifold. The unit gradient ( ∇ h ‖∇ ‖ h ) x = x -˜ x ‖ x - ‖ ˜ x which is precisely n x up to orientation is normal to σ p at x ∈ σ p .
PROOF. [ Proof of Theorem 2 ] Consider parameterizing the segment qq ′ by the length of qq ′ . Take two arbitrarily close points p = p t ( ) and p ′ = p t ( +∆ ) t in qq ′ for arbitrarily small ∆ t > 0 . Let θ t ( ) = ∠ n , n q p t ( ) and ∆ α = ∠ n , n p p ′ . Then, | θ t ( +∆ ) t -θ t ( ) | ≤ ∆ α giving
$$| \theta ^ { \prime } ( t ) | \leq \lim _ { \Delta t \rightarrow 0 } \frac { \Delta \alpha } { \Delta t }.$$
If we show that lim ∆ t → 0 ∆ α ∆ t is no more than 1 (1 -ε f ) ( q ) we are done since then
$$\underset { \Delta t \to 0 } { \mu } \overline { \Delta t } \text{ is in one una } \overline { \underset { ( 1 - \varepsilon ) f ( q ) } { \nu } } \text{ we are out} \\ \underset { \zeta } { \mu } n _ { q }, n _ { q ^ { \prime } } & \leq \int _ { q q ^ { \prime } } | \theta ^ { \prime } ( t ) | d t \\ & \leq \int _ { q q ^ { \prime } } \frac { 1 } { ( 1 - \varepsilon ) f ( q ) } d t \\ & = \frac { d ( q, q ^ { \prime } ) } { ( 1 - \varepsilon ) f ( q ) } \\ & \leq \frac { \varepsilon } { ( 1 - \varepsilon ) }. \\ \tilde { p } ) \, < \, d ( q, p ) + d ( p, \tilde { p } ) \text{ and } d ( q, p ) \, < \, \varepsilon f ( q ).$$
We have d q, p ( ˜) ≤ d q, p ( ) + d p, p ( ˜) and d q, p ( ) ≤ εf ( q ) . Since also ω = d p, p ( ˜) ≤ d p, q ( ) ≤ εf ( q ) , we have ω ≤ 2 ε 1 -2 ε f (˜) p (by a standard argument using the fact that the function f is 1 -Lipshitz). Therefore, ω < f p (˜) for ε < 1 3 / , and there is a smooth neighborhood σ p ⊂ Σ ω of p satisfying Claim 5.
Let r be the closest point to p ′ in Σ ω , and let ∆ t be small enough so that r and the geodesic between p and r in σ p lies in σ p . Notice that , by Claim 5, ∆ α = ∠ n , n p p ′ = ∠ n , n p r .
$$\text{Claim} \, 6 \, \lim _ { \Delta t \rightarrow 0 } \frac { d ( p, r ) } { \Delta t } \leq 1.$$
PROOF. Consider the triangle prp ′ . If the tangent plane to σ p at r separates p and p ′ , the angle ∠ prp ′ is obtuse. It follows that d p, r ( ) ≤ d p, p ( ′ ) = ∆ t . In the other case when the tangent plane to σ p at r does not separate p and p ′ , the angle ∠ prp ′ is non-obtuse. Let x be the foot of the perpendicular dropped from p on the line of p r ′ . Wehave d p, r ( ) cos α ≤ d p, p ( ′ ) where α is the acute angle ∠ rpx . Combining the two cases we have d p, r ( ) / ∆ t ≤ 1 cos α . Since α goes to 0 as ∆ t goes to 0 , we have lim ∆ t → 0 d p,r ( ) ∆ t ≤ 1 .
Now consider the geodesic between p and r in σ p , and let m be the point on the geodesic at which the maximum curvature κ m is realized. Recall that d σ p ( p, r ) denotes the geodesic distance between p and r on σ p . Let r m be the radius of curvature corresponding to κ m , i.e., κ m = 1 /r m . Clearly, f ( m ) ≤ r m . So, Proposition 4 tells us that
$$\Delta \alpha \leq \frac { d _ { \sigma _ { p } } ( p, r ) } { f ( m ) }.$$
Therefore,
$$\lim _ { \Delta t \to 0 } \frac { \Delta \alpha } { \Delta t } \ \leq \ \lim _ { \Delta t \to 0 } \frac { d _ { \sigma _ { p } } ( p, r ) } { \Delta t \, f ( m ) }$$
In the limit when ∆ t goes to zero, d σ p ( p, r ) approaches d p, r ( ) which in turn approaches ∆ t (Proposition 3 and Claim 6). Meanwhile, d q, m ( ) ≤ d q, p ( ) + d p, r ( ) approaches d q, p ( ) ≤ εf ( q ) as ∆ t goes to zero (again by Claim 6). So, in the limit, f ( m > ) (1 -ε f ) ( q ) (again using the fact that f is 1 -Lipshitz). Therefore,
$$\lim _ { \Delta t \rightarrow 0 } \frac { \Delta \alpha } { \Delta t } \leq \frac { 1 } { ( 1 - \varepsilon ) f ( q ) }$$
which is what we need to prove.
Remark : The bound on normal variation can be slightly improved to -ln(1 -ε ) by observing the following. We used that d q, p ( ) ≤ εf ( q ) to arrive at the bound f ( m > ) (1 -ε f ) ( q ) . In fact, one can observe that d q, p ( ) ≤ εtf ( q ) giving f ( m > ) (1 -εt f ) ( q ) . This gives | θ ′ ( ) t | ≤ 1 (1 -εt f ) ( q ) . We have
$$\zeta n _ { q }, n _ { q ^ { \prime } } \leq \int _ { q q ^ { \prime } } \frac { 1 } { ( 1 - \varepsilon t ) f ( q ) } d t = - \frac { d ( q, q ^ { \prime } ) \ln ( 1 - \varepsilon ) } { \varepsilon f ( q ) } = - \ln ( 1 - \varepsilon ).$$
Acknowledgement We thank Siu-Wing Cheng, Jian Sun, and Edgar Ramos for discussions on the normal variation problem.
## References
- [1] N. Amenta and M. Bern. Surface reconstruction by Voronoi filtering. Discr. Comput. Geom. 22 (1999), 481-504.
- [2] T. K. Dey. Curve and surface reconstruction : Algorithms with mathematical analysis. Cambridge University Press, New York, 2006. | null | [
"Nina Amenta",
"Tamal K. Dey"
] | 2014-08-01T22:10:56+00:00 | 2014-08-01T22:10:56+00:00 | [
"cs.CG"
] | Normal variation for adaptive feature size | The change in the normal between any two nearby points on a closed, smooth
surface is bounded with respect to the local feature size (distance to the
medial axis). An incorrect proof of this lemma appeared as part of the analysis
of the "crust" algorithm of Amenta and Bern. |
1408.0315v3 | ## ON BETTER-QUASI-ORDERING CLASSES OF PARTIAL ORDERS
## GREGORY MCKAY
Abstract. We provide a method of constructing better-quasi-orders by generalising a technique for constructing operator algebras that was developed by Pouzet. We then generalise the notion of σ -scattered to partial orders, and use our method to prove that the class of σ -scattered partial orders is better-quasi-ordered under embeddability. This generalises theorems of Laver, Corominas and Thomass´ e regarding σ -scattered linear orders and trees, countable forests and N -free partial orders respectively. In particular, a class of countable partial orders is better-quasi-ordered whenever the class of indecomposable subsets of its members satisfies a natural strengthening of better-quasi-order.
## Contents
| 1. | Introduction | 2 |
|------------------|-----------------------------------------------------|-----|
| 2. | Preliminaries | 3 |
| 2.1. | Basic bqo theory | 3 |
| 2.2. | Concrete categories | 5 |
| 2.3. | Partial orders | 7 |
| 3. | Operator construction | 9 |
| 3.1. | Iterating over L | 10 |
| 3.2. | Limits | 13 |
| 4. | Decomposition trees | 17 |
| 4.1. | Scattered and structured L -trees | 17 |
| 4.2. | Encoding with structured L -trees | 20 |
| 5. | The construction is bqo | 23 |
| 5.1. | The non-limit case | 23 |
| 5.2. | The limit case | 25 |
| 6. | Constructing structured L -trees | 26 |
| 6.1. | The constructed trees are σ -scattered | 28 |
| 6.2. | The class T L P of structured trees is well-behaved | 29 |
| 7. | Constructing partial orders | 32 |
| 7.1. | Intervals and indecomposable partial orders | 32 |
| 7.2. | Generalised σ -scattered partial orders | 33 |
| 7.3. | Pathological decomposition trees | 35 |
| 7.4. | Pathological partial orders | 39 |
| 7.5. | Characterising the construction | 40 |
| 8. | Corollaries and conclusions | 42 |
| | | 44 |
| Acknowledgements | Acknowledgements | |
| References | References | 44 |
2010 Mathematics Subject Classification. Primary 06A07; Secondary 03E05, 06A06, 05C05.
Key words and phrases. Better-quasi-order, well-quasi-order, scattered, structured trees, interval.
This work was carried out as part of the author's Ph.D. thesis at the University of East Anglia in the UK.
## 1. Introduction
Some of the most striking theorems in better-quasi-order (bqo) theory are that certain classes of partial orders, often with colourings, are bqo under embeddability. Indeed, the notion of bqo was first used by Nash-Williams to prove that the class R of rooted trees of height at most ω has no infinite descending chains or infinite antichains under the embeddability quasi-order [9].
Another contribution from Nash-Williams comes in [10]. He proved that if Q is bqo, then the class ˜ Q of transfinite sequences of members of Q is bqo (see [11]). This theorem can be viewed as an embeddability result on a class of coloured partial orders; since it is equivalent to saying that the ordinals preserve bqo (which is to say that if Q is bqo, then the class of ordinals coloured by Q is also bqo under a natural embeddability ordering; see Definition 2.15). In fact Nash-Williams proved a technical strengthening of this statement, equivalent to saying that the ordinals are well-behaved (see Definition 2.16). This is an important generalisation of bqo that will be crucial in this paper, because it is much more useful than bqo or even preserving bqo when constructing large bqo classes.
Perhaps the most well known result of this kind comes from Laver, who proved that the class M of σ -scattered linear orders preserves bqo, a positive result for a generalisation of Fra¨ ıss´'s e conjecture [12]. A few years after his paper on σ -scattered linear orders, Laver also showed that the class T of σ -scattered trees preserves bqo [13]. Initially we notice that there should be some connection between the theorems of σ -scattered linear orders and σ -scattered trees. In each, we first take all partial orders of some particular type (linear orders, trees) that do not embed some particular order (namely Q and 2 <ω ). In both cases, the class of countable unions of these objects turn out to preserve bqo.
We prove a general theorem of this type (Theorem 7.37) which states that given well-behaved classes L and P of linear orders and partial orders respectively, the class of 'generalised σ -scattered partial orders' M L P will be well-behaved (see Definition 7.10). We define our general 'scattered' partial orders to be those orders X such that:
- · Every indecomposable subset of X is contained in P . (See Definition 7.3.)
- · Every chain of intervals of X under ⊇ has order type in L . (See definitions 2.31 and 7.2.)
- · No 'pathological' order 2 <ω , -2 <ω or 2 <ω ⊥ embeds into X . (See definitions 4.1 and 7.9).
Our class M L P is then the class of countable increasing unions or limits of such X .
Applying this theorem with classes known to be well-behaved yields generalisations of many other known results in this area. For example, Corominas showed that the class of countable C -trees 1 preserves bqo [2] (see Definition 2.33) and Thomass´ showed that the class of countable e N -free partial orders preserves bqo [1] (see Definition 2.27). We summarise known results as applications of Theorem 7.37 in the following table. 2 In each case Theorem 7.37 tells us that the given class is well-behaved. 3
| Class | Description | P | L | Limits |
|---------------------|--------------------------------------|--------------|-----------|----------|
| N | Finite numbers under injective maps | 1, A 2 | 1 | ✗ |
| S | Scattered linear orders [12] | 1, C 2 | On ∪ On ∗ | ✗ |
| M | σ -scattered linear orders [12] | 1, C 2 | On ∪ On ∗ | ✓ |
| U On | Scattered trees [13] | 1, C 2 , A 2 | On | ✗ |
| T On | σ -scattered trees [13] | 1, C 2 , A 2 | On | ✓ |
| T C | Countable L -trees [2] | 1, C 2 , A 2 | C | ✓ |
| C { 1 , A 2 , C 2 } | Countable N -free partial orders [1] | 1, C 2 , A 2 | C | ✓ |
1 Here C is the class of countable linear orders.
2 Here A 2 and C 2 are the antichain and chain of size 2 respectively.
3 In the cases of U On , T On and T C the constructed M L P is actually a larger class of partial orders.
Applying this theorem with the largest known well-behaved classes L and P gives that some very large classes of partial orders are well-behaved (Theorem 8.7). For example, let P be the set of indecomposable partial orders of cardinality less than some n ∈ ω , and L = M . Then for n > 2 the well-behaved class M L P contains the σ -scattered linear orders, σ -scattered M -trees, countable N -free partial orders, and generalisations of such objects.
Crucial to the ideas in this paper are those of constructing objects with so called 'structured trees'. Put simply, these are trees with some extra structure (usually a partial order) given to the set of successors of each element. Embeddings between structured trees are then required to induce embeddings of this extra structure.
Theorems on structured trees also appear throughout the literature on bqo theory (cf. [1, 2, 3, 5, 6, 8]). The rationale for their usefulness is explained by Pouzet in [5]. His method is to take a 'simple' class of objects (e.g. partial orders) and a bqo class of multivariate functions sending a list of objects to a new object (so called 'operator algebras'). Closing the class under these functions then yields a new class, which one can prove to be bqo. The crucial step is to show that this construction can be encoded as a structured tree, contained inside a class which is known to preserve bqo. Pouzet's method however was limited in that the structured trees that he used were only 'chain-finite' (i.e. those trees for which every chain is finite).
Since then, larger classes of structured trees have been shown to be preserve bqo. In particular, using a modification of the Minimal Bad Array Lemma (see [11]), Kˇ r´ ıˇ managed to prove that z if Q is well-behaved then R Q (the class of Q -structured trees of R , see Definition 4.7) is wellbehaved [3]. We give a generalisation of Pouzet's method, that incorporates these larger classes of structured trees. This allows for iterating functions over a general linear order and for taking countable limits. Using this coding, these more complex classes can be shown to be bqo (even well-behaved).
So we aim to show that some large classes of partial orders are well-behaved, the main theorem being Theorem 7.37 and its main application Theorem 8.7. The general method of the proof will be as follows. In Section 3 we define an operator algebra construction, similar to ideas explained in [5] but with our two generalisations. We also give an example of how to construct partial orders. In Section 4 we encode this construction in terms of structured trees. In Section 5 we prove that such a construction, under the correct conditions, will be bqo. In Section 6, using the structured tree theorem of Kˇ r´ ıˇ from [3] (Theorem 4.11), in conjunction with our construction z theorem, we construct a more general well-behaved class of ' σ -scattered structured L -trees'. This serves to supercharge the construction theorem. In Section 7 we prove a generalisation of Hausdorff's theorem on scattered order types, which characterises the class of partial orders that we constructed as precisely M L P . This completes the proof of our main theorem, that this class is well-behaved. Finally in Section 8 we explain how this result expands all known bqo results on embeddability of coloured partial orders.
## 2. Preliminaries
## 2.1. Basic bqo theory.
Definition 2.1. If A is an infinite subset of ω , let [ A ] ω = { X ⊆ A | | X | = ℵ } 0 and [ A ] <ω = { X ⊆ A | | X < | ℵ } 0 . We equate X ∈ [ A ] ω with the increasing enumeration of elements of X .
Definition 2.2. · A class Q with a binary relation /lessorequalslant Q on Q is called a quasi-order whenever /lessorequalslant Q is transitive and reflexive.
- · If Q is a quasi-order with /lessorequalslant Q antisymmetric, then we call Q a partial order .
/negationslash
/negationslash
- · For a, b ∈ Q we write a < Q b iff a /lessorequalslant Q b and b /lessorequalslant Q a . We write a ⊥ Q b and call a and b incomparable iff a /lessorequalslant Q b and b /lessorequalslant Q a .
/negationslash
- · We write /lessorequalslant , < and ⊥ in place of /lessorequalslant Q , < Q and ⊥ Q when the context is clear.
4
- · A function f : [ ω ] ω → Q is called a Q -array (or simply an array ) if f is continuous (giving [ ω ] ω the product topology and Q the discrete topology).
- · An array f : [ ω ] ω → Q is called bad if ∀ X ∈ [ ω ] ω we have
$$f ( X ) \, \ll f ( X \, \sum \{ \min X \} ).$$
/negationslash
- · An array f : [ ω ] ω → Q is called perfect if ∀ X ∈ [ ω ] ω we have
$$f ( X ) \leqslant f ( X \, \sum \{ \min X \} ).$$
- · A quasi-order Q is called a better-quasi-order (bqo) if there is no bad Q -array.
Remark 2.3 . We note that we could replace 'continuous' in the definition of a Q -array with 'Borel measurable' and this would make no difference to the definition of bqo (see [11]). We can also consider bad arrays with domain [ A ] ω for some A ∈ [ ω ] ω .
The following is a well-known Ramsey-theoretic result due to Galvin and Prikry.
Theorem 2.4 (Galvin, Prikry [15]) . Given X ∈ [ ω ] ω and a Borel set B in [ X ] ω , there exists A ∈ [ X ] ω such that either [ A ] ω ⊆ B or [ A ] ω ∩ B = ∅ .
Proof. See [15] or [11].
$$\Box$$
Theorem 2.5 (Nash-Williams [9]) . If f is a Q -array, then there is A ∈ [ ω ] ω such that f /harpoonupright [ A ] ω is either bad or perfect.
Proof. Let B = { X ∈ [ ω ] ω | f ( X ) /lessorequalslant f ( X \ { min X } } ) . If B is Borel, then by Theorem 2.4 we will be done. Let S : [ ω ] ω → [ ω ] ω be the function S X ( ) = X \ { min X } and let g = f × ( f ◦ S ). Then g is continuous, since f and S are continuous. We also have that B = g -1 ( /lessorequalslant ), considering the relation /lessorequalslant as a subset of the discrete space Q × Q . Therefore B is open and we are done. /square
Definition 2.6. Given two sets x and y we write x /unionsq y for the disjoint union of x and y . And given a set X of sets, we define ⊔ x ∈ X x as the disjoint union of the sets in X . Definition 2.7. Let Q 0 and Q 1 be quasi-orders, we define new quasi-orders:
- · Q 0 ∪ Q 1 = ( Q 0 /unionsq Q , 1 /lessorequalslant ) where for p, q ∈ Q 0 ∪ Q 1 , we have p /lessorequalslant q iff both p and q are in the same Q i for i ∈ { 0 1 , } , and p /lessorequalslant Q i q ;
- · Q 0 × Q 1 = {〈 q , q 0 1 〉 | q 0 ∈ Q , q 0 1 ∈ Q 1 } where for 〈 p , p 0 1 〉 , 〈 q 0 , q 1 〉 ∈ Q 0 × Q 1 we have
〈 p , p 0 1 〉 /lessorequalslant 〈 q , q 0 1 〉 iff ( p 0 /lessorequalslant Q 0 q 0 ) ∧ ( p 1 /lessorequalslant Q 1 q 1 ) .
Theorem 2.8. If there is a bad Q 0 ∪ Q 1 -array f then there is A ∈ [ ω ] ω such that f /harpoonupright [ A ] ω is either a bad Q 0 -array, or a bad Q 1 -array.
$$\text{ith } B = f ^ { - 1 } ( Q _ { 0 } ). & & \Box$$
Proof. Apply Theorem 2.4 with B = f ( Q 0
Theorem 2.9 (Nash-Williams [9]) . If there is a bad Q 0 × Q 1 -array f , then there is A ∈ [ ω ] ω and either a bad Q 0 -array, or a bad Q 1 -array g with dom ( g ) = [ A ] ω and such that g X ( ) is either the first or second component of f ( X ) for all X ∈ [ A ] ω .
Proof. Define the Q 0 -array f 0 and the Q 1 -array f 1 so that for every X ∈ [ ω ] ω we have
$$f ( X ) = \langle f _ { 0 } ( X ), f _ { 1 } ( X ) \rangle.$$
Now apply Theorem 2.5 twice to restrict so that f 0 and f 1 are both either bad or perfect. Then either we are done or they are both perfect, which contradicts that f was bad. /square
2.2. Concrete categories. Usually we will be interested in quasi-ordering classes of partial orders under embeddability, however we can keep the results more general with no extra difficulty by considering the notion of a concrete category . The idea is to add a little more meat to the notion of a quasi-order, considering classes of structures quasi-ordered by existence of some kind of embedding. This allows us to generate more complicated orders by colouring the elements of these structures with a quasi-order; enforcing that embeddings must increase values of this colouring. Then we can construct complicated objects from simple objects in a ranked way by iterating this colouring process, and the notion of well-behaved allows us to reduce back down through the ranks. We shall now formalise these notions, similar to the definitions of [3] and [4].
Definition 2.10. A concrete category is a pair O = obj( 〈 O ) , hom( O 〉 ) such that:
- (1) each γ ∈ obj( O ) has an associated underlying set U γ ;
- (2) for each γ, δ ∈ obj( O ) there are sets of embeddings hom O ( γ, δ ) consisting of some functions from U γ to U δ ;
- (3) hom O ( γ, γ ) contains the identity on γ for any γ ∈ O ;
- (4) for any γ, δ, β ∈ O , if f ∈ hom O ( γ, δ ) and g ∈ hom O ( δ, β ) then f ◦ g ∈ hom O ( γ, β );
- (5) hom( O ) = { hom O ( γ, δ ) | γ, δ ∈ obj( O } ) .
Elements of obj( O ) are called objects and elements of O ( γ, δ ) are called O -morphisms or embeddings . To simplify notation we write γ ∈ O for γ ∈ obj( O ) and equate γ with U γ .
Remark 2.11 . Similar definitions to 2.10 appear in [3], [4] and [8]. The first two enjoy a more category theoretic description, and the last is in terms of structures and morphisms.
With the following definition, concrete categories will turn into quasi-ordered sets under embeddability ; (3) and (4) of Definition 2.10 guaranteeing the reflexivity and transitivity properties respectively. This allows us to consider the bqo properties of concrete categories.
Definition 2.12. For γ, δ ∈ O , we say that
/negationslash
$$\gamma \leqslant _ { \mathcal { O } } \delta \text{ iff hom} _ { \mathcal { O } } ( \gamma, \delta ) \neq \emptyset$$
i.e. γ /lessorequalslant O δ iff there is an embedding from γ to δ . If f ∈ hom O ( γ, δ ) then we call f a witnessing embedding of γ /lessorequalslant O δ .
Example 2.13. Let obj( P ) be the class of partial orders. For any two partial orders x, y , let:
- (1) U x = x ,
- (2) hom P ( x, y ) = { ϕ : x → y | ( ∀ a, b ∈ x , a ) /lessorequalslant x b ←→ ϕ a ( ) /lessorequalslant y ϕ b ( ) } ,
- (3) hom( P ) = { hom P ( γ, δ ) | γ, δ ∈ obj( P } ) ,
- (4) P = obj( 〈 P ) , hom( P 〉 ) .
The category P of partial orders with embeddings is then a pragmatic example of a concrete category, and the order /lessorequalslant P is the usual embeddability ordering on the class of partial orders. We keep this example in mind since the majority of concrete categories used in this paper are either subclasses of P or are derived from P . We note that all P -morphisms are injective.
Definition 2.14. Given a quasi-order Q and a concrete category O , we add colours to O , by defining the new concrete category O ( Q ) as follows.
- · The objects of O ( Q ) are pairs 〈 γ, c 〉 for γ ∈ O and c : γ → Q .
- · For 〈 γ, c 〉 ∈ obj( O ( Q )), we let V 〈 γ,c 〉 = V γ , we call c a Q -colouring of γ and for each v ∈ γ , we call c v ( ) the colour of v . To simplify notation we equate 〈 γ, c 〉 with γ , and write c γ = c and γ ∈ O ( Q ).
- · We define morphisms of O ( Q ) from γ to δ to be embeddings ϕ : γ → δ such that
$$\mathbf c _ { \gamma } ( x ) \leqslant _ { Q } \mathbf c _ { \delta } ( \varphi ( x ) )$$
for every x ∈ γ .
Given γ ∈ O ( Q ) and v ∈ γ , we will sometimes write c ( v ), to be read as 'the colour of v ', in place of c γ ( v ) when this is unambiguous.
We are now able to define the bqo preservation properties mentioned in Section 1, allowing us to pass from bad O ( Q )-arrays to bad Q -arrays.
Definition 2.15. Let O be a concrete category, then O preserves bqo iff
$$Q \text{ is a bqo } \longrightarrow \mathcal { O } ( Q ) \text{ is a bqo}.$$
Unfortunately, this simple definition fails to be particularly useful. Given a bad O ( Q )-array, preservation of bqo ensures the existence of a bad Q -array, but no link between these two arrays is guaranteed. The following definition remedies this situation and is extremely important for the rest of the paper.
Definition 2.16. Let O be a concrete category, then O is well-behaved iff for any quasi-order Q and any bad array f : [ ω ] ω →O ( Q , ) there is an M ∈ [ ω ] ω and a bad array
$$g \colon [ M ] ^ { \omega } \rightarrow Q$$
such that for all X ∈ [ M ] ω there is some v ∈ f ( X ) with
$$g ( X ) = \mathbf c _ { f ( X ) } ( v ).$$
We call g a witnessing bad array for f .
Warning: this notion of well-behaved is the same as from [3]; it is different from the definition of well-behaved that appears in [4] which is in fact equivalent to Louveau and Saint-Raymond's notion of reflecting bad arrays [8].
Lemma 2.17. Let P be a finite set of finite partial orders, then P is well-behaved.
Proof. Let Q be an arbitrary quasi-order and let f be a bad P ( Q )-array. Then since P is finite, we can repeatedly apply the Galvin and Prikry Theorem 2.4 to find A ∈ [ ω ] ω such that for each X,Y ∈ [ A ] ω , we have that f ( X ) and f ( Y ) have the same underlying finite partial order P . Then applying Theorem 2.5 at most | P | many times, restrict in turn so that the colours of each point of f ( X ) give either a bad array or a perfect array. They cannot all be perfect otherwise f would also be perfect on some restriction to an infinite set. Therefore one of these arrays is bad, and this is clearly a witnessing array for f . /square
Proposition 2.18. O is well-behaved -→ O preserves bqo -→ O is bqo.
Proof. If O is well-behaved then given a bad O ( Q )-array f , we have a bad Q -array. If Q were bqo this would give a contradiction and hence there is no such bad array f .
Now let 1 = { 0 } be the singleton quasi-order, clearly then 1 is bqo. Thus if O preserves bqo then O (1) is bqo. Clearly O (1) is isomorphic to O , therefore O is also bqo. /square
Remark 2.19 . Note that the converse O is bqo -→ O preserves bqo does not hold. For a counter example let Z be the partial order consisting of points 0 n and 1 n for n ∈ ω ; ordered so that for a, b ∈ Z , we have a /lessorequalslant b iff there is some n ∈ ω such that a ∈ { 0 n , 0 n +1 } and b = 1 n . Then { Z } is clearly bqo but it does not preserve bqo.
It is not known whether or not the other converse holds, i.e. is it the case that
$$\mathcal { O } \text{ preserves } b \alpha \, \longrightarrow \mathcal { O } \text{ is well-behaved?}$$
This is an interesting technical question, which was asked by Thomas in [4].
Figure 1. The partial order N .
Definition 2.20. We give some notation for the set of all of possible coloured copies of some γ ∈ obj( O ). Given a concrete category O , a quasi-order Q and γ ∈ obj( O ) we define
$$Q ^ { \gamma } = \{ \langle \gamma, c \rangle \ | \ c \colon \gamma \to Q \} \subseteq \mathcal { O } ( Q ).$$
If γ , γ 0 1 ∈ O ( Q ) are such that there is some γ ∈ O with γ , γ 0 1 ∈ Q γ , then we say that γ 0 and γ 1 have the same structure .
## 2.3. Partial orders.
Definition 2.21. We define Card as the class of cardinals, On as the class of ordinals and On ∗ = { α ∗ : α ∈ On , where } α ∗ is a reversed copy of α for every α ∈ On.
Theorem 2.22 (Nash-Williams [10]) . On is
Proof. See [11, 10].
$$\text{). On is well-behaved.}$$
Definition 2.23. If P is a partial order, a chain of P is a subset with no incomparable elements. An antichain of P is a pairwise incomparable subset.
Definition 2.24. We let 1 = { 0 } be the partial order consisting of a single point. For κ ∈ Card we let A κ be the antichain of size κ . For n ∈ ω we let C n be the chain of length n .
We will now define some notation for traversing partial orders.
Definition 2.25. Let P be a partial order and x ∈ P , we define:
$$\downarrow x = \{ y \in P \ | \ y \leqslant x \}, \ \uparrow x = \{ y \in P \ | \ y \geqslant x \},$$
$$\mathbb { J } x = \{ y \in P \ | \ y < x \}, \quad \mathbb { J } x = \{ y \in P \ | \ y > x \}.$$
$$\begin{matrix} x \}, \\ x \}. \end{matrix}$$
For x, y ∈ P , if it exists, we define the meet x ∧ y to be the supremum of ↓ x ∩ ↓ y .
Definition 2.26. Let P be a partial order and P ′ ⊆ P . Then:
- · we call P ′ ↓ -closed if ( ∀ x ∈ P ′ ), ↓ x = { y ∈ P | y /lessorequalslant x } ⊆ P ′ ,
- · we call P ′ ↑ -closed if ( ∀ x ∈ P ′ ), ↑ x = { y ∈ P | y /greaterorequalslant x } ⊆ P ′ .
Definition 2.27. We define the partial order N = { 0 1 2 3 , , , } as follows. For a, b ∈ N we let a < b iff a = 1 and b ∈ { 0 2 , } or a = 3 and b = 2, (see Figure 1). A partial order is called N -free if it contains no subset isomorphic to N .
Definition 2.28.
- · A linear order is a partial order L with no incomparable elements.
- · A linear order L is well-founded if it has no infinite descending sequence.
- · A linear order L is scattered if Q /lessorequalslant L .
/negationslash
- · A linear order L is σ -scattered iff L can be partitioned into countably many scattered linear orders.
- · We denote the class of all linear orders as L .
- · We denote the class of scattered linear orders as S .
- · We denote the class of σ -scattered linear orders as M .
- · We denote the class of countable linear orders as C .
Definition 2.29. Given r, r ′ ∈ L and linear sequences k = 〈 k i : i ∈ r 〉 and k ′ = 〈 k ′ j : j ∈ r ′ 〉 , we denote by /subsetsqequal the initial segment relation, and /squareimage the strict initial segment relation. That is
$$k \sqsubseteq k ^ { \prime } \text{ iff } k = k ^ { \prime } \text{ or } k = \langle k ^ { \prime } _ { i } \colon i < j \rangle \text{ or } k = \langle k ^ { \prime } _ { i } \colon i \leqslant j \rangle \text{ for some } j \in r ^ { \prime }$$
/negationslash and k /squareimage k ′ iff k /subsetsqequal k ′ and k = k . ′ We denote by k /slurabove k ′ the concatenation of k and k ′ . We also define ot( k ) = r , set k ′ \ k = 〈 k ′ i : i ∈ r ′ \ r 〉 , and call k i the i th element of k .
Definition 2.30. Let P be a partial order, and for each p ∈ P , let P p be a partial order. We define the P -sum of the P p denoted by ∑ p ∈ P P p as the set ⊔ p ∈ P P p ordered by letting a /lessorequalslant b iff · there is some p ∈ P such that a, b ∈ P p and a /lessorequalslant P p b , or
- · there are p, q ∈ P such that a ∈ P p , b ∈ P q and p < P q .
Definition 2.31. If L is a class of linear orders, we define L as the least class containing L ∪ ω and closed under L -sums for all L ∈ L .
$$\text{Theorem } 2. 3 2 \ ( \text{Hausdorff } [ 1 4 ] ). \ I f \, \mathbb { L } = \text{On} \cup \text{On} ^ { * } \text{ then } \overline { \mathbb { L } } = \mathcal { S }.$$
Proof.
$$P r o o f. \, & \text{See } [ 14, 11 ].$$
Definition 2.33. A partial order T is called a tree iff ( ∀ t ∈ T ), ↓ t is a well-founded linear order. If L is a class of linear orders, we call T a L -tree iff every chain of T has order type in L and for every x, y ∈ T , we have ↓ x is a linear order and the meet x ∧ y exists.
Note that On-trees are simply trees, and L -trees are the most general 'tree-like' partial orders.
Remark 2.34 . We make explicit the definition of L -tree, in order to clarify because in the literature on bqo theory, the term 'tree' has varied quite significantly. Indeed, the historical time line of bqo results for trees in the authors' terminology is as follows: Nash-Williams proved that the class of all trees is bqo [9], Laver proved that the class of countable increasing unions of trees that do not embed 2 <ω is bqo [13], then Corominas proved that the class of all countable trees is bqo [2]. This is perplexing since each successive breakthrough seems to be a subclass of the previous! However the differences become clear when we use this notation. Nash-Williams proved that a class of ω +1-trees 4 is bqo, Laver proved that a class of On-trees are bqo, and Corominas proved that a class of C -trees is bqo.
Definition 2.35. For a tree T we call α ∈ On the height of T iff α = sup x ∈ T { ot( ↓ x ) } .
Definition 2.36. Let L be a class of linear orders and T be an L -tree, then we define as follows:
- · t ∈ T is called a leaf of T if there is no t ′ ∈ T such that t ′ > t .
- · T is rooted iff T has a minimal element, denoted root( T ).
- · T is chain-finite iff every chain of T is finite.
Definition 2.37. Given a rooted chain-finite tree T , and some t ∈ T we define inductively 5
$$\text{rank} ( t ) = \sup \{ \text{rank} ( s ) + 1 \, | \, t < _ { T } s \}.$$
We then define the tree rank of T as rank( T ) = rank(root( T )).
4 Here ω +1 is the set of its predecessors.
5 For the base case we have that sup( ∅ ) = 0.
## 3. Operator construction
In this section we give the definitions required to translate more complicated structured trees (i.e. not necessarily chain-finite) into an operator algebra construction similar to Pouzet's [5]. First we must give the parameters of our construction. We will always let:
- · Q be concrete category;
- · B be a subset of Q ;
- · A be a concrete category;
- · F be a quasi-ordered class of functions f with range in Q ;
- · L be a non-empty class of linear orders that is closed under taking non-empty subsets;
- · C denote the whole system 〈Q B A F L〉 , , , , .
Intuitively, Q is going to be the class of objects for which we will be constructing a bqo subclass; B will be a class of 'simple' objects that we will start the construction from; A is a class of possible arities which we will use to generalise the notion of a multivariate function; F is a class of functions which we will be applying to elements of Q in order to construct more complex elements of Q ; and L is a class of linear orders which we will allow iteration of functions from F over. We will keep these standard symbols when using this construction.
We restrict our attention to such Q F A , , and L , so that for every f ∈ F there is some a ( f ) ∈ A and b ( f ) ⊆ a ( f ) with
$$\text{dom} ( f ) = \{ a \in \mathcal { Q } ^ { \mathfrak { a } ( f ) } \ | \ ( \forall i \in \mathbf b ( f ) ), \mathbf c _ { a } ( i ) \in \mathcal { B } \}.$$
(Here Q a ( f ) is as from Definition 2.20.) We call a ( f ) the arity of f . We think of the functions of F as having arguments structured by a ( f ). For example, if a ( f ) is a finite linear order of length n then the arguments of f are linearly ordered and f has the form f ( x , ..., x 1 n ); if a ( f ) were an antichain, then the order on the x i would not matter; and if a ( f ) was the binary tree of height 2 then we think of f having form
$$f \, \left ( \begin{array} { c } x \\ y \end{array} \right ). \\ \text{tions it will be more conv}$$
We do this because for some functions it will be more convenient to think of the arguments arranged in some general partial order (particularly the sums of Definition 2.30). We include b ( f ) in the definition, since in Section 6 it will be more convenient to only allow elements of B into some arguments of our functions. All other constructions used in this paper will have b ( f ) = ∅ for every f ∈ F , in which case dom( f ) = Q a ( f ) .
Example 3.1. Let Q be a quasi-order and L be some class of linear orders. Now we let
- · Q = L ( Q ) be the class of all Q -coloured linear orders;
- · B = Q 1 , i.e. single points coloured by elements of Q ;
- · A = L ∪ ω ;
- · F be the set of L -sums for all L ∈ A , as defined in Definition 2.30, inheriting colours;
- · L = { 1 } .
We order F by letting ∑ A /lessorequalslant F ∑ B iff A /lessorequalslant B . For ∑ L ∈ F we have a ( ∑ L ) = L and b ( ∑ L ) = ∅ . Now define C Q L = 〈Q B A F L〉 , , , , . This relatively simple example will construct the class L ( Q ), which will ultimately allow us to prove that if L is well-behaved, then L is well-behaved.
Example 3.2. Let Q be a quasi-order; Q ′ be Q with an added minimal element -∞ ; let L be some class of non-empty linear orders; and let P be some class of partial orders. Now we let
- · Q be the class all of Q ′ -coloured partial orders;
- · B = Q ′ 1 ;
- · A = P ;
- · F be the set of P -sums for all P ∈ P , inheriting colours;
- · L = L .
We order F by letting ∑ A /lessorequalslant F ∑ B iff A /lessorequalslant B . Now define C Q L P , = 〈Q B A F L〉 , , , , . This example will construct our class of Q ′ -coloured partial orders, that we will characterise in Section 7.
We wish to generalise Pouzet's operator algebra construction in two ways. The first way will allow us to iterate functions over a linear order from L , and the second way will allow us to take countable limits.
3.1. Iterating over L . First we will define what is required for iteration. This will allow us to represent complicated functions in terms of simpler ones. For a basic example, suppose we would like to construct the ω -sum as from Definition 2.30. Given two linear orders L 0 and L 1 , we can define the simple function +, so that L 0 + L 1 is a copy of L 0 followed by a copy of L 1 . Now suppose we iterate this function; we can easily define L 0 + L 1 + L 2 and L 0 + L 1 + L 2 + L 3 and so on. This is easily done finitely many times (which is what Pouzet was doing when he used chain-finite trees [5]). But it is possible to iterate + over a more complex linear order. Naturally, its ω iteration would be ⋃ i ∈ ω L i ordered by a < b if a ∈ L i , b ∈ L j , i < j or i = j and a < L i b (i.e. the ω -sum). We need not stop there though, we could define this iteration over a larger linear order, e.g. its Q iteration could be defined similarly.
Here + is really the function ∑ C 2 . We think of this function as having arguments arranged in arity C 2 . At each successive stage of the iteration we apply the next function into argument corresponding to the larger of the two points in C 2 . 6 If we repeatedly compose ∑ C 2 in this way, we get the functions ∑ C 3 , ∑ C 4 , and so on. If we allowed ourselves to compose over a linear order we could get ∑ ω or ∑ Q as in the previous paragraph. So in general we want to be able to turn linearly ordered lists of functions, each with a distinguished argument, into a new function that acts as their composition. This gives rise to the definition of a composition sequence , which will be a list of functions f i indexed by a linear order r , each with a distinguished argument s i .
/negationslash
Definition 3.3. We call η a composition sequence iff η = 〈〈 f , s i i 〉 : i ∈ r 〉 where r ∈ L , and for all i ∈ r , we have f i ∈ F and s i ∈ a ( f i ) such that if i = max( ) then r s i / ∈ b ( f ).
We call r the length of the sequence η . For i ∈ r let
/negationslash
We also define A = i r a i and B = i r b ( f i ) so that B ⊆ A . For any j ∈ r , let
$$a _ { i } ^ { \eta } = & \begin{cases} \mathbf a ( f _ { i } ) \sum \{ s _ { i } \} & \text{if} \quad i \neq \max ( r ) \\ \mathbf a ( f _ { i } ) & \text{if} \quad i = \max ( r ) \end{cases}. \\ \bigsqcup _ { i \neq r } a _ { i } ^ { \eta } \text{ and } B ^ { \eta } = \bigsqcup _ { i \neq r } \mathbf b ( f _ { i } ) \text{ so that } B ^ { \eta } \subseteq A ^ { \eta }. \text{ For a}$$
$$\mathfrak { v } \stackrel { \sim } { \lambda \cdot \nu } \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \colon \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \,. } \end{cases}$$
For each composition sequence η we will define a function f η that will act as the composition of the functions f i (in order type r ). We want f j to be applied to the composition of all of the f i (for i > j ) in the argument s j .
/negationslash
Remark 3.4 . We require that for i ∈ r , if i = max( ) then r s i / ∈ b ( f ), because otherwise it could be that the composition of the f j for j > i is not allowed into the domain of f i in position s i .
Example 3.5. Suppose we want to define iteration of the sums of Definition 2.30 over a general linear order. Let η = 〈〈 f , s i i 〉 : i ∈ r 〉 be a composition sequence, and for each i ∈ r , suppose a ( f i ) is a partial order, f i is the a ( f i )-sum, and s i ∈ a ( f i ). We turn this composition sequence into a new function f η that acts as the composition of the f i as follows. First define the partial
6 Note that since we are dealing with multivariate functions, we are required to distinguish an argument in order to know in which position to compose further sums inside.
Figure 2. The partial order H η , for η = 〈〈 ∑ N , 3 〉 : i ∈ ω 〉 .
- · i = j and u < a ( f i ) v ;
order H η as the set ⊔ i ∈ r a η i ordered so that for u, v ∈ H η we let u < v iff u ∈ a η i , v ∈ a η j one of the following occurs:
- · i < j and u < a ( f i ) s i ;
- · i > j and v > a ( f j ) s j .
(See Figure 2.) We can then define our composition f η to be the H η -sum. Notice that for finite r , this is equivalent to the usual finite composition of sums, each composed in the argument s i .
The next definition allows us, in a general setting, to turn composition sequences into functions that will act as the linear composition of the functions in the composition sequence.
Definition 3.6. F is called L -iterable if we distinguish a class F L consisting of composition functions f η , for every composition sequence η = 〈〈 f , s i i 〉 : i ∈ r 〉 where
$$f ^ { \eta } \colon \{ \langle q _ { u } \colon u \in A ^ { \eta } \rangle \in \mathcal { Q } ^ { A ^ { \eta } } \ | \ ( \forall u \in B ^ { \eta } ), q _ { u } \in \mathcal { B } \} \longrightarrow \mathcal { Q },$$
and F L satisfies the following properties.
- (i) For all f 0 ∈ F and s 0 ∈ a ( f 0 ), we have f 〈〈 f ,s 0 0 〉〉 = f 0 .
- (ii) Given 〈 q u : u ∈ A η 〉 ∈ dom( f η ) and some j ∈ r , set q s j = f η + j ( 〈 q u : u ∈ a , i > j η i 〉 ), then
$$f ^ { \eta } ( \langle q _ { u } \colon u \in A ^ { \eta } \rangle ) = f ^ { \eta _ { j } ^ { - } } ( \langle q _ { u } \colon u \in A ^ { \eta _ { j } ^ { - } } \rangle ).$$
So we require that when we split η up into initial and final sections η -j and η + j , the composition will behave as expected (i.e. f η is f η -j applied to f η + j ). The remaining arguments of f i are then those in positions corresponding to elements of a η i (see Figure 3), so we consider f η as a multivariate function from Q , which has arguments for each element of a η i , ( i ∈ r ), and only allows values admissible into dom( f i ) (i.e. those respecting that if u ∈ b ( f i ) then q u ∈ B ).
Remark 3.7 . For a composition sequence η of length r , notice that elements of A η can be indexed by i ∈ r and u ∈ a η i , so we will sometimes write elements of dom( f η ) as 〈 q i,u : i ∈ r, u ∈ a η i 〉 .
Lemma 3.8. Let C = C Q L P , = 〈Q B A F L〉 , , , , for some quasi-order Q , some class of linear orders L and some class of partial orders P . Then F is L -iterable .
Proof. Given a composition sequence η , define f η = ∑ H η as in Example 3.5. Consider Definition 3.6, clearly f η satisfies (i). It remains to show (ii), so let 〈 q u : u ∈ A η 〉 ∈ dom( f η ) and for some
Figure 3. The arrangement of the arguments of f η .
i ∈ r , let q s i = f η + i ( 〈 q u : u ∈ a , j > i η i 〉 ). We have that H η + i = ⊔ j>i a η j and H η -i = { s i }/unionsq ⊔ j /lessorequalslant i a η j . So that the H η -i -sum of H η + i in position s i and single points everywhere else is precisely H η . Similarly, we see that
$$f ^ { \eta } ( \langle q _ { u } \colon u \in A ^ { \eta } \rangle ) = \sum _ { u \in H _ { \eta } } q _ { u } = \sum _ { u \in H _ { \eta _ { i } } ^ { - } } q _ { u } = f ^ { \eta _ { i } ^ { - } } ( \langle q _ { u } \colon u \in a _ { i } ^ { \eta }, j \leqslant i \rangle ).$$
So we have (ii) as required.
/square
/negationslash
Definition 3.9. Let η = 〈〈 f , s i i 〉 : i ∈ r 〉 and ν = 〈〈 f , s ′ i ′ i 〉 : i ∈ r ′ 〉 be composition sequences. We define η /lessorequalslant ν iff there is an embedding ϕ : r → r ′ such that for every i ∈ r we have f i /lessorequalslant F f ′ ϕ i ( ) and an embedding ϕ i witnessing a ( f i ) /lessorequalslant a ( f ′ ϕ i ( ) ), such that whenever i = max( ), we have r ϕ i ( s i ) = s ′ ϕ i ( ) . If η /lessorequalslant ν we define ϕ η,ν : A η → A ν so that when u ∈ a η i we have ϕ η,ν ( u ) = ϕ i ( u ).
Definition 3.10. We call f ∈ F extensive if for all q ∈ Q and x ∈ dom( f ) with i ∈ x such that c ( ) = i q , we have that q /lessorequalslant Q f ( x ).
The idea of the next definition is to express the notion that applying 'short' lists of 'small' functions to 'small' objects will give a smaller result than applying 'long' lists of 'large' functions to 'large' objects.
Definition 3.11. We call C = 〈Q B A F L〉 , , , , infinitely extensive iff:
- · F is L -iterable;
- · every f ∈ F is extensive;
- · for any two composition sequences η /lessorequalslant ν and any k = 〈 q u : u ∈ A η 〉 ∈ dom( f η ), k ′ = 〈 q ′ u : u ∈ A ν 〉 ∈ dom( f ν ), if for all u ∈ A η we have embeddings ϕ u witnessing q u /lessorequalslant q ′ ϕ η,ν ( u ) , then we have a corresponding embedding ϕ k,k ′ η,ν witnessing f η ( k ) /lessorequalslant Q f ν ( k ′ ).
Lemma 3.12. Let L be a class of linear orders and Q be an arbitrary quasi-order, then C Q L is infinitely extensive.
Proof. Let C Q L = 〈Q B A F L〉 , , , , . Since L = { 1 } it is clear that F is L -iterable. We have that every f ∈ F is extensive, because if we take the sum of some linear orders, then each of the linear orders embeds into the sum. Now suppose as in Definition 3.11 we have composition sequences
$$\eta = \langle \langle \sum _ { A }, s \rangle \rangle \text{ and } \nu = \langle \langle \sum _ { B }, s ^ { \prime } \rangle \rangle$$
with η /lessorequalslant ν and k = 〈 q u : u ∈ A η 〉 ∈ dom( f η ), k ′ = 〈 q ′ u : u ∈ A ν 〉 ∈ dom( f ν ) such that q u /lessorequalslant q ′ ϕ η,ν ( u ) for each u ∈ A η , with ϕ u a witnessing embedding.
So we have that f η ( k ) = ∑ u ∈ A q u and f ν ( k ) = ∑ u ∈ B q ′ u . Moreover, ϕ η,ν is an embedding from A to B . So we define ϕ k,k ′ η,ν : f η ( k ) → f ν ( k ′ ) so that for a ∈ f η ( k ), with a ∈ q u we have
$$\varphi _ { \eta, \nu } ^ { k, k ^ { \prime } } ( a ) = \varphi _ { u } ( a ) \in q _ { \varphi _ { \eta, \nu } ( u ) } ^ { \prime } \subseteq f ^ { \nu } ( k ^ { \prime } ).$$
Then ϕ k,k ′ η,ν is clearly an embedding. Hence C Q L is infinitely extensive.
/square
Lemma 3.13. Suppose C is infinitely extensive. Let η be a composition sequence and k = 〈 q u : u ∈ A η 〉 ∈ dom ( f η ) . Then for any u ∈ A η , we have q u /lessorequalslant f η ( k ) .
Proof. Let η = 〈〈 f , s i i 〉 : i ∈ r 〉 . Pick some u ∈ A η and let i ∈ r be such that u ∈ a η i . Set η ′ = 〈〈 f , s i i 〉〉 and q s i = f η + i ( 〈 q v : v ∈ a , j > i η j 〉 ). Since f i ∈ F we have f i is extensive, therefore
$$q _ { u } \leqslant f _ { i } ( \langle q _ { v } \, \colon v \in \mathfrak { a } ( f _ { i } ) \rangle ) = f ^ { \eta ^ { \prime } } ( \langle q _ { v } \, \colon v \in A ^ { \eta ^ { \prime } } \rangle ).$$
Then by a simple application of the infinite extensivity of C , we see that
$$f ^ { \eta ^ { \prime } } ( \langle q _ { v } \, \colon v \in A ^ { \eta ^ { \prime } } \rangle ) \leqslant f ^ { \eta _ { i } ^ { - } } ( \langle q _ { v } \, \colon v \in A ^ { \eta _ { i } ^ { - } } \rangle ).$$
Now by L -iterability of F , we have
$$f ^ { \eta _ { i } ^ { - } } ( \langle q _ { v } \colon v \in A ^ { \eta _ { i } ^ { - } } \rangle ) = f ^ { \eta } ( k ).$$
So that q u /lessorequalslant f η ( k ), which gives the lemma.
/square
Definition 3.14. When F is L -iterable we define ˜ C 0 ⊆ Q as the smallest class containing B and closed under applying f η for any composition sequence η .
Example 3.15. Following Definitions 2.28 and 2.30. Let Q be the class of linear orders; B = { 1 } ⊆ Q ; A = On ∪ On ∗ and F = { ∑ α | α ∈ A} be the class of ordinal sums and reversed ordinal sums; finally set L = { 1 } . Then ˜ C 0 = S the class of scattered linear orders. This is precisely Theorem 2.32, Hausdorff's theorem on scattered order types [14].
Remark 3.16 . We can construct ˜ C 0 level by level. Starting with B ; at successor stages applying f η (for every composition sequence η ) to every element of the previous level; and at limit stages taking unions. This allows the definition of an ordinal rank of an element of ˜ . C 0 So for x ∈ C ˜ , 0 we denote by rank( x ) the least α ∈ On such that x appears at level α of this construction.
Definition 3.17. We define Q α = { x ∈ C ˜ 0 | rank( x ) = α } and Q <α = { x ∈ C ˜ 0 | rank( x ) < α } . For each composition sequence η we also define,
$$\text{dom} ( f ^ { \eta } ) _ { < \alpha } = \{ \langle q _ { u } \, \colon u \in A ^ { \eta } \rangle \in \text{dom} ( f ^ { \eta } ) \ | \ ( \forall u \in A ^ { \eta } ), \text{rank} ( q _ { u } ) < \alpha \}.$$
3.2. Limits. Now we will define limits, which will allow us to extend the construction further. For example, when we choose our parameters so that ˜ C 0 gives us scattered linear orderings (Example 3.15), taking limits will give us the σ -scattered linear orderings of [12]. The next definition will allow us to chain together many embeddings eventually allowing us to produce embeddings between limits.
Definition 3.18. We call C extendible iff C is infinitely extensive and satisfies the following property. For any two composition sequences η /lessorequalslant ν , and any k 0 = 〈 q 0 u : u ∈ A η 〉 ∈ dom( f η ), k 1 = 〈 q 1 u : u ∈ A η 〉 ∈ dom( f η ), and k = 〈 q u : u ∈ A ν 〉 ∈ dom( f ν ); if for all u ∈ A η we have:
- · q 0 u /lessorequalslant q 1 u with µ u a witnessing embedding;
- · q 0 u /lessorequalslant q ϕ η,ν ( u ) with ϕ u a witnessing embedding;
- · q 1 u /lessorequalslant q ϕ η,ν ( u ) with ψ u a witnessing embedding;
- · ϕ u = ψ u ◦ µ u ;
then we have
$$\varphi _ { \eta, \nu } ^ { k _ { 0 }, k } = \psi _ { \eta, \nu } ^ { k _ { 1 }, k } \circ \mu _ { \eta, \eta } ^ { k _ { 0 }, k _ { 1 } }.$$
Theorem 3.19. Let C = C Q L P , = 〈Q B A F L〉 , , , , for some quasi-order Q , some class of linear orders L and some class of partial orders P . Then C is extendible.
Proof. By Lemma 3.8 we know that F is L -iterable. If we take the sum of some partial orders, then each of these partial orders embeds into the sum and therefore every f ∈ F is extensive.
Suppose now as in Definition 3.11 that we have composition sequences
$$\eta = \langle \langle f _ { i }, s _ { i } \rangle \, \colon i \in r \rangle \text{ and } \nu = \langle \langle g _ { i }, s _ { i } ^ { \prime } \rangle \, \colon i \in r ^ { \prime } \rangle$$
with η /lessorequalslant ν and k = 〈 q u : u ∈ A η 〉 ∈ dom( f η ) and k ′ = 〈 q ′ u : u ∈ A ν 〉 ∈ dom( f ν ) such that q u /lessorequalslant q ′ ϕ η,ν ( u ) for each u ∈ A η , with ϕ u a witnessing embedding.
We have that f η ( k ) = ∑ u ∈ H η q u and f ν ( k ′ ) = ∑ u ∈ H ν q ′ u . Moreover ϕ η,ν is an embedding from H η to H ν . So we define ϕ k,k ′ η,ν : f η ( k ) → f ν ( k ′ ) so that for a ∈ f η ( k ), with a ∈ q u we have
$$\varphi _ { \eta, \nu } ^ { k, k ^ { \prime } } ( a ) = \varphi _ { u } ( a ) \in q _ { \varphi _ { \eta, \nu } ( u ) } ^ { \prime }.$$
Which is clearly an embedding, hence C is infinitely extensive.
Now suppose that ˆ = k 〈 ˆ q u : u ∈ A η 〉 is such that q u /lessorequalslant ˆ q u for all u ∈ A η , with some witnessing embedding µ u . Suppose also that ˆ q u /lessorequalslant q ′ ϕ η,ν ( ) for each u ∈ A η , with ψ u a witnessing embedding, and that ϕ u = ψ u ◦ µ u . Then for a ∈ f η ( k ), with a ∈ q u we have
$$\varphi _ { \eta, \nu } ^ { k, k ^ { \prime } } ( a ) = \varphi _ { u } ( a ) = \psi _ { u } \circ \mu _ { u } ( a ) \in q _ { \varphi _ { \eta, \nu } ( u ) } ^ { \prime }$$
and µ k,k ′ η,ν ( a ) = µ u ( a ) ∈ ˆ , so that q u
$$\psi _ { \eta, \nu } ^ { \hat { k }, k ^ { \prime } } \circ \mu _ { \eta, \nu } ^ { k, k ^ { \prime } } ( a ) = \psi _ { u } \circ \mu _ { u } ( a ) \in q ^ { \prime } _ { \varphi _ { \eta, \nu } ( u ) }.$$
Thus we have verified the conditions of Definition 3.18, and C is extendible.
/square
Remark 3.20 . Let x ∈ C ˜ 0 with rank( x ) = α . So for some composition sequence η , we have that x = f η ( 〈 q u : u ∈ A η 〉 ), with q u ∈ Q <α for all u ∈ A η . Applying the same reasoning to each of the q u , it is then possible to view x as a composition of some more composition functions, applied to further lower ranked elements or fragments (see Figure 4).
We repeat this process, splitting fragments into more fragments. Because the ranks are wellfounded we will eventually obtain g , a composition of many f η , and an element d of dom( g ), which is a configuration of elements of B such that g d ( ) = x . Since g is just a composition of f η , it will usually be possible to compose further still.
We now make precise the notion of composing many f η .
Definition 3.21. Let I be a tree of finite sequences of elements of ⋃ L × ⋃ A under /subsetsqequal , and let D be the set of leaves of I . Suppose that I has a /subsetsqequal -minimal element, 7 and that we have F = { η /vector ( p ) : /vector p ∈ I \ D } with each η /vector ( p ) a composition sequence. We write r /vector ( p ) for the length of η /vector ( p ). Now suppose that for each /vector p ∈ I we have
$$\vec { p } \langle i, u \rangle \in \mathfrak { I } \text{ iff } i \in r ( \vec { p } ) \text{ and } u \in a _ { i } ^ { \eta ( \vec { p } ) }$$
then we call F an composition set and I a set of position sequences of F . If additionally I is a chain-finite tree under /subsetsqequal then we call F admissible .
7 For this definition we denote the minimal element of I by 〈〉 , but in general it may not be 〈〉 .
Figure 4. The arrangement of the composition of f η and f ν . Here f ν is one of the arguments of f i .
If F is admissible we define the composition of F , a new function g F : Q D →Q . To do so, we determine the value of g F ( 〈 q /vector p : /vector p ∈ D 〉 ). When /vector p ∈ I \ D we define inductively
$$k ^ { \vec { p } } = \langle k ^ { \vec { p } } _ { i, u } \colon i \in r ( \vec { p } ), u \in a _ { i } ^ { \eta ( \vec { p } ) } \rangle \in \text{dom} ( f ^ { \eta ( \vec { p } ) } ),$$
such that for each i ∈ r /vector ( p ) and each u ∈ a η /vector ( p ) i we have
$$k _ { i, u } ^ { \dagger } & = \begin{cases} \ f ^ { \eta ( \vec { p } ^ { \wedge } \langle i, u \rangle ) } ( k ^ { \vec { p } ^ { \wedge } \langle i, u \rangle } ) & \colon & \vec { p } ^ { \wedge } \langle i, u \rangle \notin \mathfrak { D } \\ q _ { \vec { p } ^ { \wedge } \langle i, u \rangle } & \colon & \vec { p } ^ { \wedge } \langle i, u \rangle \in \mathfrak { D } \end{cases}. \\ \intertext { a n i t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t b a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a c t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t { a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a s t t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a a a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t t t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a s a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a e t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t a t
} \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolis _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nolimits _ { \nuclas _ { \nolimits _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ } { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ {.nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuch _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nuclas _ { \nu {$$
Since I was a chain-finite tree, these k /vector p are well-defined. Now when F = , we define ∅
/negationslash
$$g ^ { \tilde { 8 } } ( \langle q _ { \vec { p } } \colon \vec { p } \in \mathfrak { D } \rangle ) = f ^ { \eta ( \langle \rangle ) } ( k ^ { \langle \rangle } ),$$
and when I = D = {〈〉} so that F = , we define ∅ g F ( 〈 x 〉 ) = x for every x ∈ Q .
Definition 3.22. Let F be an admissible composition set and I be a set of position sequences of F . For /vector p ∈ I we define
$$\mathfrak { F } ( \vec { p } ) = \{ \eta ( \vec { u } ) \in \mathfrak { F } \ | \ \vec { p } \sqsubseteq \vec { u } \}$$
and I ( /vector p ) = { /vector u ∈ I | /vector p /subsetsqequal /vector u } . We notice that F ( /vector p ) is an admissible composition set and I ( /vector p ) is a set of position sequences of F ( /vector p ). We also define D ( /vector p ) to be the set of leaves of I ( /vector p ).
Definition 3.23. Let F be an admissible composition set. We call g F a decomposition function for x ∈ C ˜ , whenever there is some 0 〈 q /vector p : /vector p ∈ D 〉 ∈ dom( g F ), such that ( ∀ /vector p ∈ D ), q /vector p ∈ B and
$$x = g ^ { \widetilde { \mathfrak { S } } } ( \langle q _ { \vec { p } } \colon \vec { p } \in \mathfrak { D } \rangle ).$$
Proposition 3.24. Let x ∈ Q α then for some composition sequence η and k ∈ dom ( f η ) <α , we have x = f η ( k ) .
Proof. Clear by Remark 3.16.
/square
Lemma 3.25. For any x ∈ C ˜ 0 , there is a decomposition function g for x .
Proof. Let x ∈ C ˜ , then we will define a decomposition function 0 g for x inductively on the rank of x as follows. If rank( x ) = 0 then x ∈ B so set F = ∅ and I = D = {〈〉} , so that g F ( 〈 x 〉 ) = x is a decomposition function for x .
Suppose for induction that for every q ∈ Q <α there is a decomposition function for x . If rank( x ) = α > 0, then x = f η ( k ) for some composition sequence η of length r , and
$$k = \langle q _ { i, u } \colon i \in r, u \in a _ { i } ^ { \eta } \rangle \in \text{dom} ( f ^ { \eta } ) _ { < \alpha }.$$
So by the induction hypothesis, for each i ∈ r and u ∈ a η i , we see that q i,u = g F i,u ( d i,u ) for some admissible composition set F i,u = { η i,u ( /vector p ) | /vector p ∈ I i,u } with I i,u a set of position sequences for F i,u and some d i,u = 〈 d i,u /vector p : /vector p ∈ D i,u 〉 , where D i,u is the set of leaves of I i,u . Let
$$\mathfrak { I } = \{ \langle i, u \rangle \cap \vec { p } \colon i \in r, u \in a _ { i } ^ { \eta }, \vec { p } \in \mathfrak { F } _ { i, u } \} \cup \{ \langle \rangle \},$$
and let D be the set of leaves of I . We set η ( 〈〉 ) = η and for i ∈ r and u ∈ a η i , we set η ( 〈 i, u 〉 /slurabove /vector p ) = η i,u ( /vector p ). Now let F = { η /vector ( p ) | /vector p ∈ I } , so that clearly F is an admissible composition set. Finally we set d 〈 i,u 〉 /slurabove /vector p = d i,u /vector p and d = 〈 d /vector p : /vector p ∈ D 〉 now we have by construction
$$g ^ { \widetilde { s } } ( d ) = f ^ { \eta } ( \langle g ^ { \widetilde { s } _ { i, u } } ( d _ { i, u } ) \, \colon i \in r, u \in a _ { i } ^ { \eta } \rangle ) = f ^ { \eta } ( k ) = x.$$
Thus g F is a decomposition function for x , which completes the induction. /square
Definition 3.26. We call a decomposition function standard if it can be constructed by the method of Lemma 3.25.
Definition 3.27. Let ( x n n ) ∈ ω be a sequence of elements of ˜ . C 0 Suppose that Q has a minimal element q 0 and for every n ∈ ω , there is a standard decomposition function g n = g F n for x n , and some k n = 〈 k /vector p n : /vector p ∈ D n 〉 ∈ dom( g F n ) with g n ( k n ) = x n . Let I n be the set of position sequences of F n and D n be the set of leaves of I n . Then we call ( x n n ) ∈ ω a limiting sequence if there are such I n , F n , D n and k n such that the following properties hold for every n ∈ ω .
- (1) I n is a ↓ -closed subset of I n +1 .
- (2) F n +1 = { η n ( /vector p ) | /vector p ∈ I n \ D n } .
- (3) η n ( /vector p ) /subsetsqequal η n +1 ( /vector p ) for every /vector p ∈ I n \ D n .
- (4) If /vector u ∈ D m for all m /greaterorequalslant n , then for all m /greaterorequalslant n we have k /vector u n = k /vector u m .
- (5) If /vector u ∈ D n and /vector v ∈ D m with n < m and /vector u /squareimage /vector v , then k /vector u n = q 0 , and ∀ /vector p ∈ D n +1 such that /vector u /subsetsqequal /vector p we have in fact /vector u /squareimage /vector p .
Limiting sequences ( x n n ) ∈ ω are those sequences with an increasing construction; that is to say that x n +1 is produced by the same set of functions as x n applied to increasingly more functions. We could perhaps be slightly less restrictive in the definition, particularly in conditions (4) and (5), however these conditions simplify some of the work later on and do not reduce the set of limits we can produce. 8 This condition determines the characteristics of limits, in particular it enforces that our bqo class of σ -scattered trees will be those covered by ↓ -closed scattered trees (as opposed to any scattered trees).
Remark 3.28 . Suppose C is infinitely extensive, let ( x n n ) ∈ ω be a limiting sequence, and η = η ( 〈〉 ). So for each n ∈ ω we have x n = f η ( k n ) and x n +1 = f η ( k n +1 ), for some k n = 〈 q n u : u ∈ A η 〉 , k n +1 = 〈 q n +1 u : u ∈ A η 〉 . It can be seen by an easy induction 9 that q n u /lessorequalslant q n +1 u with ψ u a witnessing embedding, for all u ∈ A η . So to simplify notation, we let µ n = ψ k n ,k n +1 η,η .
Note that in all of the applications in this paper, we consider x n as a subset of x n +1 , in this case µ n just acts as the identity on elements of x n , and we can define the limit to be the union.
Definition 3.29. Suppose that to every limiting sequence ( x n n ) ∈ ω we associate a unique limit x ∈ Q . Let η be a composition sequence, and for each n ∈ ω let η n /subsetsqequal η be a composition sequence so that η n /subsetsqequal η n +1 , and η = ⋃ n ∈ ω η n . Also let k n = 〈 q n u : u ∈ A η 〉 ∈ dom( f η n ) be such
8 This is the case at least in the applications used within this paper.
9 Using that q 0 was minimal and repeatedly applying Lemma 3.13.
that for every u ∈ A η we have ( q n u ) n ∈ ω is a limiting sequence with limit q u . We say that C has limits if the limit of ( f η n ( k n )) n ∈ ω is precisely f η ( 〈 q u : u ∈ A η 〉 ) for all such η , η n and k n ( n ∈ ω ).
Example 3.30. Let C = C Q L P , = 〈Q B A F L〉 , , , , for some quasi-order Q , some class of linear orders L and some class of partial orders P .
Given a limiting sequence ( x n n ) ∈ ω , we have by Remark 3.28 that µ n was the embedding given by infinite extensiveness of C using the construction of each x n . Each x n +1 is x n with some elements coloured by -∞ replaced by a larger partial order. The embedding µ n then acts as the identity on everything that is not changed, and maps a point a coloured by -∞ that is replaced, into some element b of the order that replaces it. Since it makes no difference to the order, we equate a and b , and consider the underlying set of x n as a subset of the underlying set of x n +1 , possibly with some colours changing from -∞ . In this way each µ n ( n ∈ ω ) becomes the identity map from the underlying set of x n to the underlying set of x n +1 .
Now given a limiting sequence ( x n n ) ∈ ω we have that x n ⊆ x n +1 for all n ∈ ω . Hence we can define the limit x to be the union of all of the x n . This means that C has limits, because an H η -sum of limits, is the limit of the H η -sums of the elements of the limiting sequence.
Definition 3.31. Suppose that C has limits. Let ˜ C ⊆ Q be the class containing all elements of ˜ C 0 and all limits of limiting sequences in ˜ . C 0 We also define ˜ C ∞ = ˜ C \ C ˜ 0 .
Finally the following condition allows us to produce embeddings between limits.
Definition 3.32. Suppose that C has limits. Then we say that C has nice limits iff for any x, y ∈ C ˜ with x the limit of ( x n n ) ∈ ω , and for any n ∈ ω ; if there are embeddings ϕ n : x n → y such that ϕ n = ϕ n +1 ◦ µ n , then x /lessorequalslant y . 10
Theorem 3.33. Let C = C Q L P , = 〈Q B A F L〉 , , , , for some quasi-order Q , some class of linear orders L and some class of partial orders P . Then C has nice limits.
Proof. Let x, y ∈ C ˜ be such that x is the limit of ( x n n ) ∈ ω and for any n ∈ ω we have embeddings ϕ n : x n → y such that ϕ n = ϕ n +1 ◦ µ n . We want to show that x /lessorequalslant y .
By Remark 3.30 we consider each x n ⊆ x n +1 and each µ n ( n ∈ ω ) to be the identity map. So ϕ n = ϕ n +1 ◦ µ n is equivalent to a ∈ x n implies ϕ n +1 ( a ) = ϕ n ( a ). Hence it is possible to define ϕ : x → y as the union of all of the ϕ n . We claim that ϕ is an embedding.
Let a, b ∈ x = ⋃ n ∈ ω x n , and let n be least such that a, b ∈ x n and c x n ( a ) = c x ( a ) (such an n exists since either the colour of a is always -∞ or changes at some n , but then stays this colour in x ). In order to show that x /lessorequalslant y we need to verify the following properties of ϕ .
- (1) a /lessorequalslant b iff ϕ n ( a ) /lessorequalslant ϕ n ( b ) iff ϕ a ( ) /lessorequalslant ϕ b ( ) (since ϕ n is an embedding).
- (2) c x ( a ) = c x n ( a ) /lessorequalslant c y ( ϕ n ( a )) = c y ( ϕ a ( )).
So indeed ϕ is an embedding from x to y , and thus x /lessorequalslant y as required.
/square
## 4. Decomposition trees
The aim of this section is to encode the elements of ˜ in C terms of structured trees. This reduces the problem of showing that ˜ is bqo to showing that a class of structured trees is bqo. C
## 4.1. Scattered and structured L -trees.
Definition 4.1. We define the tree 2 <ω , (the infinite binary tree of height ω ), as the tree of finite sequences of elements of 2 = { 0 1 , } ordered by /subsetsqequal .
Definition 4.2. Let L be a class of linear orders and T be an L -tree, then we define as follows:
10 Here µ n is as from Remark 3.28.
Figure 5. A ζ -tree-sum of the T γ i .
. . .
Figure 6. Scattered L -trees of increasing scattered rank, and a σ -scattered L -tree.
- · T is scattered iff 2 <ω /lessorequalslant T .
/negationslash
- · T is L -σ -scattered iff there are countably many ↓ -closed subsets of T that are each scattered L -trees, and every point of T is contained in one of these subsets.
- · We let U L be the class of scattered L -trees.
- · We let T L be the class of L -σ -scattered L -trees.
- · We let R be the class of rooted ω +1-trees.
Remark 4.3 . When the context is clear we will call elements of T L σ -scattered. Notice that elements of T L are not necessarily L -trees as they could contain chains with order type not in L . 11 We note also that different classes of linear orders can generate the same classes of σ -scattered trees, for example T ω = T ω +1 .
Definition 4.4. Given a chain ζ , and L -trees T γ i for each i ∈ ζ , and γ < κ i ∈ Card, we define the ζ -tree-sum of the T γ i (see Figure 5) as the set ζ /unionsq ⊔ i ∈ ζ,γ<κ i T γ i ordered by letting a /lessorequalslant b iff · a, b ∈ ζ with a /lessorequalslant ζ b ;
- · or a ∈ ζ and b ∈ T γ i for some i ∈ ζ with a < ζ i and γ < κ i .
- · or for some i ∈ ζ , γ < κ i we have a, b ∈ T γ i with a /lessorequalslant T γ i b ;
Definition 4.5. Let L be a class of linear orders that is closed under finite sums. Let U L 0 = {∅ , 1 } , and for α ∈ On let U L α +1 be the class of ζ -tree-sums of trees of U L α for ζ ∈ L . For limit λ ∈ On we let U L λ = ⋃ γ<λ U L γ , and finally set U L ∞ = ⋃ γ ∈ On U L γ . For T ∈ U L ∞ define the scattered rank of T , denoted rank U ( T ) as the least ordinal α such that T ∈ U L α . (See Figure 6.)
Theorem 4.6. Let L be a class of linear orders that is closed under non-empty subsets and finite sums. Then U L = U L ∞ .
Proof.
We leave the proof as an exercise since it is similar to Theorem 6.7. /square
Definition 4.7. Let T be a class of L -trees, and let O be a concrete category. We define the new concrete category of O -structured trees of T , denoted T O as follows. The objects of T O consist of pairs 〈 T, l 〉 such that:
- · T ∈ T .
11 Elements of T L are are still L -trees.
Figure 7. A tree structured by { C 2 , A 2 } .
- · U 〈 T,l 〉 = T .
- · l = { l | v ∈ T } , where for each v ∈ T there is some γ ∈ obj( O ) such that
- v v l v : /arrowtailright v -→ γ v
and if x, y ∈ T with x > y > v then l v ( x ) = l v ( y ).
For O -structured trees 〈 T, l 〉 and 〈 T , l ′ ′ 〉 , we let ϕ : T → T ′ be an embedding whenever:
- (1) x /lessorequalslant y iff ϕ x ( ) /lessorequalslant ϕ y ( ),
- (2) ϕ x ( ∧ y ) = ϕ x ( ) ∧ ϕ y ( ),
- (3) for any v ∈ T , set θ : range( l v ) → range( l ′ ϕ v ( ) ) such that
$$\theta ( l _ { v } ( x ) ) = l ^ { \prime } _ { \varphi ( v ) } ( \varphi ( x ) ) ;$$
then θ is an embedding of O .
We call l v ( x ) the v -label of x . To simplify notation, we write T in place of 〈 T, l 〉 and always use l v ( x ) to denote the v -label of x , regardless of which T ∈ T Q we are considering (it will be unambiguous since v ∈ T ).
Intuitively T O is obtained by taking T ∈ T and for each vertex v ∈ T , and ordering the successors of v by some order in O as in Figure 7. Embedding for T O is then tree embedding that preserves this ordering on the successors of v for every v ∈ T . However, general L -trees may contain points with no immediate successors. To accommodate this, our labelling functions l v have domain /arrowtailright v and we enforce that if x, y ∈ T with x > y > v then l v ( x ) = l v ( y ).
Definition 4.8. Let T be an O -structured L -tree, with x ∈ T and p ∈ range( l x ) then we define p /arrowtailright x = { y ∈ T | ( y > x ) ∧ ( l x ( y ) = p ) } .
It is clear that T Q is a concrete category and hence we also have defined the Q -coloured, O -structured L -trees of T , denoted T O ( Q ). Finally we mention a theorem of Kˇ r´ ıˇ that z is fundamental to the results of this paper.
Theorem 4.9 (Kˇ r´ ˇ, ız [3]) . If O is a well-behaved concrete category with injective morphisms, then R O is well-behaved.
$$P r o o f. \, \text{See} \, [ 3 ].$$
/square
Remark 4.10 . Louveau and Saint-Raymond proved, using a modification of Nash-Williams' original method, that if Q satisfies a slight weakening of well-behaved (that is stronger than preserving bqo) then R Q , the class of Q -structured trees of R , satisfies this same property [8]. They were unable to attain full well-behavedness and Nash-Williams' method seems to be insufficient.
Theorem 4.11. If O is a well-behaved concrete category with injective morphisms, then T ω O is well-behaved.
/negationslash
Proof. Notice that any T ∈ T ω O is a tree of height at most ω , but not necessarily rooted. Consider { t | ∀ s ∈ T, s < t } , let κ be the cardinality of this set, and enumerate its elements as t i for i ∈ κ . Given a quasi-order Q , let τ : T ω O ( Q ) → On( R O ( Q )) be the function sending T to 〈 κ, c 〉 where c i ( ) = ↑ t i for each i ∈ κ . Thus we have that if τ ( S ) /lessorequalslant τ ( T ) then S /lessorequalslant T . So given a bad T ω O ( Q )-array f , we see that τ ◦ f is a bad On( R O )-array, then by Theorem 2.22 we have a witnessing bad R O ( Q )-array, and by Theorem 4.9 we have a witnessing bad Q -array for f . /square
- 4.2. Encoding with structured L -trees. We now want to take an element x ∈ C ˜ and construct from it a structured L -tree T x that contains all of the information required to describe how x is built up from elements of B , using functions from F . We assume for the rest of this section that F is L -iterable and C has limits.
Definition 4.12. Let F be an composition set with I a set of position sequences for F , and d = 〈 d /vector p : /vector p ∈ D 〉 ∈ dom( g F ). Suppose that for each /vector p ∈ I we have η /vector ( p ) = 〈〈 f /vector p i , s /vector p i 〉 : i ∈ r /vector ( p ) . 〉 We define the F ∪ B -coloured A -structured L -tree T F d whose underlying set is
$$\{ \vec { p } \langle i \rangle \ | \ \vec { p } \in \mathfrak { I } \ \mathfrak { D }, i \in r ( \vec { p } ) \} \cup \mathfrak { D } ;$$
ordering /vector p /slurabove 〈 〉 i /lessorequalslant /vector q /slurabove 〈 j 〉 or /vector p /slurabove 〈 〉 i /lessorequalslant /vector q /slurabove 〈 j, u 〉 ∈ D iff either:
- (1) /vector p = /vector q and j /greaterorequalslant i .
- (2) /vector p /squareimage /vector q and the first element of /vector q \ /vector p is 〈 j , u ′ ′ 〉 with j ′ /greaterorequalslant i .
We colour all /vector p /slurabove 〈 〉 ∈ i T F d by letting
$$\mathbf c ( \vec { p } \, \langle i \rangle ) = f _ { i } ^ { \vec { p } },$$
and for /vector p ∈ D ⊆ T F d we let
$$\mathbf c ( \vec { p } ) = d ^ { \vec { p } }.$$
For all /vector p ∈ I \ D , i ∈ r /vector ( p ) and t > /vector p /slurabove 〈 〉 i we define the labels l /vector p /slurabove 〈 〉 i ( ) t ∈ a ( f /vector p i ) so that
$$l _ { \vec { p } ^ { \sim } ( i ) } ( t ) = \begin{cases} \ s _ { i } ^ { \vec { p } } & \colon & \vec { p } ^ { \sim } \langle j \rangle \subseteq t \text{ or } \vec { p } ^ { \sim } \langle j, u \rangle \subseteq t \text{ for some } j > i \text{ and } u \in a _ { j } ^ { \eta ( \vec { p } ) } &. \\ u & \colon & \vec { p } ^ { \sim } \langle i, u \rangle \subseteq t &. \\ \tilde { \sim } & \tilde { \sim } & \tilde { \sim } &. \end{cases}.$$
Definition 4.13. If g F is a standard decomposition function for x ∈ C ˜ 0 and d is such that g F ( d ) = x , then we call T F d a decomposition tree for x .
Let ( x n n ) ∈ ω be a limiting sequence and x ∈ C ˜ ∞ be the limit of this sequence. Let q 0 , F n and k /vector p n be as in Definition 3.27. We set F = ⋃ n ∈ ω F n (so F is a composition set). Let /vector p ∈ D then there must be some least m such that k /vector p m ∈ D , and we let d /vector p = k /vector p m . 12 Set d = 〈 d /vector p : /vector p ∈ D 〉 . When F and d are defined in this way we call T F d a decomposition tree for x .
Lemma 4.14. If x ∈ C ˜ then there is a decomposition tree T = T F d for x such that T ∈ T L A ( F∪B ) . Moreover if x ∈ C ˜ 0 then there is a decomposition tree T ∈ U L A ( F ∪ B ) for x with rank U ( T ) = rank ( x ) .
Proof. Suppose x ∈ C ˜ 0 then let g F be the decomposition function for x as in Lemma 3.25, and let d = 〈 q /vector p : /vector p ∈ D 〉 ∈ dom( g F ) be such that x = g F ( d ) and q /vector p ∈ B for each /vector p ∈ D . In this case we set T = T F d , so T is a decomposition tree for x . We claim that rank U ( T ) = rank( x ), and prove this by induction on the rank of x .
If rank( x ) = 0 then x ∈ B so that T is just a single point coloured by x , and hence rank U ( T ) = 0. Suppose that rank( x ) = α and for each y ∈ Q <α , and each decomposition tree T ′ for y that was constructed from the decomposition function given by Lemma 3.25, that we have rank( y ) =
12 Note that by (4) of Definition 3.27, we have k /vector p m = k /vector p n for any n /greaterorequalslant m .
rank U ( T ′ ). Then let ζ = {〈 〉 | i i ∈ r ( 〈〉 } ⊆ ) T so that T is a ζ -tree-sum of decomposition trees T u i for some q i,u ∈ Q <α ( i ∈ ζ, u ∈ a η ( 〈〉 ) i ). Then using the induction hypothesis, we have
$$\text{rank} _ { \mathcal { Z } } ( T ) \ & = \ \sup \{ \text{rank} _ { \mathcal { Z } } ( T _ { i } ^ { u } ) + 1 \ | \ i \in \zeta, u \in a _ { i } ^ { \eta ( \langle ) } \} \\ & = \ \sup \{ \text{rank} ( q _ { i, u } ) + 1 \ | \ i \in \zeta, u \in a _ { i } ^ { \eta ( \langle ) } \} \\ & = \ \text{rank} ( x ).$$
This completes the induction, and the lemma holds for all x ∈ C ˜ . 0
If x ∈ C ˜ ∞ then a decomposition tree T for x was of the form T = T F d for some d and F = ⋃ n ∈ ω F n , with the F n as from Definition 3.27. By what we just proved, we have that the underlying set of T is covered by ↓ -closed subsets consisting of the underling sets of T n = T F n d n ∈ U L A ( F ∪ B ) for some d n ; hence T ∈ T L A ( F ∪ B ). /square
Lemma 4.15. Let x ∈ C ˜ and T = T F d be a decomposition tree for x . For any t = /vector p /slurabove 〈 〉 ∈ i T and u ∈ range ( l t ) , there exists some x t, u ( ) ∈ C ˜ with a decomposition tree T x t,u ( ) such that
$$^ { u } \uparrow t = T _ { x ( t, u ) }.$$
/negationslash
Moreover if x ∈ C ˜ 0 and /vector p = 〈〉 or u = s 〈〉 i ; then x t, u ( ) ∈ C ˜ 0 with rank ( x t, u ( )) < rank ( x ) .
/negationslash
/negationslash
Proof. Let I be a set of position sequences for F and fix t = /vector p /slurabove 〈 〉 ∈ i T and u ∈ range( l t ). We will find x t, p ( ). First, if u = s /vector p i we define:
- · I t,u = { /vector q ∈ I | /vector p /slurabove 〈 i, u 〉 /subsetsqequal /vector q } ,
- · F t,u = { η /vector ( q ) | /vector p /slurabove 〈 i, u 〉 /subsetsqequal /vector q } ⊆ F ,
and if u = s /vector p i we define:
/negationslash
- · I t,u = { /vector q ∈ I | ( ∃ j > i )( ∃ v ∈ a η /vector ( p ) j ) , /vector p /slurabove 〈 j, v 〉 /subsetsqequal /vector q } ∪ { /vector p } ,
- · η ′ ( /vector p ) = η /vector ( p ) + i , and η ′ ( /vector q ) = η /vector ( q ) whenever /vector p = /vector q ∈ I ,
- · F t,u = { η ′ ( /vector q ) | /vector q ∈ I t,u } .
Now let
- · D t,u be the set of leaves of I t,u ,
- · d t,u = 〈 d /vector q : /vector q ∈ D t,u 〉 , (where d = 〈 d /vector q : /vector q ∈ D 〉 ),
- · T t,u = T F t,u d t,u .
Then in either case, by construction we have that T t,u = u /arrowtailright t .
/negationslash
If g F t,u ( d t,u ) ∈ C ˜ 0 then set x t, u ( ) = g F t,u ( d t,u ), so we have that T t,u is by definition a decomposition tree for x t, u ( ). Moreover if /vector p = 〈〉 or u = s 〈〉 i then there is some i 0 ∈ r ( 〈〉 ) and u 0 ∈ range( l 〈 i 0 〉 ) such that T t,u ⊆ u 0 /arrowtailright i 0 and hence by Lemma 4.14 we have
/negationslash
$$\text{rank} ( x ( t, u ) ) = \text{rank} _ { \mathcal { A } } ( T _ { t, u } ) \leqslant \text{rank} _ { \mathcal { A } } ( ^ { u _ { 0 } } \uparrow i _ { 0 } ) < \text{rank} _ { \mathcal { A } } ( T ) = \text{rank} ( x ).$$
It just remains to find x t, u ( ) when x ∈ C ˜ ∞ . First suppose that x is the limit of ( x n n ) ∈ ω , then consider x n ( t, u ) for every n large enough so that t ∈ T F n . Then since ( x n n ) ∈ ω was a limiting sequence; for some m ∈ ω we have that ( x n ( t, u )) n /greaterorequalslant m is a limiting sequence. We let x t, u ( ) be its limit, then by construction x t, u ( ) has a decomposition tree equal to
$$T _ { d _ { t, u } } ^ { \mathbb { U } _ { n \in \omega } \, \mathfrak { F } _ { t, u } ^ { n } } = T _ { d _ { t, u } } ^ { \mathfrak { F } _ { t, u } } = T _ { t, u } = { ^ { u } } \uparrow t.$$
Definition 4.16. For a given x ∈ C ˜ and a decomposition tree T for x , let ζ be a ↓ -closed chain of T that contains no leaf. For each i ∈ ζ , if i is not the maximum element of ζ then let j i be an arbitrary element of ζ ∩↑ i , otherwise let j i be an arbitrary element of ↑ i . We then define
$$\eta ( \zeta ) = \langle \langle \mathbf c ( i ), l _ { i } ( j _ { i } ) \rangle \, \colon i \in \zeta \rangle$$
which is a composition sequence since for each i ∈ ζ we have c ( ) is i a function of F , and since the labels below i are elements of a c ( ( )). i We note that the choice of j i made no difference to l i ( j i ), except possibly when i is maximal, in which case only s i is ambiguous - but this makes no difference to f η ζ ( ) .
To simplify notation, for each i ∈ ζ , we let f ζ i = c ( ), i s ζ i = l i ( j i ) and a ζ i = a η ζ ( ) i . We also define k ζ = 〈 x i, u ( ) : i ∈ ζ, u ∈ a η ζ ( ) i 〉 ∈ dom( f η ζ ( ) ), where x i, u ( ) is as from Lemma 4.15.
We now require the following lemma, which will allow us to swap composition sequences.
Lemma 4.17. Let F be L -iterable, x ∈ C ˜ 0 have a decomposition tree T = T F d and let ζ be a ↓ -closed chain of T , that is either maximal or has a maximum element, then
$$x = f ^ { \eta ( \zeta ) } ( k ^ { \zeta } ).$$
Moreover if C has limits, then the conclusion holds for all x ∈ C ˜ .
/negationslash
/negationslash
Proof. Let ξ = {〈 〉 | i i ∈ r ( 〈〉 } ) be the chain satisfying η ξ ( ) = η ( 〈〉 ). We assume without loss of generality that ζ /subsetsqequal ξ , since otherwise the lemma follows from a simple application of L -iterability (noticing that η ζ ( ) = η ξ ( ) -max( ) ζ ). We also assume that ξ /subsetsqequal ζ since ξ /subsetsqequal ζ is impossible unless ξ has a maximum element, 13 in which case the lemma follows similarly by L -iterability. Clearly the Lemma holds when rank( x ) = 0 so suppose for induction that rank( x ) = α and the lemma holds for all y ∈ Q <α .
By Lemma 4.14 we have T ∈ T L A ( F ∪ B ) and thus ξ ∩ ζ has a maximal element m . By Proposition 3.24, we have that x = f η ξ ( ) ( k ξ ) and k ξ ∈ dom( f η ξ ( ) ) <α . For each χ ∈ { ζ, ξ } and u ∈ A η χ ( ) we define w χ u = x i, u ( ) whenever i is such that u ∈ a η χ ( ) i , and
$$w _ { s _ { m } ^ { \chi } } ^ { \chi } = f ^ { \eta ( \chi ) _ { m } ^ { + } } ( \langle w _ { u } ^ { \chi } \, \colon u \in A ^ { \eta ( \chi ) _ { m } ^ { + } } \rangle ).$$
We then set w χ = 〈 w χ u : u ∈ A η χ ( ) -m i 〉 ∈ dom( f η χ ( ) ). Thus, since F is L -iterable, we have
$$f ^ { \eta ( \xi ) } ( k ^ { \xi } ) = f ^ { \eta ( \xi ) _ { m } ^ { - } } ( w ^ { \xi } ) \quad \text{and} \quad f ^ { \eta ( \zeta ) } ( k ^ { \zeta } ) = f ^ { \eta ( \zeta ) _ { m } ^ { - } } ( w ^ { \zeta } ).$$
/negationslash
Clearly η ξ ( ) -m = η ξ ( ∩ ζ ) = η ζ ( ) -m since m was the maximal element of ξ ∩ ζ . It remains to verify that w ξ = w ζ . By definition of m , for all i < m and u ∈ a ξ i = a ζ i , we have w ξ u = x i, u ( ) = w ζ u . We also have w ξ u = x i, u ( ) = w ζ u for u ∈ a ξ m ∩ a ζ m . So the only possible cases in which w ξ u = w ζ u are when u ∈ { s ζ m , s ξ m } , hence it only remains to verify that they agree here too.
/negationslash
Since ξ /subsetsqequal ζ and ζ /subsetsqequal ξ we have s ξ m = s ζ m , and therefore s ξ m ∈ a ζ m . Thus
/negationslash
/negationslash
$$w _ { s _ { m } ^ { \xi } } ^ { \zeta } = x ( m, s _ { m } ^ { \xi } ).$$
We also know
$$w ^ { \xi } _ { s ^ { \xi } _ { m } } = f ^ { \eta ( \xi ) ^ { + } _ { m } } ( \langle w ^ { \xi } _ { u } \, \colon u \in A ^ { \eta ( \xi ) ^ { + } _ { m } } \rangle ) = f ^ { \eta ( \xi ) ^ { + } _ { m } } ( \langle x ( i, u ) \, \colon i \in \xi, i > m, u \in a ^ { \eta ( \xi ) ^ { - } _ { m } } _ { i } \rangle )$$
and hence by the construction of x m,s ( ξ m ) from Lemma 4.15, we see that
$$w _ { s _ { m } ^ { \xi } } ^ { \xi } = x ( m, s _ { m } ^ { \xi } ).$$
Hence indeed we have w ξ s ξ m = w ζ s ξ m .
13 This is by definition of the order on decomposition trees.
Figure 8. The method of the proof of Lemma 4.17.
/negationslash
Similarly, we have that w ξ s ζ m = x m,s ( ζ m ) and w ζ s ζ m = f η ζ ( ) + m ( 〈 w ζ u : u ∈ A η ζ ( ) + m 〉 ). It remains only to show that these are equal. Note that by Lemma 4.15, since s ζ m = s ξ m we have that rank( x m,s ( ζ m )) < α . We also have that η ζ ( ) + m = η ζ ( ∩ ↑ m ) and ζ ∩ ↑ m is a ↓ -closed chain of s ζ m /arrowtailright m , which is a decomposition tree for x m,s ( ζ m ). So by the induction hypothesis we have
$$w ^ { \xi } _ { s ^ { \zeta } _ { m } } = x ( m, s ^ { \zeta } _ { m } ) = f ^ { \eta ( \zeta ) ^ { + } _ { m } } ( \langle w ^ { \zeta } _ { u } \colon u \in A ^ { \eta ( \zeta ) ^ { + } _ { m } } \rangle ) = w ^ { \zeta } _ { s ^ { \zeta } _ { m } }.$$
So we have verified that f η ξ ( ) -m ( w ξ ) = f η ζ ( ) -m ( w ζ ), and therefore x = f η ξ ( ) ( k ξ ) = f η ζ ( ) ( k ζ ).
Now suppose that C has limits and let x be the limit of ( x n n ) ∈ ω . For each n ∈ ω let T n be the corresponding decomposition tree for x n , such that the underlying set of T is the union of the underlying sets of the T n ( n ∈ ω ). Then ζ = ⋃ n ∈ ω ζ n where for each n ∈ ω ζ , n is a ↓ -closed, chain of T n which is either maximal or has a maximal element. So by what we just proved, we have that x n = f η ζ ( n ) ( k ζ n ). But we know by Definition 3.29, that the limit of the limiting sequence ( f η ζ ( n ) ( k ζ n )) n ∈ ω is precisely f η ζ ( ) ( k ζ ). Therefore since limits were unique, we have f η ζ ( ) ( k ζ ) = x as required. /square
## 5. The construction is bqo
Now we aim to prove that if x, y ∈ C ˜ have decomposition trees T x and T y respectively, then x /lessorequalslant y . This will allow us to reflect bad ˜ -arrays to bad arrays for the trees. C Then we can use a structured tree theorem such as Theorem 4.11 or Theorem 6.12 in order to show that ˜ is bqo. C
## 5.1. The non-limit case.
Theorem 5.1. Suppose C is infinitely extensive. If x ∈ C ˜ 0 and y ∈ C ˜ have decomposition trees T x and T y respectively such that T x /lessorequalslant T y , then x /lessorequalslant y .
Proof. Suppose x, y are as in the statement of the theorem. We prove x /lessorequalslant y by induction on rank( x ). So for the base case we assume rank( x ) = 0, i.e. that x ∈ B , so we have that T x is just a single point coloured by x . If T x /lessorequalslant T y , then there must be some t 0 ∈ T y such that x /lessorequalslant c ( t 0 ) and thus necessarily t 0 = /vector p /slurabove 〈 i, u 〉 is a leaf of T y . Let ξ = ↓ t 0 , thus by Lemma 4.17 we have that y = f η ξ ( ) ( 〈 y j, u ( ) : j ∈ ξ, u ∈ a ξ j 〉 ). Now since y /vector ( p /slurabove 〈 〉 i , u ) = c ( t 0 ) we have by Lemma 3.13 that
$$x \leqslant \mathbf c ( t _ { 0 } ) \leqslant f ^ { \eta ( \xi ) } ( k ^ { \xi } ) = y$$
which gives the base case.
Now suppose that rank( x ) = α and whenever rank( x ′ ) < α , T x ′ /lessorequalslant T y ′ and y ′ ∈ C ˜ we have x ′ /lessorequalslant y ′ . We set η = η ( 〈〉 ) as from Proposition 3.24, so for some k = 〈 q u : u ∈ A η 〉 ∈ dom( f η ) <α , we have that x = f η ( k ). We also let χ = {〈 〉 | i i ∈ r ( 〈〉 } ⊆ ) T x , η = 〈〈 f , s i i 〉 : i ∈ r 〉 and ϕ be an embedding witnessing T x /lessorequalslant T y . We notice that η = η χ ( ).
For 〈 〉 ∈ i χ we denote by ϕ i the embedding from range( l 〈 〉 i ) to range( l ϕ ( 〈 〉 i ) ) induced by the structured tree embedding ϕ . We have, since ϕ was an embedding, that for each p ∈ range( l 〈 〉 i ),
$$\varphi ( ^ { p } \uparrow \langle i \rangle ) \subseteq ^ { \varphi _ { i } ( p ) } \uparrow \varphi ( \langle i \rangle ).$$
/negationslash
By Lemma 4.15, this implies T x ( 〈 〉 i ,p ) /lessorequalslant T y ϕ ( ( 〈 〉 i ) ,ϕ i ( p )) where T x ( 〈 〉 i ,p ) and T y ϕ ( ( 〈 〉 i ) ,ϕ i ( p )) are decomposition trees for x ( 〈 〉 i , p ) and y ϕ ( ( 〈 〉 i ) , ϕ i ( p )) respectively. Moreover, whenever p = s i , or 〈 〉 i is the maximum element of χ , we have p /arrowtailright i ∩ χ = ∅ ; hence rank( x i, p ( )) < α by Lemma 4.15. Therefore, by the induction hypothesis, whenever p ∈ a η i we have that
$$x ( \langle i \rangle, p ) \leqslant y ( \varphi ( \langle i \rangle ), \varphi _ { i } ( p ) )$$
Now let so that by Lemma 4.17 we have that y = f η ζ ( ) ( k ζ ). Let ϕ ′ be the embedding from ot( χ ) to ot( ζ ) induced by ϕ . It then simple to verify that ϕ ′ and the ϕ i witness the fact that η = η χ ( ) /lessorequalslant η ζ ( ). Then using (1) and since C is infinitely extensive, we have
$$\zeta = \bigcup _ { t \in \chi } \downarrow \varphi ( t ) \\ = \ f ^ { \eta ( \zeta ) ( k ^ { \zeta } ) }. \ F e t. \, \varphi$$
$$x = f ^ { \eta } ( k ) \leqslant f ^ { \eta ( \zeta ) } ( k ^ { \zeta } ) = y$$
as required.
/square
We immediately obtain the following bqo result.
Corollary 5.2. Suppose C is infinitely extensive. Also suppose that T L A is well-behaved, F is bqo and B is bqo. Then ˜ C 0 is bqo.
Proof. Suppose we had a bad array f : [ ω ] ω → C ˜ . 0 Let g : [ ω ] ω → T L A ( F ∪ B ) be defined by letting g X ( ) be a decomposition tree for f ( X ). Then by Theorem 5.1, g must be bad. Thus since T L A is well-behaved there must be a bad F ∪ B -array, which is a contradiction of Theorem 2.8 since F and B were bqo. /square
We now present a simple application of Theorem 5.1. We will show that if a class of linear orders L is well-behaved, then L is well-behaved. This will allow us to give nicer descriptions of the classes that we will construct in section 7, as well as expanding on some classical results. So for the rest of this subsection we let L be a class of linear orders and Q be a quasi-order.
Definition 5.3. We define L 0 = L ∪ ω , and for α ∈ On,
$$\L _ { \alpha + 1 } = \left \{ \sum _ { i \in L } ^ { ^ { \prime } } L _ { i } \, | \, L \in \mathbb { L } _ { 0 }, L _ { i } \in \mathbb { L } _ { \alpha } \right \} \\ \L _ { \lambda } \, \mathbb { L } _ { \sim } \, \text{final} v \, \text{set} \, \mathbb { L } _ { \sim } = | | _ { \lambda \sim } \, \mathbb { L } _ { \lambda }. \text{ For } x \in \mathbb { L } _ { \sim } \, \text{w}$$
Proposition 5.4. Let C = C Q L , then L ∞ ( Q ) = ˜ C 0 .
and for limit λ , L λ = ⋃ γ<λ L γ finally set L ∞ = ⋃ α ∈ On L α . For x ∈ L ∞ we say that rank L ( x ) = α iff α is least such that x ∈ L α .
Proof. This is by construction, considering Remark 3.16.
/square
## Lemma 5.5. L = L ∞ .
Proof. Clearly L ∪ ω ⊆ L ∞ ⊆ L . So if we can show that L ∞ is closed under sums, we will have the lemma. So let y ∈ L ∞ , and y j ∈ L ∞ for each j ∈ y , and we claim that ∑ j ∈ y y j ∈ L ∞ . Let y 〈〉 = y and suppose inductively that we have defined y s ∈ L ∞ \ L 0 for some sequence s . Then we can find y ′ s ∈ L 0 and for each i ∈ y ′ s , we find y /slurabove s 〈 〉 i ∈ L ∞ with rank L ( y /slurabove s 〈 〉 i ) < rank L ( y s ); such that y s = ∑ i ∈ y ′ s y /slurabove s 〈 〉 i . Thus ∑ j ∈ y s y j = ∑ i ∈ y ′ s ∑ j ∈ y s /slurabove 〈 〉 i y j . So if every y /slurabove s 〈 〉 i ∈ L 0 then
∑ i ∈ y s y i ∈ L ∞ . Since the ranks are well-founded y s ∈ L 0 for all of the longest sequences s generated by this induction. So by induction, ∑ j ∈ y y j ∈ L ∞ , which gives the lemma. /square
Theorem 5.6. If L is well-behaved, then L is well-behaved.
Proof. Let C = C Q L = 〈Q B A F L〉 , , , , . By Lemma 5.5 and Proposition 5.4 we have that L ( Q ′ ) = ˜ . C 0 Suppose we have a bad L -array f , then define g X ( ) to be a decomposition tree for f ( X ). First we note that by Definition 4.12, it is clear that the leaves of the tree g X ( ) are coloured by some element of B used to construct f ( X ). Thus the colours of the linear order f ( X ) are colours of colours of elements of g X ( ). Now using Lemma 3.12 we can apply Theorem 5.1, in order to see that g is a bad array with range in U L A ( F ∪ B ).
We know that L = ω and hence is well-behaved. We also have that A = P is well-behaved and has injective morphisms; hence by Theorem 4.11, there is a witnessing bad F ∪ B -array g ′ . Since the order on F is isomorphic to the order on L , we know that F is bqo. Hence by Theorem 2.8, we can find a witnessing bad B -array f ′ . Since the colours from each g X ( ) that were in B were points of f ( X ) we now know that f ′ is a witnessing bad array for f . Then since B = Q 1 this clearly gives a witnessing bad Q -array just passing to colours. Hence any bad L ( Q )-array admits a witnessing bad Q -array, i.e. L is well-behaved. /square
Remark 5.7 . Without much more difficulty, using we could show that the class of countable unions of elements of L is also well-behaved, by considering limiting sequences ( x n n ) ∈ ω to be such that x n ⊆ x n +1 for each n ∈ ω and defining limits as the countable union. Then it is relatively simple to verify that C is infinitely extensive and has nice limits. The result then follows similarly to Theorem 5.6 (using Theorem 5.8 in place of Theorem 5.1). We omit the proof because this is implied by Theorem 7.37.
## 5.2. The limit case.
Theorem 5.8. Let C be extendible and have nice limits. If x, y ∈ C ˜ have decomposition trees T x and T y respectively such that T x /lessorequalslant T y , then x /lessorequalslant y .
Proof. By Theorem 5.1, it only remains to check that the theorem holds when x ∈ C ˜ ∞ . So suppose that x is a limit of the limiting sequence ( x n n ) ∈ ω . For each n ∈ ω let T n = T F n d n be a decomposition tree for x n . Since T n ⊆ T x /lessorequalslant T y and each x n ∈ C ˜ 0 we have by Theorem 5.1 that x n /lessorequalslant y for every n . Let ϕ n : x n → y be the embedding granted by infinite extensivity that was used in the final line of the proof of Theorem 5.1.
By the definition of a limiting sequence, we see that for each n ∈ ω there are k, k ′ ∈ dom( g n ) such that x n = g n ( k ) and x n +1 = g n ( k ′ ). If k = 〈 q u : u ∈ D n 〉 and k ′ = 〈 q ′ u : u ∈ D n 〉 then for each u ∈ D n we have q u /lessorequalslant q ′ u since either q u = q ′ u or q u is the minimum element q 0 . Set η = η ( 〈〉 ) ∈ F , then we can inductively find, using k and k ′ , elements k n , k n +1 of dom( f η ) such that x n = f η ( k n ), x n +1 = f η ( k n +1 ). Thus we have that if k n = 〈 q n u : u ∈ A η 〉 and k n +1 = 〈 q n +1 u : u ∈ A η 〉 then q n u /lessorequalslant q n +1 u , by repeated applications of Lemma 3.13. So let ψ u be the embedding witnessing q n u /lessorequalslant q n +1 u for u ∈ A η , granted by this induction. Then since C is extendible, we see that
$$\varphi _ { n } = \varphi _ { n + 1 } \circ \psi _ { \eta, \eta } ^ { k ^ { n }, k ^ { n + 1 } } = \varphi _ { n + 1 } \circ \mu _ { n }$$
with µ n as from Remark 3.28. Therefore, since Q has nice limits we have x /lessorequalslant y . /square
Again we immediately obtain a bqo result, we also mention some further corollaries.
Corollary 5.9. Let C be extendible and have nice limits. Suppose that T L A is well-behaved, F is bqo and B is bqo. Then ˜ C is bqo.
Proof. Suppose we had a bad ˜ -array C f . Let define the bad T L A ( F ∪ B )-array g by letting g X ( ) be a decomposition tree for f ( X ). Then by Theorem 5.8, g must be bad. Thus since T L A is well-behaved there must be a bad function to F ∪ B , which is a contradiction of Theorem 2.8 since F and B were bqo. /square
Corollary 5.10 (Laver [12]) . M , the class of σ -scattered linear orders is bqo.
Proof. Set Q B F A , , , and L as in Example 3.15. For a limiting sequence ( x n n ) ∈ ω , we consider x n ⊆ x n +1 and define the limit to be the union, in this way µ n acts as the identity on elements of x n . We then have that ˜ = C M . We have A = On ∪ On ∗ and L = { 1 } hence L = ω so since A has injective morphisms and by theorems 2.22, 2.8 and 4.11 we know T L A is well-behaved. By Corollary 5.9 it just remains to show that C is extendible and has nice limits. Since L = { 1 } it is easily verified that C is extendible. Finally if we have embeddings ϕ n for n ∈ ω as in the definition of nice limits, then we let ϕ be the union of the ϕ n , this clearly satisfies the definition of nice limits. The result now follows from Corollary 5.9. /square
Corollary 5.11 (Laver [13]) . T On , the class of σ -scattered trees is bqo.
Proof. Let Q be the class of all trees, B = {∅} ⊆ Q , A = On and L = 1 . We let { } F = { c α : α ∈ On } ∪ { d κ : κ ∈ Card , ordered so that } c α < c β and d α < d β for α < β ; where we define:
- · c α : Q α →Q such that c α ( 〈 T γ : γ < α 〉 ) is the α -tree-sum of the T γ ( γ < α ).
- · d κ : Q →Q κ such that d κ ( 〈 T γ : γ < κ 〉 ) is the disjoint union of the T γ .
Then ˜ C 0 is U On , the class of scattered trees (as in [13]) by Theorem 4.6. Define the limit of a limiting sequence to be the union so that ˜ = C T On . Now proceed similarly to Corollary 5.10. /square
The following theorem is our main application of Theorem 5.8, which is the crucial step in showing that the large classes of partial orders in Section 7 are well-behaved.
Theorem 5.12. Let Q be a quasi-order, L be a class of linear orders, P be a bqo class of partial orders and C = C Q L P , . If T L P is well-behaved, then any bad ˜ C -array admits a witnessing bad Q -array.
Proof. Let C = 〈Q B A F L〉 , , , , . Suppose we have a bad ˜ -array C f then for each X ∈ [ ω ] ω , let g X ( ) be a decomposition tree for f ( X ). First we note that by Definition 4.12, it is clear that the leaves of the tree g X ( ) are coloured by some element of B used to construct f ( X ). Thus the colours of the partial order f ( X ) are colours of colours of leaves of g X ( ). Now by Theorem 5.8 we have that g is a bad U L A ( F ∪ B )-array.
Since we assumed that T L P is well-behaved, there is a witnessing bad F ∪ B -array g ′ . Since the order on F is isomorphic to the order on P , we know that F is bqo. Hence we can restrict g ′ as in the proof of Theorem 2.8, to find a witnessing bad B -array f ′ . Since the colours from each g X ( ) that were in B were points of f ( X ) we now know that f ′ is a witnessing bad array for f . Then since B = Q ′ 1 we can restrict again to be inside the complement of f ′-1 ( -∞ ) in order to find a witnessing bad Q 1 -array. This clearly gives a witnessing bad Q -array just passing to colours. Hence any bad ˜ -array admits a witnessing bad C Q -array. /square
## 6. Constructing structured L -trees
The aim of this section is to show that whenever L is a class of linear orders and P is a class of partial orders both of which are well-behaved, then we have T L P is also well-behaved. We can then combine this result with Theorem 5.12. The proof will consist of an application of Theorem 5.8, but now we will construct structured trees instead of partial orders.
ε
Figure 9. The order on E ′ , for given E ∈ E .
Definition 6.1. We define E ⊆ L P ( (A )) such that whenever 2 〈 E i : i ∈ r 〉 ∈ E and i ∈ r there is a unique e ∈ E i with c ( e ) = 1.
We define from E the new concrete category E ′ of A -coloured 2 P -structured L -trees. Let E ∈ E be such that E = 〈 E i : i ∈ r 〉 , for each i ∈ r we let
- · E ′ i = E i if i is not the maximum element of r ,
- · E ′ i = E i ∪ { ε } if i is the maximum element of r .
Then we define and let E ′ = { E ′ : E ∈ E } . We order E ′ as in Figure 9 by letting ε > e ∈ E ′ iff c ( e ) = 1; and for d, e ∈ E ′ we have d < e iff d = ε and either
$$E ^ { \prime } = \coprod _ { i \in r } E ^ { \prime } _ { i } \\ \text{as in Figure 9 1}$$
/negationslash
- · d, e ∈ E i , c ( d ) = 1, c ( e ) = 0.
- · d ∈ E i , e ∈ E j , i < j and c ( d ) = 1.
The colouring of elements of the E i induces a colouring of E ′ (with colours in A ), we also set 2 c ( ε ) = 0 when ε ∈ E ′ . For each i ∈ r and d ∈ E i with c ( d ) = 1 and d < e we define the label l d ( e ) to be d if c ( e ) = 1 and e otherwise. This gives us a P -structured L -tree E ′ . We call the chain of elements of E ′ coloured by 1 the central chain of E ′ .
We define morphisms of E ′ to be maps induced by embeddings of E (we can also find a place for the possibly new maximum ε ). Thus morphisms of E ′ will also be structured tree embeddings.
We will now define the parameters of our construction. Throughout this section, we let:
- (1) Q be an arbitrary quasi-order;
- (2) Q ′ be Q with an added minimal element -∞ ;
- (3) P be an arbitrary concrete category;
- (4) L be an arbitrary class of linear orders;
- (5) Q = T L P ( Q ′ );
- (6) B = Q 1 ∪ {∅} ⊆ Q ;
- (7) A = E ′ ;
- (8) L = { 1 } .
It remains to define F , now we will define its functions.
Definition 6.2. For each E = 〈 E i : i ∈ r 〉 ∈ E we define the function S E : dom( S E ) → Q as follows. First let a ( S E ) = E ′ and b ( S E ) = { e ∈ E ′ | c ( e ) = 1 } be the central chain of E ′ . Define
the function τ : dom( S E ) →Q E ′ so that τ ( 〈 x i : i ∈ E ′ 〉 ) = 〈 x ′ i : i ∈ E ′ 〉 where x ′ i is a single point coloured by -∞ if x i = ∅ and i is on the central chain of E ′ ; and x ′ i = x i otherwise. Now we let S E = E ′ ◦ τ . We inherit labels and colours in this sum. We define S E /lessorequalslant F S F iff E /lessorequalslant E F .
∑ Throughout this section we will also let F = { S E : E ∈ E } .
Remark 6.3 . Clearly the minimal element of Q is ∅ . Therefore elements of limiting sequences are constructed by replacing ∅ arguments with larger L -trees. So limiting sequences will be sequences of coloured and structured L -trees, whose underlying sets are each ↓ -closed subsets of the next (as in Figure 6). Thus it is possible to define limits as unions of such sequences.
We now have two goals. Firstly we will show that ˜ = C T L P ( Q ′ ), and secondly we will verify the conditions of Theorem 5.8. Using this theorem we will then see that T L P is well-behaved whenever L and P are well-behaved.
6.1. The constructed trees are σ -scattered. Now we aim to show that ˜ = C T L P ( Q ′ ).
Definition 6.4. Let T ∈ T L P ( Q ′ ), ξ be a chain of T and t ∈ ξ . We define l t ( ξ ) to be ∅ whenever t is the maximal element of ξ , and if there is some t ′ ∈ ξ with t ′ > t then we let l t ( ξ ) = { l t ( t ′ ) } . This is well-defined by the definition of l t .
Lemma 6.5. Suppose that T ∈ T L P ( Q ′ ) \ C ˜ 0 , and ξ is an ↓ -closed chain of T that contains no leaves. Then either ⋂ j ∈ ξ ↑ j / ∈ C ˜ 0 or there is some t ∈ ξ and p ∈ range ( l t ) \ l t ( ξ ) such that p /arrowtailright t / ∈ C ˜ 0 .
Proof. Let T and ξ ⊆ T be as described. Suppose ⋂ j ∈ ξ ↑ j ∈ C ˜ 0 and there are no such t and p . Pick a maximal chain ζ of ⋂ j ∈ ξ ↑ j , and let ξ ′ = ζ ∪ ξ . Then for each i ∈ ot( ξ ′ ) let ξ ′ i be the i th element of ξ ′ and we define E i as an A -coloured copy of range( 2 l ξ ′ i ) ∈ P , that we colour as follows. When ot( ξ ′ ) has a maximum element m , we let every colour of E m be 0. For nonmaximal i ∈ ot( ξ ′ ) we let u ∈ E i be coloured with 1 iff there is some j ∈ ot( ξ ′ ) with i < j and l ξ ′ i ( ξ ′ j ) = u . We set E = 〈 E i : i ∈ ot( ξ ′ ) , then we have that 〉 E ∈ E .
/negationslash
Now for a ∈ E ′ we define K a ∈ T L P ( Q ′ ) as follows. If a is the i th point on the central chain of E ′ and c T ( ξ ′ i ) = -∞ we let K a be a single point coloured by c T ( ξ ′ i ), if c T ( ξ ′ i ) = -∞ then we let K a = , hence ∅ K a ∈ B . Otherwise a is not on the central chain of E ′ and we can let i be largest such that the i th point w of the central chain of E ′ is < a (there was such a largest element, by definition of the order on E ′ ). Now let u ∈ range( l w ) be such that l w ( a ) = u . Then we set
$$K _ { a } = { ^ { u } } \uparrow \xi _ { i } ^ { \prime } \subseteq T.$$
We then have by construction that
$$S _ { E } ( \langle K _ { a } \colon a \in E ^ { \prime } \rangle ) = T.$$
But we know that each K a was in ˜ C 0 by our original assumption. Hence since ˜ C 0 was closed under applying any function in F it must be that T ∈ C ˜ 0 which is a contradiction. /square
Lemma 6.6. Suppose that T ∈ T L P ( Q ′ ) \ C ˜ 0 , then there is some u ∈ T and two disjoint ↑ -closed subtrees S , S 1 2 ⊆ /arrowtailright u such that S , S 1 2 / ∈ C ˜ 0 .
Proof. Let T ∈ T L P ( Q ′ ) \ C ˜ . 0 Pick a chain ξ 0 of T that contains no leaves and is maximal. Then by Lemma 6.5 there is some t 0 ∈ ξ 0 and p 0 ∈ range( l t 0 ) \ l t 0 ( ξ 0 ) such that p 0 /arrowtailright t 0 / ∈ C ˜ . 0 Suppose for induction that for some α ∈ On we have defined t 0 < ... < t α ∈ T and p , ..., p 0 α such that for any γ < α we have l t γ ( t α ) = p γ and p γ /arrowtailright t α / ∈ C ˜ . 0 We now apply Lemma 6.5 to a maximal chain ξ α of p α /arrowtailright t α that contains no leaves, whence we find t α +1 ∈ p α /arrowtailright t α and p α +1 such that p α +1 /arrowtailright t α +1 / ∈ C ˜ , 0 in other words t α +1 and p α +1 satisfy the induction hypothesis.
/negationslash
Now suppose for some limit λ ∈ On we have defined such t γ and p γ for every γ < λ . Let ζ λ = ⋃ γ<λ ↓ t γ , then ζ λ is an ↓ -closed chain of T that contains no leaves because every element has some t γ larger than it. By Lemma 6.5 either ⋂ j ∈ ζ λ ↑ j / ∈ C ˜ 0 or there is some t ∈ ζ λ and p ∈ range( l t ) \ l t ( ζ λ ) such that p /arrowtailright t / ∈ C ˜ . 0 Suppose the latter, then let δ be least such that t < t δ . Since t δ ∈ ζ λ we have that l t ( t δ ) ∈ l t ( ζ λ ) = ∅ . Now we also had that p / l ∈ t ( ζ λ ), hence the trees S 1 = p /arrowtailright t and S 2 = p δ /arrowtailright t δ are disjoint. Moreover each is clearly ↑ -closed and we also have S , S 1 2 / ∈ C ˜ 0 we now set u = t and we are done.
/negationslash
Finally suppose that ⋂ j ∈ ζ λ ↑ j / ∈ C ˜ , 0 then pick a maximal chain ξ λ of ⋂ j ∈ ξ ↑ j , so by Lemma 6.5 there is some t λ ∈ ξ λ and p λ ∈ range( l t λ ) \ l t λ ( ξ λ ) such that p λ /arrowtailright t λ / ∈ C ˜ , and we can continue 0 the induction. Now the induction must stop at some point, otherwise for all limit λ ∈ On it must be that ⋂ j ∈ ζ λ ↑ j = ∅ , and therefore T is a proper class! This contradiction ensures that we will always find u , S 1 and S 2 as required. /square
$$\text{Theorem } 6. 7. \ \tilde { \mathcal { C } } _ { 0 } = \mathcal { V } _ { \mathbb { P } } ^ { \overline { \mathbb { L } } } ( Q ^ { \prime } ).$$
/negationslash
/negationslash
Proof. First we will show that ˜ C 0 ⊆ U L P ( Q ′ ). It is clear that ˜ C 0 consists of Q ′ -coloured, P -structured L -trees, so we prove by induction on the rank of x ∈ C ˜ 0 that 2 <ω /lessorequalslant x . Clearly if x ∈ B then 2 <ω /lessorequalslant x , so we have the base case.
So let x ∈ Q α for some α ∈ On. Then for some E ∈ E we have that x = S E ( 〈 x i : i ∈ E ′ 〉 ) with x i ∈ Q <α for each i ∈ E ′ . Hence by the induction hypothesis, x is a ζ -tree-sum of the scattered trees x i , ( i ∈ E ′ ) of lower rank for some chain ζ ⊆ x .
/negationslash
For the other direction we want to show ˜ C 0 ⊇ U L P ( Q ′ ). So suppose that T ∈ U L P ( Q ′ ) \ C ˜ , 0 we will show that 2 <ω /lessorequalslant T . First we set T 〈〉 = T . Now suppose that we have already defined T s / ∈ C ˜ 0 for some s ∈ 2 <ω . We apply Lemma 6.6 to T s to obtain corresponding u s , and S , S s 1 s 2 / ∈ C ˜ . 0 Now let T s /slurabove 〈 0 〉 = S s 1 and T s /slurabove 〈 1 〉 = S s 2 .
Suppose 2 <ω /lessorequalslant x and let ψ be a witnessing embedding. Then either range( ψ ) is contained entirely inside the chain ζ (which is a contradiction) or there is some z ∈ 2 <ω with ψ z ( ) inside x i for some i ∈ E ′ . But each x i is ↑ -closed, hence ψ ( ↑ z ) ⊆ x i . But for any point t ∈ 2 <ω we have that ↑ t is a copy of 2 <ω , and therefore 2 <ω /lessorequalslant x i which contradicts that x i was scattered. Therefore 2 <ω /lessorequalslant x , and hence we have ˜ C 0 ⊆ U L P ( Q ′ ).
We now define the embedding ϕ : 2 <ω → T by letting ϕ s ( ) = u s . We claim that this is a tree embedding. For any s ∈ 2 <ω we have that the ↑ -closed trees T s /slurabove 〈 0 〉 and T s /slurabove 〈 1 〉 are disjoint, hence u /slurabove s 〈 0 〉 and u /slurabove s 〈 1 〉 are incomparable elements of T . It remains to check that for i ∈ { 0 1 , } we have u s < u /slurabove s 〈 〉 i , but this is clear since we had that
$$u _ { s \widehat { \mu } \langle i \rangle } \in T ^ { s \widehat { \mu } \langle i \rangle } = S _ { i } ^ { s } \subseteq \uparrow u _ { s }.$$
So indeed ϕ is an embedding and 2 <ω /lessorequalslant T . Therefore T / ∈ U L P ( Q ′ ), and we conclude that ˜ C 0 ⊇ U L P ( Q ′ ). So we have shown both directions, which completes the proof. /square
## 6.2. The class T L P of structured trees is well-behaved.
Lemma 6.8. If L and P are well-behaved, then A is well-behaved and F is bqo.
Proof. Suppose L and P are well-behaved. We have that A = E ′ and E ⊆ L P ( (A )). 2 Consider the function
$$\tau \colon \mathbb { E } ^ { \prime } ( Q ) \rightarrow \mathbb { L } ( \mathbb { P } ( \mathtt A _ { 2 } \times Q ) ) \times Q ^ { \prime }$$
defined as follows. First if ˆ E ′ ∈ E ′ ( Q ), (so ˆ E ′ is an A 2 × Q -coloured copy of E ′ ∈ E ′ ), then let ˆ E be a coloured copy of E = 〈 E i : i ∈ r 〉 ∈ E such that c ˆ E i ( z ) = c ˆ E ′ ( z ) for every i ∈ r and z ∈ E i . If we added ε to E ′ then we let σ = c ( ε ), otherwise we let σ = -∞ . Then we define
τ ( ˆ ) = E ′ 〈 ˆ E,σ 〉 . Now if τ ( E ′ ) /lessorequalslant τ ( F ′ ) then there is an embedding of E that increases the colours that are in Q . Hence in this case E ′ /lessorequalslant F ′ .
So suppose there were a bad E ′ -array f , hence τ ◦ f is a bad L P ( (A 2 × Q )) × Q ′ -array and we notice that τ ◦ f ( X ) has all of its colours in Q from f ( X ), and all its colours in Q ′ either from f ( X ) or -∞ . By Theorem 2.9 we either get either a witnessing bad Q ′ -array or L P ( (A 2 × Q ))-array. In the first case applying the Galvin and Prikry Theorem 2.4, restricting to be inside the complement of f -1 ( -∞ ) would give a bad Q -array which would be witnessing for f .
So we have a witnessing bad L P ( (A 2 × Q ))-array. Since L is well-behaved, we have by Theorem 5.6 that L is well-behaved, and hence since P is also well-behaved, we have a witnessing bad A 2 × Q -array f ′ . But each of the colours in Q of f ′ ( X ) were from f ( X ), hence using Theorem 2.9 we find a witnessing bad Q -array for f . Therefore A = E ′ is well-behaved. Clearly this means that E is bqo, and hence also F is bqo. /square
We now verify the conditions of Theorem 5.8.
## Lemma 6.9. Every f ∈ F is extensive.
Proof. Let S E ∈ F and consider x = S E ( 〈 T u : u ∈ E ′ 〉 ). Either u ∈ E ′ is on the central chain of E ′ and T u is the minimal element ∅ ; or a copy of T u appears precisely in x . /square
## Theorem 6.10. C is extendible.
Proof. Since L = { 1 } it is clear that F is L -iterable. By Lemma 6.9 we know that every f ∈ F is extensive. So suppose as in Definition 3.11 that we are given composition sequences
$$\eta = \langle \langle S _ { E }, s \rangle \rangle \text{ and } \nu = \langle \langle S _ { F }, s ^ { \prime } \rangle \rangle$$
with η /lessorequalslant ν and k = 〈 q u : u ∈ A η 〉 ∈ dom( S E ), k ′ = 〈 q ′ u : u ∈ A ν 〉 ∈ dom( S F ) such that q u /lessorequalslant q ′ ϕ η,ν ( u ) for each u ∈ A η with ϕ u a witnessing embedding. So we have that f η ( k ) = S E ( k ) and f ν ( k ) = S F ( k ′ ). We also have that ϕ η,ν is an embedding from E ′ = A η to F ′ = A ν . We define ϕ k,k ′ η,ν : f η ( k ) → f ν ( k ′ ) as follows.
If u is on the central chain of E ′ , then u ∈ b ( S E ), so that q u ∈ B . We see that ϕ η,ν sends elements of the central chain of E ′ to the central chain of F ′ since it was induced from an embedding of E , thus also q ′ ϕ η,ν ( u ) ∈ B . Let a be the point of f η ( k ) that corresponds to u , i.e. either it is the single point of q u in the sum or q u = ∅ and a is a single point coloured by -∞ . In this case we let ϕ k,k ′ η,ν ( a ) be the point b ∈ f ν ( k ′ ) that corresponds to ϕ η,ν ( u ). Notice that the embedding ϕ u gives us that c ( a ) /lessorequalslant c ( b ).
Suppose u is not on the central chain of E ′ and that a ∈ q u ⊆ f η ( k ), then we let
$$\varphi _ { \eta, \nu } ^ { k, k ^ { \prime } } ( a ) = \varphi _ { u } ( a ) \in q _ { \varphi _ { \eta, \nu } ( u ) } \subseteq f ^ { \nu } ( k ^ { \prime } ).$$
Then clearly ϕ k,k ′ η,ν is a structured tree embedding, since ϕ η,ν , and each of the ϕ u ( u ∈ A η ) are embeddings. Therefore C is infinitely extensive.
Now suppose that ˆ = k 〈 ˆ q u : u ∈ A η 〉 is such that q u /lessorequalslant ˆ q u for all u ∈ A η , with some witnessing embedding µ u . Suppose also that ˆ q u /lessorequalslant q ′ ϕ η,ν ( ) for each u ∈ A η , with ψ u a witnessing embedding, and that ϕ u = ψ u ◦ µ u . Then for a ∈ f η ( k ), with a ∈ q u , we have
$$\varphi _ { \eta, \nu } ^ { k, k ^ { \prime } } ( a ) = \varphi _ { u } ( a ) = \psi _ { u } \circ \mu _ { u } ( a ) \in q ^ { \prime } _ { \varphi _ { \eta, \nu } ( u ) }$$
and µ k,k ′ η,ν ( a ) = µ u ( a ) ∈ ˆ , so that q u
$$\psi _ { \eta, \nu } ^ { \hat { k }, k ^ { \prime } } \circ \mu _ { \eta, \eta } ^ { k, \hat { k } } ( a ) = \psi _ { u } \circ \mu _ { u } ( a ) \in q ^ { \prime } _ { \varphi _ { \eta, \nu } ( u ) }.$$
It remains to check when q u = ∅ and a is the point corresponding to u . In this case ϕ k,k ′ η,ν ( a ) is the point b ∈ f ν ( k ′ ) corresponding to ϕ η,ν ( u ). Now µ k,k ˆ η,η ( a ) is the point of f η ( ˆ ) k corresponding
to u , so that ψ ˆ k,k ′ η,ν ◦ µ k,k ˆ η,η ( a ) = b . Thus we have verified the conditions of Definition 3.18, and therefore C is extendible. /square
## Theorem 6.11. C has nice limits.
Proof. First it is easily verified that C has limits. Let x, y ∈ C ˜ be such that x is the limit of ( x n n ) ∈ ω and for any n ∈ ω we have embeddings ϕ n : x n → y such that ϕ n = ϕ n +1 ◦ µ n . We want to show that x /lessorequalslant y .
We have by Remark 3.28 that µ n was the embedding given by the infinite extensiveness of F , using the construction of each x n . By the same argument as Remark 6.3; the underlying set of x n is a ↓ -closed subset of the underlying set of x n +1 where we consider µ n : x n → x n +1 acting as the identity on x n . Now we have that ϕ n = ϕ n +1 ◦ µ n is equivalent to saying that when a ∈ x n we have ϕ n +1 ( a ) = ϕ n ( a ). Hence it is possible to define ϕ : x → y as the union of all of the ϕ n . We claim that ϕ is an embedding.
Let a, b ∈ x = ⋃ n ∈ ω x n , and let n be least such that a, b ∈ x n . Let l x a be the a -label of x , and l x n +1 a be the a -label of x n +1 . If b > a we define θ : range( l x a ) → range( l ϕ a ( ) ) such that if l x a ( b ) = u then θ u ( ) = l ϕ a ( ) ( ϕ b ( )) (these are labels from x ). In order to show that x /lessorequalslant y we verify the following properties of ϕ (see Definition 4.7).
- (1) a /lessorequalslant b iff ϕ n ( a ) /lessorequalslant ϕ n ( b ) iff ϕ a ( ) /lessorequalslant ϕ b ( ) (since ϕ n is an embedding).
- (2) ϕ a ( ∧ b ) = ϕ a ( ) ∧ ϕ b ( ). Note that each x n is ↓ -closed in x . So that ϕ n ( a ∧ b ) is defined, and hence since ϕ n is an embedding, we have
$$\varphi ( a \wedge b ) = \varphi _ { n } ( a \wedge b ) = \varphi _ { n } ( a ) \wedge \varphi _ { n } ( b ) = \varphi ( a ) \wedge \varphi ( b ).$$
- (3) If a /lessorequalslant b then θ is an embedding. By property (5) of Definition 3.27, we have range( l x a ) = range( l x n +1 a ). So θ is an A -morphism since it is a map induced by the embedding ϕ n +1 .
- (4) c ( a ) /lessorequalslant c ( ϕ a ( )). By property (4) of Definition 3.27, we have
$$\mathbf c _ { x } ( a ) = \mathbf c _ { x _ { n } } ( a ) \leqslant \mathbf c _ { y } ( \varphi _ { n } ( a ) ) = \mathbf c _ { y } ( \varphi ( a ) ).$$
Hence ϕ is an embedding and x /lessorequalslant y as required.
/square
Theorem 6.12. Suppose that L and P are well-behaved, then T L P is well-behaved. 14
Proof. By Theorem 6.7 we have that ˜ = C T L P ( Q ′ ). Suppose we have a bad T L P ( Q ′ )-array f then define g X ( ) to be a decomposition tree for f ( X ). First we note that by Definition 4.12, it is clear that the leaves of the tree g X ( ) are coloured by some element of B used to construct f ( X ). Thus the colours of the tree f ( X ) are colours of colours of elements of g X ( ). Now by Theorem 5.8 we have that g is a bad array with range in T L A ( F ∪ B ).
We know that L = ω and A has injective morphisms. By Lemma 6.8 we know that A is well-behaved; hence by Theorem 4.11, there is a witnessing bad F ∪ B -array g ′ . Using Lemma 6.8 we know that F is bqo, hence we can restrict g ′ using Theorem 2.8, to find a witnessing bad array f ′ to B . Since the colours from each g X ( ) that were in B were points of f ( X ) we now know that f ′ is a witnessing bad array for f . Then since B = Q 1 ∪ {∅} we can restrict again to find a witnessing bad array to Q 1 , which clearly gives a witnessing bad array to Q just passing to colours. Hence any bad array admits a witnessing bad array, i.e. T L P is well-behaved. /square
14 This result was obtained independently by Christian Delhomme in as yet unpublished work. The author thanks him for his private communication.
## 7. Constructing partial orders
In this section we aim to show that some large classes of partial orders are well-behaved. Theorems 5.12 and 6.12 together, essentially already give us that some classes of partial orders are well-behaved. The challenge now is to characterise the partial orders that we have constructed.
Our method is similar to Thomass´ e's in [1], but expanded to make use of the two generalisations of section 3. This allows us to generalise the bqo class of partial orders not only to allow embeddings of the N partial order, but also into the transfinite. In order to prove his main result of [1], Thomass´ e used a structured tree theorem: that T C C 2 (A ) preserves bqo. 3 We will use what is essentially the same method, but with the more general trees T L P for general wellbehaved classes of partial orders P and linear orders L . So that, in particular when L = M and P = { 1 A , 2 , C 2 } we obtain a transfinite version of Thomass´'s result. e We will prove the general theorem (Theorem 7.37) in this section, and explore applications of this theorem in Section 8.
- 7.1. Intervals and indecomposable partial orders. We will now borrow some definitions from [1]. We want to define the indecomposable partial orders which will serve as building blocks for larger partial orders, in order to do so we first require the notion of an interval.
Definition 7.1. Suppose that a, b, c ∈ x ∈ Q . We say that a shares the same relationship to b and c , and write SSR( a b, c ; ) iff for all R ∈ { <,>, ⊥} we have aRb iff aRc.
/negationslash
Definition 7.2. Let P be a partial order and I ⊆ P , then we call I = ∅ an interval of P if ∀ x, y ∈ I and ∀ p ∈ P \ I we have SSR( p x, y ; ) .
Definition 7.3. Let P be a partial order. Then P is called indecomposable if every interval of P is either P itself, or a singleton. P is called decomposable if it is not indecomposable.
Proposition 7.4. Let η be a composition sequence of length r and b 0 , b 1 , b 2 ∈ H η , such that for each i ∈ { 0 1 2 , , } we have b i ∈ a η j i for j i ∈ r then
$$j _ { 0 } < j _ { 1 }, j _ { 2 } \longrightarrow \text{SSR} ( b _ { 0 } ; b _ { 1 }, b _ { 2 } ).$$
Proof. This is clear by the definition of H η
$$\text{ion of } H _ { \eta }. & & \Box$$
Lemma 7.5. Let 〈 I j : j ∈ r 〉 be a chain of intervals of a partial order P under ⊇ . Then ⋃ j ∈ r I j and ⋂ j ∈ r I j are intervals.
Proof. Let a ∈ P \ ⋃ j ∈ r I j and b, c ∈ ⋃ j ∈ r I j . Then b, c ∈ I i for some i ∈ r , we know that I i is an interval hence SSR( a b, c ; ) as required. The case of intersection is similar. /square
Definition 7.6. Let P be a partial order and I ⊆ P be an interval. We define the new partial order P/I to be the partial order obtained from P by removing all but one point of I . This is well-defined since I is an interval. If I is a set of disjoint intervals of P , then let P/ I be the partial order obtained from P by removing all but one point of each element of I .
Definition 7.7. Let η = 〈〈 f , s i i 〉 : i ∈ r 〉 be a composition sequence, k ∈ dom( f η ) and x ∈ Q α . We call 〈 η, k 〉 maximal for x iff for each i ∈ r we have a ( f i ) is indecomposable and there is a maximal chain 〈 I j : j ∈ r 〉 of intervals of x under ⊇ such that for some
$$\colon a ^ { \eta } \vee \varepsilon \, \text{dom} \, f ^ { \eta } \vee \varepsilon$$
we have x = f ( k ), and if η = 〈〈 f , s 〉 : i /greaterorequalslant j 〉 then we have for all j ∈ r ,
$$\cdot \, \check { \ } k = \langle q _ { i, u } \, \colon i \in r, u \in a _ { i } ^ { \eta } \rangle \in \text{dom} ( f ^ { \eta } ) _ { < \alpha }, \\ \intertext { a v e x } \, x = f ^ { \eta } ( k ), \, \text{and if } \eta _ { j } = \langle \langle f _ { i }, s _ { i } \rangle \, \colon i \geqslant j \rangle \, \text{then we have for all } j \in r, \\ I _ { j } = f ^ { \eta _ { j } } ( \langle q _ { i, u } \, \colon i \geqslant j, u \in a _ { i } ^ { \eta } \rangle ).$$
.
Figure 10. The partial order 2 <ω ⊥ .
Lemma 7.8. Let η be a composition sequence of length r , and k ∈ dom ( f η ) be such that 〈 η, k 〉 is maximal for x = f η ( k ) . Then any r ′ ⊆ r has an infimum and a supremum in r .
Proof. Let η = 〈〈 f , s i i 〉 : i ∈ r 〉 , η j = 〈〈 f , s i i 〉 : i /greaterorequalslant j 〉 , k = 〈 q i,u : i ∈ r, u ∈ a η i 〉 and I j = f η j ( 〈 q i,u : i /greaterorequalslant j, u ∈ a η i 〉 ). Consider I = i ∈ r ′ I i , this is an interval by Lemma 7.5.
- ⋂ 7.2. Generalised σ -scattered partial orders. We will now begin to characterise the partial orders that we were constructing earlier. For the rest of this section we let:
⋃ For any j ∈ r we have that I j is comparable with I under ⊆ , because if j > i ′ for some i ′ ∈ r ′ then I ⊇ I i ′ ⊇ I j , and if j < i ′ for all i ′ ∈ r ′ then for all such i ′ we have I j ⊇ I i ′ and this implies I j ⊇ I . Thus since η was maximal, we have that 〈 I i : i ∈ r 〉 is a maximal chain and thus I = I j 0 for some j 0 ∈ r . Clearly then j 0 is a greatest lower bound of r ′ , i.e. r ′ has an infimum in r . For the supremum, consider i ∈ r ′ I i , then the argument is similar. /square
- (1) Q be an arbitrary quasi-order;
- (2) Q ′ be Q with an added minimal element -∞ ;
- (3) P be a class of non-empty partial orders that is closed under non-empty subsets;
- (4) L be a class of linear orders that is closed under subsets;
- (5) C = C Q L P , = 〈Q B A F L〉 , , , , as in Example 3.2.
Definition 7.9. We let -2 <ω be the partial order obtained by reversing the order on 2 <ω . We also define the partial order 2 <ω ⊥ as follows. Elements of 2 <ω ⊥ are finite sequences of elements of 2, for s, t ∈ 2 <ω ⊥ , we define s < t iff there are some sequences u, s , t ′ ′ such that s = u /slurabove 〈 0 〉 /slurabove s ′ and t = u /slurabove 〈 1 〉 /slurabove t ′ . (See Figure 10.)
The following definition of S L P will characterise the orders of ˜ C 0 (see Theorem 7.35). We think of these as a generalisation of scattered linear orders S . We will think of M L P as a generalisation of σ -scattered orders M . Our aim is to show that ˜ C 0 = S L P and ˜ = C M L P .
Definition 7.10. We define S L P to be the class of non-empty partial orders X with the following properties.
- (i) If Y /lessorequalslant X is indecomposable, then Y ∈ P .
- (ii) Every chain of intervals of X with respect to ⊇ has order type in L .
- (iii) 2 <ω , -2 <ω and 2 <ω ⊥ do not embed into X .
We let P L P be the class of those non-empty X satisfying (i) and (ii). We call a sequence ( x n n ) ∈ ω increasing if for each n ∈ ω , we have that x n ∈ S L P and x n +1 is an x n -sum of some partial orders from S L P (so we consider x n ⊆ x n +1 similarly to Remark 3.30). We let M L P be the class containing all of S L P and unions of increasing sequences.
Remark 7.11 . For any limiting sequence ( x n n ) ∈ ω , let y n be the underlying set of x n for each n ∈ ω . Then ( y n n ) ∈ ω is an increasing sequence. Furthermore, for any union y of an increasing
sequence, we could construct a limiting sequence ( x n n ) ∈ ω with a limit x whose underlying set is y and has any Q ′ -colouring that we desire.
In order to prove that ˜ = C M L P , we first require several lemmas. Firstly given an x ∈ C ˜ , we now want to fine tune the construction of a decomposition tree for x by making more explicit the choice of η used to construct a decomposition function. (See Lemma 3.25 and Definition 4.12.) We want to only take η that are maximal in the sense of Definition 7.7.
Lemma 7.12. Let x ∈ Q α ∩ P L P ( Q ) . Then there is a composition sequence η and k ∈ dom ( f η ) , such that 〈 η, k 〉 is maximal for x .
Proof. Let x ∈ Q α so by Proposition 3.24, there is some η and k = 〈 q u : u ∈ A η 〉 ∈ dom( f η ) <α such that x = f η ( k ) = ∑ u ∈ H η q u . We will define a maximal chain 〈 I ′ i : i ∈ r ′ 〉 of intervals of x , a composition sequence η ′ and some k ′ ∈ dom( f η ′ ), so that 〈 η , k ′ ′ 〉 is maximal for x .
Let r be the length of η , then for j ∈ r define
$$I _ { j } = \{ d \in x \ | \ d \in q _ { u }, u \in a _ { i } ^ { \eta }, i \geqslant j \}.$$
By Proposition 7.4 and since x is an H η -sum, we have for all j ∈ r that I j is an interval of x ; and for i < j we have I i ⊃ I j . So 〈 I j : j ∈ r 〉 is a chain of intervals under ⊇ . Consider a maximal chain 〈 I ′ i : i ∈ r ′ 〉 of intervals under ⊇ that includes all of the I j . For i ∈ r we define the set
$$D _ { i } = \sum _ { u \in a _ { i } ^ { \eta } } q _ { u } = I _ { i } \, \sum \{ I _ { i ^ { \prime } } \ | \ i ^ { \prime } < i \}$$
and for j ∈ r ′ we define the set
$$D _ { j } ^ { \prime } = I _ { j } ^ { \prime } \, \bigcup \{ I _ { j ^ { \prime } } ^ { \prime } \, | \, j ^ { \prime } < j \}. \\ \text{moreover since each of the $I_{i}$ were} \\ \text{more over $i$} \,. \,. \,. \,. \,. \,. \,. \,. \,.$$
Then D ′ j ⊆ x = ⋃ i ∈ r D i , and moreover since each of the I i were equal to some I ′ j it must be that each D i is a union of some of the D ′ j . In particular, for all j ∈ r ′ there is some i ∈ r such that D ′ j ⊆ D i . Therefore, for some b j ⊆ a η i and some q ′ u ⊆ q u for u ∈ b j , we can write
$$D ^ { \prime } _ { j } & = \sum _ { u \in b _ { j } } q ^ { \prime } _ { u }. \\ - \frac { 1 } { 2 } \sim \frac { 1 } { 2 } \sim \dots \dots \dots.$$
/negationslash
For j ∈ r ′ , if j = max( r ′ ) we put b ′ j = b j , otherwise we can set b ′ j = b j /unionsq { s ′ j } . We define the order on b ′ j by inheriting from b j , and for u ∈ b ′ j with u = s ′ j we put
- · u < s ′ j iff there is some w ∈ x \ ⋃ j ′ /lessorequalslant j D ′ j ′ and v ∈ q ′ w such that v < w and · u > s ′ j iff there is some w ∈ x \ ⋃ j ′ /lessorequalslant j D ′ j ′ and v ∈ q ′ w such that v > w . This order is well-defined since 〈 I ′ i : i ∈ r ′ 〉 was a chain of intervals.
$$D ^ { \prime } _ { j } = \sum _ { u \in a _ { j } \langle \{ s ^ { \prime } _ { j } \} } y _ { u }. \\ \omega \quad \cdot \, - \, \omega \quad \omega \quad.$$
Let Z = { Z γ : γ ∈ κ } be the set of all non-singleton unions of maximal chains of intervals of b ′ j that do not contain s ′ j ; so by Lemma 7.5 for each γ ∈ κ we have Z γ is an interval. Now let a j = b / ′ j Z , letting e γ ∈ a j be the point remaining from Z γ . So the only non-singleton intervals of a j contain s ′ j . For each u ∈ a j \ ( { e γ | γ < κ } ∪ { s ′ j } ) define y u = q ′ u and for each γ ∈ κ define y e γ = ∑ u ∈ Z γ q ′ u . Thus we can write
We set f ′ j = ∑ a j and η ′ = 〈〈 f , s ′ j ′ j 〉 : j ∈ r ′ 〉 . For each i ∈ r , D i was a union of some D ′ j , so since 〈 I ′ i : i ∈ r ′ 〉 was a maximal chain of intervals, we can see that x = ⋃ j ∈ r ′ D ′ j . We chose η ′ precisely so that H η ′ consisted of copies of the a j arranged in the same order as the D ′ j were arranged; we also know that D ′ j was a sum of the y u and therefore x is an H η ′ sum of the y u .
In other words x = f η ′ ( 〈 y u : u ∈ A η ′ 〉 ). Since for each u ∈ A η we have q ′ u ⊆ q u ∈ Q <α , we know that q ′ u ∈ Q <α (by applying the same construction to q ′ u as to q u , taking smaller sums where necessary we see that q ′ u must have rank < α ). Let η j = 〈〈 f , s ′ i ′ i 〉 : i /greaterorequalslant j 〉 then we have I ′ j = f η j ( 〈 q ′ u : u ∈ a , i η i /greaterorequalslant j 〉 ), since the I ′ j were just the corresponding portion of the H η ′ sum.
Now we set k ′ = 〈 y u : u ∈ A η ′ 〉 ∈ dom( f η ′ ) and thus in order to show that 〈 η , k ′ ′ 〉 is maximal, it remains only to show that a i is indecomposable for each i ∈ r ′ . So suppose for contradiction that there is an i ∈ r ′ with a i decomposable, so we can let Z be an interval of a i with 1 < Z < a | | | i | . Thus we have from before that s ′ i ∈ Z . Now set
$$J = \left ( \sum _ { u \in Z \vee \{ s _ { i } ^ { \prime } \} } y _ { u } \right ) \cup \bigcup _ { j > i } I _ { j } ^ { \prime }. \\ \text{id we are taking the sum, it is simple}.$$
Since Z is an interval of a i and we are taking the sum, it is simple to verify that J is an interval of x . Moreover, J is a strict subset of I ′ i = ( ∑ u ∈ a i \{ s ′ i } y u ) ∪ ⋃ j>i I ′ j which strictly contains ⋃ j>i I ′ j . But then existence of such a J contradicts that 〈 I ′ j : j ∈ r 〉 was a maximal chain of intervals. Thus no such Z exists and each a i is indecomposable. This completes the proof. /square
Remark 7.13 . When we constructed decomposition functions (Lemma 3.25), the only condition on the composition sequence η that we used at each stage is that when rank( x ) = α we have x = f η ( k ) for some k ∈ dom( f η ) <α . Hence by Lemma 7.12, we can always assume without loss of generality that at every stage of the construction of a decomposition tree for x (Definition 4.12) we chose η with a corresponding maximal chain of intervals so that η was maximal. Doing so makes no difference to the results of Section 5, since the choice of η was arbitrary so long as η satisfies the condition above. For the rest of this section, we always assume that any decomposition tree was constructed by choosing such maximal composition sequences.
Lemma 7.14. If x ∈ C ∩ ˜ P L P ( Q ) has a decomposition tree T , then for every non-leaf t ∈ T , range ( l t ) is indecomposable.
Proof. Let t ∈ T x then either t is a leaf or t = /vector p /slurabove 〈 〉 i for some /vector p and i . If η /vector ( p ) = 〈〈 f /vector p i , s /vector p i 〉 : i ∈ r /vector ( p ) 〉 then range( l t ) = a ( f /vector p i ). Following Remark 7.13, we have assumed that η /vector ( p ) is maximal, and therefore a ( f /vector p i ) is indecomposable, which gives the lemma. /square
- 7.3. Pathological decomposition trees. We now continue with lemmas that mainly address which types of embeddings of 2 <ω can appear within decomposition trees, and what affect such embeddings will have on the partial order. First we have a lemma that tells us that we can always find a copy of 2 <ω in decomposition trees for elements of ˜ C ∞ .
Lemma 7.15. Let x ∈ P L P ( Q ) with x / ∈ C ˜ 0 , then any decomposition tree for x is not scattered.
Proof. We will prove this by induction on the scattered rank of possible decomposition trees for x . Since the only non-empty scattered tree of rank 0 is a single point, clearly x has no scattered decomposition tree of rank 0. Suppose for any y ∈ P L P ( Q ) \C ˜ 0 that y has no decomposition tree T ′ with rank U ( T ′ ) < α and that x / ∈ C ˜ 0 has a scattered decomposition tree T with rank U ( T ) = α . Thus there is a chain ζ of T such that T is a ζ -tree-sum of the lower ranked trees p /arrowtailright t ( t ∈ ζ, p ∈ a ζ t ).
Then x = f η ζ ( ) ( k ζ ) by Lemma 4.17. For each t ∈ ζ and p ∈ a ζ t we have a decomposition tree T x t,p ( ) = p /arrowtailright t for x t, p ( ), with rank U ( T x t,p ( ) ) < α . Thus by the induction hypothesis it must be that x t, p ( ) ∈ C ˜ . 0 But then since x = f η ζ ( ) ( k ζ ) we see that x is f η ζ ( ) applied to elements of ˜ C 0 hence it must be that x ∈ C ˜ . 0 This is a contradiction, therefore x has no scattered decomposition tree of rank α . This completes the induction and gives the lemma. /square
Figure 11. The structured trees B + , B -, C Q , and A .
Lemma 7.16. Let x ∈ C ˜ , and T be a decomposition tree for x . If y ⊆ x is indecomposable and | y | > 1 , then there exists t ∈ T such that the underlying partial order of y embeds into range ( l t ) .
Proof. First we suppose x ∈ C ˜ . 0 If rank( x ) = 0 then x is a single point, hence | y | /lessorequalslant | x | = 1 which gives the base case. Suppose now that for some α ∈ On we have the lemma for all x 0 ∈ Q <α and that some x ∈ Q α has an indecomposable subset y ⊂ x with | y | > 1. Then we know that x = f η ( k ) = ∑ u ∈ H η q u for some k = 〈 q u : u ∈ A η 〉 ∈ dom( f η ) <α . If any of these q u contains a subset isomorphic to a copy of y ; then since a decomposition tree for x is a ζ -tree-sum of decomposition trees for the q u (for some chain ζ ), by the induction hypothesis we are done.
So suppose for every u ∈ A η , q u does not have a subset isomorphic to y . We claim that if η = 〈〈 f , s i i 〉 : i ∈ r 〉 then for some i ∈ r , and some partial order P with y /lessorequalslant P , we have f i = ∑ P . If for some i ∈ r at least two points of y were in ∑ u ∈ a η i q u and another point of y was in ∑ u ∈ a η j q u for j < i , then the interval
/negationslash
$$y \cap \bigcup _ { j ^ { \prime } > j } \sum _ { u \in a _ { i } ^ { \eta } } q _ { u } & \subset y \\ \intertext { w o c a n l o t i b o l a e t e n c h }$$
shows that y is decomposable. So we can let i be least such that y ∩ a η i = ∅ , and we know that there is at most one point v ∈ y contained inside some q u for u ∈ a η j with j > i ; in this case we let ϕ v ( ) = s i . If two points of y were in some q u ⊂ x then there is another point of y not in here, so that the interval q u ∩ y ⊂ y shows that y is decomposable. Thus all w ∈ y \ { v } are inside ∑ u ∈ a η i q u , with at most one point in each q u . So we let ϕ w ( ) = u whenever w ∈ q u . Thus we have defined ϕ : y → a ( f i ) and it is simple to verify that ϕ is an embedding. Now without loss of generality we have η = η ( 〈〉 ) so that a ( f i ) = range( l 〈 〉 i ). This completes the proof for x ∈ C ˜ . 0
Suppose that x ∈ C ˜ ∞ is the limit of ( x n n ) ∈ ω . Then we claim that for some n ∈ ω there is some y ′ ⊆ x n such that the underlying orders of y ′ and y are isomorphic. Let n be least such that | y ∩ x n | > 1. If y is not a subset of x n then for some m>n , some points of x n were replaced by larger partial orders; equating the original point to a point of this partial order as in Remark 3.30. If at least two points of y were in a partial order that replaced a single point, then these points form a proper interval of y which contradicts that y is indecomposable. So let y ′ consist of those points of x n that are either in y or are replaced by a partial order containing a point of y in some x m ( m > n ). Thus the underlying orders of y ′ and y are isomorphic. Hence for some T n a decomposition tree for x n we have t ∈ T n and the underlying partial order of y ′ = y embeds into range( l t ). Hence also t ∈ T . This completes the proof. /square
We next define the { C 2 , A 2 } -structured, { ∑ C 2 , ∑ A 2 } -coloured trees B + , B -, C , Q and A as in Figure 11. These will be pathological decomposition trees for the partial orders 2 <ω , -2 <ω , 2 <ω ⊥ , Q and A ℵ 0 respectively (we deal with these in 7.4). It will turn out that Q and A can never be decomposition trees under the assumption granted by Remark 7.13.
Definition 7.17. B + has underlying set consisting of all finite sequences s = 〈 s i : i /lessorequalslant n 〉 of elements of { 0 1 2 3 , , , } such that:
/negationslash
- · if s = 〈〉 then s 0 ∈ { 2 3 , } ,
- · if s i ∈ { 0 1 , } and i < n then s i +1 ∈ { 2 3 , } ,
- · if s i = 2 and i < n then s i +1 ∈ { 0 1 , } ,
- · if s i = 3 then i = n .
Thus B + is a tree under /subsetsqequal . If n is even, then we set c ( s ) = C 2 and label so that
$$\text{ree under $\sqsubset$}. \text{ If $n$ is even, then we set $c(s)=\sum_{C_{2}}$ and label so that} \\ l _ { s } ( \widehat { s } \langle 3 \rangle ) = \min ( C _ { 2 } ) \text{ and } l _ { s } ( s \langle 2 \rangle \widehat { s } t ) = \max ( C _ { 2 } ) \\ \mu _ { s } \quad - \mu _ { s }.$$
for every possible sequence t . If n is odd, the we set c ( s ) = ∑ A 2 and range( l s ) = A . We define 2 the tree B -in the same way as B + , with the only difference that
$$l _ { s } ( \widehat { s } \langle 3 \rangle ) = \max ( C _ { 2 } ) \text{ and } l _ { s } ( \widehat { s } \langle 2 \rangle \widehat { s } t ) = \min ( C _ { 2 } )$$
for every possible sequence t . We also define the tree C in the same way as B + , but change the labels and colours as follows. If n is even, then we set c ( s ) = ∑ A 2 and label so that range( l s ) = A . If 2 n is odd then we set c ( s ) = C 2 and for every possible sequence t
$$\dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \delta _ { 2 } \dots \dots \dots \dots \dots \delta _ { 3 } \dots \dots \dots \dots \delta _ { 4 } \dots \dots \dots \delta _ { 5 } \dots \dots \dots \delta _ { 6 } \dots \dots \dots \delta _ { 7 } \dots \dots \dots \delta _ { 8 } \dots \dots \dots \delta _ { 9 } \dots \dots \dots \delta _ { 9 } \dots \dots \dots \delta _ { 9 } \dots \dots \delta _ { 9 } \dots \dots \delta _ { 9 } \dots \dots \delta _ { 9 } \dots \dots \delta _ { 9 } \dots \dots \delta _ { 9 } \dots \dots \delta _ { 9 } \dots \dots \delta _ { 9 } \dots \dots \delta _ { 9 } \dots \dots \delta _ { 9 } \dots \delta _ { 9 } \dots \dots \delta _ { 9 } \dots \delta _ { 9 } \dots \delta _ { 9 } \dots \delta _ { 9 } \dots \delta _ { 9 } \dots \delta _ { 9 } \dots \delta _ { 9 } \dots \delta _ { 9 } \dots \delta _ { 9 } \dots \delta _ { 9 } \dots \delta _ { 9 } \dots \delta _ { 9 } \dots \delta _ { 9 } \dots \delta _ { 9 } \dots \delta _ { 9 } \dots \delta _ { 9 } \dots \delta _ { 9 } \dots \delta _ { 9 } \dots \delta _ { 9 } \dots \delta _ { 9 } \dots \delta _ { 9 - 1 } \dots \delta _ { 9 - 2 } \dots \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \ dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta_ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 3 } \dots \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots \delta _ { 9 - 2 } \dots.$$
We now define Q as a copy of 2 <ω , coloured and labelled so that for each t ∈ Q we have c ( ) = t ∑ C 2 and range( l t ) = C . 2 Finally we define A as a copy of 2 <ω , coloured and labelled so that for each t ∈ A we have c ( ) = t ∑ A 2 and range( l t ) = A . 2
Lemma 7.18. Let x ∈ C ˜ have a decomposition tree T . Then if B + , B -/lessorequalslant T and A /lessorequalslant T , then there is a ↑ -closed subset A ⊆ T such that A /lessorequalslant A and for each t ∈ A , we have range ( l t ) = A 2 .
/negationslash
/negationslash
Proof. Suppose B + , B -/lessorequalslant T and A /lessorequalslant T , so let ϕ : A → T be a witnessing embedding. For each t = ϕ s ( ) ∈ im( ϕ ), we have that P t = range( l t ) is some partial order that embeds A . 2 Let a t = l t ( ϕ s ( /slurabove 〈 0 )) and 〉 b t = l t ( ϕ s ( /slurabove 〈 1 )). 〉 Hence a , b t t ∈ P t with a t ⊥ b t . Now for each t , either P t = { a , b t t } or there is some c t ∈ P t such that without loss of generality (swapping the names of a t and b t if necessary) one of the following cases occurs:
- (1) a , b t t ⊥ c t ,
- (2) a t < c t ,
- (3) a t > c t .
/negationslash
$$\text{If } s = s ^ { \prime } \cap \langle 2 \rangle \, \text{ we let } \delta _ { 0 } \text{ and } \delta _ { 1 } \text{ be elements of }$$
Suppose that there is some u 0 ∈ im( ϕ ) such that for every u ∈ im( ϕ ) ∩↑ u 0 there is t ( u ) /greaterorequalslant u and some c t ( u ) satisfying case 2, (i.e. a t ( u ) < c t ( u ) ). Set τ ( 〈〉 ) = t ( u 0 ). Suppose we have defined τ ( s ) for some s ∈ B -of length n . To simplify notation, we let s ′ be s with its last element removed and π = τ ( s ). Now suppose further that whenever s = s ′ /slurabove 〈 3 , we have we have 〉 τ ( s ) ∈ im( ϕ ); and whenever s = 〈〉 , or s = s ′ /slurabove 〈 〉 i for i ∈ { 0 1 , } we have that τ ( s ) satisfies case 2.
$$\lim ( \varphi ) \cap ^ { a _ { \pi } } \updownarrow \pi \text{ and } \, \lim ( \varphi ) \cap ^ { b _ { \pi } } \updownarrow \pi$$
respectively, and set τ ( s /slurabove 〈 0 ) = 〉 t ( δ 0 ) and τ ( s /slurabove 〈 1 ) = 〉 t ( δ 1 ). These exist and are incomparable since A was a copy of 2 <ω . If s = 〈〉 or s = s ′ /slurabove 〈 〉 i for i ∈ { 0 1 , } , then pick the values of τ ( s /slurabove 〈 2 ) 〉 and τ ( s /slurabove 〈 3 ) to be elements of 〉
$$\lim ( \varphi ) \cap ^ { a _ { \pi } } \uparrow \pi \text{ and } ^ { c _ { \pi } } \uparrow \pi$$
respectively. So by construction, the map τ : B -→ T is an embedding, which is a contradiction. Thus no such u 0 exists and there is u 1 ∈ im( ϕ ) such that no t > u 1 satisfies case 2.
/negationslash
Now using that B + /lessorequalslant T , we can apply a similar argument to the tree ↑ u 1 (in place of T ) and case 3 (in place of case 2). So we find a u 2 ∈ im( ϕ ) such that for every t > u 2 we have that t does not satisfy cases 2 or 3 for any choice of c t . Let A = ↑ u 2 , so that for each t ∈ A we have either P t = { a , b t t } or t satisfies case 1, for any choice of c t . Hence P t is an antichain, and therefore by
Lemma 7.14 we always have P t = { a , b t t } = A , since any indecomposable antichain is either 1 2 or A . It remains to check that 2 A /lessorequalslant A , but this is clear since A = ↑ u 2 , with u 2 ∈ im( ϕ ). /square
/negationslash
Lemma 7.19. Let T be a decomposition tree for x ∈ C ∩ ˜ P L P ( Q ) . If B + , B -/lessorequalslant T , then A /lessorequalslant T .
/negationslash
/negationslash
Proof. Suppose for contradiction that B + , B -/lessorequalslant T and A /lessorequalslant T . By Lemma 7.18 there is some ↑ -closed subset A ⊆ T such that for each t ∈ A we have range( l t ) = A . Consider 2 y = x t, p ( ) for arbitrary t ∈ A and p ∈ range( l t ). We claim that y is an antichain. Suppose not, then there is some chain y ′ ⊆ y of size 2. Hence by Lemma 7.16, since y ′ is an indecomposable subset of y with | y ′ | > 1, there must be some t ∈ T y ⊆ A such that y ′ embeds into P t = range( l t ). But this is a contradiction since each P t was an antichain, therefore indeed we have that y is an antichain.
Let /vector q be shortest such that there is t ′ = /vector q /slurabove 〈 i 0 〉 ∈ p /arrowtailright t for some i 0 , let w ∈ a η /vector ( q ) i 0 , then set z = x t , w ( ′ ). Then z has decomposition tree w /arrowtailright t ′ which is a ↑ -closed subset of A and hence | T z | > 1 and therefore | z | > 1. Let η /vector ( q ) = η = 〈〈 f , s i i 〉 : i ∈ r 〉 and for j ∈ r let η j = 〈〈 f , s i i 〉 : i /greaterorequalslant j 〉 . By Remark 7.13 can assume that there is a maximal chain of intervals 〈 I j : j ∈ r 〉 of y , and some k = 〈 q u : u ∈ A η 〉 ∈ dom( f η ) <α such that x = f η ( k ) and I j = f η j ( 〈 q u : u ∈ a , i η i /greaterorequalslant j 〉 ) for each j ∈ r .
$$\bigcup _ { i > i _ { 0 } } I _ { i } \subset I \subset I _ { i _ { 0 } }. \\ \text{i as an anti-chain } \text{Than}.$$
Thus we have that z = q w so that z ⊆ I i 0 = ∑ u ∈ H η i 0 q u , since w ∈ H η i 0 . For each i ∈ r we have f i = ∑ A 2 , therefore H η is an antichain of size | r | . Let I = I j \ { ρ } for some ρ ∈ z then since | z | > 1 we have
But I ⊆ y is also an interval, since y is an antichain. Therefore 〈 I j : j ∈ r 〉 was not a maximal chain of intervals of y . This contradiction gives the lemma. /square
/negationslash
Lemma 7.20. Let x ∈ C ˜ have a decomposition tree T . Then if C /lessorequalslant T and Q /lessorequalslant T , then there is a ↑ -closed subset A ⊆ T such that Q /lessorequalslant A and for each t ∈ A , we have range ( l t ) = C 2 .
Proof. Similar to Lemma 7.18, replacing ⊥ with ⊥ and < with ⊥
/negationslash
$$\text{ith $\perp$.}$$
/negationslash
Lemma 7.21. Let x ∈ C ∩ ˜ P L P ( Q ) have a decomposition tree T . Then if C /lessorequalslant T then Q /lessorequalslant T .
/negationslash
/negationslash
Proof. Suppose for contradiction that C /lessorequalslant T and Q /lessorequalslant T . By the reasoning of Remark 7.13 we can assume without loss of generality that each chain used to construct T was maximal. By Lemma 7.20 there is some ↑ -closed subset A ⊆ T such that for each t ∈ A we have range( l t ) = C . 2 We can now proceed as in Lemma 7.19, replacing the word chain with antichain, and vice versa. /square
Lemma 7.22. If x ∈ C ∩ ˜ P L P ( Q ) has a decomposition tree T such that B + /lessorequalslant T ( resp. B -, C ) , then 2 <ω /lessorequalslant x ( resp. -2 <ω , 2 <ω ⊥ ) .
Proof. Let ϕ : B + → T be an embedding. Let W be the subset of B + consisting of 〈〉 , u /slurabove 〈 0 〉 and u /slurabove 〈 1 〉 for all possible sequences u . For each s ∈ W , let t s = ϕ s ( ), and u s = l t s ( ϕ s /slurabove ( 〈 3 )). 〉 Then let y s be an element of
$$x ( t _ { s }, l _ { t _ { s } } ( u _ { s } ) ) \subseteq x.$$
/negationslash
We now define an embedding ψ : 2 <ω → x . Given a = 〈 a , a 0 1 , ..., a n -1 〉 ∈ 2 <ω , if a = 〈〉 then let a ′ = a , and if a = 〈〉 let a ′ = 2 〈 , a 0 , 2 , a 1 , ..., 2 , a n -1 〉 . Now set ψ a ( ) = y a ′ , so if a, b ∈ 2 <ω are such that a /lessorequalslant b then a /subsetsqequal b so that a ′ /subsetsqequal b ′ and thus t a ′ /lessorequalslant t b ′ and l t a ′ ( u a ′ ) /lessorequalslant l t b ′ ( u b ′ ) which means that ψ a ( ) /lessorequalslant ψ b ( ). If a ⊥ b then similarly t a ′ /lessorequalslant t b ′ and l t a ′ ( u a ′ ) ⊥ l t b ′ ( u b ′ ) so that ψ a ( ) ⊥ ψ b ( ). Thus ψ is an embedding and witnesses 2 <ω /lessorequalslant x . The cases for -2 <ω and 2 <ω ⊥ are similar. /square
## 7.4. Pathological partial orders.
Definition 7.23. Set M = { 2 <ω , -2 <ω , 2 <ω ⊥ } . We call elements of M pathological partial orders.
/negationslash
Lemma 7.24. Suppose we have x ∈ Q and for each i ∈ x , we have x i ∈ Q . Suppose that for y ∈ M and all z ∈ { x } ∪ { x i | i ∈ x } we have y /lessorequalslant z . Then y /lessorequalslant i ∈ x x . i
/negationslash
∑ Proof. Fix y ∈ M and suppose that y /lessorequalslant ∑ i ∈ x x i , with ϕ a witnessing embedding. For all t ∈ y , let i t be such that ϕ s ( ) ∈ x i t . Let F be a finite subset of x , i ∈ F , and s ∈ y .
/negationslash
/negationslash
We claim that there is some s ′ ∈ y with s /subsetsqequal s ′ such that for all z ∈ y with s ′ /subsetsqequal z we have i z = . So suppose for contradiction that for all i s ′ ∈ y with s /subsetsqequal s ′ there is some Z s ( ′ ) ∈ y with s ′ /subsetsqequal Z s ( ′ ) and i Z s ( ′ ) = . i We define ψ : y → x i inductively as follows. Let ψ ( 〈〉 ) = Z s ( ) and if for t ∈ y we have defined ψ t ( ) = ϕ t ( ′ ) then for m ∈ { 0 1 , } let ψ t ( /slurabove 〈 m 〉 ) = ϕ Z t ( ( ′ /slurabove 〈 m 〉 )) . It is easily verified then that ψ is an embedding, which is a contradiction and we have the claim. Applying the claim repeatedly for each i ∈ F , we then see that for all s ∈ y there is s F ∈ y with s /subsetsqequal s F such that for all i ∈ F and all z ∈ y with s F /subsetsqequal z we have i z = . i
Now let µ ( 〈〉 ) = ϕ ( 〈〉 ). Suppose inductively we have defined µ on some sequences t ∈ y so that µ t ( ) = ϕ t ( ′ ) for some t ′ , and let G be the set of these t such that µ t ( ) is already defined. Let u ∈ y be the lexicographically least element of y \ G and let v ∈ y and m ∈ { 0 1 , } be such that u = v /slurabove 〈 m 〉 ; so µ v ( ) is already defined. Now let u ′ = v ′ /slurabove 〈 m 〉 and let µ u ( ) = ϕ s ( G ). Now µ is an embedding and i µ t ( ) is distinct for distinct t . So let µ ′ : y → x be given by µ ′ ( ) = t i µ t ( ) , then µ ′ is an embedding, which is a contradiction. /square
Proposition 7.25. Let y ∈ M , and s, s , t ′ ∈ y be such that s /subsetsqequal s ′ and s and t are incomparable under /subsetsqequal . Then ¬ SSR( ; s s , t ′ ) and ¬ SSR( s ′ ; s, t ) .
Proof. Let y ∈ M and s, s , t ′ be as described. Suppose that y = 2 <ω , then since 2 <ω is just ordered by /subsetsqequal we have s /lessorequalslant s ′ and s ⊥ t hence ¬ SSR( ; s s , t ′ ). We also have that s ′ ⊥ t , and therefore ¬ SSR( s ′ ; s, t ). If y = -2 <ω then we have s /greaterorequalslant s ′ , s ⊥ t and s ′ ⊥ t , and again we can conclude that ¬ SSR( ; s s , t ′ ) and ¬ SSR( s ′ ; s, t ). If y = 2 <ω ⊥ then we have s ⊥ s ′ , and either t > s and t > s ′ or t < s and t < s ′ . Hence again we can conclude ¬ SSR( ; s s , t ′ ) and ¬ SSR( s ′ ; s, t ). /square
/negationslash
Lemma 7.26. Suppose that no element of M embeds into any element of P . Then for all y ∈ M , and for every composition sequence η we have y /lessorequalslant H η .
Proof. Let η = 〈〈 f , s i i 〉 : i ∈ r 〉 be a composition sequence, then for each i ∈ r , we have that a ( f i ) ∈ A . Since A = P we know that
/negationslash
$$( \forall i \in r ) ( \forall y \in \mathbb { M } ), y \ll \mathfrak { a } ( f _ { i } ).$$
/negationslash
/negationslash
/negationslash
Let a = ϕ ( 〈〉 ) and choose a 0 = ϕ s ( 0 ) ∈ a η i 0 and a 1 = ϕ s ( 1 ) ∈ a η i 1 as in the claim. Now choose a 00 = ϕ s ( 00 ) ∈ a η i 00 and a 11 = ϕ s ( 11 ) ∈ a η i 11 by applying the claim to a 0 and a 1 . Notice that by a similar argument to before, we can assume that every element of I = { i, i 0 , i 1 , i 00 , i 11 } is distinct, otherwise we would be able to embed y into a ( f j ) for some j ∈ I .
Suppose that for some y ∈ M , we have y /lessorequalslant H η with ϕ a witnessing embedding. We claim that for any a = ϕ s ( ) ∈ a η i , there are a 0 = ϕ s ( 0 ) ∈ a η i 0 and a 1 = ϕ s ( 1 ) ∈ a η i 1 with s /subsetsqequal s , s 0 1 , and s , s 0 1 incomparable under /subsetsqequal , with i = i 0 = i 1 = . Suppose not, then for some i s ∈ y and i ∈ r , we have ϕ s ( /slurabove t ) ⊆ a η i for every sequence t . This is a contradiction, since then y embeds into { t ′ ∈ y | s /subsetsqequal t ′ } /lessorequalslant a η i /lessorequalslant a ( f i ). This gives the claim.
We now use Proposition 7.4 in the following cases:
- · i 0 < i , i 1 00 which implies SSR( a 0 ; a , a 1 00 ),
- · i 00 < i , i 0 1 which implies SSR( a 00 ; a , a 0 1 ),
- · i 1 < i , i 0 00 and i 1 < i , i 0 11 which implies SSR( a 1 ; a , a 0 11 ),
- · i 1 < i , i 0 00 and i 11 < i , i 0 1 which implies SSR( a 11 ; a , a 0 1 ),
- · i 1 < i , i 0 00 and i 0 < i , i 1 11 which implies i 0 < i 1 < i 0 which is a contradiction.
Now note that any of the first four cases contradict Proposition 7.25. So we have a contradiction in every case, and our assumption that y /lessorequalslant H η must have been false. This gives the lemma. /square
/negationslash
Lemma 7.27. If x ∈ C ˜ 0 and for all y ∈ M , z ∈ P , we have y /lessorequalslant z , then for all y ′ ∈ M , y ′ /lessorequalslant x .
/negationslash
/negationslash
/negationslash
Proof. Clearly this holds when rank( x ) = 0, since then x is a single point and so does not embed any element of M . Suppose the statement holds for all x ′ ∈ Q <α for some α ∈ On, and that x ∈ Q α . Then x = f η ( k ) = ∑ u ∈ H η q u for some composition sequence η and k = 〈 q u : u ∈ A η 〉 ∈ dom( f η ) <α . By the induction hypothesis, for each u ∈ H η and y ∈ M we know that y /lessorequalslant q u , and by Lemma 7.26 we know that y /lessorequalslant H η . So by Lemma 7.24, we have that y /lessorequalslant x as required. /square
/negationslash
## 7.5. Characterising the construction.
## Lemma 7.28. Suppose that:
- · U is an indecomposable partial order with | U > | 2 ;
/negationslash
Then I = P .
- · P = ∑ u ∈ U P u for some non-empty partial orders P u ( u ∈ U ) ; · I ⊆ P is an interval of P with I ∩ P v 0 = ∅ and I ∩ P v 1 = ∅ for some v 0 = v 1 ∈ U .
## Proof. First we claim that
/negationslash
$$J = \{ u \, | \, I \cap P _ { u } \neq \emptyset \}$$
is an interval of U . To see this, we let v ∈ U \ J , u , u 0 1 ∈ J , a ∈ P v and b 0 ∈ I ∩ P u 0 , b 1 ∈ I ∩ P u 1 . Then a / I ∈ (otherwise v ∈ J ), so we have SSR( a b , b ; 0 1 ) because I was an interval. But this implies SSR( v u , u ; 0 1 ) since P = ∑ u ∈ U P u , hence we have the claim that J is an interval of U . Since U was indecomposable, we either have | J | = 1 or J = U . If | J | = 1 then this contradicts our assumption that I ∩ P v 0 = ∅ and I ∩ P v 1 = ∅ for some v 0 = v 1 ∈ U . Hence J = U .
/negationslash
/negationslash
/negationslash
Suppose for contradiction that there is some a ∈ P \ I , then a ∈ P v for some v ∈ U . For arbitrary u , u 0 1 ∈ U = I ′ with v / ∈ { u , u 0 1 } we have b 0 ∈ I ∩ P u 0 and b 1 ∈ I ∩ P u 1 . Since a ∈ P \ I , b 0 , b 1 ∈ I and I is an interval, we then know that SSR( a b , b ; 0 1 ). So since P = ∑ u ∈ U P u and v / ∈ { u , u 0 1 } we see that SSR( v u , u ; 0 1 ). But then since u 0 and u 1 were arbitrary, it must be that U \ { v } is an interval. Hence since U was indecomposable we have | U \ { v }| = 1 which means | U | = 2 which is a contradiction. /square
## Lemma 7.29. Suppose that:
- · U = { 0 1 , } with 0 R 1 for some R ∈ { <,>, ⊥} , so U ∈ { A 2 , C 2 } ;
/negationslash
/negationslash
- · there are no non-empty subsets K ,K 0 1 of P 0 with K 0 ∩ K 1 = ∅ and P 0 = K 0 ∪ K 1 , such that for all a 0 ∈ K 0 and a 1 ∈ K 1 we have a Ra 0 1 .
- · P = ∑ u ∈ U P u for some non-empty partial orders P , u ( u ∈ U ) ; · I ⊆ P is an interval of P with I ∩ P 0 = ∅ and I ∩ P 1 = ∅ ;
## Then P 0 ⊆ I .
/negationslash
/negationslash
Proof. Suppose that P 0 ⊆ I so we can let K 0 = I ∩ P 0 , K 1 = P 0 \ I . So there are a ∈ I ∩ P 0 and b ∈ P 0 \ I , such that a ⊥ b if R = ⊥ and a ⊥ b if R = ⊥ , so in either case we have ¬ ( bRa ). Let c ∈ P 1 ∩ I , then bRc since P = ∑ u ∈ U P u and b ∈ P 0 . So we have shown ¬ SSR( ; b a, c ). But a, c ∈ I and b / I ∈ with I an interval, hence we have SSR( ; b a, c ), which is a contradiction. /square
/negationslash
In the next definition and following few lemmas (7.30, 7.31, 7.32, 7.33 and 7.34) we fix a composition sequence η = 〈〈 f , s i i 〉 : i ∈ r 〉 and k ∈ dom( f η ) so that 〈 η, k 〉 is maximal for P ∈ C ˜ . 0
Definition 7.30. Let k = 〈 P i,u : i ∈ r, u ∈ a η i 〉 and define:
- · P i = ∑ u ∈ a η i P i,u ;
/negationslash
/negationslash
- · P /greaterorequalslant i = ⋃ j /greaterorequalslant u P j ⊆ f η ( k ); · P >i = P /greaterorequalslant i \ P i .
Lemma 7.31. Suppose that I is an interval of P and i ∈ r be such that I ∩ P i = ∅ and I ∩ P >i = ∅ . If | a ( f i ) | > 2 then P /greaterorequalslant i ⊆ I , and if | a ( f i ) | = 2 then P i ⊆ I .
/negationslash
/negationslash
/negationslash
If | a ( f i ) | = 2 then set R ∈ { <,>, ⊥} such that for a ∈ a η i we have aRs i . Then using Lemma 7.29, we either have either that P i ⊆ I or there are non-empty K ,K 0 1 ⊆ P i with K 0 ∩ K 1 = ∅ and P i = K 0 ∪ K 1 such that for all a 0 ∈ K 0 and a 1 ∈ K 1 we have a Ra 0 1 . Consider J = K 1 ∪ P >i , let a ∈ P \ J and z , z 0 1 ∈ J then either a ∈ P \ P /greaterorequalslant j in which case we have SSR( a z , z ; 0 1 ) or a ∈ K 0 in which case aRz 0 and aRz 1 , so that SSR( a z , z ; 0 1 ). Thus J is an interval. But we have that P >i ⊂ J ⊂ P /greaterorequalslant i , which contradicts that η was maximal. So it must be that P i ⊆ I . /square
Proof. We have that P = f η ( k ) is an H η -sum of the non-empty partial orders P i,a . By definition of H η we have that P /greaterorequalslant i is an a ( f i )-sum of the P i,u in position u ∈ a η i , and P >i in position s i . We know that a ( f i ) is indecomposable since 〈 η, k 〉 was maximal. So since I ∩ P i = ∅ , we can apply Lemma 7.28 to see that either | a ( f i ) | /lessorequalslant 2 or P /greaterorequalslant i ⊆ I .
Lemma 7.32. Let I be an interval of P , then there are j 0 , j 1 ∈ r and an X ⊆ P j 1 either empty or an interval of P j 1 such that I = P /greaterorequalslant j 0 or I = ( P /greaterorequalslant j 0 \ P /greaterorequalslant j 1 ) ∪ X.
Proof. If for any i ∈ r we have | a ( f i ) | = 1 then f i = ∑ 1 which makes no difference to the sum over H η , so without loss of generality we assume this does not occur.
Let j 0 = inf { i ∈ r | I ∩ P i = ∅} , and j 1 = sup { i ∈ r | I ∩ P i = ∅} , then j 0 , j 1 ∈ r by Lemma 7.8. Suppose that for some n ∈ r with j 0 /lessorequalslant n < j 1 we have | a ( f n ) | /greaterorequalslant 2. We claim that in this case I = P /greaterorequalslant i 0 . For each such n we have by Lemma 7.31 that P /greaterorequalslant n ⊆ I . Therefore for all m ∈ r with j 0 /lessorequalslant m<n we have again by Lemma 7.31 that P m ⊆ I , hence I = P /greaterorequalslant j 0 by definition of j 0 .
/negationslash
/negationslash
/negationslash
/negationslash
Suppose now that the only n ∈ r such that j 0 /lessorequalslant n /lessorequalslant j 1 with | a ( f n ) | > 2 is n = j 1 , or that no such n exists. Let X = I ∩ P j 1 , then X is either empty or an interval of P j 1 . If j 0 = j 1 we are done, otherwise for each j 0 /lessorequalslant m<j 1 we have | a ( f n ) | = 2. We either have P j 1 = ∅ or P j 1 = ∅ in which case by definition of j 1 there must be some m ′ ∈ r with m < m < j ′ 1 and P m ′ = ∅ . So in either case we can apply Lemma 7.31 to see that P m ⊆ I . Using the definition of j 0 we also have that I ⊆ P /greaterorequalslant j 0 . Therefore we have shown that I = ( P /greaterorequalslant j 0 \ P /greaterorequalslant j 1 ) ∪ X as required. /square
Lemma 7.33. Suppose that ∀ i ∈ r and ∀ a ∈ a η i , every chain of intervals of P i,a under ⊇ has order type in L . Then every chain of intervals of f η ( k ) under ⊇ has order type in L .
Proof. Let 〈 J α : α ∈ σ 〉 be a chain of intervals of P under ⊇ . If for some α ∈ σ we have some j α ∈ r with P /greaterorequalslant j α = J α then for every β /lessorequalslant α , by Lemma 7.32 and since J β ⊇ J α , there must be some j β with J β = P /greaterorequalslant j β . Hence the order type of the chain 〈 J β : β /lessorequalslant α 〉 must be a subset of the order type of { i ∈ r | j /lessorequalslant j α } , and hence this order type is in L . Therefore since L is closed under finite sums, it only remains to show that the final segment
/negationslash
$$F = \{ J _ { \gamma } \, | \, ( \forall i \in r ), J _ { \gamma } \neq P _ { \geqslant i } \}$$
has order type in L under ⊇ .
′ such that
$$\dots \dots \dots \dots \dots \dots = \cdot \\ \text{Let $J_{\gamma} \in F$, then by Lemma $7.32$, there are some $j_{\gamma}, j_{\gamma}^{\prime} \in r$ and $X$ an interval of $P_{j_{\gamma}}$ such that} \\ J_{\gamma} = ( P _ { \geqslant j_{\gamma } } \, \langle P _ { \geqslant j_{\gamma } ^ { \prime } } ) \cup X _ { \gamma } \\ \cdot \cdot \cdot$$
Now if γ < υ then j γ /lessorequalslant j υ , j ′ γ /greaterorequalslant j ′ υ and X γ ⊇ X υ , and if j ′ γ > j ′ υ then X υ ⊆ J γ . Let F δ = { J γ ∈ F | j γ = j δ } . Then ot( F ) is an ot( { j δ | δ ∈ σ } )-sum of the ot( F δ ). But { j δ | δ ∈ σ } ⊆ r hence this subset has order type in L . So it remains only to show that each ot( F δ ) is in L . We also have that each ot( F δ ) is an ot( { j ′ τ | τ ∈ σ, j τ = j δ } )-sum of the order types of F δ,τ = { X λ | λ ∈ σ, j λ = j δ , j ′ λ = j ′ τ } . But { j ′ τ | τ ∈ σ, j τ = j δ } ⊆ r and F δ,τ is a chain of intervals of P τ under ⊇ , whence both are in L and thus ot( F δ ) ∈ L which gives the lemma. /square
Lemma 7.34. Suppose that for any indecomposable partial order y and any i ∈ r and a ∈ a η i we have that y /lessorequalslant P i,a -→ y ∈ P . Then for any indecomposable y ′ we have y ′ /lessorequalslant f η ( k ) -→ y ′ ∈ P .
Proof. Since 〈 η, k 〉 was maximal for P we know that for each i ∈ r we have a ( f i ) is indecomposable. Now any subset A of H η = ⊔ i ∈ r a η i that contains points of a η i and at least two points of ⋃ j>i a η j is such that A ∩ ⋃ j>i a η j is an interval with at least two points inside. So here A cannot be indecomposable. Thus any indecomposable subset I of H η is a subset of a η i /unionsq a η j for some i, j ∈ r with i < j , moreover, | I ∩ a η j | /lessorequalslant 1. So I has the same order type as a subset of a ( f i ), which shows that I has order type in P .
Thus if we take a subset A ⊆ P with at least two points inside a single P i,a and at least one point not in P i,a then A ∩ P i,a is an interval which contradicts that A is indecomposable. We know that P is an H η -sum of the partial orders P i,a . So let J be an indecomposable subset J of the sum P . Then either J is entirely contained within some P i,a and hence J has order type in P ; or J contains at most one point of each of the P i,a that it intersects, and hence has the same order type as an indecomposable order of H η . Hence by the previous paragraph J has order type in P , which completes the proof. /square
We are now ready to prove the following generalisation of Hausdorff's famous theorem on scattered linear orders (Theorem 2.32).
## Theorem 7.35. S L P ( Q ′ ) = ˜ C 0 .
Proof. First we claim that ˜ C 0 ⊆ S L P ( Q ′ ). By Lemma 7.27, we have that ˜ C 0 satisfies (iii) of Definition 7.10. We will show (i) and (ii) by induction on the rank of x ∈ C ˜ . 0 If x has rank 0 then x is just a single point and so satisfies (i) and (ii) since both P and L contained 1. Now suppose that any y ∈ Q <α satisfies (i) and (ii). Then if x ∈ Q α , we have that x = f η ( k ) where k ∈ dom( f η ) <α . Hence by Lemma 7.34 we have that x satisfies (i), and by Lemma 7.33 we have that x satisfies (ii). So indeed we have ˜ C 0 ⊆ S L P ( Q ).
We will now show S L P ( Q ′ ) ⊆ C ˜ . 0 Suppose we have some non-empty Q ′ -coloured partial order x / ∈ C ˜ . 0 We claim that either (i), (ii) or (iii) fails for x . Suppose that (i) and (ii) hold, we will show that (iii) fails. By Lemma 7.15 we have that any decomposition tree T for x embeds 2 <ω . So let T be a decomposition tree for x and B ⊆ T be a copy of 2 <ω . Then we can either find a copy of A inside B , or below some point of B there is no point coloured by ∑ A 2 , and hence below this point there is a copy of Q . Thus by Lemmas 7.19 and 7.21 we have that either B + /lessorequalslant T , B -/lessorequalslant T or C /lessorequalslant T . Hence by Lemma 7.22 either 2 <ω , -2 <ω or 2 <ω ⊥ embeds into x . Thus x fails (iii), and x / ∈ S L P ( Q ′ ), which gives the theorem. /square
## Corollary 7.36. M L P ( Q ′ ) = ˜ C .
Proof. We have that M L P ( Q ′ ) is the class containing S L P ( Q ′ ) and countable unions of increasing sequences and ˜ is the class containing C ˜ C 0 and countable unions of limiting sequences. The result then follows from Theorem 7.35, considering Remark 7.11. /square
Theorem 7.37. If L and P are well-behaved, then M L P is well-behaved.
Proof. By Corollary 7.36 we have that M L P ( Q ′ ) = ˜ . C Suppose there is a bad M L P ( Q )-array; so by theorems 6.12 and 5.12, we have a witnessing bad Q -array. Hence M L P is well-behaved. /square
## 8. Corollaries and conclusions
We now present some applications of Theorem 7.37.
Theorem 8.1 (Kˇ r´ ˇ [3]) ız . M , the class of σ -scattered linear orders is well-behaved.
Proof. Set L = On ∪ On , set ∗ P = 1 C { , 2 } . Then L = S by Theorem 2.32. L is well-behaved, by theorems 2.8 and 2.22. We also have that P is well-behaved by Lemma 2.17. Hence by Theorem 7.37 we have that M L P is well-behaved.
We claim that M L P = M . Recall the definition of S L P , 7.10. Let X ∈ S , then clearly X satisfies (i) and (iii) since if either failed then X would contain two incomparable elements. A chain of intervals of X does not embed Q , otherwise X would embed Q . Hence every chain of intervals of X under ⊇ has scattered order type. If we take the chain of intervals of X under ⊇ consisting of final segments, this chain has order type precisely the same as X . So we have that S L P = S which implies that M L P = M , and therefore M is well-behaved. /square
Definition 8.2. Let P be a set of countable indecomposable partial orders. Then C P is the class of countable partial orders such that every indecomposable subset is in P .
Theorem 8.3. If C ⊆ L then C P ⊆ M L P .
Proof. Suppose that C ⊆ L . Let X ∈ C P , we claim that X ∈ M L P . Fix an enumeration of X in order type ω , so that X = { x n : n ∈ ω } . (If | X < | ℵ 0 then the argument is essentially the same.) We will write X as the countable union of some limiting sequence ( X n n ) ∈ ω that we will define.
Let X 〈〉 = X , I 0 = {〈〉} and suppose for some sequence /vector t of elements of ⋃ L × ⋃ P we have defined some X /vector t ⊆ X . When | X /vector t | > 1, pick a maximal chain 〈 I /vector t i : i ∈ r /vector t 〉 of intervals of X /vector t that contains { x n } , where n is least such that x n ∈ X /vector t . Since X /vector t ⊆ X we have that r /vector t ∈ L . For each i ∈ r /vector t , let D /vector t i be the set of unions of maximal chains of intervals of I /vector t i that do not contain I /vector t i .
/negationslash
Now let f /vector t i = ∑ a i and s /vector t i be the point of a i that was also contained in ⋃ j>i I /vector t j ∈ D /vector t i . We set η /vector ( ) = t 〈〈 f , s /vector t i /vector t i 〉 : i ∈ r /vector t 〉 . For i ∈ r /vector t and u ∈ a η i ⊆ a i , let X /slurabove /vector t 〈 i,u 〉 be the element of D /vector t i that contains the point u . Then if k /vector t = 〈 X /slurabove /vector t 〈 i,u 〉 : i ∈ r , u /vector t ∈ a η i 〉 we have by construction that 〈 η, k /vector t 〉 is maximal for X /vector t .
We claim that a i = I / /vector t i D /vector t i is indecomposable. If Z = a i is a non-singleton interval of a i , then let Z ′ ⊆ I /vector t i be the union of the sets of D /vector t i that contain points of Z . Then Z ′ is an interval of I /vector t i not equal to I /vector t i , and not in any of the maximal chains used to form D /vector t i ; this is a contradiction which gives the claim. Now, we see that a i is an indecomposable subset of X and thus a i ∈ P .
For m ∈ ω we let I m +1 = I m ∪ { /vector t /slurabove 〈 i, u 〉 | /vector t ∈ I m , i ∈ r , u /vector t ∈ a , η i | X /vector t | > 1 } , D m be the set of leaves of I m and F m = { η /vector ( ) t | /vector t ∈ I m \ D m } . Then F m is an admissible composition set and I m is a set of position sequences for F m . Now let d m = 〈 d /vector p m : /vector p ∈ D m 〉 ∈ dom( f F m ) be such that each d /vector p m is a single point, coloured either by -∞ if | X /vector p | > 1, and coloured by c X /vector p ( x ) when X /vector p = { x } . Now define X m = g F m ( d m ), so that by construction ( X m m ω ) ∈ is a limiting sequence. Moreover, for each m ∈ ω we can consider the underlying set of each X m as a subset of the underlying set of X . Note also that the underling set of X m contains the point x m , and since the relevant chain of intervals contained { x m } the colour of this point in X m will be the same as from X . Thus the limit of ( X m m ω ) ∈ is precisely X , so that X ∈ M L P as required. /square
Corollary 8.4. If P is a well-behaved set of countable indecomposable partial orders, then C P is well-behaved. 15
Proof. By Theorems 8.1, 8.3 and 7.37.
$$\Box$$
Remark 8.5 . If P is a set of finite partial orders then let ˜ be the class of countable partial orders P whose every finite restriction is in P . In [5], Pouzet asked: if P preserves bqo, then is ˜ P bqo? As we have seen, well-behaved is a more useful concept than preserving bqo, so we modify the
15 This result was obtained independently by Christian Delhomme in as yet unpublished work. The author thanks him for his private communication.
Figure 12. An antichain of ⋃ n ∈ ω I n (A ). 2
question to if P well-behaved. Corollary 8.4 brings us closer to a result of this kind, however fails to account for possible infinite indecomposable subsets of orders in ˜ . P If we could prove that for any infinite indecomposable order X the set of finite indecomposable subsets of X is not well-behaved, then we would answer this version of Pouzet's question positively.
Definition 8.6. For n ∈ ω let I n denote the set of indecomposable partial orders whose cardinality is at most n .
Theorem 8.7. For any n ∈ ω , the class M M I n is well-behaved.
Proof. I n is a finite set of finite partial orders so by Lemma 2.17, is well-behaved. Furthermore, M is well-behaved by Theorem 8.1, so using Theorem 7.37 completes the proof. /square
Corollary 8.8. For any n ∈ ω , the class C I n is well-behaved.
Proof. By Theorems 8.3 and 8.7.
$$\Box$$
Note that ⋃ n ∈ ω I n (A ) contains an antichain as in Figure 12, and as described by Pouzet in 2 [5]. Hence ⋃ n ∈ ω I n does not preserve bqo and is certainly not well-behaved. Thus Theorem 8.7 is maximal in some sense. In order to improve this result we would like to know the answers to questions such as:
- (1) Is there consistently a well-behaved class of linear orders larger than M ? E.g. is the class of Aronszajn lines from [16] well-behaved under PFA?
- (2) Is there an infinite well-behaved class of indecomposable partial orders?
- (3) Is there an infinite indecomposable partial order P such that { P } is well-behaved?
A positive answer to any of these questions would immediately improve Theorem 8.7.
## Acknowledgements
The author thanks Mirna Dˇ zamonja, Yann Pequignot and Rapha¨ el Carroy for extremely helpful discussions and remarks.
## References
- [1] S. Thomass´ e, On better-quasi-ordering countable series-parallel orders . Trans. Amer. Math. Soc., Vol. 352, 6, (1999), pp 2491-2505.
- [2] E. Corominas, On better-quasi-ordering countable countable trees . Discrete Mathematics 53, North-Holland, (1985), pp 35-53.
- [3] I. Kˇ r ´ ˇ, ız Proving a witness lemma in better-quasiordering theory: the method of 'extensions' . Math. Proc. Camb. Phil. Soc. 106, (1989), pp 253-262.
- [4] R. Thomas, Well-quasi-ordering infinite graphs with forbidden finite planar minor . Trans. Amer. Math. Soc., Vol. 312, 1, (1989), pp 279-313.
- [5] M. Pouzet, Applications of well-quasi-ordering and better-quasi-ordering . Graphs and Orders, (I. Rival ed.), D. Reidel Publishing Company, (1985), pp 503-519.
- [6] J. B. Kruskal, The theory of well-quasi-ordering: a frequently discovered concept . J. Comb. Th. (A) 13, (1972), pp 297-306.
- [7] F. van Engelen, A. W. Miller and J. Steel, Rigid Borel sets and better quasiorder theory . Contemp. Math. 65, (1987), pp 119-222.
- [8] A. Louveau and J. Saint-Raymond, On the quasi-ordering of borel linear orders under embeddability . J. Symbolic Logic, Vol. 55, 2 (1990), pp 449-896.
- [9] C. St. J.A. Nash-Williams, On well-quasi-ordering infinite trees . Proc. Camb. Phil. Soc. 61, (1965), pp 697-720.
- [10] , On well-quasi-ordering transfinite sequences . Proc. Camb. Phil. Soc. 64, (1968), 273-290.
- [11] S. G. Simpson, BQO theory and Fra¨ ıss´'s conjecture e . Chapter 9 of Recursive aspects of descriptive set theory, Oxford Univ. Press, New York, (1985), pp. 124-138.
- [12] R. Laver, On Fra¨ ıss´'s e order type conjecture . Ann. of Math. 93, (1971), pp 89-111.
- [13] , Better-quasi-orderings and a class of trees . Studies in Foundations and Combinatorics, Adv. in Math. Supplementary Studies 1 (1978), pp 31-48.
- [14] F. Hausdorff, Grundz¨ uge einer Theorie der geordneten Mengen . Math. Ann. 65 (1908), pp 435-505.
- [15] F. Galvin and K. Prikry, Borel sets and Ramsey's theorem . J. Symb. Logic 38 (1973), pp 193-198.
- [16] C. Martinez-Ranero, Well-quasi-ordering Aronszajn lines . Fund. Math. 213 (2011), pp 197-211.
School of Mathematics, University of East Anglia, Norwich Research Park, Norwich, NR4 7TJ,
UK
E-mail address : [email protected] | null | [
"Gregory McKay"
] | 2014-08-01T22:23:53+00:00 | 2014-10-01T14:47:32+00:00 | [
"math.LO",
"06A07 (Primary) 03E05, 06A06, 05C05 (Secondary)"
] | On better-quasi-ordering classes of partial orders | We provide a method of constructing better-quasi-orders by generalising a
technique for constructing operator algebras that was developed by Pouzet. We
then generalise the notion of $\sigma$-scattered to partial orders, and use our
method to prove that the class of $\sigma$-scattered partial orders is
better-quasi-ordered under embeddability. This generalises theorems of Laver,
Corominas and Thomass\'{e} regarding $\sigma$-scattered linear orders and
trees, countable forests and N-free partial orders respectively. In particular,
a class of countable partial orders is better-quasi-ordered whenever the class
of indecomposable subsets of its members satisfies a natural strengthening of
better-quasi-order. |
1408.0316v1 | ## Computation of potentials from current electrodes in cylindrically stratified media: A stable, rescaled semi-analytical formulation
H. Moon a, ∗ , F. L. Teixeira a , B. Donderici b
a ElectroScience Laboratory, The Ohio State University, Columbus, OH 43212, USA b Sensor Physics & Technology, Halliburton Energy Services, Houston, TX 77032, USA
## Abstract
We present an efficient and robust semi-analytical formulation to compute the electric potential due to arbitrary-located point electrodes in three-dimensional cylindrically stratified media, where the radial thickness and the medium resistivity of each cylindrical layer can vary by many orders of magnitude. A basic roadblock for robust potential computations in such scenarios is the poor scaling of modified-Bessel functions used for computation of the semi-analytical solution, for extreme arguments and/or orders. To accommodate this, we construct a set of rescaled versions of modified-Bessel functions, which avoids underflows and overflows in finite precision arithmetic, and minimizes round-off errors. In addition, several extrapolation methods are applied and compared to expedite the numerical evaluation of the (otherwise slowly convergent) associated Sommerfeld-type integrals. The proposed algorithm is verified in a number of scenarios relevant to geophysical exploration, but the general formulation presented is also applicable to other problems governed by Poisson equation such as Newtonian gravity, heat flow, and potential flow in fluid mechanics, involving cylindrically stratified environments.
Keywords: Poisson equation, steady-state diffusion equation, discontinuous coefficients, stratified media, resistivity logging, electric potential
## 1. Introduction
Resistivity logging is extensively used for detecting, characterizing, and analyzing hydrocarbon-bearing zones in the subsurface earth [30, 17, 7, 4, 13, 14, 35, 3]. This sensing modality employs electrode-type devices mounted on a mandrel that inject electric currents into the surrounding earth formation [5, 29]. The ensuing electric potential is then measured at different locations to provide estimates for the surrounding resistivity. Many numerical techniques such as finite-differences, finite elements, numerical mode-matching, and finite volumes method can be used to model the response of resistivity logging tools [12, 8, 10, 26, 23, 24, 20, 25, 21, 22, 11, 6, 9]. Brute-force techniques are rather versatile and applicable to arbitrary resistivity distributions; however, at the same time, this precludes optimality in particular cases of special interest, such as when resistivity logging environment can be represented as a cylindrically stratified medium [2]. Depending on the implementation, brute-force techniques may have difficulties handling extreme sharp discontinuities in the coefficients, as is the case for the resistivity parameter for the physical scenario considered here, which can change by many orders of magnitude across adjacent layers.
In this paper, a robust semi-analytical formulation for computing the electric potential due to arbitrarylocated point electrodes in three-dimensional cylindrically stratified media is proposed. The present formulation is based on a series expansion in terms of azimuth Fourier modes and a spectral integral over the vertical wavenumber along the axial direction. The resulting problem in terms of the radial variable yields
∗ Corresponding author
Email addresses:
[email protected] [email protected]
(H. Moon),
(B. Donderici)
[email protected]
(F. L. Teixeira),
a set of modified Bessel equations. The present formulation removes roadblocks for numerical computations associated with the poor scaling of modified-Bessel functions for very small and/or very large arguments and/or orders [27, 31, 1]. This is done by constructing a set of rescaled, modified-Bessel functions that can be stably evaluated under double-precision arithmetic, akin to what has been done in the past for ordinary (non-modified) Bessel functions [16] . The present formulation also carefully manipulates the analytical formulae for the potential in such media to yield a set of integrand expressions can be computed in a robust manner under double-precision for a wide range of layer thicknesses, layer resistivities, and source and observation point separations. Finally, a number of acceleration techniques are implemented and compared to effect the efficient numerical integration of the Sommerfeld-type (spectral) integrals, which otherwise suffer from slow convergence. The proposed algorithm is verified in a number of practical scenarios relevant to geophysical exploration. More generally, the mathematical setting here corresponds to the classical problem of obtaining the Green's function for the steady diffusion equation (Poisson problem) with discontinuous coefficients in a separable geometry. As such, the general formalism presented here is also applicable to other problems governed by Poisson equation such as Newtonian gravity, heat flow, elasticity, neutron transport, and potential flows in fluid mechanics, in cylindrically stratified geometries.
## 2. Formulation
## 2.1. Electric potential in homogeneous media
In a homogeneous medium, the electric potential ψ from a current electrode at the origin writes as [31]
$$\psi = \frac { \mathcal { I } } { 4 \pi \sigma \sqrt { \rho ^ { 2 } + z ^ { 2 } } } = \frac { \mathcal { I } } { 2 \pi ^ { 2 } \sigma } \int _ { 0 } ^ { \infty } K _ { 0 } ( \lambda \rho ) \cos ( \lambda z ) d \lambda,$$
where I is the electric current flowing into the medium from the electrode, σ is the conductivity of the medium, and K 0 ( ) · is the modified-Bessel function of the second kind of the zeroth order. For the second equality, the complete Lipschitz-Hankel integral [32] is employed. When the source is off the origin, higher order azimuthal modes appear. Using the addition theorem for K 0 , (1) is modified to
$$\psi & = \frac { \mathcal { I } } { 2 \pi ^ { 2 } \sigma } \int _ { 0 } ^ { \infty } K _ { 0 } ( \lambda | \rho - \rho ^ { \prime } | ) \cos ( \lambda ( z - z ^ { \prime } ) ) d \lambda \\ & = \frac { \mathcal { I } } { 2 \pi ^ { 2 } \sigma } \sum _ { n = - \infty } ^ { \infty } e ^ { i n ( \phi - \phi ^ { \prime } ) } \int _ { 0 } ^ { \infty } I _ { n } ( \lambda \rho _ { < } ) K _ { n } ( \lambda \rho _ { > } ) \cos ( \lambda ( z - z ^ { \prime } ) ) d \lambda,$$
in terms of modified-Bessel functions of both first, I n ( ) and second, · K n ( ), kinds. · In the above, primed coordinates ( ρ , φ , z ′ ′ ′ ) represent the source location and unprimed coordinates ( ρ, φ, z ) represent the observation point. Also, ρ < = min( ρ, ρ ′ ) and ρ > = max( ρ, ρ ′ ).
## 2.2. Electric potential in cylindrically stratified media
In a cylindrically stratified medium, boundary conditions at the interfaces need to be incorporated. Let us first consider the case with two distinct cylindrical layers, as depicted in Fig. 1. When the source is embedded in layer 1, we denote it the outgoing-potential case. In this case, the primary potential ψ p is a function of K n ( λρ ) because I n ( λρ ) diverges as ρ goes to infinity. On the other hand, when the source is embedded in layer 2, we denote it the standing-potential case and ψ p is a function of I n ( λρ ) instead of K n ( λρ ) because K n ( λρ ) diverges when ρ goes to zero. For the outgoing-potential case, the n -th harmonic with e i n φ ( -φ ′ ) dependence in layer 1 and layer 2 can be expressed, resp., as
$$\psi _ { 1 } = \left [ K _ { n } ( \lambda \rho ) + R _ { 1 2 } I _ { n } ( \lambda \rho ) \right ] A _ { 0 },$$
ψ
2
=
T
12
K
n
(
λρ A ,
)
0
(3b)
and
ρ
Figure 1: Schematic description of two layers with associated coefficients in the ρz -plane.
where R 12 and T 12 are the (local) reflection and transmission coefficients at the boundary a 1 , and A 0 is an arbitrary amplitude of the primary potential. Applying the boundary conditions [31] at the interface, we obtain
$$R _ { 1 2 } = \frac { ( \sigma _ { 2 } - \sigma _ { 1 } ) K _ { n } ( \lambda a _ { 1 } ) K _ { n } ^ { \prime } ( \lambda a _ { 1 } ) } { \sigma _ { 1 } I _ { n } ^ { \prime } ( \lambda a _ { 1 } ) K _ { n } ( \lambda a _ { 1 } ) - \sigma _ { 2 } I _ { n } ( \lambda a _ { 1 } ) K _ { n } ^ { \prime } ( \lambda a _ { 1 } ) }, \quad \tilde { \sigma }.$$
$$T _ { 1 2 } = \frac { \mu _ { 1 } ^ { \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \colon } ^ { \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \colon } } { \lambda a _ { 1 } \left [ \sigma _ { 1 } I _ { n } ^ { \prime } ( \lambda a _ { 1 } ) K _ { n } ( \lambda a _ { 1 } ) - \sigma _ { 2 } I _ { n } ( \lambda a _ { 1 } ) K _ { n } ^ { \prime } ( \lambda a _ { 1 } ) \right ] }.$$
For the standing-potential case, we similarly have
$$\psi _ { 1 } & = T _ { 2 1 } I _ { n } ( \lambda \rho ) B _ { 0 },$$
$$\psi _ { 2 } = \left [ R _ { 2 1 } K _ { n } ( \lambda \rho ) + I _ { n } ( \lambda \rho ) \right ] B _ { 0 },$$
$$R _ { 2 1 } = \frac { ( \sigma _ { 2 } - \sigma _ { 1 } ) I _ { n } ( \lambda a _ { 1 } ) I _ { n } ^ { \prime } ( \lambda a _ { 1 } ) } { \sigma _ { 1 } I _ { n } ^ { \prime } ( \lambda a _ { 1 } ) K _ { n } ( \lambda a _ { 1 } ) - \sigma _ { 2 } I _ { n } ( \lambda a _ { 1 } ) K _ { n } ^ { \prime } ( \lambda a _ { 1 } ) },$$
$$T _ { 2 1 } = \frac { \mu _ { 1 } ^ { \nu - n } \cdots \nu _ { 1 } ^ { \nu - \mu _ { 2 } ^ { \nu } } \cdots \nu _ { 2 } ^ { \nu - \mu _ { 3 } ^ { \nu - n } } \cdots \nu _ { 3 } ^ { \nu - \mu _ { 4 } ^ { \nu } } } { \lambda a _ { 1 } \left [ \sigma _ { 1 } I _ { n } ^ { \prime } ( \lambda a _ { 1 } ) K _ { n } ( \lambda a _ { 1 } ) - \sigma _ { 2 } I _ { n } ( \lambda a _ { 1 } ) K _ { n } ^ { \prime } ( \lambda a _ { 1 } ) \right ] }.$$
When more than two distinct layers are present, multiple reflections and transmissions occur. Therefore, generalized reflection and transmission coefficients should be determined. The procedure to obtain these coefficients is very similar to the one used for time-harmonic case in [2] and will not be derived in detail here. Fig. 2 depicts the relevant coefficients to the outgoing-potential case in the medium consisting of three cylindrical layers. The potentials in the three layers can be expressed as
$$\psi _ { 1 } & = \left [ K _ { n } ( \lambda \rho ) + \widetilde { R } _ { 1 2 } I _ { n } ( \lambda \rho ) \right ] A _ { 1 },$$
$$\psi _ { 2 } = \left [ K _ { n } ( \lambda \rho ) + R _ { 2 3 } I _ { n } ( \lambda \rho ) \right ] A _ { 2 },$$
$$\psi _ { 3 } & = T _ { 2 3 } K _ { n } ( \lambda \rho ) A _ { 3 }.$$
Applying proper constraint equations, we obtain
$$\widetilde { R } _ { 1 2 } = R _ { 1 2 } + T _ { 2 1 } R _ { 2 3 } ( 1 - R _ { 2 1 } R _ { 2 3 } ) ^ { - 1 } T _ { 1 2 }.$$
If another layer is added beyond layer 3, only R 23 needs to be replaced by ˜ R 23 . Therefore, the generalized reflection coefficient in cylindrically stratified media for the outgoing-potential case is
$$\widetilde { R } _ { i, i + 1 } & = R _ { i, i + 1 } + T _ { i + 1, i } \widetilde { R } _ { i + 1, i + 2 } ( 1 - R _ { i + 1, i } \widetilde { R } _ { i + 1, i + 2 } ) ^ { - 1 } T _ { i, i + 1 }.$$
Layer 2
R
21
23
R
Figure 2: Schematic description of three layers with associated coefficients for the outgoing-potential case: (a) R 12 and (b) S 12 .
Figure 3: Generalized transmission coefficient for the outgoing-potential case.
All amplitudes A i 's as well as generalized reflection coefficients should be determined in order to obtain the potential everywhere. The relationship between two successive amplitudes in cylindrically stratified media can be written as
$$A _ { i + 1 } & = T _ { i, i + 1 } A _ { i } + R _ { i + 1, i } R _ { i + 1, i + 2 } A _ { i + 1 }. & ( 1 0 )$$
From (10), a new coefficient denoted by S is defined as
$$S _ { i, i + 1 } = ( 1 - R _ { i + 1, i } \widetilde { R } _ { i + 1, i + 2 } ) ^ { - 1 } T _ { i, i + 1 } & & ( 1 1 )$$
such that A i +1 = S i,i +1 A i . The above coefficient can be regarded as a 'local' transmission coefficient between two adjacent layers, as depicted in Fig. 2b. Generalized transmission coefficient described in Fig. 3 can be defined using the S -coefficients (11) through
$$\widetilde { T } _ { j i } = T _ { i - 1, i } S _ { i - 2, i - 1 } \cdots S _ { j, j + 1 } = T _ { i - 1, i } \prod _ { k = j } ^ { i - 2 } S _ { k, k + 1 } = T _ { i - 1, i } X _ { j, i - 1 }.$$
Note that i > j for the outgoing-potential case. When i = j +1, X j,i -1 = 1 in (12). For the standing-potential case depicted in Fig. 4, the potentials in three cylindrical layers can be written a 2
La
a 1
Layer 2
T
12
R
21
R
23
Figure 4: Schematic description of three layers with associated coefficients for the standing-potential case: (a) R 32 and (b) S 32 .
as
$$\psi _ { 1 } & = I _ { n } ( \lambda \rho ) B _ { 1 }, & ( 1 3 \tt a )$$
$$\psi _ { 2 } & = \left [ R _ { 2 1 } K _ { n } ( \lambda \rho ) + I _ { n } ( \lambda \rho ) \right ] B _ { 2 }, & ( 1 3 b )$$
$$\psi _ { 3 } & = \left [ \widetilde { R } _ { 3 2 } K _ { n } ( \lambda \rho ) + I _ { n } ( \lambda \rho ) \right ] B _ { 3 }. & ( 1 3 c )$$
Similarly, applying proper constraint conditions, we obtain
$$\widetilde { R } _ { 3 2 } = R _ { 3 2 } + T _ { 2 3 } R _ { 2 1 } ( 1 - R _ { 2 3 } R _ { 2 1 } ) ^ { - 1 } T _ { 3 2 }.$$
and, more generally,
$$\widetilde { R } _ { i, i - 1 } & = R _ { i, i - 1 } + T _ { i - 1, i } \widetilde { R } _ { i - 1, i - 2 } ( 1 - R _ { i - 1, i } \widetilde { R } _ { i - 1, i - 2 } ) ^ { - 1 } T _ { i, i - 1 }.$$
To obtain all amplitudes B i 's, it is convenient to define the S coefficient with decreasing subscripts S i,i -1 . Since the relation between two successive amplitudes is
$$B _ { i - 1 } = T _ { i, i - 1 } B _ { i } + R _ { i - 1, i } R _ { i - 1, i - 2 } B _ { i - 1 },$$
the S coefficient in this case is defined as
$$S _ { i, i - 1 } = ( 1 - R _ { i - 1, i } \widetilde { R } _ { i - 1, i - 2 } ) ^ { - 1 } T _ { i, i - 1 }$$
such that B i -1 = S i,i -1 B i . The generalized transmission coefficient depicted in Fig. 5 can then be written as
$$\widetilde { T } _ { j i } = T _ { i + 1, i } S _ { i + 2, i + 1 } \cdots S _ { j, j - 1 } = T _ { i + 1, i } \prod _ { k = j } ^ { i + 2 } S _ { k, k + 1 } = T _ { i + 1, i } X _ { j, i + 1 }.$$
Note that i < j for the outgoing-potential case. When j = i +1, X j,i +1 = 1 in (18).
Using generalized reflection and transmission coefficients, we can extend (2) to incorporate multiple reflections and transmissions. The integral part of (2) can be modified to
$$\int _ { 0 } ^ { \infty } [ I _ { n } ( \lambda \rho _ { < } ) K _ { n } ( \lambda \rho _ { > } ) + K _ { n } ( \lambda \rho ) A _ { j n } ( \rho ^ { \prime } ) + I _ { n } ( \lambda \rho ) B _ { j n } ( \rho ^ { \prime } ) ] \cos ( \lambda ( z - z ^ { \prime } ) ) d \lambda,$$
Figure 5: Generalized transmission coefficient for the standing-potential case.
where the two unknowns are A jn ( ρ ′ ) and B jn ( ρ ′ ). Applying two constraint conditions at ρ = a j -1 and ρ = a j , we obtain
$$A _ { j n } ( \rho ^ { \prime } ) = ( 1 - \widetilde { R } _ { j, j - 1 } \widetilde { R } _ { j, j + 1 } ) ^ { - 1 } \widetilde { R } _ { j, j - 1 } \left [ K _ { n } ( \lambda \rho ^ { \prime } ) + I _ { n } ( \lambda \rho ^ { \prime } ) \widetilde { R } _ { j, j + 1 } \right ],$$
$$B _ { j n } ( \rho ^ { \prime } ) = ( 1 - \widetilde { R } _ { j, j + 1 } \widetilde { R } _ { j, j - 1 } ) ^ { - 1 } \widetilde { R } _ { j, j + 1 } \left [ I _ { n } ( \lambda \rho ^ { \prime } ) + K _ { n } ( \lambda \rho ^ { \prime } ) \widetilde { R } _ { j, j - 1 } \right ].$$
Substituting (20a) and (20b) into (19) and rearranging the integrand excluding the cosine factor gives
[
I
n
(
λρ
<
)
K
n
(
λρ
>
) +
K
n
(
λρ A
)
jn
(
ρ
′
) +
I
n
(
λρ B
)
jn
(
ρ
′
)]
$$& \stackrel { \iota \nolimits \nu \nu ^ { \prime } \nolimits \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nolimits \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu\nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu\nu \nu \nolimits \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nolimits \nu \nu\nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nu \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nulimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimim\nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimim\nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimim\nolimits \nolimits \nolimits \nolimits \nolimits \nolimim\nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimim\nolimits \nolimits \nolimits \nolimim\nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimim\nolimits \nolimits \nolimits \nolimim\nolimits \nolimits \nolimits \nolimimits \nolimits \nolimits \nolimim\nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimits \nolimim\nolimits \nolimits \nolimits \nolimim\nolimits \nolimim\nolimits \nolimits \nolimits \nolimim\nolimits \nolimits \nolimim\nolimits \nolimits \nolimim\nolimits \nolimits \nolimim\nolimits \nolimits \nolimim\nolimits \nolimits \nolimim\nolimits \nolimim\nolimits \nolimim\nolimits \nolimits \nolimim\nolimits \nolimim\nolimits \nolimim\nolimits \nolimits \nolimits \nolimim\nolimits \nolimim\nolimits \nolimits \nolimim\nolimits \nolimim\nolimits \nolimim\nolimits \nolimits \nolimim\nolimits \nolim\nolimits \nolimim\nolimits \nolimim\nolimits \nolimits \nolimim\nolimits \nolimim\nolimits \nolimim\nolimits \nolimim\nolimits \nolimim\nolimits \nolimim\nolimits \nolimim\nolimits \nolimits \nolimim\nolimits \nolimim\nolimits \nolimim\nolimits \nolimim\nolimits \nolimim\nolimim\nolimits \nolimim\nolimits \nolimim\nolimits \nolimim\nolimits \nolimim\nolimim\nolimits \nolim\nolim\nolimits \nolimim\nolim\nolimits \nolim\nolim\nolim\nolim\nolimits \nolim\nolim\nolim\nolimits \nolim\nolim\nolim\nolimits \nolimim\nolimits \nolim\nolim\nolimits \nolimim\nolim\nolimits \nolim\nolimits \nolim\nolimits \nolimim\nolimits \nolim\nolim\nolim\nolim\nolimits \nolimim\nolim\nolimits \nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolimits \nolim\nolim\nolim\nolim\nolimits \nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolimits \nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolimits \nolim\nolimits \nolimim\nolimits \nolimits \nolimim\nolimits \nolimim\nolim\nolim\nolimits \nolimim\nolimits \nolimim\nolimits \nolimim\nolimits \nolimim\nolimim\nolimits \nolimim\nolimits \nolimim\nolimim\nolimits \nolimim\nolimits \nolimim\nolim\nolimits \nolimim\nolimits \nolimim\nolim\nolimim\nolimits \nolimim\nolim\nolimits \nolimim\nolimim\nolimits \nolimim\nolimits \nolim\nolim\nolim\nolimits \nolimim\nolimim\nolim\nolimits \nolimim\nolimits \nolimim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolimits \nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolimits \nolimim\nolimits \nolim\nolim\nolim\nolim\nolimits \nolimim\nolimim\nolim\nolim\nolim\nolim\nolimits \nolimim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolimits \nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim\nolim
& \quad = \begin{cases} \\ & \quad \left [ K _ { n } ( \lambda \rho ) + I _ { n } ( \lambda \rho ) + I _ { n } ( \lambda \rho ) ) \end{cases} \right ] \left [ I _ { n } ( \lambda \rho ) + I _ { n } ( \lambda \rho ) + I _ { n } ( \lambda \rho ) + I _ { n } ( \lambda \rho ) + I _ { n } ( \lambda \rho ) + I _ { n } ( \lambda \rho ) + I _ { n } ( \lambda \rho ) + I _ { n } ( \lambda \rho ) + I _ { n } ( \lambda \rho ) + I _ { n } ( \lambda \rho ) + I _ { n } ( \lambda \rho ) + I _ { n } ( \lambda \rho ) + I _ { n } ( \lambda \rho ) + I _ { n } ( \lambda \rho ) + I _ { n } ( \lambda \rho ) + I _ { n } ( \lambda \rho ) + I _ { n, j - 1 } ( \lambda \rho ) + I _ { n, j - 1 } ) \end{cases} \right ] \left [ K _ { n } ( \lambda \rho ) + I _ { n } ( \lambda \rho ) + I _ { n, j - 1 } ) \end{cases} \right ] \left [ K _ { n } ( \lambda \rho ) + I _ { n, j - 1 } \end{cases$$
where M j = ( 1 -˜ R j,j -1 ˜ R j,j +1 ) -1 . When i > j , the potential in layer i is expressed as
$$\psi _ { i } = \frac { I } { 2 \pi ^ { 2 } \sigma _ { j } } \sum _ { n = - \infty } ^ { \infty } e ^ { i n ( \phi - \phi ^ { \prime } ) } \int _ { 0 } ^ { \infty } \left [ K _ { n } ( \lambda \rho ) + I _ { n } ( \lambda \rho ) \widetilde { R } _ { i, i + 1 } \right ] A _ { i n } \cos ( \lambda ( z - z ^ { \prime } ) ) d \lambda,$$
where A in is the amplitude of the outgoing-potential in layer i and expressed as
$$A _ { i n } = \left ( 1 - R _ { i, i - 1 } \widetilde { R } _ { i, i + 1 } \right ) \widetilde { T } _ { j i } A _ { j n } = N _ { i + } \widetilde { T } _ { j i } A _ { j n } = N _ { i + } \widetilde { T } _ { j i } \left [ I _ { n } ( \lambda \rho ^ { \prime } ) + K _ { n } ( \lambda \rho ^ { \prime } ) \widetilde { R } _ { j, j - 1 } \right ] M _ { j }. \quad \ ( 2 3 )$$
Therefore, the integrand factor for layer i in (22) is
$$\left [ K _ { n } ( \lambda \rho ) + I _ { n } ( \lambda \rho ) \widetilde { R } _ { i, i + 1 } \right ] \left [ I _ { n } ( \lambda \rho ^ { \prime } ) + K _ { n } ( \lambda \rho ^ { \prime } ) \widetilde { R } _ { j, j - 1 } \right ] N _ { i + } \widetilde { T } _ { j i } M _ { j }.$$
When i < j , the potential in layer i is expressed as
$$\psi _ { i } = \frac { I } { 2 \pi ^ { 2 } \sigma _ { j } } \sum _ { n = - \infty } ^ { \infty } e ^ { i n ( \phi - \phi ^ { \prime } ) } \int _ { 0 } ^ { \infty } \left [ I _ { n } ( \lambda \rho ) + K _ { n } ( \lambda \rho ) \widetilde { R } _ { i, i - 1 } \right ] A _ { i n } \cos ( \lambda ( z - z ^ { \prime } ) ) d \lambda,$$
where A in is the amplitude of the standing-potential in layer i and expressed as
$$A _ { i n } = \left ( 1 - R _ { i, i + 1 } \widetilde { R } _ { i, i - 1 } \right ) \widetilde { T } _ { j i } A _ { j n } = N _ { i - } \widetilde { T } _ { j i } A _ { j n } \left [ K _ { n } ( \lambda \rho ^ { \prime } ) + I _ { n } ( \lambda \rho ^ { \prime } ) \widetilde { R } _ { j, j + 1 } \right ] M _ { j }$$
The integrand of the integral for potential in layer i in (25) is
$$\left [ I _ { n } ( \lambda \rho ) + K _ { n } ( \lambda \rho ) \widetilde { R } _ { i, i - 1 } \right ] \left [ K _ { n } ( \lambda \rho ^ { \prime } ) + I _ { n } ( \lambda \rho ^ { \prime } ) \widetilde { R } _ { j, j + 1 } \right ] N _ { i - } \widetilde { T } _ { j i } M _ { j }.$$
In summary, the potential in cylindrically stratified media admits four different expressions depending on the relative position of ρ and ρ ′ , which can be written as
$$\psi _ { i } = \frac { \mathcal { I } } { 2 \pi ^ { 2 } \sigma _ { j } } \sum _ { n = - \infty } ^ { \infty } e ^ { i n ( \phi - \phi ^ { \prime } ) } \int _ { 0 } ^ { \infty } F _ { n } ( \rho, \rho ^ { \prime } ) \cos ( \lambda ( z - z ^ { \prime } ) ) d \lambda,$$
where
Case 1 : ρ and ρ ′ are in the same layer and ρ ≥ ρ ′
$$F _ { n } ( \rho, \rho ^ { \prime } ) = \left [ K _ { n } ( \lambda \rho ) + I _ { n } ( \lambda \rho ) \widetilde { R } _ { j, j + 1 } \right ] \left [ I _ { n } ( \lambda \rho ^ { \prime } ) + K _ { n } ( \lambda \rho ^ { \prime } ) \widetilde { R } _ { j, j - 1 } \right ] M _ { j },$$
Case 2 : ρ and ρ ′ are in the same layer and ρ < ρ ′
$$F _ { n } ( \rho, \rho ^ { \prime } ) = \left [ I _ { n } ( \lambda \rho ) + K _ { n } ( \lambda \rho ) \widetilde { R } _ { j, j - 1 } \right ] \left [ K _ { n } ( \lambda \rho ^ { \prime } ) + I _ { n } ( \lambda \rho ^ { \prime } ) \widetilde { R } _ { j, j + 1 } \right ] M _ { j },$$
Case 3 : ρ and ρ ′ are in different layers and ρ > ρ ′
$$F _ { n } ( \rho, \rho ^ { \prime } ) = \left \lceil K _ { n } ( \lambda \rho ) + I _ { n } ( \lambda \rho ) \widetilde { R } _ { i, i + 1 } \right \rceil \left \lceil I _ { n } ( \lambda \rho ^ { \prime } ) + K _ { n } ( \lambda \rho ^ { \prime } ) \widetilde { R } _ { j, j - 1 } \right \rceil N _ { i + } \widetilde { T } _ { j i } M _ { j },$$
Case 4 : ρ and ρ ′ are in different layers and ρ < ρ ′
$$F _ { n } ( \rho, \rho ^ { \prime } ) = \left [ I _ { n } ( \lambda \rho ) + K _ { n } ( \lambda \rho ) \widetilde { R } _ { i, i - 1 } \right ] \left [ K _ { n } ( \lambda \rho ^ { \prime } ) + I _ { n } ( \lambda \rho ^ { \prime } ) \widetilde { R } _ { j, j + 1 } \right ] N _ { i - } \widetilde { T } _ { j i } M _ { j }.$$
## 2.3. Rescaled modified-Bessel functions
The electric potential involves products of the modified-Bessel function of the first and second kind, viz. I n and K n . Those products can involve disparate values due to the exponential behavior of the functions. For example, when | z | << 1, K n ( z ) has very large value whereas I n ( z ) has very small value. This disparity becomes progressively worse for higher order modes. On the other hand, when glyph[Rfractur] e z [ ] >> 1, K n ( z ) has very small value while I n ( z ) has very large value. This leads to unreliable results under double-precision computations. To eliminate this problem, a new set of rescaled modified-Bessel functions are defined in a similar fashion to what has been done in [16] for Bessel and Hankel functions, viz. J n and H (1) n . In order to apply such rescaled functions, the analytical expressions for the potential need to modified accordingly, as described next.
When | z | glyph[lessmuch] 1, I n ( z ) and K n ( z ) can be expressed via small argument approximations for n > 0. Noting that, for -π < arg z ( ) < π/ 2, the relationship between cylindrical functions and modified-cylindrical functions reads as
$$I _ { n } ( z ) = \mathfrak { i } ^ { - n } J _ { n } ( i z ),$$
$$K _ { n } ( z ) & = \frac { \pi } { 2 } \mathfrak { i } ^ { n + 1 } H _ { n } ^ { ( 1 ) } ( i z ). & ( 3 0 b )$$
Thus, we have the following small argument approximations for I n and K n and their derivatives.
$$I _ { n } ( \lambda a _ { i } ) \approx \frac { 1 } { n! } \left ( \frac { \lambda a _ { i } } { 2 } \right ) ^ { n } = \frac { 1 } { n! } \left ( \frac { \lambda } { 2 } \right ) ^ { n } \cdot a _ { i } ^ { n } \cdot 1 = G a _ { i } ^ { n } \hat { I } _ { n } ( \lambda a _ { i } ),$$
$$I ^ { \prime } _ { n } ( \lambda a _ { i } ) \approx \frac { \stackrel { \cdot } { 1 } } { 2 ( n - 1 )! } \left ( \frac { \lambda a _ { i } } { 2 } \Big ) ^ { n - 1 } = \frac { 1 } { n! } \left ( \frac { \lambda } { 2 } \right ) ^ { n } \cdot a _ { i } ^ { n } \cdot \frac { n } { \lambda a _ { i } } = G a _ { i } ^ { n } \hat { I } ^ { \prime } _ { n } ( \lambda a _ { i } ),$$
$$K _ { n } ( \lambda a _ { i } ) \approx \frac { \theta ^ { \prime } - 1 )! } { 2 } \left ( \frac { \theta ^ { \prime } } { \lambda a _ { i } } \right ) ^ { n } = n! \left ( \frac { 2 } { \lambda } \right ) ^ { n } \cdot a _ { i } ^ { - n } \cdot \left ( \frac { 1 } { 2 n } \right ) = G ^ { - 1 } a _ { i } ^ { - n } \hat { K } _ { n } ( \lambda a _ { i } ),$$
$$K _ { n } ^ { \prime } ( \lambda a _ { i } ) \approx - \frac { n! } { 4 } \left ( \frac { 2 } { \lambda a _ { i } } \right ) ^ { n + i } = n! \left ( \frac { 2 } { \lambda } \right ) ^ { n } \cdot a _ { i } ^ { - n } \cdot \left ( - \frac { 1 } { 2 \lambda a _ { i } } \right ) = G ^ { - 1 } a _ { i } ^ { - n } \hat { K } _ { n } ^ { \prime } ( \lambda a _ { i } ).$$
It should be noted that the multiplicative factor G above is chosen so as to not depend on the radial distance, a i . Also, G is identical for a function and its derivative, and the multiplicative factors appearing in I n and K n are reciprocal to each other. This will facilitate some computations later on.
When glyph[Rfractur] e z [ ] glyph[greatermuch] 1 and z = λa i = ( λ ′ +i λ ′′ ) a i , the large argument approximations for the modified-Bessel functions write as [34]
$$I _ { n } ( \lambda a _ { i } ) & = \frac { e ^ { \lambda a _ { i } } } { \sqrt { 2 \pi \lambda a _ { i } } } \left [ 1 - \frac { ( \mu - 1 ) } { 1! ( 8 \lambda a _ { i } ) } + \frac { ( \mu - 1 ) ( \mu - 9 ) } { 2! ( 8 \lambda a _ { i } ) ^ { 2 } } + \cdots \right ] \\ & = e ^ { \lambda ^ { \prime } a _ { i } } \hat { I } _ { n } ( \lambda a _ { i } ), \\ & \longmapsto \frac { r } { r } \quad \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \colon$$
$$K _ { n } ( \lambda a _ { i } ) & = \sqrt { \frac { \pi } { 2 \lambda a _ { i } } } \stackrel { \dots } { e } ^ { - \lambda a _ { i } } \left [ 1 + \frac { ( \mu - 1 ) } { 1! ( 8 \lambda a _ { i } ) } + \frac { ( \mu - 1 ) ( \mu - 9 ) } { 2! ( 8 \lambda a _ { i } ) ^ { 2 } } + \cdots \right ] \\ & = e ^ { - \lambda ^ { \prime } a _ { i } } \hat { K } _ { n } ( \lambda a _ { i } ),$$
where µ = 4 n 2 . Again, the associated multiplicative factors are reciprocal to each other. The derivatives of rescaled modified-cylindrical functions for large arguments can be derived through the recursive formulas such that
$$I _ { n } ^ { \prime } ( \lambda a _ { i } ) = I _ { n - 1 } ( \lambda a _ { i } ) - \frac { n } { \lambda a _ { i } } I _ { n } ( \lambda a _ { i } ) = e ^ { \lambda ^ { \prime } a _ { i } } \hat { I } _ { n - 1 } ( \lambda a _ { i } ) - e ^ { \lambda ^ { \prime } a _ { i } } \frac { n } { \lambda a _ { i } } \hat { I } _ { n } ( \lambda a _ { i } ) = e ^ { \lambda ^ { \prime } a _ { i } } \hat { I } _ { n } ^ { \prime } ( \lambda a _ { i } ),$$
$$K _ { n } ^ { \prime } ( \lambda a _ { i } ) = - K _ { n - 1 } ( \lambda a _ { i } ) - \frac { \ddots \vdots } { \lambda a _ { i } } K _ { n } ( \lambda a _ { i } ) = - e ^ { - \lambda ^ { \prime } a _ { i } } \hat { K } _ { n - 1 } ( \lambda a _ { i } ) - e ^ { - \lambda ^ { \prime } a _ { i } } \frac { n } { \lambda a _ { i } } \hat { K } _ { n } ( \lambda a _ { i } ) = e ^ { - \lambda ^ { \prime } a _ { i } } \hat { K } _ { n } ^ { \prime } ( \lambda a _ { i } ).$$
If the argument is neither small nor large, rescaled modified-cylindrical functions are defined, in analogy to small and large arguments, as
$$I _ { n } ( \lambda a _ { i } ) = P _ { i } \hat { I } _ { n } ( \lambda a _ { i } ),$$
$$I _ { n } ^ { \prime } ( \lambda a _ { i } ) = P _ { i } \hat { I } _ { n } ^ { \prime } ( \lambda a _ { i } ),$$
$$K _ { n } ( \lambda a _ { i } ) & = P _ { i } ^ { - 1 } \hat { K } _ { n } ( \lambda a _ { i } ), & ( 3 5 c )$$
$$K _ { n } ^ { \prime } ( \lambda a _ { i } ) = P _ { i } ^ { - 1 } \hat { K } _ { n } ^ { \prime } ( \lambda a _ { i } ),$$
where the multiplicative factor P i is defined as in [16], i.e.,
$$\text{If } | I _ { n } ( \lambda a _ { i } ) | ^ { - 1 } < T _ { m }, \ \ P _ { i } = 1.$$
$$\text{If } | I _ { n } ( \lambda a _ { i } ) | ^ { - 1 } \geq T _ { m }, \ \ P _ { i } = | I _ { n } ( \lambda a _ { i } ) |.$$
with T m = 10 00 used for double-precision arithmetic computations. 1
and its subscript is linked to a i . As Table 1 shows, the argument for rescaled modified-cylindrical functions can be categorized into small, moderate, and large, with different and appropriate multiplicative factors defined accordingly.
The numerical threshold values used here to define small, moderate, and large argument ranges are identical to those used for the time-harmonic case involving cylindrical Bessel and Hankel functions detailed in [16] and not repeated here 1 .
## 2.4. Rescaled reflection and transmission coefficients
We can classify the multiplicative factors associated with the rescaled modified-cylindrical functions into two types, denoted as α and β , and shown in Table 2. The factor α i is associated with K n ( λa i ), whereas β i is associated with I n ( λa i ). Note again that the subscript i refers to to the index of the radial factor a i in the argument. There are two important properties to note for α and β : ( ) i Reciprocity : α i = 1 /β i and ( ii ) Boundness : | β α i j | ≤ 1 , for i < j . As we will see below, these two properties are important in ensuring stable numerical computations.
1 A third type of threshold, used in connection with modified-Bessel functions magnitudes (not arguments) is also necessary within the moderate argument range. Again, this threshold value is identical to the one applied to ordinary Bessel functions in [16]
Table 1: Definition of rescaled modified-cylindrical functions for all types of arguments.
| | small arguments | moderate arguments | large arguments |
|----------------|--------------------------------|--------------------------|------------------------------|
| I n ( λa i ) | Ga n i ˆ I n ( λa i ) | P i ˆ I n ( λa i ) | e λ ′ a i ˆ I n ( λa i ) |
| I ′ n ( λa i ) | Ga n i ˆ I ′ n ( λa i ) | P i ˆ I ′ n ( λa i ) | e λ ′ a i ˆ I ′ n ( λa i ) |
| K n ( λa i ) | G - 1 a - n i ˆ K n ( λa i ) | P - 1 i ˆ K n ( λa i ) | e - λ ′ a i ˆ K n ( λa i ) |
| K ′ n ( λa i ) | G - 1 a - n i ˆ K ′ n ( λa i ) | P - 1 i ˆ K ′ n ( λa i ) | e - λ ′ a i ˆ K ′ n ( λa i ) |
Table 2: Definition of α i and β i .
| argument type | α i | β i |
|-----------------|---------------|-----------|
| small | G - 1 a - n i | Ga n i |
| moderate | P - 1 i | P i |
| large | e - λ ′ a i | e λ ′ a i |
Recalling the expressions for the reflection and transmission coefficients obtained before, the reflection coefficient R 12 for the outgoing-potential case is modified to
$$\text{or use outgoing-posteria case is motionive } \\ R _ { 1 2 } & = \frac { ( \sigma _ { 2 } - \sigma _ { 1 } ) K _ { n } ( \lambda a _ { 1 } ) K _ { n } ^ { \prime } ( \lambda a _ { 1 } ) } { \sigma _ { 1 } I _ { n } ^ { \prime } ( \lambda a _ { 1 } ) K _ { n } ( \lambda a _ { 1 } ) - \sigma _ { 2 } I _ { n } ( \lambda a _ { 1 } ) K _ { n } ^ { \prime } ( \lambda a _ { 1 } ) } \\ & = \alpha _ { 1 } ^ { 2 } \frac { ( \sigma _ { 2 } - \sigma _ { 1 } ) \hat { K } _ { n } ( \lambda a _ { 1 } ) \hat { K } _ { n } ^ { \prime } ( \lambda a _ { 1 } ) } { \sigma _ { 1 } \hat { I } _ { n } ^ { \prime } ( \lambda a _ { 1 } ) \hat { K } _ { n } ( \lambda a _ { 1 } ) - \sigma _ { 2 } \hat { I } _ { n } ( \lambda a _ { 1 } ) \hat { K } _ { n } ^ { \prime } ( \lambda a _ { 1 } ) } \\ & = \alpha _ { 1 } ^ { 2 } \hat { R } _ { 1 2 }.$$
Similarly, it can be shown that the reflection coefficient R 21 for the standing-potential case is modified to R 21 = β R 2 1 ˆ 21 , and that the transmission coefficient T 12 for the outgoing-potential case and the transmission coefficient T 21 for the standing-potential case simply recover the original ones without any multiplicative factors, i.e. T 12 = ˆ T 12 and T 21 = ˆ T 21 .
Based on the above modifications for the reflection and transmission coefficients, generalized reflection and transmission coefficients for thre or more layers can be similarly modified. After some algebra, it can be shown that ˜ R i,i +1 = α R 2 i ˆ ˜ i,i +1 and ˜ R i +1 ,i = β R 2 i ˆ ˜ i +1 ,i , and that, for both the outgoing-potential and standing-potential cases, ˜ T ij = ˆ ˜ T ij .
In addition to generalized reflection and transmission coefficients, the factors M j and N i ± considered before are also required to compute the potential. The basic difference between the two types of coefficients is that M j involves two generalized reflection coefficients whereas N i ± involves only one generalized reflection coefficient. All these auxiliary coefficients can be redefined accordingly using rescaled reflection coefficients, i.e.,
$$M _ { j } = \left [ 1 - \widetilde { R } _ { j, j - 1 } \widetilde { R } _ { j, j + 1 } \right ] ^ { - 1 } = \left [ 1 - \beta _ { j - 1 } ^ { 2 } \alpha _ { j } ^ { 2 } \hat { \widetilde { R } } _ { j, j - 1 } \hat { \widetilde { R } } _ { j, j + 1 } \right ] ^ { - 1 },$$
$$N _ { i + } & = \left [ 1 - R _ { i, i - 1 } \widetilde { R } _ { i, i + 1 } \right ] ^ { - 1 } = \left [ 1 - \beta _ { i - 1 } ^ { 2 } \alpha _ { i } ^ { 2 } \hat { R } _ { i, i - 1 } \hat { \widetilde { R } } _ { i, i + 1 } \right ] ^ { - 1 },$$
$$N _ { i - } = \left [ 1 - R _ { i, i + 1 } \widetilde { R } _ { i, i - 1 } \right ] ^ { - 1 } = \left [ 1 - \beta _ { i - 1 } ^ { 2 } \alpha _ { i } ^ { 2 } \hat { R } _ { i, i + 1 } \hat { \widetilde { R } } _ { i, i - 1 } \right ] ^ { - 1 }.$$
## 2.5. Rescaled integrand
Nest step is to modify the full integrand using rescaled modified-cylindrical functions. Since there are four integrand expressions, depending of the relative position of ρ and ρ ′ , each case is considered separately. For Case 1 , there are four radial parameters of interest: a j -1 , ρ ′ , ρ , and a j . For convenience, we let a j -1 = a 1 , ρ ′ = a 2 , ρ = a 3 , and a j = a 4 so that a 1 < a 2 < a 3 < a 4 , and the integrand rewrites as
$$\begin{array} { c c c c c c } \, \cdot & \, \cdot & \, \cdot & \, \cdot & \, \cdot & \, \cdot \\ F _ { n } ( \rho, \rho ^ { \prime } ) & = \left [ K _ { n } ( \lambda \rho ) + I _ { n } ( \lambda \rho ) \widetilde { R } _ { j, j + 1 } \right ] \left [ I _ { n } ( \lambda \rho ^ { \prime } ) + K _ { n } ( \lambda \rho ^ { \prime } ) \widetilde { R } _ { j, j - 1 } \right ] M _ { j } \\ & = \left [ \beta _ { 2 } \alpha _ { 3 } \hat { K } _ { n } ( \lambda \rho ) + ( \beta _ { 2 } \alpha _ { 4 } ) ( \beta _ { 3 } \alpha _ { 4 } ) \hat { I } _ { n } ( \lambda \rho ) \hat { \widetilde { R } } _ { j, j + 1 } \right ] \left [ \hat { I } _ { n } ( \lambda \rho ^ { \prime } ) + ( \beta _ { 1 } \alpha _ { 2 } ) ^ { 2 } \hat { K } _ { n } ( \lambda \rho ^ { \prime } ) \hat { \widetilde { R } } _ { j, j - 1 } \right ] M _ { j } \\ & = \left [ A _ { 1 } \hat { K } _ { n } ( \lambda \rho ) + A _ { 2 } \hat { I } _ { n } ( \lambda \rho ) \hat { \widetilde { R } } _ { j, j + 1 } \right ] \left [ A _ { 3 } \hat { I } _ { n } ( \lambda \rho ^ { \prime } ) + A _ { 4 } \hat { K } _ { n } ( \lambda \rho ^ { \prime } ) \hat { \widetilde { R } } _ { j, j - 1 } \right ] M _ { j }, \end{array}$$
where β -1 2 = α 2 has been used. All multiplicative factors A 1 , A 2 , A 3 , A 4 have magnitudes no larger than one due to the boundness property discussed above.
Similarly, for Case 2 , there are four radial parameters of interest: a j -1 , ρ , ρ ′ , and a j . For convenience, we let a j -1 = a 1 , ρ = a 2 , ρ ′ = a 3 , and a j = a 4 so that a 1 < a 2 < a 3 < a 4 , and the integrand rewrites as
$$\L _ { n } ^ { \ } ( \rho, \rho ^ { \prime } ) & = \left [ I _ { n } ( \lambda \rho ) + K _ { n } ( \lambda \rho ) \widetilde { R } _ { j, j - 1 } \right ] \left [ K _ { n } ( \lambda \rho ^ { \prime } ) + I _ { n } ( \lambda \rho ^ { \prime } ) \widetilde { R } _ { j, j + 1 } \right ] M _ { j } \\ & = \left [ \beta _ { 2 } \alpha _ { 3 } \hat { I } _ { n } ( \lambda \rho ) + ( \beta _ { 1 } \alpha _ { 2 } ) ( \beta _ { 1 } \alpha _ { 3 } ) \hat { K } _ { n } ( \lambda \rho ) \hat { \widetilde { R } } _ { j, j - 1 } \right ] \left [ \hat { K } _ { n } ( \lambda \rho ^ { \prime } ) + ( \beta _ { 3 } \alpha _ { 4 } ) ^ { 2 } \hat { I } _ { n } ( \lambda \rho ^ { \prime } ) \hat { \widetilde { R } } _ { j, j + 1 } \right ] M _ { j } \\ & = \left [ B _ { 1 } \hat { I } _ { n } ( \lambda \rho ) + B _ { 2 } \hat { K } _ { n } ( \lambda \rho ) \hat { \widetilde { R } } _ { j, j - 1 } \right ] \left [ B _ { 3 } \hat { K } _ { n } ( \lambda \rho ^ { \prime } ) + B _ { 4 } \hat { I } _ { n } ( \lambda \rho ^ { \prime } ) \hat { \widetilde { R } } _ { j, j + 1 } \right ] M _ { j }.$$
Again, all multiplicative factors B 1 , B 2 , B 3 , B 4 magnitudes are bounded by one.
For Case 3 , there are six radial parameters of interest: a j -1 , ρ ′ , a j , a i -1 , ρ , and a i . For convenience, we let a j -1 = a 1 , ρ ′ = a 2 , a j = a 3 , a i -1 = a 4 , ρ = a 5 , and a i = a 6 so that so that a 1 < a 2 < a 3 ≤ a 4 < a 5 < a 6 , and the integrand rewrites as
$$& \Lambda _ { n } ( \rho, \rho ^ { \prime } ) = \left [ K _ { n } ( \lambda \rho ) + I _ { n } ( \lambda \rho ) \widetilde { R } _ { i, i + 1 } \right ] \left [ I _ { n } ( \lambda \rho ^ { \prime } ) + K _ { n } ( \lambda \rho ^ { \prime } ) \widetilde { R } _ { j, j - 1 } \right ] N _ { i + \widetilde { T } _ { j i } M _ { j } } \\ & \quad = \left [ \beta _ { 2 } \alpha _ { 5 } \hat { K } _ { n } ( \lambda \rho ) + ( \beta _ { 2 } \alpha _ { 6 } ) ( \beta _ { 5 } \alpha _ { 6 } ) \hat { I } _ { n } ( \lambda \rho ) \hat { \widetilde { R } } _ { i, i + 1 } \right ] \left [ \hat { I } _ { n } ( \lambda \rho ^ { \prime } ) + ( \beta _ { 1 } \alpha _ { 2 } ) ^ { 2 } \hat { K } _ { n } ( \lambda \rho ^ { \prime } ) \hat { \widetilde { R } } _ { j, j - 1 } \right ] N _ { i + \widetilde { T } _ { j i } M _ { j } } \\ & \quad = \left [ C _ { 1 } \hat { K } _ { n } ( \lambda \rho ) + C _ { 2 } \hat { I } _ { n } ( \lambda \rho ) \hat { \widetilde { R } } _ { i, i + 1 } \right ] \left [ C _ { 3 } \hat { I } _ { n } ( \lambda \rho ^ { \prime } ) + C _ { 4 } \hat { K } _ { n } ( \lambda \rho ^ { \prime } ) \hat { \widetilde { R } } _ { j, j - 1 } \right ] N _ { i + \widetilde { T } _ { j i } M _ { j } }.$$
All multiplicative factors C 1 , C 2 , C 3 , C 4 have magnitudes never greater than unity.
Finally, for Case 4 , there are again six radial parameters of interest: a i -1 , ρ , a i , a j -1 , ρ ′ , and a j . For convenience, we let a i -1 = a 1 , ρ = a 2 , a i = a 3 , a j -1 = a 4 , ρ ′ = a 5 , and a j = a 6 so that a 1 < a 2 < a 3 ≤ a 4 < a 5 < a 6 , and the integrand rewrites as
$$\begin{array} { c c c } \cdot & \quad \cdot & \quad \cdot \\ F _ { n } ( \rho, \rho ^ { \prime } ) & = \left [ I _ { n } ( \lambda \rho ) + K _ { n } ( \lambda \rho ) \widetilde { R } _ { i, i - 1 } \right ] \left [ K _ { n } ( \lambda \rho ^ { \prime } ) + I _ { n } ( \lambda \rho ^ { \prime } ) \widetilde { R } _ { j, j + 1 } \right ] N _ { i - } \widetilde { T } _ { j i } M _ { j } \\ & = \left [ \beta _ { 2 } \alpha _ { 5 } \hat { I } _ { n } ( \lambda \rho ) + ( \beta _ { 1 } \alpha _ { 2 } ) ( \beta _ { 1 } \alpha _ { 5 } ) \hat { K } _ { n } ( \lambda \rho ) \hat { \widetilde { R } } _ { i, i - 1 } \right ] \left [ \hat { K } _ { n } ( \lambda \rho ^ { \prime } ) + ( \beta _ { 5 } \alpha _ { 6 } ) ^ { 2 } \hat { I } _ { n } ( \lambda \rho ^ { \prime } ) \hat { \widetilde { R } } _ { j, j + 1 } \right ] N _ { i - } \widetilde { T } _ { j i } M _ { j } \\ & = \left [ D _ { 1 } \hat { I } _ { n } ( \lambda \rho ) + D _ { 2 } \hat { K } _ { n } ( \lambda \rho ) \hat { \widetilde { R } } _ { i, i - 1 } \right ] \left [ D _ { 3 } \hat { K } _ { n } ( \lambda \rho ^ { \prime } ) + D _ { 4 } \hat { I } _ { n } ( \lambda \rho ^ { \prime } ) \hat { \widetilde { R } } _ { j, j + 1 } \right ] N _ { i - } \widetilde { T } _ { j i } M _ { j }. \end{array}$$
Once more, all multiplicative factors D 1 , D 2 , D 3 , D 4 have magnitudes never greater than one.
## 2.6. Numerical integration
The electric potential expression includes a semi-infinite integral and an infinite series summation, as shown in (28). Therefore, truncation errors are inevitable and an error analysis should be made to ensure reliable results. In addition, the involved Sommerfeld-type integrals can be notoriously slowly convergent. A variety of extrapolation methods are applied here in order to accelerate the numerical integration, as described in more detail in Appendix B.
## 3. Results
A number of cases of practical interest for geophysical exploration are considered next. The results below are obtained on 2.6 GHz Opteron with 8 cores and 32 GB memory. For all cases below, the relative permittivity glyph[epsilon1] r and relative permeability µ r are set to one. The innermost layer has 6 ′′ radius and represents a mud-filled borehole. The outer layers represent the adjacent Earth formations, invasion zones, and/or casing layers. Each layer assumes different resistivity values R , where R = 1 /σ . As Fig. 6 illustrates, a current source emitting a DC current of 1 A is located at the Survey Electrode position. The electric potential is measured at two points: the Measurement Electrode 1 ( V 16 ′′ ) positioned 16 inches away from the source along the vertical direction and the Measurement Electrode 2 ( V 32 ′′ ) positioned 32 inches away from the source along the vertical direction. Both the source and two measurement electrodes are spaced 5 inches away from the z -axis and their azimuthal positions are the same ( φ = φ ′ = 0). The relevant error parameters as defined in the Appendix and used in the following are e tol = 10 -4 and e thr = 10 -4 . The electric potential and resistivity units used here are [ V ] and [Ω · m ], respectively. The results of the present algorithm are compared against those from the finite element method (FEM).
## 3.1. Logging simulation results
Case 1 corresponds to a homogeneous problem, where analytical solutions are available. As Table 3 shows, the new algorithm produces very accurate results with very fast computing time. This is very important for making feasible the solution of the inverse problem (i.e., determining the unknown resistivity profile from known electric potential at the received electrodes) using iterative methods predicated on repeated forward solves. Homogeneous Formation Results
(Rm = 1 Ω-m , Rf = 1 Ω-m)
Cross-section of the formation in ρ-z plane is shown.
Survey electrode transmits a DC
current of 1A.
Both mud resistivity and formation resistivity are 1 Ω-m.
Voltages at measure electrodes and voltage difference between measure
electrodes for this case are:
V\_ME1 [V]
0.17878839
V\_ME2 [V]
∆V [V]
1
Figure 6: Case 1 in the ρz -plane with R 1 = R 2 = 1.
Table 3: Comparison of electric potential for Case 1.
| | Analytical | FEM | Present Algorithm |
|-----------------|---------------------------------|---------------------------------|---------------------------------------------------|
| V 16 ′′ V 32 ′′ | 1.9581 × 10 - 1 9.7905 × 10 - 2 | 1.7878 × 10 - 1 9.4416 × 10 - 2 | 1.9580 × 10 - 1 (2 sec.) 9.7900 × 10 - 2 (2 sec.) |
| ∆ V | 9.7905 × 10 - 2 | 8.4371 × 10 - 2 | |
| | | | 9.7902 × 10 - 2 |
0.094416797
0.084371591
Cases 2 and 3 correspond to a two-layer problem with 1:5 contrast between the adjacent borehole and formation resistivities.
with One Radial Layer with One Radial Lay
(Rm = 1 Ω-m , Rf = 5 Ω-m)
Mud resistivity is Rm = 1 Ω-m
Formation resistivity is Rf = 5 Ω- m.
Voltages at measure electrodes and voltage difference between
measure electrodes for this case are:
V\_ME1 [V]
V\_ME2 [V]
∆V [V]
(Rm = 5 Ω-m , Rf = 1 Ω-
Mud resistivity is Rm = 5 Ω-m
Formation resistivity is Rf = 1 Ω- m.
Voltages at measure electrodes and voltage difference between
measure electrodes for this case are:
V\_ME1 [V]
V\_ME2 [V]
∆V [V]
0.16945594
0.090310919
0.079145021
4
5
Figure 7: Cases 2 and 3 in the ρz -plane: (a) Case 2 with R 1 = 1, R 2 = 5 and (b) Case 3 with R 1 = 5, R 2 = 1.
Table 4: Comparison of electric potential for Case 2.
| | FEM | Present Algorithm |
|---------|-----------------|--------------------------|
| V 16 ′′ | 9.5544 × 10 - 1 | 9.7802 × 10 - 1 (2 sec.) |
| V 32 ′′ | 5.4241 × 10 - 1 | 5.4981 × 10 - 1 (2 sec.) |
| ∆ V | 4.1303 × 10 - 1 | 4.2822 × 10 - 1 |
Table 5: Comparison of electric potential for Case 3.
| | FEM | Present Algorithm |
|-----------------|-----------------------------|--------------------------|
| V 16 ′′ V 32 ′′ | 1.6945 × 10 - 1 - 2 | 2.0533 × 10 - 1 (2 sec.) |
| ∆ V | 9.0310 × 10 7.9145 × 10 - 2 | 9.7677 × 10 - 2 (2 sec.) |
| | | 1.0766 × 10 - 1 |
Cases 4 and 5 include a highly conductive casing. As Table 7 shows, there is disagreement between the semi-analytical and FEM results w.r.t. the absolute value of the electric potentials, although the computed potential differences (voltage drop) between the two electrodes are very similar. This disagreement is easy to explain as a FEM mesh truncation effect. The electric current in this case flows primarily along the thin, highly conductive casing, which does not produce sufficient decay on the electric potential before before it reaches the mesh boundary. Consequently, the FEM result has a spurious potential offset. The FEM has difficulty in simulating this problem unless a mesh truncation treatment is included on both top and bottom boundaries and/or a very long mesh is used along the z direction. Notwithstanding such discrepancy, note that, in a resistivity logging context, the primary quantity of interest is the potential difference between electrodes.
Simulation Results with Casing
Simulation Results with Casing and Formation
(Rm = 1 Ω-m , Rc = 10 -8 Ω-m)
In this case, formation is replaced with (an infinite) casing.
Mud resistivity is Rm = 1 Ω-m
Casing resistivity is Rc = 10 -8 Ω- Electrode 2
m.
Voltages at measure electrodes Electrode 1
and voltage difference between 16'
measure electrodes for this case are:
V\_ME1 [V]
V\_ME2 [V]
∆V [V]
(Rm = 1 Ω-m , Rc = 10 -8 Ω-m, Rf = 2 Ω-m )
In this simulation, casing width is shrinked to an eight of an inch and a
formation layer that fills the rest of the computational domain is added.
Mud resistivity is Rm = 1 Ω-m
Casing resistivity is Rc = 10 -8 Ω-m.
Formation resistivity is Rf = 2 Ω- m.
Voltages at measure electrodes and voltage difference between measure
electrodes for this case are:
V\_ME1 [V]
0.22658744E-03
V\_ME2 [V]
∆V [V]
7
6
Figure 8: Cases 4 and 5 in the ρz -plane: (a) Case 4 with R 1 = 1, R 2 = 10 -8 and (b) Case 5 with R 1 = 1, R 2 = 10 -8 , R 3 = 2.
Table 6: Comparison of electric potential for Case 4.
| | FEM | Present Algorithm |
|---------|-----------------|--------------------------|
| V 16 ′′ | 1.2980 × 10 - 4 | 1.3873 × 10 - 4 (2 sec.) |
| V 32 ′′ | 2.1372 × 10 - 7 | 2.1415 × 10 - 7 (2 sec.) |
| ∆ V | 1.2959 × 10 - 4 | 1.3852 × 10 - 4 |
Table 7: Comparison of electric potential for Case 5.
| | FEM | Present Algorithm |
|---------|-----------------|--------------------------|
| V 16 ′′ | 2.2658 × 10 - 4 | 1.7241 × 10 - 3 (3 sec.) |
| V 32 ′′ | 9.6331 × 10 - 5 | 1.5885 × 10 - 3 (2 sec.) |
| ∆ V | 1.3025 × 10 - 4 | 1.3562 × 10 - 4 |
0.96331022E-04
0.13025642E-03
As Fig. 9 illustrates, Cases 6 and 7 feature two layers besides the borehole, where the mud layer can represent an invasion zone with a resistivity between those of the borehole and the outer formation.
Simulation Results with Two Formation Layers
Simulation Results with Two Formation Laye
(Rm = 1 Ω-m , Rf1 = 1.5 Ω-m, Rf2 = 2 Ω-m )
(Rm = 1 Ω-m , Rf1 = 1.5 Ω-m, Rf2 = 2 Ω-
In this simulation, casing in the previous slide is replaced with a resistive layer of
resistivity Rf1 = 1.5 Ω-m.
Mud resistivity is Rm = 1 Ω-m Measure
Resistivity of second layer is Rf1 = 1.5 Ω- e 2
m. 16'
Resistivity of second layer is Rf2 = 2
Ω- m.
Voltages at measure electrodes and voltage difference between measure urvey
electrodes for this case are:
V\_ME1 [V]
V\_ME2 [V]
∆V [V]
8
9
Figure 9: Cases 6 and 7 in the ρz -plane: (a) Case 6 with R 1 = 1, R 2 = 1 5, . R 3 = 2 and (b) Case 7 with R 1 = 1, R 2 = 1 5, . R 3 = 2.
Table 8: Comparison of electric potential for Case 6.
| | FEM | Present Algorithm |
|---------|-----------------|--------------------------|
| V 16 ′′ | 3.7097 × 10 - 1 | 3.9061 × 10 - 1 (2 sec.) |
| V 32 ′′ | 1.9796 × 10 - 1 | 2.0245 × 10 - 1 (2 sec.) |
| ∆ V | 1.7301 × 10 - 1 | 1.8816 × 10 - 1 |
Table 9: Comparison of electric potential for Case 7.
| | FEM | Present Algorithm |
|---------|-----------------|--------------------------|
| V 16 ′′ | 3.6077 × 10 - 1 | 3.8160 × 10 - 1 (2 sec.) |
| V 32 ′′ | 1.9694 × 10 - 1 | 2.0156 × 10 - 1 (2 sec.) |
| ∆ V | 1.6383 × 10 - 1 | 1.8003 × 10 - 1 |
## 3.2. Electric potential maps
Plots of the spatial distributions of the electric potential of Cases 2-5 are provided next. Since the potential varies by many orders of magnitude near the electrodes, a log-scale is used; i.e. we plot the quantity 10 log 10 | V | . Fig. 10 depicts the potential distributions using a three-dimensional view. Fig. 11 and Fig. 12 show the cross-sections of the potential distributions at the z = 16 ′′ and z = 32 ′′ planes, respectively, while Fig. 13 shows the cross-sections at the y = 0 ′′ plane. In all figures, thicker black lines represent interfaces between cylindrical layers and thinner black lines represent equipotential contours. Note that the third layer of Case 5 is too thin for visualization in these plots.
In this simulation, width of the first formation layer is expanded to 3'.
Mud resistivity is Rm = 1 Ω-m
Resistivity of second layer is Rf1 = 1.5 Ω- m.
Resistivity of second layer is Rf2 = 2
Ω- m.
Voltages at measure electrodes and voltage difference between measure
electrodes for this case are:
V\_ME1 [V]
0.36077656
V\_ME2 [V]
∆V [V]
0.19694043
0.16383612
(a)
(b)
Figure 10: Electric potential around the current electrode: (a) Case 2, (b) Case 3, (c) Case 4, and (d) Case 5.
Case5
Top : 0.
Botto
Left : 0
Right :
Case7-
Top : 0.
Botto
Left : 0
Right :
Figure 11: Electric potential at z = 16 ′′ plane around the current electrode: (a) Case 2, (b) Case 3, (c) Case 4, and (d) Case 5.
Case5 - z
Top : 0.2
Bottom :
Left : 0.0
Right : 4.
Case7 - z16
Top : 0.2
Bottom : 3.
Left : 0.0
Right : 4.6 o
Figure 12: Electric potential at z = 32 ′′ plane around the current electrode: (a) Case 2, (b) Case 3, (c) Case 4, and (d) Case 5.
Case5 - z
Top : 0.2
Bottom :
Left : 0.0
Right : 4.
Case7 - z32
Top : 0.2
Bottom : 3.
Left : 0.0
Right : 4.6 o
Figure 13: Electric potential at y = 0 ′′ plane around the current electrode: (a) Case 2, (b) Case 3, (c) Case 4, and (d) Case 5.
Case5 - y
Top : 0.2
Bottom :
Left : 0.0
Right : 4.
Case7 - y
Top : 0.2
Bottom :
Left : 0.0
Right : 4.
## 4. Conclusion
We have introduced a stable semianalytical formulation to compute the potential due to arbitrarylocated electrodes in a cylindrically layered media, where the physical parameters (viz., layer resistivity) can vary by many order of magnitude. Stability was achieved by a rescaling of the modified Bessel functions and subsequent manipulation of the integrand expressions to avoid underflow and overflow problems under double-precision arithmetic. Several extrapolation methods to compute the involved Sommerfeld integrals were also considered and evaluated. The resulting algorithm was verified in a number of cases relevant to borehole geophysics. The fast speed and robustness of the algorithm makes it also quite suited as a forward solver engine for the inverse roblem, where the formation resistivity needs to be estimated from a limited numberof computed (measured) potential data. The same basic algorithm can be applied to other steadystate diffusion problems obeying Poisson's equation with discontinous coefficients in cylindrically layered geometries.
## Acknowledgments
The authors are grateful to Halliburton Energy Services for the permission to publish this work and to Dr. Baris Guner for generating validation data.
## Appendix A: Folding azimuth summation
The computation of potential in (28) involves infinite series over the azimuth mode n . For an arbitrary order ν ≥ 0 [34], we have
$$I _ { - \nu } ( z ) & = I _ { \nu } ( z ) + \frac { 2 } { \pi } \sin ( \nu \pi ) K _ { \nu } ( z ), & ( 4 3 \tt a ) \\ \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
$$K _ { - \nu } ( z ) & = K _ { \nu } ( z ).$$
If the order is a positive integer, ν = n , the above reduces to I -n ( z ) = I n ( z ) and K -n ( z ) = K n ( z ). Consequently, we can write
$$\cdot \quad \vec { \psi } _ { i } & = \frac { \mathcal { I } } { 2 \pi ^ { 2 } \sigma _ { j } } \sum _ { n = - \infty } ^ { \infty } e ^ { i n ( \phi - \phi ^ { \prime } ) } \int _ { 0 } ^ { \infty } F _ { n } ( \rho, \rho ^ { \prime } ) \cos ( \lambda ( z - z ^ { \prime } ) ) d \lambda \\ & = \frac { \mathcal { I } } { 2 \pi ^ { 2 } \sigma _ { j } } \int _ { 0 } ^ { \infty } \left [ F _ { 0 } ( \rho, \rho ^ { \prime } ) + 2 \sum _ { n = 1 } ^ { \infty } F _ { n } ( \rho, \rho ^ { \prime } ) \cos ( n ( \phi - \phi ^ { \prime } ) ) \right ] \cos ( \lambda ( z - z ^ { \prime } ) ) d \lambda.$$
## Appendix B: Extrapolation methods and convergence study
Among numerous extrapolation methods, some popular ones for Sommerfeld-type integrals are very briefly revisited here. For more details, the reader can refer, e.g., to [15, 33]. Before a given sequence is extrapolated, a Sommerfeld-type integral can be divided into a number of subintervals as
$$S = \int _ { a } ^ { \infty } g ( \lambda ) p ( \lambda ) d \lambda = \sum _ { i = 0 } ^ { \infty } \int _ { \lambda _ { i - 1 } } ^ { \lambda _ { i } } g ( \lambda ) p ( \lambda ) d \lambda = \sum _ { i = 0 } ^ { \infty } u _ { i },$$
where g λ ( ) is an exponentially decaying part and p λ ( ) is an oscillatory part. This is called the partitionextrapolation approach [28]. In general, the remainders are defined as
$$r _ { n } = S _ { n } - S = - \int _ { \lambda _ { n } } ^ { \infty } g ( \lambda ) p ( \lambda ) d \lambda.$$
Furthermore, the remainders are assumed to feature Poincar´-type asymptotic expansions [15] of the form e
$$r _ { n } \sim \omega _ { n } \sum _ { i = 0 } ^ { \infty } a _ { i } \lambda _ { n } ^ { - i }, \quad n \to \infty,$$
where ω n is the remainder estimate and a i are associated coefficients. The estimates ω n play an important role in the extrapolation and can be obtained analytically or numerically. The coefficients a i are unknowns but they are not necessary for the extrapolation itself. In our case, g λ ( ) in (45) can be asymptotically expressed as
$$g ( \lambda ) \sim \frac { e ^ { - \lambda | \rho - \rho ^ { \prime } | } } { \lambda } \sum _ { i = 0 } ^ { \infty } \frac { c _ { i } } { \lambda ^ { i } },$$
where c i are arbitrary constants. Furthermore, p λ ( ) = cos( λ z ( -z ′ )) with half-period equal to π/ ρ | -ρ ′ | . After some algebra, it can be shown that the remainder estimates write as
$$\omega _ { n } = \frac { ( - 1 ) ^ { n + 1 } } { \lambda _ { n } } e ^ { - \frac { n \pi | \rho - \rho ^ { \prime } | } { | \varepsilon - \varepsilon ^ { \prime } | } }.$$
We list next three popular extrapolation methods for a given sequence, { S n } = S , S 0 1 , S 2 , · · · , S n , where we consider S n = S (0) n .
## (i) Euler transformation
$$S _ { n } ^ { ( k + 1 ) } = \frac { 1 } { 2 } \left ( S _ { n } ^ { ( k ) } + S _ { n + 1 } ^ { ( k ) } \right ), \quad n, k \geq 0$$
The best approximation in this case is S ( k ) 0 | k = n , and this choice is most effective for logarithmic alternating sequences.
## (ii) Iterative Aitken transformation
$$S _ { n } ^ { ( k + 1 ) } = S _ { n } ^ { ( k ) } - \frac { [ \Delta S _ { n } ^ { ( k ) } ] ^ { 2 } } { \Delta ^ { 2 } S _ { n } ^ { ( k ) } }, \quad n, k \geq 0$$
Note that ∆ S ( k ) n = S ( k ) n +1 -S ( k ) n and ∆ 2 S ( k ) n = S ( k ) n +2 -2 S ( k ) n +1 + S ( k ) n . Obviously, this is a nonlinear transformation and it can be applied to both linear monotone and alternating sequences. When an odd number of sequences is given, the best approximation is S ( k ) 0 | k = n . For an even number, the best approximation is S ( k ) 1 | k = n .
## (iii) Weighted-averages method
$$S _ { n } ^ { ( k + 1 ) } = \frac { S _ { n } ^ { ( k ) } + \eta _ { n } ^ { ( k ) } S _ { n + 1 } ^ { ( k ) } } { 1 + \eta _ { n } ^ { ( k ) } }, \quad n, k \geq 0,$$
where η ( k ) is the weight and defined as
$$\eta ^ { ( k ) } = - \frac { \omega _ { n } } { \omega _ { n + 1 } } = \frac { \lambda _ { n + 1 } } { \lambda _ { n } } e ^ { \frac { \pi | \rho - \rho ^ { \prime } | } { | s - z ^ { \prime } | } }.$$
The weighted-averages method [18, 19] can be regarded as generalized Euler transformation since η = 1 recovers the Euler transformation. The relative power of this method, compared to the two other methods, comes from using remainder estimates. As pointed out in [15], the weighted-averages method is quited suited for extrapolating Sommerfeld-type integrals.
Figure 14: Two scenarios of the integrand: (a) Type 1 with | ρ -ρ ′ | = 0 05, . | z -z ′ | = 0 01 and (b) Type 2 . with | ρ -ρ ′ | = 0 05, . | z -z ′ | = 0 5. .
To evaluate the integral (45), the subinterval length q should be first determined. As suggested in [15], the half-period of the oscillating part of the integrand is a good choice for q because it makes the sequences alternating; i.e., q = π/ z | -z ′ | . However, this is not appropriate here in some circumstances. Let us consider the two scenarios depicted in Fig. 14. The two figures show the behavior of g λ ( ) and p λ ( ) for different combinations of | ρ -ρ ′ | and | z -z ′ | as λ increases. For better visualization, the functions are normalized by their respective maxima. Type 1 integrand occurs when | ρ -ρ ′ | > z | -z ′ | and shows linear monotone convergence. On the other hand, Type 2 integrand occurs when | ρ -ρ ′ | < z | -z ′ | and shows logarithmic alternating convergence. Therefore, such an expression for q is not appropriate for Type 1 because the integrand is near zero before the first half-period comes. For the Type 1 integrand, a different subinterval length can be defined as q = π/ ρ | -ρ ′ | , so that the subinterval length can be written in general as
$$q = \frac { \pi } { \max ( | \rho - \rho ^ { \prime } |, | z - z ^ { \prime } | ) }.$$
Once q is determined, it becomes necessary to determine how many subintervals are required to reach sufficient convergence. To do so, the relative error below is defined
$$e _ { i } = \frac { | T ( S _ { i + 1 } ) - T ( S _ { i } ) | } { | T ( S _ { i + 1 } ) | },$$
where T S ( i ) is the extrapolated (transformed) value for given sequences, { S i } = S , S 0 1 , · · · , S i . If e i is less than a given error tolerance e tol , the sequence is stopped and the number of subintervals is determined. Next, the number of quadrature points per the subinterval is increased until the relative error between two adjacent iterations meets the desired criterion. To distinguish it from e tol , the criterion in this step is denoted error threshold e thr . The same step is repeated for determining the number of orders. The overall procedure is schematically depicted in Fig. 15.
Some convergence tests are performed next to evaluate the numerical integration procedures above. The domain is assumed homogeneous because exact (analytical) solutions are available as reference. Three cases of source/observation distances are considered.
Case 1 : | ρ -ρ ′ | = 0 001 m, . | z -z ′ | = 0 1 m .
$$\text{Case 2} \, \colon \, | \rho - \rho ^ { \prime } | = \ 0. 1 \, m, \ | z - z ^ { \prime } | = 0. 1 \, m$$
$$\text{Case } 3 \, \colon \, \stackrel { \dots } { | \rho - \dots } \rho ^ { \prime } | = \ \ 1 0 \, m, \ \ | z - z ^ { \prime } | = 0. 1 \, m$$
Figure 15: Flowchart of the procedure for computation of electric potential using extrapolation methods.
In each case, the three extrapolation methods mentioned before are compared. The excitation current magnitude and the medium resistivity are both set to one. The relevant error parameters are e tol = 10 -6 and e thr = 10 -4 . The smaller e tol is chosen to examine the effect of the methods on the number of subintervals. Tables 10, 11, and 12 compare the results, where the first rows represent the number of subintervals needed to achieve convergence in terms of e tol and the second rows represent the relative error against the analytical solution. As Table 10 shows, the iterative Aitken method does not work for Case 1 because it corresponds to a Type 2 integrand with logarithmic alternating convergence. The Euler transformation works well for all cases, but it can be seen that the weighted-averages method provides best results, corroborating the conclusions stated in [15].
Table 10: Comparison of the three extrapolation methods for Case 1.
| | Euler | Aitken | weighted-averages |
|-------------------|-----------------|----------|---------------------|
| # of subintervals | 16 | > 100 | 10 |
| Relative error | 1.9990 × 10 - 6 | N. A. | 3.5863 × 10 - 7 |
Table 11: Comparison of the three extrapolation methods for Case 2.
| | Euler | Aitken | weighted-averages |
|-------------------|-----------------|-----------------|---------------------|
| # of subintervals | 13 | 13 | 4 |
| Relative error | 1.4131 × 10 - 6 | 1.1921 × 10 - 6 | 1.4058 × 10 - 6 |
Table 12: Comparison of the three extrapolation methods for Case 3.
| | Euler | Aitken | weighted-averages |
|-------------------|-----------------|-----------------|---------------------|
| # of subintervals | 13 | 13 | 5 |
| Relative error | 4.9227 × 10 - 7 | 4.9227 × 10 - 7 | 4.9799 × 10 - 7 |
## References
- [1] Michael Carley. Numerical solution of the modified Bessel equation. IMA Journal of Numerical Analysis , 33(3):1048-1062, Jul. 2013.
- [2] Weng Cho Chew. Waves and Fields in Inhomogeneous Media . IEEE Press Series on Electromagnetic Waves. IEEE Press, New York, 1995.
- [3] J. A. Doetsch, I. Coscia, S. Greenhalgh, N. Linde, A. Green, and T. G¨nther. u The borehole-fluid effect in electrical resistivity imaging. Geophysics , 75(4):F107-F114, Jul.-Aug. 2010.
- [4] D. Drahos. Electrical modeling of the inhomogeneous invaded zone. Geophysics , 49(10):1580-1585, Oct. 1984.
- [5] Darwin V. Ellis and Julian M. Singer. Well Logging for Earth Scientists . Springer, Dordrecht, The Netherlands, second edition, 2007.
- [6] Guo-Xin Fan, Qing Huo Liu, and Sean P. Blanchard. 3-D numerical mode-matching (NMM) method for resistivity well-logging tools. IEEE Transactions on Antennas and Propagation , 48(10):1544-1552, Oct. 2000.
- [7] S. Gianzero and B. Anderson. An integral transform solution to the fundamental problem in resistivity logging. Geophysics , 47(6):946-956, 1982.
- [8] Y. K. Hue, F. L. Teixeira, L. San Martin, and M. S. Bittar. Three-dimensional simulation of eccentric LWD tool response in boreholes through dipping formations. IEEE Transactions on Geoscience and Remote Sensing , 43(2):257-268, Feb. 2005.
- [9] Yik-Kiong Hue and F. L. Teixeira. Numerical mode-matching method for tilted-coil antennas in cylindrically layered anisotropic media with multiple horizontal beds. IEEE Transactions on Geoscience and Remote Sensing , 45(8):24512462, Aug. 2007.
- [10] Hwa Ok Lee, Fernando L Teixeira, Luis E San Martin, and Michael S Bittar. Numerical modeling of eccentered LWD borehole sensors in dipping and fully anisotropic earth formations. IEEE Transactions on Geoscience and Remote Sensing , 50(3):727-735, Mar. 2012.
- [11] Qing-Huo Liu, Barbara Anderson, and Weng Cho Chew. Modeling low-frequency electrode-type resistivity tools in invaded thin beds. IEEE Transactions on Geoscience and Remote Sensing , 32(3):494-498, May 1994.
- [12] Sixin Liu and Motoyuki Sato. Electromagnetic logging technique based on borehole radar. IEEE Transactions on Geoscience and Remote Sensing , 40(9):2083-2092, Sep 2002.
- [13] J. R. Lovell and W. C. Chew. Effect of tool eccentricity on some electrical well-logging tools. IEEE Transactions on Geoscience and Remote Sensing , 28(1):127-136, Jan. 1990.
- [14] John Richard Lovell. Finite element methods in resistivity logging . Ph.d. dissertation, Delft University, Delft, The Netherlands, 1993.
- [15] K. A. Michalski. Extrapolation methods for Sommerfeld integral tails. IEEE Transactions on Antennas and Propagation , 46(10):1405-1418, Oct. 1998.
- [16] H. Moon, F. L. Teixeira, and B. Donderici. Stable pseudoanalytical computation of electromagnetic fields from arbitrarilyoriented dipoles in cylindrically stratified media. Journal of Computational Physics , 273:118-142, 2014.
- [17] J. H. Moran and S. Gianzero. Effects of formation anisotropy on resistivity-logging measurements. Geophysics , 44(7): 1266-1286, Jul. 1979.
- [18] JR Mosig and FE Gardiol. Analytical and numerical techniques in the Green's function treatment of microstrip antennas and scatterers. In IEE Proceedings H (Microwaves, Optics and Antennas) , volume 130, pages 175-182. IET, 1983.
- [19] Juan Mosig. The weighted averages algorithm revisited. IEEE Transactions on Antennas and Propagation , 60(4):20112018, Apr. 2012.
- [20] M. J. Nam, D. Pardo, C. Torres-Verd´ ın, S. Hwang, K. G. Park, and C. Lee. Simulation of eccentricity effects on shortand long-normal logging measurements using a Fourier-hp-finite-element method. Exploration Geophysics , 41(1):118-127, 2010.
- [21] Marcela S. Novo, Luiz C. da Silva, and F. L. Teixeira. Finite volume modeling of borehole electromagnetic logging in
- 3-D anisotropic formations using coupled scalar-vector potentials. IEEE Antennas and Wireless Propagation Letters , 6: 549-552, 2007.
- [22] Marcela S. Novo, Luiz C. da Silva, and Fernando L. Teixeira. Three-dimensional finite-volume analysis of directional resistivity logging sensors. IEEE Transactions on Geoscience and Remote Sensing , 48(3):1151-1158, Mar. 2010.
- [23] D. Pardo, L. Demkowicz, C. Torres-Verd´ ın, and M. Paszynski. Two-dimensional high-accuracy simulation of resistivity logging-while-drilling (LWD) measurements using a self-adaptive goal-oriented hp finite element method. SIAM Journal on Applied Mathematics , 66(6):2085-2106, Oct. 2006.
- [24] D. Pardo, C. Torres-Verd´ ın, and L. F. Demkowicz. Simulation of multifrequency borehole resistivity measurements through metal casing using a goal-oriented hp finite-element method. IEEE Transactions on Geoscience and Remote Sensing , 44 (8):2125-2134, Aug. 2006.
- [25] Zhengyong Ren and Jingtian Tang. 3D direct current resistivity modeling with unstructured mesh by adaptive finiteelement method. Geophysics , 75(1):H7-H17, Jan.-Feb. 2010.
- [26] Yutaka Sasaki. 3-D resistivity inversion using the finite-element method. Geophysics , 59(12):1839-1848, Dec. 1994.
- [27] William Ralph Smythe. Static and Dynamic Electricity . International Series in Pure and Applied Physics. McGraw-Hill, New York, second edition, 1950.
- [28] W. Squire. Partition-extrapolation methods for numerical quadratures. International Journal of Computer Mathematics , 5(1):81-91, 1975.
- [29] W. M. Telford, L. P. Geldart, and Robert E. Sheriff. Applied Geophysics . Cambridge University Press, Cambridge England and New York, second edition, 1990.
- [30] J. R. Wait and K. R. Umashankar. Analysis of the earth resistivity response of buried cables. Pure and Applied Geophysics , 117(4):711-742, 1978.
- [31] James R. Wait. Geo-Electromagnetism . Academic Press, New York, 1982.
- [32] G. N. Watson. A Treatise on the Theory of Bessel Functions . Cambridge Mathematical Library. Cambridge University Press, New York, second edition, 1995.
- [33] Ernst Joachim Weniger. Nonlinear sequence transformations for the acceleration of convergence and the summation of divergent series. Computer Physics Reports , 10:189-371, 1989.
- [34] Shanjie Zhang and Jian-Ming Jin. Computation of Special Functions . John Wiley, New York, 1996.
- [35] Z. Zhang and Z. Zhou. Real-time quasi-2-D inversion of array resistivity logging data using neural network. Geophysics , 67(2):517-524, Mar.-Apr. 2002. | 10.1016/j.jcp.2014.10.015 | [
"Haksu Moon",
"Fernando L. Teixeira",
"Burkay Donderici"
] | 2014-08-01T22:31:30+00:00 | 2014-08-01T22:31:30+00:00 | [
"physics.comp-ph"
] | Computation of potentials from current electrodes in cylindrically stratified media: A stable, rescaled semi-analytical formulation | We present an efficient and robust semi-analytical formulation to compute the
electric potential due to arbitrary-located point electrodes in
three-dimensional cylindrically stratified media, where the radial thickness
and the medium resistivity of each cylindrical layer can vary by many orders of
magnitude. A basic roadblock for robust potential computations in such
scenarios is the poor scaling of modified-Bessel functions used for computation
of the semi-analytical solution, for extreme arguments and/or orders. To
accommodate this, we construct a set of rescaled versions of modified-Bessel
functions, which avoids underflows and overflows in finite precision
arithmetic, and minimizes round-off errors. In addition, several extrapolation
methods are applied and compared to expedite the numerical evaluation of the
(otherwise slowly convergent) associated Sommerfeld-type integrals. The
proposed algorithm is verified in a number of scenarios relevant to geophysical
exploration, but the general formulation presented is also applicable to other
problems governed by Poisson equation such as Newtonian gravity, heat flow, and
potential flow in fluid mechanics, involving cylindrically stratified
environments. |
1408.0317v1 | ## Some sample path properties of multifractional Brownian motion
## Paul Balança
e-mail: [email protected]
École Centrale Paris Laboratoire MAS, ECP Grande Voie des Vignes 92295 Châtenay-Malabry, France
Abstract: The geometry of the multifractional Brownian motion (mBm) is known to present a complex and surprising form when the Hurst function is greatly irregular. Nevertheless, most of the literature devoted to the subject considers sufficiently smooth cases which lead to sample paths locally similar to a fractional Brownian motion (fBm). The main goal of this paper is therefore to extend these results to a more general frame and consider any type of continuous Hurst function. More specifically, we mainly focus on obtaining a complete characterization of the pointwise Hölder regularity of the sample paths, and the Box and Hausdorff dimensions of the graph. These results, which are somehow unusual for a Gaussian process, are illustrated by several examples, presenting in this way different aspects of the geometry of the mBm with irregular Hurst functions.
AMS 2000 subject classifications: 60G07, 60G17, 60G22, 60G44.
Keywords and phrases: 2-microlocal analysis, Box dimension, Hausdorff dimension, Hölder regularity, multifractional Brownian motion.
## 1. Introduction
↦
The multifractional Brownian motion (mBm) has been independently introduced by Peltier and Lévy Véhel [29] and Benassi et al. [6] as a natural extension of the well-known fractional Brownian motion (fBm). The main idea behind these two works was to drop the stationary assumption on the process, and allow the Hurst exponent to change as time passes. In this way, the mBm is parametrized by a function t → H t ( ), usually continuous, and happens to be an interesting stochastic model for non-stationary phenomena (e.g. signal processing, traffic on internet, terrain modelling, . . . ).
Since its introduction, several authors have investigated sample path properties of the mBm. For instance, Benassi et al. [6] and Peltier and Lévy Véhel [29] have respectively studied the law of iterated logarithm and the Hölder regularity of its trajectories. In the latter, the Box and Hausdorff dimensions of the graph have also been determined. Moreover, the fine covariance structure have been analysed and refined by Ayache et al. [3], Herbin [15] and Stoev and Taqqu [35]. Finally, the local time and the Hausdorff dimension of the level sets have recently been considered by Boufoussi et al. [10]. Several multiparameter extensions, which generalise the classic Lévy fractional Brownian motion and fractional Brownian sheet, have also been introduced and investigated by Herbin [15] Meerschaert et al. [24] and Ayache et al. [4].
Fine geometric properties obtained in the aforementioned works usually rely on a key as-
sumption H 0 on the Hurst function:
$$H \text{ is a } \beta \text{-Holder continuous function such that } \sup _ { t \in R } H ( t ) < \beta. \quad \quad ( \mathcal { H } _ { 0 } )$$
Owing to this hypothesis, the study of the H 0 -mBm is usually easier and leads to results closely related to their counterparts on fractional Brownian motion. This structure of the mBm is due to the form of its covariance which, under H 0 , is locally equivalent to fBm's one with Hurst exponent H t ( ):
$$\forall u, v \in B ( t, \rho ) ; \ \mathbb { E } \left [ ( X _ { u } - X _ { v } ) ^ { 2 } \right ] \asymp | u - v | ^ { 2 H ( t ) },$$
for any t ∈ R \ { 0 } .
The study of the more general multifractional Brownian motion, i.e. when the assumption H 0 does not hold, has only been recently considered by Herbin [15] and Ayache [1]. Both have investigated the Hölder regularity of the sample paths, and more precisely the pointwise and local Hölder exponents, which are defined by
$$\forall t \in { \mathbf R } ; \quad \alpha _ { X, t } = \sup \left \{ \alpha \, \colon \lim \sup _ { \rho \to 0 } \sup _ { u, v \in B ( t, \rho ) } \frac { | X _ { u } - X _ { v } | } { \rho ^ { \alpha } } < \infty \right \} \quad \quad ( 1. 1 )$$
and
$$\forall t \in { \mathbf R } ; \quad \widetilde { \alpha } _ { X, t } = \sup \left \{ \alpha \, \colon \lim \sup _ { \rho \to 0 } \sup _ { u, v \in B ( t, \rho ) } \frac { | X _ { u } - X _ { v } | } { | u - v | ^ { \alpha } } < \infty \right \}.$$
These two exponents aim to characterise the local Hölder behaviour of the trajectories at t , by respectively comparing its oscillations to a power of ρ or | u -v | . For instance, it is wellknown that a fractional Brownian motion B H satisfies with probability one and for all t ∈ R , α B H ,t = ˜ α B H ,t = H .
The general form of the Hölder regularity of the multifractional Brownian motion has first been obtained by Herbin [15], proving that for all t > 0
$$\alpha _ { X, t } = H ( t ) \wedge \alpha _ { H, t } \quad \text{and} \quad \widetilde { \alpha } _ { X, t } = H ( t ) \wedge \widetilde { \alpha } _ { H, t } \quad \text{a.s.}$$
The previous result only holds for a fixed t > 0, and thus not uniformly on the sample paths. We also note that the geometric properties of a generic mBm are not a simple extension of those on fBm, but display a more complicated structure where the fine geometry of the Hurst function intervene. Ayache [1] has recently extended this study, obtaining a uniform characterisation of the local Hölder exponent and proving that its pointwise exponent can behave randomly as time passes. The resulting process, with unusual sample path properties, is sometimes called the irregular multifractional Brownian motion in the literature.
The main goal of this work is therefore to push further the study of this irregular mBm. In Section 2, we first discuss the existence of an alternative deterministic representation of the fractional Brownian field, which will then be used to study more precisely the geometric properties of the mBm. Section 3 is devoted to the 2-microlocal and Hölder regularity of the trajectories (Theorem 3.1 and Proposition 3.2), and therefore extends the results obtained by Herbin [15] and Ayache [1]. Hausdorff and Box dimensions of the graph, and images of fractal sets, are investigated in Section 4 (Theorems 4.1, 4.2 and 4.3), where further connections with
the geometry of the Hurst function are presented. Appendix 5 gathers a few deterministic results related to the 2-microlocal frontier and the fractal dimensions which are used along this work. Finally, several examples are also given in these different sections to illustrate the main results and some particular aspects of the geometry of the irregular multifractional Brownian motion.
## 2. Deterministic representation
The question of the equivalence of the different mBm representations is known to be non-trivial. On one side, Peltier and Lévy Véhel [29] have introduced the mBm as the following stochastic integral
$$\forall t \in \mathbf R ; \ \ X _ { t } = \frac { 1 } { \Gamma ( H ( t ) + \frac { 1 } { 2 } ) } \int _ { \mathbf R } \left [ ( t - u ) _ { + } ^ { H ( t ) - 1 / 2 } - ( - u ) _ { + } ^ { H ( t ) - 1 / 2 } \right ] \mathrm d W _ { u }.$$
Whereas on the other hand, Benassi et al. [6] have considered a real-harmonizable definition:
$$\forall t \in \mathbf R ; \ \widetilde { X } _ { t } = \int _ { \mathbf R } \frac { e ^ { i u t } - 1 } { | u | ^ { H ( t ) + 1 / 2 } } \, \mathrm d \widehat { W } _ { u }.$$
Stoev and Taqqu [35] have proved that the covariance structure is slightly different between these two definitions. Considering this fact, they have suggested to name multifractional Brownian motion any process X which has the following form
$$\forall t \in \mathbf R ; \ \ X _ { t } = a ^ { + } B ^ { + } ( t, H ( t ) ) + a ^ { - } B ^ { - } ( t, H ( t ) ),$$
↦
where ( a + , a -) ∈ R 2 \ { 0 0 , } and the fractional Brownian fields ( t, H ) → B ± ( t, H ) are defined by
$$B ^ { \pm } ( t, H ) = \frac { 1 } { \Gamma ( H + \frac { 1 } { 2 } ) } \int _ { \mathbf R } \left [ ( t - u ) _ { \pm } ^ { H - 1 / 2 } - ( - u ) _ { \pm } ^ { H - 1 / 2 } \right ] \mathrm d W _ { u },$$
for all t ∈ R and H ∈ (0 , 1).
In the next proposition, we present an alternative deterministic representation for these fractional Brownian fields which will be very useful for the study of sample path properties.
Proposition 2.1. Suppose ( B t ) t ∈ R is a Brownian motion such that B t := W ([0 , t ]) , where W is a Wiener measure. Then, for all t ∈ R and any H ∈ (0 1) , , the fractional Brownian field B ± ( t, H ) is almost surely equal to:
- · if ( t, H ) ∈ R × ( 1 2 , 1 ) ,
$$B ^ { \pm } ( t, H ) = \frac { \pm 1 } { \Gamma ( H - \frac { 1 } { 2 } ) } \int _ { \mathbf R } B _ { u } \left [ ( t - u ) _ { \pm } ^ { H - 3 / 2 } - ( - u ) _ { \pm } ^ { H - 3 / 2 } \right ] \mathrm d u. \quad \quad ( 2. 2 )$$
- · if ( t, H ) ∈ R × ( 0 , 1 2 ] and T is a fixed number such that T ∈ ±∞ ( , t ) ,
$$B ^ { \pm } ( t, H ) & = \frac { \pm ( H - \frac { 1 } { 2 } ) } { \Gamma ( H + \frac { 1 } { 2 } ) } \int _ { \mathbf R } ( B _ { u } - B _ { t } \mathbf 1 _ { \{ t \geq \pm T \} } ) ( t - u ) _ { \pm } ^ { H - 3 / 2 } - B _ { u } ( - u ) _ { \pm } ^ { H - 3 / 2 } \mathbf d u \\ & \pm B _ { t } ( t - T ) _ { \pm } ^ { H - 1 / 2 },$$
or equivalently,
$$\cdots \, \Psi ^ { \pm } ( t, H ) & = \frac { \pm 1 } { \Gamma ( H + \frac { 1 } { 2 } ) } \frac { \mathrm d } { \mathrm d s } \left ( \int _ { [ \pm T, s ] } B _ { u } ( s - u ) _ { \pm } ^ { H - \frac { 1 } { 2 } } \mathrm d u \right ) \{ ( t ) - ( 0 ) \} \\ & \pm \frac { H - \frac { 1 } { 2 } } { \Gamma ( H + \frac { 1 } { 2 } ) } \int _ { ( \pm \infty, \pm T ] } B _ { u } \left [ ( t - u ) _ { \pm } ^ { H - 3 / 2 } - ( - u ) _ { \pm } ^ { H - 3 / 2 } \right ] \mathrm d u. \quad \ ( 2. 4 ) \\ \cdots \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm. \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm + \, u } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \ \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ \ \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \cdots \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ { \pm } \, \Psi ^ \ \Ps }$$
Proof. Without any loss of generality, we may only prove the equality for the term B + ( t, H ). Let first consider the case t ∈ R and H ∈ ( 1 2 , 1). Using Ito's lemma on the product B u ( t -u ) H -1 / 2 + , we obtain for any x ≤ t
$$B _ { t } ( t - t ) ^ { H - 1 / 2 } = B _ { x } ( t - x ) ^ { H - 1 / 2 } + \int _ { x } ^ { t } ( t - u ) ^ { H - 1 / 2 } \mathrm d B _ { u } - ( H - \frac { 1 } { 2 } ) \int _ { x } ^ { t } B _ { u } ( t - u ) ^ { H - 3 / 2 } \mathrm d u.$$
Hence, subtracting the case t = 0,
$$\lambda _ { x } \left [ ( t - x ) _ { + } ^ { H - 1 / 2 } - ( - x ) _ { + } ^ { H - 1 / 2 } \right ] & = - \int _ { x } ^ { t } \left [ ( t - u ) _ { + } ^ { H - 1 / 2 } - ( - u ) _ { + } ^ { H - 1 / 2 } \right ] \mathrm d B _ { u } \\ & \quad + ( H - \frac { 1 } { 2 } ) \int _ { x } ^ { t } B _ { u } \left [ ( t - u ) _ { + } ^ { H - 3 / 2 } - ( - u ) _ { + } ^ { H - 3 / 2 } \right ] \mathrm d u.$$
$$- \, \frac { 1 } { 2 } ) \int _ { x } \, B _ { u } \left [ \left ( t - u \right ) _ { + } ^ { H - 3 / 2 } - \left ( - u \right ) _ { + } ^ { H - 3 / 2 } \right ] \mathrm d u.$$
The Brownian motion is known to satisfy lim u →±∞ | B u | / u | | 1 / 2+ ε = 0. Moreover, we observe that [ ( t -x ) H -1 / 2 + - -( x ) H -1 / 2 + ] ∼ -∞ ( H -1 ) t H -1 / 2 ( -x ) H -3 / 2 . Therefore, as 3 -H > 1 ,
$$x \rightarrow - \infty$$
2 2 2 B x [ ( t -x ) H -1 / 2 + - -( x ) H -1 / 2 + ] a.s. - - - - -→ 0 .
Similarly, using the dominated convergence theorem,
$$\int _ { x } ^ { t } B _ { u } \left [ ( t - u ) _ { + } ^ { H - 3 / 2 } - ( - u ) _ { + } ^ { H - 3 / 2 } \right ] \mathrm d u \xrightarrow { a. s. } \int _ { - \infty } ^ { t } B _ { u } \left [ ( t - u ) _ { + } ^ { H - 3 / 2 } - ( - u ) _ { + } ^ { H - 3 / 2 } \right ] \mathrm d u,$$
Finally, owing to the L 2 -continuity of the stochastic integral,
$$\int _ { x } ^ { t } \left [ ( t - u ) _ { + } ^ { H - 1 / 2 } - ( - u ) _ { + } ^ { H - 1 / 2 } \right ] \mathrm d B _ { u } \xrightarrow { L ^ { 2 } ( \Omega ) } \int _ { - \infty } ^ { t } \left [ ( t - u ) _ { + } ^ { H - 1 / 2 } - ( - u ) _ { + } ^ { H - 1 / 2 } \right ] \mathrm d B _ { u },$$
which proves the expected equality.
Let now consider the second case H ∈ (0 , 1 2 ). The calculus is quite similar, with only a slight modification in the application of Ito's lemma between T and t -ε .
$$& ( B _ { t - \varepsilon } - B _ { t } ) \varepsilon ^ { H - 1 / 2 } + B _ { t } ( t - T ) ^ { H - 1 / 2 } \\ & = B _ { T } ( t - T ) ^ { H - 1 / 2 } + \int _ { T } ^ { t - \varepsilon } ( t - u ) ^ { H - 1 / 2 } \mathrm d B _ { u } - ( H - \frac { 1 } { 2 } ) \int _ { T } ^ { t - \varepsilon } ( B _ { u } - B _ { t } ) ( t - u ) ^ { H - 3 / 2 } \mathrm d u, \\ \hat { \nu } & \hat { \nu } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim } \hat { \sim }$$
for any ε > 0. Owing to the Hölder continuity of the sample paths of B , the previous expression converges when ε → 0. In addition, the calculus presented in the case H > 1 2 holds as well on the interval [ x, T ], therefore proving Equation (2.3). Finally, the second form (2.4) of the fractional Brownian field is obtained using an alternative representation of the RiemannLiouville fractional derivative of a function on the interval [ T, t ] (see the book of Samko et al. [33] for more information on the subject).
This type of representation has already been exhibited and studied by Picard [32] for the fractional Brownian motion, even though his proof is based on a different approach (approximation of the Brownian motion with Lipschitz functions). The case H > 1 2 has also been considered by Takashima [36].
↦
As presented in the next simple lemma and owing to the Hölder continuity of the sample paths of Brownian motion, we obtain a deterministic representation of the fractional Brownian field ( t, H ) → B t, H ( ) which is continuous in both variables. It will be very useful in the two next sections to study in the sample path properties of the multifractional Brownian motion.
Lemma 2.1. Let ( B t ) t ∈ R be a continuous Brownian motion and B t, H ( ) be the fractional Brownian field defined in Proposition 2.1.
↦
Then, with probability one for all ( t 0 , H 0 ) ∈ R × (0 , 1) , the field ( t, H ) → B t, H ( )( ω ) is continuous at ( t 0 , H 0 ) .
Proof. As previously, we only need to consider the component B + ( t, H ). The continuity on the domains R × (0 , 1 2 ) and R × ( 1 2 , 1) is a simple consequence of the dominated convergence theorem: the deterministic integral is split in two parts on which the theorem can be applied with different bounds.
To threat the case H 0 = 1 2 , we observe that the representation (2.3) in fact holds for all H ∈ (0 1). Then, similarly to the previous case, the dominated convergence theorem implies , that the first term converges to zero when H → 1 2 whereas the second one simply converges to B t 0 .
## 2.1. Well-balanced case
As observed by Stoev and Taqqu [35], the introduction of the mBm is more delicate in the well-balanced case a + = a -. Indeed, when H = 1 2 , the aforementioned definition (2.1) of the fractional Brownian field B t, H ( ) implies that B , ( · 1 2 ) = 0 almost surely. Hence, as presented in [35], to obtain a well-defined mBm, one needs to change the normalisation term, and more specifically, add a diverging factor: 1 / H ( -1 2 ).
Considering this slightly modified definition of the well-balanced mBm, we observe that it corresponds to a particular deterministic expression.
Proposition 2.2. Suppose ( B t ) t ∈ R + is a continuous Brownian motion and a + = a -= 1 . Then, for all ( t, H ) ∈ R × (0 , 1) ,
$$B ( t, H ) = \frac { 1 } { \Gamma ( H + \frac { 1 } { 2 } ) } \int _ { \mathbf R } B _ { u } \left [ ( t - u ) ^ { < H - 3 / 2 > } - ( - u ) ^ { < H - 3 / 2 > } \right ] \mathrm d u \ \ a. s. \quad \ \ ( 2. 5 )$$
where for any x ∈ R , x <α> := sign( x ) | x | α .
Proof. Let ( t, H ) ∈ R × (0 , 1), δ > 0 and T ± = t ∓ δ . Owing to the representation obtained in Proposition 2.1 and using the well-balanced re-normalisation,
$$B ( t, H ) = \frac { 1 } { \Gamma ( H + \frac {$$
Without any restriction, we may assume that 0 / B t, δ ∈ ( ). Then, let us observe the component of the integral corresponding to the interval [ t -δ, t + ]. It is equal to δ
$$\int _ { B ( t, \delta ) } ( B _ { u } - B _$$
For any ε > 0, we note that ∫ B t,δ ( ) \ B t,ε ( ) B t t ( -u ) <H -3 / > 2 d u = 0. Furthermore, the other part converges as well to
$$\begin{array} {$$
since B is Hölder continuous at t . Hence, the integral converges when ε → 0 and we obtain the representation presented in Equation (2.5).
On the contrary to the formulas presented in Proposition 2.1, we observe that the representation (2.5) is an improper deterministic integral when H ≤ 1 2 .
Also note that we may check that when H = 1 2 , the corresponding Gaussian process is, up to a constant, a Brownian motion. More precisely, Stoev and Taqqu [35] have shown that it has the following stochastic integral representation
$$B ( t, \frac { 1 } { 2 } ) = \frac { 1 } { \Gam$$
As proved in the next proposition, the well-balanced case is in fact a classic mBm defined with respect to the latter Brownian motion.
Proposition 2.3. Suppose B is a continuous Brownian motion and B t, H ( ) denotes the wellbalanced fractional Brownian field. Define the Brownian motion ˜ B by
$$\forall t \in \mathbf R ; \ \ \widetilde { B } _ {$$
Then, for all t ∈ R and H ∈ (0 , 1) ,
$$B ( t, H ) = \frac { 1 } { ( H - \frac { 1 } { 2 } ) \tan \left ( ( H + \frac { 1 } { 2 } ) \frac { \pi } { 2 } \right ) } \left ( \widetilde { B } ^ { + } ( t, H ) - \widetilde { B } ^ { - } ( t, H ) \right ) \ \ a. s.$$
where ˜ B ± ( t, H ) designates the fractional Brownian field with respect to ˜ B . Note that the normalising function is continuous at H = 1 2 .
Proof. There exist several ways to prove this result. One might directly obtain the expression of the fractional Brownian field ˜ B ± ( t, H ) using Propositions 2.1 and 2.2, but the calculus is unfortunately a bit tricky.
The proof happens to be simpler if we consider the Fourier representation of B t, H ( ) obtained by Stoev and Taqqu [35]. For all t ∈ R and H ∈ (0 , 1),
$$B ^ { \pm } ( t, H ) = \frac { 1 } { 2 \sqrt { 2 \pi } } \int _ { \mathbf R } \frac { e ^ { i t \xi } - 1 } { | \xi | ^ { H + 1 / 2 } } e ^ { \mp i \text{sign} ( \xi ) ( H + 1 / 2 ) \pi / 2 } \text{d} \widehat { W } ( \xi ),$$
where ̂ W is a complex Gaussian measure such that ̂ W is a complex Wiener measure on R + and ̂ W A ( ) = ̂ W ( -A ) for any A ∈ B ( R + ). The well-balanced fractional Brownian field then has the following representation
$$B ( t, H ) = \frac { 2 \cos \left ( ( H + \frac { 1 } { 2 } ) \frac { \pi } { 2 } \right ) } { \sqrt { 2 \pi } ( H - \frac { 1 } { 2 } ) } \int _ { \mathbf R } \frac { e ^ { i t \xi } - 1 } { | \xi | ^ { H + 1 / 2 } } \text{d} W ( \xi ).$$
We introduce a slightly modified complex Gaussian measure ˜ W , given by
$$\forall A \in \mathcal { B } ( \mathbf R _ { + } ) ; \ \ W ( A ) \coloneqq i W ( A ) \ \text{ and } \ W ( - A ) \coloneqq W ( A ) = - i W ( - A ).$$
Then, let ˜ B ± ( t, H ) denote the fractional Brownian field defined with respect to ˜ W . Since ˜ W (d ξ ) = i sign( ξ W ) ̂ (d ξ ),
$$\widetilde { B } ^ { + } ( t, H ) - \widetilde { B } ^ { - } ( t, H ) & = \frac { 2 \sin \left ( ( H + \frac { 1 } { 2 } ) \frac { \pi } { 2 } \right ) } { \sqrt { 2 \pi } } \int _ { \mathbf R } \frac { e ^ { i t \xi } - 1 } { | \xi | ^ { H + 1 / 2 } } ( - i ) \, \text{sign} ( \xi ) \, \text{d} \, \widetilde { W } ( \xi ) \\ & = ( H - \frac { 1 } { 2 } ) \tan \left ( ( H + \frac { 1 } { 2 } ) \frac { \pi } { 2 } \right ) \, B ( t, H ).$$
To conclude the proof, we need to check that the field ˜ B ± ( t, H ) corresponds to the same process defined in the proposition, i.e. the integral with respect to the Brownian motion ˜ B . For this purpose, we simply need to consider the previous equality when H = 1 2 . We first note that lim H → 1 / 2 ( H -1 2 ) tan ( ( H + 1 2 ) π 2 ) = 1. Hence,
$$\widetilde { B } _ { t } \coloneqq \frac { 1 } { 2 } B ( t, \frac { 1 } { 2 } ) = \frac { 1 } { 2 } \left ( \widetilde { B } _ { + } ( t, \frac { 1 } { 2 } ) - \widetilde { B } _ { - } ( t, \frac { 1 } { 2 } ) \right ).$$
According to Equation (2.1), this last equality proves that ˜ B is the Brownian motion with respect to which are defined the fractional Brownian fields, and thus we clearly obtain the expected representation of B t, H ( ).
The equality in law between the two processes can be obtained more directly using Theorem 4.1 from [35]. Nevertheless, Proposition 2.3 indicates precisely how these two fractional Brownian fields are related, and proves that the well-balanced one is simply a classic field with a change of Brownian motion. Besides, this statement allows us to simply the study of the sample path properties of the mBm since we do not need to consider separately the specific well-balanced case a + = a -.
## 3. Hölder and 2-microlocal regularity
Following the convention initiated by [35], we will call multifractional Brownian motion a process X which has the following form:
$$\forall t \in \mathbf R ; \ \ X _ { t } = a ^ { + } B ^ { + } \left ( t, H ( t ) \right ) + a ^ { - } B ^ { - } \left ( t, H ( t ) \right ),$$
/negationslash
↦
where ( a -, a + ) ∈ R 2 \ { (0 0) , } . As previously said, we may also suppose without any loss of generality that a + = a -. The Hurst function t → H t ( ) ∈ (0 1) will be assume to be at least a , deterministic càdlàg function.
/negationslash
We may nevertheless note that several of the following results remain valid if H ( ) is replaced · by a stochastic process with càdlàg sample paths. This last model has already been considered by Papanicolaou and Sølna [28] and Ayache and Taqqu [2], where the latter has obtained some general sample path properties. For any càdlàg function H ( ), · the set { t ∈ R : ∆ H t ( ) = 0 } is countable, and therefore we can easily see from the definition of the mBm that with probability one,
/negationslash
$$\forall t \in \mathbf R ; \ \Delta H ( t ) \neq 0 \ \iff \ \Delta X _ { t } \neq 0.$$
/negationslash
Furthermore, ∆ X t is a centered Gaussian variable whose variance is proportional to ∆ H t ( ) 2 .
In this section, we will use specific functional spaces, called 2-microlocal spaces, to characterize the fine sample path properties of the mbm. The use of the unusual framework is particularly motivated by the deterministic fractional integrals and derivatives of Brownian motion exhibited in Proposition 2.1. Indeed, the latter spaces behave particularly well under the action of fractional operators, which will allow us to deduce many properties on the mBm from the geometry of the Brownian motion.
Historically, the 2-microlocal formalism has been introduced by Bony [8] to study the singularities of the solutions of PDEs, and has only been recently introduced in a stochastic frame by Herbin and Lévy Véhel [16] to characterise the fine regularity of processes. Several characterizations of the 2-microlocal spaces have been investigated in the literature, including the following time-domain one due to Kolwankar and Lévy Véhel [22] and Seuret and Lévy Véhel [34].
Definition 3.1. Let t ∈ R , s ′ ≤ 0 and σ ∈ (0 , 1) such that σ -s ′ / ∈ N . A function f : R → R belongs to the 2-microlocal space C σ,s ′ t if there exist C > 0 , ρ > 0 and a polynomial P t such that
$$| (.$$
$$\iota ^ { \prime } \quad \left | \left ( f ( u ) - P _ { t } ( u ) \right ) - \left ( ( f ( v ) - P _ { t } ( v ) ) \right | \leq C | u - v | ^ { \sigma } \left ( | u - t | + | v - t | \right ) ^ { - s ^ { \prime } },$$
$$( 3. 1 )$$
for all u, v ∈ B t, ρ ( ) .
In this work, it will only be necessary to consider the case σ ∈ (0 , 1). Nevertheless, Definition 3.1 can be extended σ ∈ R Z \ . We refer to [34, 5] for a more complete presentation on the subject.
As previously outlined, a very useful properties of the 2-microlocal spaces is their stability under the application of fractional integrations and derivations:
$$\forall \alpha > 0 ; \ \ f \in C _ { t } ^ { \sigma, s ^ { \prime } } \ \iff \ I _ { \pm } ^ { \alpha } f \in C _ { t } ^ { \sigma + \alpha, s ^ { \prime } },$$
where I α ± f denotes the fractional integral of f of order α : ( I α ± f ) ( ) := t ∫ R ( t -u ) α -1 ± f ( u ) d u . Following the presentation of these functional spaces, we can introduce a related regularity tool, named the 2-microlocal frontier and defined by
$$\forall s ^ { \prime } \in \mathbf R ; \quad \sigma _ { f, t } ( s ^ { \prime } ) = \sup \{ \sigma \in \mathbf R \, \colon f \in C _ { t } ^ { \sigma, s ^ { \prime } } \}.$$
↦
We observe that when using the 2-microlocal formalism, the local regularity at t is characterised by a function, not a single coefficient. Owing to inclusion properties of the 2-microlocal spaces, the map s ′ → σ f,t ( s ′ ) is well-defined and satisfies several interesting properties:
- · σ f,t ( ) is a concave non-decreasing function; ·
## · σ f,t ( ) has left and right derivatives between 0 and 1. ·
Furthermore, if the modulus of continuity of f is such that ω f ( h ) = O (1 / | log( h ) ), the pointwise | and local Hölder exponents can be retrieved from the frontier:
$$\alpha _ { f, t } = - \inf \{ s ^ { \prime } \, \colon \sigma _ { f, t } ( s ^ { \prime } ) \geq 0 \} \ \text{ and } \ \widetilde { \alpha } _ { f, t } = \sigma _ { f, t } ( 0 ).$$
Finally, owing to the stability of the 2-microlocal spaces, the application of a fractional integration simply corresponds to a translation of the 2-microlocal frontier along the vertical axis:
$$\forall s ^ { \prime } \in \mathbf R ; \ \ \sigma _ { I _ { \pm } ^ { \alpha } f, t } ( s ^ { \prime } ) = \sigma _ { f, t } ( s ^ { \prime } ) + \alpha.$$
The 2-microlocal frontier of some Gaussian processes has been determined by Herbin and Lévy Véhel [16]. For instance, a fractional Brownian motion B H , H ∈ (0 1) , satisfies with probability one and for all t ∈ R :
$$\forall s ^ { \prime } \in \mathbf R ; \ \ \sigma _ { B ^ { H }, t } ( s ^ { \prime } ) = \left ( H + s ^ { \prime } \right ) \wedge H.$$
To investigate the 2-microlocal and Hölder regularity of the multifractional Brownian motion, we first need to properly study the behaviour of the fractional Brownian field and its successive partial derivatives. Hence, in the next proposition, we consider uniformly on the variables t and H the 2-microlocal frontier of ∂ k H B t, H ( ) along the time axis.
↦
Proposition 3.1. Suppose B t, H ( ) is a fractional Brownian field. For any H ∈ (0 1) , and any k ∈ N , let B H,k denotes the process t → ∂ k H B t, H ( ) . Then, with probability one, for all H ∈ (0 , 1) , k ∈ N and t ∈ R
$$\forall s ^ { \prime } \in \mathbf R ; \quad \sigma _ { B ^ { H, k }, t } ( s ^ { \prime } ) = \left ( H + s ^ { \prime } \right ) \wedge H.$$
Proof. Since we use the deterministic representation obtained in Proposition 2.1, we can set a fix ω ∈ Ω in the proof. For the sake of readability, let us first consider H > 1 2 and remove the normalisation constant 1 / Γ ( H -1 2 ) . Then, for any k ∈ N , the partial derivative with respect to H is equal to
$$\partial _ { H } ^ { k } \, B ^ { \pm } ( t, H ) = \int _ { \mathbf R } B _ { u } \left [ ( t - u ) _ { \pm } ^ { H - 3 / 2 } \log ^ { k } ( t - u ) _ { \pm } - ( - u ) _ { \pm } ^ { H - 3 / 2 } \log ^ { k } ( - u ) _ { \pm } \right ] \mathrm d u.$$
Suppose ( T + , T -) is a non-empty open interval. Since the 2-microlocal frontier is a purely local property of the sample paths, we may restrict our study to t ∈ ( T + , T -) with any loss of generality. Then, we note that the component
↦
$$t \mapsto \int _ { ( \mp \infty, T _ { \pm } ] } B _ { u } \left [ ( t - u ) _ { \pm } ^ { H - 3 / 2 } \log ^ { k } ( t - u ) _ { \pm } - ( - u ) _ { \pm } ^ { H - 3 / 2 } \log ^ { k } ( - u ) _ { \pm } \right ] \mathrm d u$$
is a smooth function, and therefore does not have any influence on the 2-microlocal regularity. Hence, we focus on the study of the remaining integral
↦
$$t \in ( T _ { + }, T _ { - } ) \mapsto \int _ { [ T _ { + }, T _ { - } ] } B _ { u } ( t - u ) _ { \pm } ^ { H - 3 / 2 } \log ^ { k } ( t - u ) _ { \pm } \mathrm d u,$$
when H > 1 2 . Similarly, the case H < 1 2 is reduced to the analysis of
↦
$$t \in ( T _ { + }, T _ { - } ) \mapsto \frac { \mathrm d } { \mathrm d s } \left ( \int _ { [ T _ { \pm }, s ] } B _ { u } ( s - u ) _ { \pm } ^ { H - \frac { 1 } { 2 } } \log ^ { k } ( t - u ) _ { \pm } \mathrm d u \right ) ( t ).$$
If we define the process Y t = B t 1 t ∈ [ T + ,T -] , we observe that the two previous formulas correspond respectively to fractional-like integral and derivative of Y on the real axis. Compared to the usual definition, there is only a slight modification in the kernel with a logarithmic term.
Furthermore, when H < 1 2 , the fractional-like derivative is defined as the classic derivative of a integral of order H + 1 2 . Since we know the behaviour of the 2-microlocal regularity under derivation, we may restrict our study to the fine analysis of the following process
↦
$$t \in ( T _ { + }, T _ { - } ) \longmapsto ( I _ { \pm } ^ { \alpha, k } Y ) ( t ) \coloneqq \int _ { \mathbf R } Y _ { u } \, ( t - u ) _ { \pm } ^ { \alpha - 1 } \log ^ { k } ( t - u ) _ { \pm } \mathrm d u,$$
for any α ∈ (0 1) and , k ∈ N .
We expect the 2-microlocal frontier of this integral to behave similarly to the one of a classic fractional integral of Y , since the only difference lies in a logarithmic term. This easiest way to prove this statement is to compute the Fourier transform of the corresponding pseudodifferential operator and show it has an effect equivalent to the fractional integration in the Fourier space.
The computation of Fourier transform of the tempered distribution u α -1 ± log k u ± is similar to the classic calculus on u α -1 ± . We first consider the Laplace transform, with /Rfractur ( z ) > 0. Since this transform is analytic on the previous domain, we can restrict ourselves to the case that z > 0. Then,
$$\psi _ { z } \, \dots \, \dots \, \dots, \\ \forall z > 0 ; \quad \int _ { \mathbf R _ { + } } u ^ { \alpha - 1 } \log ^ { k } ( u ) \, e ^ { - z u } \, d u & = z ^ { - \alpha } \int _ { \mathbf R _ { + } } v ^ { \alpha - 1 } \left ( \log v - \log z \right ) ^ { k } e ^ { - v } \, d v \\ & = z ^ { - \alpha } \sum _ { j = 0 } ^ { k } ( - 1 ) ^ { j } \log ^ { j } ( z ) \, C _ { k } ^ { j } \underbrace { \int _ { \mathbf R _ { + } } v ^ { \alpha - 1 } \log ^ { k - j } ( v ) \, e ^ { - v } } _ { F ( \alpha, k - j ) }.$$
As the last term is also analytic, it implies the equality for any z ∈ C such that /Rfractur ( z ) > 0.
To obtain the Fourier transform of u α -1 ± log k u ± as a tempered distribution, we simply have to consider z = iξ + ε and let ε → 0. The scalar product with a test function converges, and proves that the generalized Fourier transform is equal to
$$& \mathcal { F } _ { g } ( u _ { + } ^ { \alpha - 1 } \log ^ { k } u _ { + } ) ( \xi ) \\ & = e ^ { - i \, \text{sign} ( \xi ) \frac { \pi } { 2 } \alpha } | \xi | ^ { - \alpha } \frac { 1 } { \sqrt { 2 \pi } } \sum _ { j = 0 } ^ { k } ( - 1 ) ^ { j } ( \log | \xi | + i \, \text{sign} ( \xi ) \frac { \pi } { 2 } ) ^ { j } \cdot C _ { k } ^ { j } F ( \alpha, k - j ).$$
Therefore, the Fourier transform corresponds to a weighted sum of elements | ξ | -α log n | ξ | , with a phase correction term.
To express the action of these pseudo-differential operators on the regularity of the Brownian motion, we rely on the original definition of the 2-microlocal spaces in the Fourier space and
make use of the arguments presented by Jaffard [18] to prove the stability. More precisely, the aforementioned definition is based on the Littlewood-Paley decomposition of tempered distributions. Shortly, consider φ ∈ S ( R ) such that its Fourier transform satisfies
$$\widehat { \phi } ( \xi ) = \begin{cases} 1 & \text{si} & | \xi | \leq 1 / 2 ; \\ 0 & \text{si} & | \xi | \geq 1, \end{cases}$$
and define for any tempered distribution f ∈ S ′ ( R ) and for all j ∈ N , φ j ( x ) = 2 j φ (2 j x ), ψ j = φ j +1 -φ j , S f 0 = φ ∗ f and ∆ j f = ψ j ∗ f . Then, the Littlewood-Paley decomposition tells us that f can be expressed as following
$$f = S _ { 0 } f + \sum _ { j = 0 } ^ { \infty } \Delta _ { j } f,$$
where we observe that every term in the sum belongs to L 2 ( R ). Roughly speaking, this decomposition corresponds to the application of band-pass filters in the Fourier space. As presented by Bony [8], Meyer [25], f ∈ C σ,s ′ t if and only there exists C > 0 such that for all j > 0,
$$| \Delta _ { j } f ( u ) | \leq C \, 2 ^ { - j \sigma } \left ( 2 ^ { - j } + | u - t | \right ) ^ { - s ^ { \prime } }.$$
Observe that, up to a simple translation, we may assume that t = 0. Then, define the rescaled function U j ( x ) = (∆ j f )(2 -j u ). Its Fourier transform ̂ U j is carried by { ξ : 1 2 < ξ | | < 2 } and U j is such that | U j ( u ) | ≤ C 2 -j ( σ -s ′ ) ( 1+ | u | ) -s ′ . Applying the operator | ξ | -α log k | ξ | in the Fourier space on f , we obtain a corresponding collection of rescaled functions ( V j ) j> 0 which satisfy for every j > 0,
$$\widehat { V } _ { j } ( \xi ) = \widehat { U } _ { j } ( \xi ) \cdot \left [ \sum _ { \ell = 0 } ^ { k } c ( \ell, k ) \, | \xi | ^ { - \alpha } \log ^ { \ell } | \xi | \cdot j ^ { k - \ell } \, 2 ^ { - j \alpha } \right ],$$
where c /lscript, ( k ) is a constant only depending on /lscript and k . Note that similarly to the classic fractional integral, the Fourier multipliers ̂ K α,/lscript ( ξ ) := | ξ | -α log /lscript | ξ | coincide with Schwartz functions on 1 2 < ξ | | < 2. Coming back into the time space, V j is given by
$$V _ { j } = \sum _ { \ell = 0 } ^ { k } j ^ { k - \ell } 2 ^ { - j \alpha } \, c ( \ell, k ) K _ { \alpha, \ell } \ast U _ { j },$$
where K α,/lscript ∈ S ( R ) for every /lscript . The convolution with a Schwartz function K α,/lscript preserves the polynomial decay, hence for every j > 0, | V j ( u ) | ≤ C j k 2 -j ( σ + α -s ′ ) ( 1 + | u | ) -s ′ . Rescaling by a factor 2 , the latter upper bound leads to the characterization of 2-microlocal spaces, proving j that
$$\forall s ^ { \prime }, \sigma \in \mathbf R ; \ \ Y \in C _ { t } ^ { \sigma, s ^ { \prime } } \Longrightarrow I _ { \pm } ^ { \alpha, k } Y \in C _ { t } ^ { \sigma - \varepsilon, s ^ { \prime } }$$
for any ε > 0. Note that the property is slightly weaker than the classic stability under the action of fractional integration due to the logarithmic terms.
The converse case is obtained similarly,
$$\forall s ^ { \prime }, \sigma \in \mathbf R ; \ \ I _ { \pm } ^ { \alpha, k } Y \in C _ { t } ^ { \sigma, s ^ { \prime } } \Longrightarrow Y \in C _ { t } ^ { \sigma - \varepsilon, s ^ { \prime } }.$$
The same property holds for any weighted sum of I α,k -Y and I α,k + Y , and therefore, the 2microlocal frontier of I α,k ± Y satisfies for any H ∈ (0 , 1) and k ∈ N
$$\forall s ^ { \prime } \in \mathbf R ; \ \ \sigma _ { I _ { \pm } ^ { \alpha, k } Y, t } ( s ^ { \prime } ) = \sigma _ { Y, t } ( s ^ { \prime } ) + H - \frac { 1 } { 2 } = ( s ^ { \prime } + H ) \wedge H.$$
It concludes the proof, since we have shown previously that the 2-microlocal frontier of B H,k is equal to the regularity I α,k ± Y .
↦
Ayache [1] has obtained a similar result on the field ( t, H ) → B + ( t, H ) based on a multiresolution analysis of the latter.
As outlined in the introduction, the Hölder regularity of a general multifractional Brownian motion has already been investigated by Herbin [15] and Ayache [1]. In particular, the latter has proved that with probability one, the local Hölder exponent is given by
$$\forall t \in \mathbf R \, \sum \{ 0 \} ; \ \widetilde { \alpha } _ { X, t } = H ( t ) \wedge \widetilde { \alpha } _ { H, t }.$$
Note that the behaviour at t = 0 is slightly different as the form of the variance of increments differs at this point.
↦
Let us first obtain a estimate for the general 2-microlocal frontier. For this purpose, we need to introduce the multiplicity of the component h → ( B t, h ( ) -B t, H ( ) ) of the fractional field at ( t, H ) ∈ R × (0 1), i.e. ,
/negationslash
$$m _ { t, H } = \inf \{ k \in \mathbf N \, \sum \{ 0 \} \colon \partial _ { H } ^ { k } B ( t, H ) \neq 0 \}.$$
Theorem 3.1. Suppose X is a multifractional Brownian motion with Hurst function H . Then, with probability one, the 2-microlocal frontier of X at any t ∈ R satisfies
$$\sigma _ { X, t } ( s ^ { \prime } ) \geq \left ( s ^ { \prime } + H ( t ) \right ) \wedge H ( t ) \wedge \sigma _ { H, t } ( s ^ { \prime } + H ( t ) ) \wedge m _ { t, H ( t ) } \sigma _ { H, t } ( s ^ { \prime } / m _ { t, H ( t ) } ).$$
for all s ′ ≥ -α X,t . Furthermore, when m t,H t ( ) = 1 , the frontier is equal to
$$\forall s ^ { \prime } \geq - \alpha _ { X, t } ; \ \ \sigma _ { X, t } ( s ^ { \prime } ) = \left ( s ^ { \prime } + H ( t ) \right ) \wedge H ( t ) \wedge \sigma _ { H, t } ( s ^ { \prime } ).$$
Proof. Without any loss of generality, we can assume that ˜ α H,t > 0, since otherwise the lower bound is equal to 0. Then, let ε > 0, ρ > 0, t ∈ R and u, v be in the neighbourhood B t, ρ ( ). We first estimate the size of the increments X u -X v :
$$B ( u, H _ { u } ) - B ( v, H _ { v } ) = \left ( B ( v, H _ { u } ) - B ( u, H _ { u } ) \right ) + \left ( B ( v, H _ { v } ) - B ( v, H _ { u } ) \right ).$$
Using a simple integration by parts, we observe that the first term is equal to
$$B ( v, H _ { u } ) - B ( u, H _ { u } ) = B ( v, H _ { t } ) - B ( u, H _ { t } ) + \int _ { H _ { t } } ^ { H _ { u } } \left ( \partial _ { H } B ( v, h ) - \partial _ { H } B ( u, h ) \right ) \text{d} h.$$
Owing to Lemma 3.1, for any ε > 0, there exists c 1 > 0 such that
$$\forall u, v \in B ( t, \rho ), \ \forall h \in [ H _ { t }, H _ { u } ] ; \ \left | \partial _ { H } B ( v, h ) - \partial _ { H } B ( u, h ) \right | \leq c _ { 1 } | u - v | ^ { h - \varepsilon }.$$
Therefore, the integral term satisfies
$$\left | \int _ { H _ { t } } ^ { H _ { u } } ( \partial _ { H } B ( v, h ) - \partial _ { H } B ( u, h ) ) d h \right | \leq c _ { 1 } | H _ { t } - H _ { u } | \cdot | v - u | ^ { H _ { t } \wedge H _ { u } - \varepsilon },$$
and is thus negligible in front of the first one. Still using Lemma 3.1, we know that for all u, v ∈ B t, ρ ( )
$$^ { \cdot \cdot \cdot } _ { \cdot } \,, & \quad | B ( v, H _ { t } ) - B ( u, H _ { t } ) | \leq c _ { 2 } | u - v | ^ { \sigma } \left ( | u - t | + | v - t | \right ) ^ { - s ^ { \prime } },$$
where s ′ ≥ -H t and σ < ( H t + s ′ ) ∧ H t . This inequality corresponds to the first term in the lower bound.
Using Taylor's formula, we can obtain an estimate of the second component ( B v, H ( v ) -B v, H ( u )).
$$B ( v, H _ { v } ) - B ( v, H _ { u } ) = \sum _ { k = 1 } ^ { m - 1 } \frac { \partial _ { H } ^ { k } B ( v, H _ { u } ) } { k! } ( H _ { v } - H _ { u } ) ^ { k } + \int _ { H _ { u } } ^ { H _ { v } } \frac { \partial _ { H } ^ { m } B ( v, h ) } { ( m - 1 )! } ( H _ { v } - h ) ^ { m - 1 } d h,$$
where m denotes the coefficient m t,H t previously introduced. There exists a neighbourhood B t, ρ ( ) of t such that for every k ∈ { 1 , . . . , m -1 } and for all v ∈ B t, ρ ( ), | ∂ k H B v, H ( u ) | ≤ c 3 | v - | t H t -ε . Hence the term k = 1 is dominant in the sum and for all u, v ∈ B t, ρ ( )
$$| \partial _ { H } B ( v, H _ { u } ) | \cdot | H _ { v } - H _ { u } | \leq c _ { 4 } | u - v | ^ { \sigma } \left ( | u - t | + | v - t | \right ) ^ { - s + H _ { t } - \varepsilon },$$
for any s ′ ≥ -α H,t and σ < σ H,t ( s ′ ). This inequality leads to the second term in the lower bound.
Finally, let consider the last integral. ∂ m H B v, h ( ) can be simply bound by a constant, implying that the integral is upper bounded by | H u -H v | m . The third lower bound is a direct consequence of this estimate.
To end the proof, we study the case m t,H t ( ) = 1. The lower bound is clearly equal to ( H t + s ′ ) ∧ H t ∧ σ H,t ( s ′ ). Let set s ′ ≥ -α X,t . Suppose first that ( H t + s ′ ) ∧ H < σ t H,t ( s ′ ). Then, the term B v, H ( u ) -B u,H ( u ) will dominate in the increment X u -X v , therefore implying the inequality σ X,t ( s ′ ) ≤ ( H t + s ′ ) ∧ H t . In the other case, we observe that
$$B ( v, H _ { v } ) - B ( v, H _ { u } ) = \partial _ { H } B ( v, H _ { v } ) ( H _ { u } - H _ { v } ) + \int _ { H _ { v } } ^ { H _ { u } } \partial _ { H } ^ { 2 } B ( v, h ) ( H _ { v } - h ) \text{d} h,$$
There exists ρ > 0 and c 5 > 0 such that for all v ∈ B t, ρ ( ), | ∂ H B v, H ( v ) | > c 5 , implying that the regularity of this component is σ H,t ( s ′ ) Furthermore, as previously noted, the second term is upper-bounded by ( H u -H v ) 2 , and thus is negligible in front of the first one. Therefore, the increment X u -X v behave similarly to H u -H v , proving that σ X,t ( s ′ ) ≤ σ H,t ( s ′ ).
As observed in the previous proof, the study of the general form of the 2-microlocal frontier is quite technical, in particular because of the interaction which may appear between the oscillations of the Hurst function H and the derivatives of the fractional Brownian field B t, H ( ). Nevertheless, a complete estimate can be obtained for the pointwise Hölder regularity.
Proposition 3.2. Suppose X is a multifractional Brownian motion with Hurst function H . Then, with probability one, the pointwise exponent is equal to
$$\forall t \in \mathbf R ; \quad \alpha _ { X, t } = H ( t ) \wedge m _ { t, H ( t ) } \alpha _ { H, t }.$$
Proof. Let us proceed similarly to the proof of Theorem 3.1 and study the increment X u -X t = ( B u,H ( u ) -B u,H ( t ) ) + ( B u,H ( t ) -B t, H ( t ) ) . Owing to Lemma 3.1, the pointwise regularity of the second term is exactly H t ( ).
Then, we consider a slightly different Taylor expansion on the first increment,
$$B ( u, H _ { u } ) - B ( u, H _ { t } ) = \sum _ { k = 1 } ^ { m } \frac { \partial _ { H } ^ { k } B ( u, H _ { t } ) } { k! } ( H _ { u } - H _ { t } ) ^ { k } + \int _ { H _ { t } } ^ { H _ { u } } \frac { \partial _ { H } ^ { m + 1 } B ( u, h ) } { m! } ( H _ { u } - h ) ^ { m } d h.$$
For every k ∈ { 1 , . . . , m - } 1 , there exists a simple upper bound : | ∂ k H B u,H ( t ) || H u -H t | k ≤ c 1 | u - | t H t + kα H,t -ε . Hence, the pointwise regularity of this type of term is strictly greater than H t ( ), they are therefore negligible. In the case k = m , we note that there exists ρ > 0 and c 2 > 0 such that for all u ∈ B t, ρ ( ), | ∂ k H B u,H ( t ) | ≥ c 2 . Hence, the pointwise exponent of this component is exactly m α · X,t . Finally, similarly to the previous proof, the last term can be simply upper bounded by c 3 | H u -H v | m +1 ≤ c 3 | u - | t ( m +1) α H,t . This last estimate is sufficient to conclude the proof.
The previous result constitutes an extension of the estimate obtained by Ayache [1]. As pointed out by the latter, it shows that the multifractional Brownian motion can have a nondeterministic pointwise regularity. More precisely, Ayache [1] has proved that for any ε > 0, with positive probability,
$$\dim _ { \mathrm H } \{ t \in [ a, b ] \, \colon m _ { t, H ( t ) } > 1 \} \geq 1 - \inf _ { u \in [ a, b ] } H ( u ) - \varepsilon.$$
In the view of the result obtained in Proposition 3.2, it would be interesting to study the Hausdorff dimension of the following sets
$$\forall k \in \mathbf N ; \ \ E _ { k } = \left \{ t \in [ a, b ] \, \colon m _ { t, H ( t ) } = k \right \}.$$
This question seems to be much more difficult than the previous estimate, since it involves the successive derivatives ∂ k H B t, H ( ) of the fractional field and thus the correlation existing between them. Finally, we also note that in terms of regularity, the behaviour at t = 0 is specific, since the multiplicity m 0 ,H (0) is equal to infinity.
To end this section, we present a couple of examples of Hurst functions which illustrate this particular geometry of the multifractional Brownian motion.
Example 1. Suppose H : R → ( 1 2 , 3 4 ) is an Hurst function such that for all t ∈ R , the 2-microlocal local frontier of H at t is equal to
$$\forall s ^ { \prime } \in \mathbf R ; \ \ \sigma _ { H, t } ( s ^ { \prime } ) = \left ( s ^ { \prime } + \frac { 3 } { 8 } \right ) \wedge \frac { 3 } { 8 }.$$
This kind of function can be easily constructed from the sample paths of a fractional Brownian motion with parameter H = 3 8 .
Then, the 2-microlocal frontier of the corresponding multifractional Brownian motion satisfies with probability one:
- · if m t,H t = 1,
$$\forall s ^ { \prime } \in \mathbf R ; \ \ \sigma _ { X, t } ( s ^ { \prime } ) = \sigma _ { H, t } ( s ^ { \prime } ) = \left ( s ^ { \prime } + \frac { 3 } { 8 } \right ) \wedge \frac { 3 } { 8 } ;$$
- · if m t,H t > 1,
$$\forall s ^ { \prime } \in \mathbf R ; \ \ \sigma _ { X, t } ( s ^ { \prime } ) = \left ( s ^ { \prime } + H ( t ) \right ) \wedge \frac { 3 } { 8 }.$$
These two cases are illustrated by Figure 1a.
To obtain the complete determination of 2-microlocal frontier, we use Theorem 3.1, observing that it implies the inequality
$$\sigma _ { X, t } ( s ^ { \prime } ) & \geq \left ( s ^ { \prime } + H ( t ) \right ) \wedge H ( t ) \left ( s ^ { \prime } + \frac { 3 } { 8 } + H ( t ) \right ) \wedge \frac { 3 } { 8 } \wedge \left ( s ^ { \prime } + \frac { 3 m _ { t, H ( t ) } } { 8 } \right ) \wedge \frac { 3 m _ { t, H ( t ) } } { 8 } \\ & = \left ( s ^ { \prime } + H ( t ) \right ) \wedge \frac { 3 } { 8 },$$
since H t ( ) ∈ ( 1 2 , 3 4 ). Then, the upper bound is a consequence of the pointwise exponent obtained in Proposition 3.2 and the following property of the 2-microlocal frontier:
$$\forall s ^ { \prime } \in \mathbf R ; \ \sigma _ { X, t } ( s ^ { \prime } ) \leq \liminf _ { s \to t } \widetilde { \alpha } _ { X, t },$$
proved in the Appendix (Lemma 5.1).
′
Figure 1: 2-microlocal frontiers of multifractional Brownian motion
In the following second example, we illustrate the particular Hölder regularity which may appear at t = 0.
↦
Example 2. Consider the Hurst function H : t → 3 4 + sin t ( 1 t 2 ) , which locally corresponds to the "chirp" function. Then, the 2-microlocal frontier of the multifractional Brownian motion at 0 is equal to
$$\forall s ^ { \prime } \in \mathbf R ; \ \ \sigma _ { X, 0 } ( s ^ { \prime } ) = \left ( s ^ { \prime } + \frac { 3 } { 4 } \right ) \wedge \left ( \frac { s ^ { \prime } } { 3 } + \frac { 7 } { 1 2 } \right ) \wedge \frac { 3 } { 4 }.$$
Figure 1b gives an illustration of this particular and hybrid 2-microlocal frontier.
To obtain this estimate, we need to refine the proof of Theorem 3.1. Let first recall that the 2-microlocal frontier of H at 0 is equal to
$$\forall s ^ { \prime } \in \mathbf R ; \ \ \sigma _ { H, 0 } ( s ^ { \prime } ) = \frac { 1 } { 3 } ( s ^ { \prime } + 1 ).$$
If we consider the regularity of the term ∂ H B v, H ( 0 )( H v -H v ), we know there exists for any ε > 0 a sequence u n → 0 such that | ∂ H B u ,H ( n 0 ) | ≥ u H 0 -ε n . Then, owing to form of the Hurst function, there exists v n → 0 such that | H u ( n ) -H v ( n ) | ≥ | u n -v n | σ ( u n + v n ) -s ′ where σ < σ H, 0 ( s ′ ). Combining the two, we obtain that the 2-microlocal frontier of the component ∂ H B v, H ( 0 )( H v -H v ) is
$$\forall s ^ { \prime } \geq - \frac { 7 } { 1 2 } ; \ \ \sigma ( s ^ { \prime } ) = \sigma _ { H, 0 } ( s ^ { \prime } + H ( 0 ) ) = \left ( \frac { s ^ { \prime } } { 3 } + \frac { 7 } { 1 2 } \right ).$$
Furthermore, we know that m 0 ,H (0) = ∞ , inducing that the other terms in the Taylor expansion are negligible.
Wenote that the 2-microlocal frontier obtained in this example is consistent with the estimate of the Hölder exponents proved in Proposition 3.2.
## 4. Fractal dimensions of the graph and images
In this section, we aim to extend our work on the sample path geometry of mBm by studying the fractal dimension of its graph and the images of sets. As previously, we will be particularly interested in the study of the irregular case in order to generalize the classic results of Peltier and Lévy Véhel [29].
We first need to recall a few notations on fractal dimensions. For any set A ⊂ R d , we respectively denote by dim H A and dim B A (or dim B A ) the Hausdorff and the lower (or upper) Box dimensions of A . In the case dim B A = dim B A , we denote by dim B A the previous quantity. The reader might refer to the book of Falconer [13] for the precise definition and the basic properties of these fractal dimensions.
As we investigate the fractal geometry of the multifractional Brownian motion, which is a purely non-stationary process, it is more relevant to consider the localised fractal dimension. For all A ⊂ R d and x ∈ R d , it is defined by
$$\dim _ { *, x } A = \lim _ { \rho \to 0 } \dim _ { * } \left ( A \cap B ( x, \rho ) \right ),$$
↦
where dim ∗ denotes any fractal dimension. The previous limit always exists as the right term is decreasing when ρ → 0. We note that the map x → dim ∗ ,x A is upper semi-continuous. Finally, we also observe that the fractal dimension can always be retrieve from the localised values: dim ∗ A = sup x ∈ R d dim ∗ ,x A .
In the first part of this section, we investigate the local fractal dimension of the graph of the mBm. For any function f , its graph on the set V will be denoted Gr( f, V ), i.e.
$$\text{Gr} ( f, V ) = \{ ( t, f ( t ) ) \colon t \in V \}.$$
In the case V = R , we will simply use the notation Gr( f ). To simplify, we will write dim ∗ ,t Gr( f ) for the local fractal dimension at ( t, f ( )). The latter is therefore equivalently defined by t
$$\dim _ { *, t } \text{Gr} ( f, V ) \coloneqq \lim _ { \wedge \to \theta } \dim _ { * } \text{Gr} ( f, V \cap B ( t, \rho ) ).$$
$$\rho \rightarrow 0$$
Several deterministic properties satisfied by the local graph dimension are presented in the Appendix 5, including some connections with the 2-microlocal frontier.
The graph dimension of an H 0 -multifractional Brownian motion has already been determined by Peltier and Lévy Véhel [29], proving that with probability one
$$\forall t \in \mathbf R ; \quad \dim _ { H, t } \mathbf G r ( f ) = \dim _ { B, t } \mathbf G r ( f ) = 2 - H ( t ).$$
In the next result, we investigate the extension of this formula to the Box dimension of the graph of a general multifractional Brownian motion.
Theorem 4.1. Suppose X is a multifractional Brownian motion. Then, with probability one, it satisfies
$$\forall t \in \mathbf R \, \sum \{ 0 \} ; \ \underline { \dim } _ { B, t } \, \text{Gr} ( X ) = \left ( 2 - H ( t ) \right ) \vee \underline { \dim } _ { B, t } \, \text{Gr} ( H )$$
and
$$\forall t \in \mathbf R \, \sum \{ 0 \} ; \ \overline { \dim } _ { B, t } \, \text{Gr} ( X ) = \left ( 2 - H ( t ) \right ) \vee \overline { \dim } _ { B, t } \, \text{Gr} ( H ).$$
Hence, the graph has a local Box dimension at t when the previous two quantities are equal.
Proof. We first obtain the upper bound for both the lower and upper Box dimensions. Let t ∈ R and ρ > 0. Owing to Lemma 3.1, with probability one, there exist c 1 ( ω ) and ε > 0 such that for all u, v ∈ B t, ρ ( ),
$$| X _ { u } - X _ { v } | \leq c _ { 1 } | u - v | ^ { H ( t ) - \varepsilon } + c _ { 1 } | H ( u ) - H ( v ) |.$$
Let N δ ( H ) denotes the smallest number of balls of diameter δ necessary to cover the graph of H on the interval [ t -ρ, t + ]. Up to a constant, we can assume that this cover is organised ρ on columns of size δ , as usually presented in the case of a graph covering. Owing to the upper bound of the increment X u -X v , we know that by adding /ceilingleft δ H t ( ) -ε δ -1 /ceilingright balls to every column covering, we obtain a covering of the graph of the process X . Hence, there exists a constant c 2 such that
$$\forall \delta > 0 ; \ \ N _ { \delta } ( X ) \leq c _ { 2 } \left ( \rho \delta ^ { H ( t ) - \varepsilon - 2 } + N _ { \delta } ( H ) \right )$$
and therefore
$$\frac { \log ( N _ { \delta } ( X ) ) } { - \log ( \delta ) } & \leq \frac { \log ( c _ { 2 } \rho \delta ^ { H ( t ) - \varepsilon - 2 } + c _ { 2 } N _ { \delta } ( H ) ) } { - \log ( \delta ) } \\ & \leq \max \left ( \frac { \log ( c _ { 3 } \delta ^ { H ( t ) - \varepsilon - 2 } ) } { - \log ( \delta ) }, \frac { \log ( c _ { 3 } N _ { \delta } ( H ) ) } { - \log ( \delta ) } \right ).$$
Considering the inferior and superior limits of this inequality, we respectively obtain the upper bound of the lower and upper Box dimensions.
To prove the lower bound 2 -H t ( ), we use classic potential methods that have already been presented by Peltier and Lévy Véhel [29] and Herbin et al. [17] for the multifractional Brownian
motion. Briefly, for any interval [ a, b ] ⊂ R + , there exists C > 0 such that E [ X u -X v ] 2 ≥ C u | - | v 2 H , where H = min u ∈ [ a,b ] H u ( ). This estimate is sufficient, using a standard potential argument to show that almost surely,
$$2 - \min _ { u \in [ a, b ] } H ( u ) \leq \dim _ { H } G r ( X, [ a, b ] ) \leq \underline { \dim } _ { B } G r ( X, [ a, b ] ) \leq \overline { \dim } _ { B } G r ( f, [ a, b ] ).$$
Considering a, b ∈ Q , we obtain the inequality with probability one for any rational interval. Then, we note that Equation (4.1) holds as well if we use in the limit rational intervals containing t . Hence, owing to the continuity of H , with probability one and for all t ∈ R , we get 2 -H t ( ) ≤ dim B ,t Gr( X ) and 2 -H t ( ) ≤ dim B ,t Gr( X ).
To prove the remaining lower bound, let us first consider the case m t,H t ( ) = 1. The constant ρ can be chosen small enough such that for all ( v, h ) in neighbourhood of ( t, H ( )), t | ∂ H B v, h ( ) | > c 4 , where c 4 is a positive constant. Furthermore, for all u, v ∈ B t, ρ ( ),
$$X _ { u } - X _ { v } = B ( u, H ( u ) ) - B ( v, H ( v ) ) + \int _ { H ( v ) } ^ { H ( u ) } \partial _ { H } B ( v, h ) \, \mathrm d h.$$
Hence, in the neighbourhood of t ,
$$c _ { 4 } | H ( u ) - H ( v ) | \leq \left | \int _ { H ( v ) } ^ { H ( u ) } \partial _ { H } B ( v, h ) \, d h \right | \leq | X _ { u } - X _ { v } | + c _ { 5 } | u - v | ^ { H ( t ) - \varepsilon }.$$
Using the covering argument described at the beginning of the proof, we obtain that the inequality dim B ,t Gr( H ) ≤ ( 2 -H t ( ) ) ∨ dim B ,t Gr( X ) and dim B ,t Gr( H ) ≤ ( 2 -H t ( ) ) ∨ dim B ,t Gr( X ). Since we already know that the lower and upper Box dimensions are greater than 2 -H t ( ), it induces the expected lower bound.
↦
Finally, let us extend the inequality to any t ∈ R \ { 0 } and any value of m t,H t ( ) . We have already noted that the function t → dim ∗ ,t Gr( H ) is upper semi-continuous for every dimension considered. As a consequence, there exists a deterministic countable set E ⊂ R \ { 0 } such that for all t , there is a sequence ( t n n ) ∈ N ∈ E satisfying
$$\dim _ { *, t } \text{Gr} ( H ) = \lim _ { n \rightarrow \infty } \dim _ { *, t _ { n } } \text{Gr} ( H ),$$
/negationslash
Since for every s ∈ E ∂ , H B s, H s ( ( )) is a centered Gaussian variable with a positive variance, with probability one and for all s ∈ E , ∂ H B s, H s ( ( )) = 0. Hence, as the local dimension is upper semi-continuous,
$$\underline { \dim } _ { B, t } \, \text{Gr} ( X ) & \geq \lim _ { n \in N } \underline { \dim } _ { B, t _ { n } } \, \text{Gr} ( X ) \\ & = \lim _ { n \in N } \left ( 2 - H ( t _ { n } ) \right ) \vee \underline { \dim } _ { B, t _ { n } } \, \text{Gr} ( H ) = \left ( 2 - H ( t ) \right ) \vee \underline { \dim } _ { B, t } \, \text{Gr} ( H ).$$
The argument holds as well for the upper Box dimension, therefore uniformly proving the result.
Interestingly, we have obtained an expression of the Box dimension of the graph which is similar to the behaviour observed on a fractional Brownian motion with variable drift (see the recent works [11, 30]). This common property happens to extend as well to the Hausdorff
dimension of the graph, recently obtained by Peres and Sousi [31] on the fBm with variable drift.
The Hausdorff dimension of the graph is a bit more complicate to analyse. For this purpose, we have to introduce a more general dimension called the parabolic Hausdorff dimension . We first need to define a parabolic metric /rho1 H on R 2 , with H > 0:
$$\varrho _ { H } \left ( ( u, x ) \, ; ( v, y ) \right ) \coloneqq \max \left ( | u - v | ^ { H }, | x - y | \right ).$$
For any set A ⊂ R 2 , we denote by dim H ( A /rho1 ; H ) the parabolic Hausdorff dimension of A . It is defined similarly to the classic Hausdorff dimension using covering balls relatively to the metric /rho1 H , i.e. it corresponds to the infimum of s ≥ 0 for which
$$\lim _ { \delta \to 0 } \inf \left \{ \sum _ { i = 0 } ^ { \infty } \text{diam} ( O _ { i } \, ; \, \varrho _ { H } ) ^ { s } \, \colon ( O _ { i } ) _ { i \in N } \text{ is a $\delta$-cover of $A$} \right \} < \infty$$
Note that given the form of the parabolic metric, a /rho1 H -ball B (( t, x ) , δ ) has the following form: B (( t, x ) , δ ) = [ t -δ 1 /H , t + δ 1 /H ] × [ x -δ, x + ]. δ
The parabolic Hausdorff dimension has first been considered by Taylor and Watson [37] for the study of polar sets of the heat equation. Recently, it has been proved to be useful by Khoshnevisan and Xiao [20] to analyse the geometry of the images of Brownian motion. Finally, as previously outlined, Peres and Sousi [31] have shown it is also a natural idea for the study of fractional Brownian motion with variable drift. Note that the convention used in the latter work is slightly different from ours.
Theorem 4.2. Suppose X is a multifractional Brownian motion. Then, with probability one, it satisfies
$$\forall t \in \mathbf R \, \sum \{ 0 \} ; \quad \dim _ { H, t } \mathbf G r ( X ) = 1 + H ( t ) \left ( \dim _ { H, t } \left ( \mathbf G r ( H ) \, ; \, \varrho _ { H ( t ) } \right ) - 1 \right ).$$
Proof. To begin with, we prove that with probability one,
$$\forall t \in \mathbf R \, \sum \{ 0 \} ; \ \dim _ { H, t } ( \mathbf G r ( X ) \, ; \, \varrho _ { H ( t ) } ) = \dim _ { H, t } ( \mathbf G r ( H ) \, ; \, \varrho _ { H ( t ) } )$$
The sketch of the proof of this equality is similar to Theorem 4.1, as it uses as well an ad-hoc covering argument. Let us set δ > 0, ρ > 0 and γ > dim H ,t ( Gr( H ) ; /rho1 H t ( ) ) . Suppose ( O i ) i ∈ N is a δ -cover of Gr( H, t [ -ρ, t + ]) with of ρ /rho1 H t ( ) -balls and such that
$$\sum _ { i = 0 } ^ { \infty } \delta _ { i } ^ { \gamma } < \infty \ \text{ where } \delta _ { i } \coloneqq \text{diam} ( O _ { i } \, ; \varrho _ { H ( t ) } ) \ \forall i \in \mathbf N.$$
We observe that | X u -X v | ≤ c 1 | u - | v H t ( ) -ε + c 1 | H u ( ) -H v ( ) | for all u, v ∈ B t, ρ ( ). For any i ∈ N , we need at most 2 δ 1 -ε/H t ( ) i · δ -1 i /rho1 H t ( ) -balls of diameter δ i to cover the part of the graph related to the term | u - | v H t ( ) -ε . Hence, from ( O i ) i ∈ N , we construct a δ -cover ( V k ) k ∈ N of Gr( X, t [ -ρ, t + ]) which satisfies ρ
$$\sum _ { k = 0 } ^ { \infty } \text{diam} ( V _ { k } \, ; \, \varrho _ { H ( t ) } ) ^ { s } \leq c _ { 2 } \sum _ { i = 0 } ^ { \infty } \delta _ { i } ^ { - \varepsilon / H ( t ) } \cdot \delta _ { i } ^ { s }.$$
The last series converges when s ≥ γ + ε/H t ( ). Hence, considering the limits ε → 0, ρ → 0 and γ → dim H ,t ( Gr( H ) ; /rho1 H t ( ) ) , we obtain dim H ,t ( Gr( X ) ; /rho1 H t ( ) ) ≤ dim H ,t ( Gr( H ) ; /rho1 H t ( ) ) .
↦
To prove the other side inequality, we proceed similarly to the proof of Theorem 4.1. We first assume that m t,H t ( ) = 1. In this case, the estimate of the increments presented in Theorem 4.1 and the previous reasoning similarly imply that dim H ,t ( Gr( H ) ; /rho1 H t ( ) ) ≤ dim H ,t ( Gr( X ) ; /rho1 H t ( ) ) . Then, we use the upper semi-continuity of the function t → dim H ,t ( Gr( H ) ; /rho1 H t ( ) ) , obtained in Lemma 5.5, to extend the inequality to any t ∈ R \ { 0 } .
The upper bound of dim H ,t Gr( X ) is then a consequence of the deterministic Lemma 5.4, with H 1 = 1 and H 2 = H t ( ). With probability one, for all t ∈ R , we have dim H Gr( X, t [ -ρ, t + ]) ρ ≤ 1+ H t ( ) ( dim H ( Gr( X, t [ -ρ, t + ]); ρ /rho1 H t ( ) ) -1 . Therefore, when ) ρ → 0, we obtain the first upper bound with probability one: for all t ∈ R \{ } 0 , dim H ,t Gr( X ) ≤ 1+ H t ( ) ( dim H ,t ( Gr( H ) ; /rho1 t ) -1 . )
To prove the lower bound, we make use of potential arguments on the parabolic and euclidean metrics. We first set t ∈ R \ { 0 } . Then, let ρ > 0 and γ < dim H ( Gr( X,B t,ρ ( )) ; /rho1 H t ( ) ) . Using Frostman's lemma applied on the parabolic metric /rho1 t , there exists a probability measure µ on B t, ρ ( ) such that
$$\mathcal { E } _ { \gamma } ( \mu _ { H } ) = \iint _ { B ( t, \rho ) ^ { 2 } } \frac { \mu ( \mathrm d u ) \, \mu ( \mathrm d v ) } { \left ( | u - v | ^ { H ( t ) } + | H ( u ) - H ( v ) | \right ) ^ { \gamma } } < \infty.$$
↦
Suppose µ X denotes the image of the measure µ by the map u → ( u, X u ). To prove the lower on the dimension of the graph of X , we need to study s -energy of µ X , i.e.
$$\mathcal { E } _ { s } ( \mu _ { X } ) = \iint _ { \mathbf R ^ { 4 } } \frac { \mu _ { X } ( \mathrm d x ) \, \mu _ { X } ( \mathrm d y ) } { \| x - y \| ^ { s } } = \iint _ { B ( t, \rho ) ^ { 2 } } \frac { \mu ( \mathrm d u ) \, \mu ( \mathrm d v ) } { \left ( | u - v | ^ { 2 } + | X _ { u } - X _ { v } | ^ { 2 } \right ) ^ { - s / 2 } },$$
where s > 1. Let us set ε > 0. To study the previous energy, we need to split the integral into two components, respectively denoted E 1 s and E 2 s :
$$\mathcal { E } _ { s } ^ { 1 } + \mathcal { E } _ { s } ^ { 2 } = \iint _ { B ( t, \rho ) ^ { 2 } } \frac { \mu ( \mathrm d u ) \, \mu ( \mathrm d v ) } { \left ( | u - v | ^ { 2 } + | X _ { u } - X _ { v } | ^ { 2 } \right ) ^ { - s / 2 } } \left ( \mathbf 1 _ { \left \{ | H ( u ) - H ( v ) | \leq | u - v | ^ { H ( t ) ( 1 - \epsilon ) } \right \} } \\ + \mathbf 1 _ { \left \{ | H ( u ) - H ( v ) | > | u - v | ^ { H ( t ) ( 1 - \epsilon ) } \right \} } \right ).$$
Let us prove E 1 s is finite. For this purpose, we estimate its expectation
$$\mathbb { E } [ \mathcal { E } _ { s } ^ { 1 } ] = \iint _ { B ( t, \rho ) ^ { 2 } } \mathbb { E } \left [ \left ( | u - v | ^ { 2 } + | X _ { u } - X _ { v } | ^ { 2 } \right ) ^ { - s / 2 } \right ] \mathbf 1 _ { \left \{ | H ( u ) - H ( v ) | \leq | u - v | ^ { H ( t ) ( 1 - t ) } \right \} } \mu ( \mathrm d u ) \, \mu ( \mathrm d v ).$$
Using the notation σ 2 u,v := E [ X u -X v ] 2 , the inner term is equal to
$$\mathbb { E } \left [ \left ( | u - v | ^ { 2 } + | X _ { u } - X _ { v } | ^ { 2 } \right ) ^ { - s / 2 } \right ] & = \frac { 2 } { \sigma _ { u, v } \sqrt { 2 \pi } } \int _ { \mathbf R _ { + } } \left ( x ^ { 2 } + | u - v | ^ { 2 } \right ) ^ { - s / 2 } \exp \left ( - \frac { x ^ { 2 } } { 2 \sigma _ { u, v } ^ { 2 } } \right ) \mathrm d x \\ & = c _ { 1 } \int _ { \mathbf R _ { + } } \left ( r \sigma _ { u, v } ^ { 2 } + | u - v | ^ { 2 } \right ) ^ { - s / 2 } r ^ { - 1 / 2 } e ^ { - r / 2 } \, \mathrm d r.$$
Splitting the integral into two terms, and using classic approximations (see e.g. [13]), the previous term is upper bounded by
$$\int _ { 0 } ^ { | u - v | ^ { 2 } / \sigma _ { u, v } ^ { 2 } } | u - v | ^ { - s } r ^ { - 1 / 2 } \, \mathrm d r + \int _ { | u - v | ^ { 2 } / \sigma _ { u, v } ^ { 2 } } ^ { \infty } r ^ { - s / 2 } \sigma _ { u, v } ^ { - s } r ^ { - 1 / 2 } \, \mathrm d r \leq c _ { 2 } | u - v | ^ { 1 - s } \sigma _ { u, v } ^ { - 1 }.$$
We recall that Herbin [15] has proved that
$$\sigma _ { u, v } ^ { 2 } = \mathbb { E } [ X _ { u } - X _ { v } ] ^ { 2 } \asymp | u - v | ^ { 2 H ( t ) } + | H ( u ) - H ( v ) | ^ { 2 }$$
in the neighbourhood of t . Hence, on the domain { | H u ( ) -H v ( ) | ≤ | u - | v H t ( )(1 -ε ) } , we observe that
$$| u - v | & = \left ( | u - v | ^ { 2 H ( t ) ( 1 - \varepsilon ) } \right ) ^ { 1 / 2 H ( t ) ( 1 - \varepsilon ) } \\ & \geq c _ { 3 } \left ( | u - v | ^ { 2 H ( t ) ( 1 - \varepsilon ) } + | H ( u ) - H ( v ) | ^ { 2 } \right ) ^ { 1 / 2 H ( t ) ( 1 - \varepsilon ) } \\ & \geq c _ { 4 } \, \sigma _ { u, v } ^ { 1 / H ( t ) ( 1 - \varepsilon ) }. \\.$$
The expectation of E 1 s is therefore upper bounded by
$$\mathbb { E } [ \mathcal { E } _ { s } ^ { 1 } ] & \leq c _ { 4 } \iint _ { B ( t, \rho ) ^ { 2 } } \sigma _ { u, v } ^ { - 1 + ( 1 - s ) / H ( t ) ( 1 - \varepsilon ) } \, \mu ( \mathrm d u ) \, \mu ( \mathrm d v ) \\ & \leq c _ { 5 } \iint _ { B ( t, \rho ) ^ { 2 } } \left ( | u - v | ^ { H ( t ) } + | H ( u ) - H ( v ) | \right ) ^ { - 1 + ( 1 - s ) / H ( t ) ( 1 - \varepsilon ) } \mu ( \mathrm d u ) \, \mu ( \mathrm d v ).$$
The term E [ E 1 s ] is thus finite when -1 + (1 -s /H t ) ( )(1 -ε ) > -γ , i.e. when
$$s < 1 + ( \gamma - 1 ) \left ( H ( t ) ( 1 - \varepsilon ) \right ) \text{ \ where } \gamma < \dim _ { H } \left ( \text{Gr} ( X, B ( t, \rho ) ) \, ; \, \varrho _ { H ( t ) } \right ).$$
Hence, considering the limits on γ and ε , the upper bound obtained on s corresponds to the expected estimate.
Let now consider the second term E 2 s . We know that m t,H t ( ) = 1 almost surely. Recall that the increment X u -X v has the following form
$$X _ { u } - X _ { v } = B ( u, H ( u ) ) - B ( v, H ( u ) ) + \int _ { H ( v ) } ^ { H ( u ) } \partial _ { H } B ( v, h ) \, \mathrm d h.$$
If ρ is sufficiently small, there exists a positive constant c 1 > 0 such that for all s ∈ B t, ρ ( ) and any h ∈ [min B t,ρ ( ) H u , ( ) max B t,ρ ( ) H u ( )], we have | ∂ H B v, h ( ) | ≥ c 1 . Furthermore, there also exists c 2 > 0 such that
$$\forall u, v \in B ( t, \rho ) ; \ \left | B ( u, H ( u ) ) - B ( v, H ( u ) ) \right | \leq c _ { 2 } | u - v | ^ { H ( t ) ( 1 - \varepsilon / 2 ) }.$$
Using the previous estimates,
$$| X _ { u } - X _ { v } | & \geq \left | \int _ { H ( v ) } ^ { H ( u ) } \partial _ { H } B ( v, h ) \, \mathrm d h \right | - \left | B ( u, H ( u ) ) - B ( v, H ( u ) ) \right | \\ & \geq c _ { 1 } | H ( u ) - H ( v ) | \left ( 1 - c _ { 2 } | H ( u ) - H ( v ) | ^ { - 1 } \cdot | u - v | ^ { H ( t ) ( 1 - \varepsilon / 2 ) } \right )$$
Hence, on the domain { | H u ( ) -H v ( ) | > u | - | v H t ( )(1 -ε ) } , we obtain
$$| X _ { u } - X _ { v } | \geq c _ { 1 } | H ( u ) - H ( v ) | \left ( 1 - c _ { 2 } | u - v | ^ { H ( t ) \varepsilon / 2 } \right ) \geq c _ { 3 } | H ( u ) - H ( v ) |,$$
when ρ is sufficiently small. Therefore, the energy term E 2 s satisfies
$$\mathcal { E } _ { s } ^ { 2 } & \leq c _ { 4 } \iint _ { B ( t, \rho ) } | H ( u ) - H ( v ) | ^ { - s } \mathbf 1 _ { \{ | H ( u ) - H ( v ) | > | u - v | ^ { H ( v ) ( 1 - s ) } \} } \, \mu ( \mathrm d u ) \, \mu ( \mathrm d v ) \\ & \leq c _ { 5 } \iint _ { B ( t, \rho ) ^ { 2 } } \left ( | u - v | ^ { H ( t ) } + | H ( u ) - H ( v ) | \right ) ^ { - s } \mu ( \mathrm d u ) \, \mu ( \mathrm d v ).$$
The last integral is finite when s < γ . But, since γ > 1, we remark that γ (1 -H t ( )) > 1 -H t ( ) and thus γ > 1+ H t ( )( γ -1), showing that E 2 s is always finite when s < 1+( γ -1) H t ( ). Therefore, we have proved that the s -energy E s ( µ X ) is almost surely finite for all s < 1+( γ -1) ( H t ( )(1 -ε ) ) . Considering the limits ε → 0 and γ → dim H ,t ( Gr( X ) ; /rho1 H t ( ) ) , we obtain
$$\dim _ { H, t } \| \text{Gr} ( X ) \geq 1 + H ( t ) \| ( \dim _ { H, t } \| ( \text{Gr} ( H ) \, ; \, \varrho _ { H ( t ) } ) - 1 ) \ \text{ a.s.}$$
↦
The lower bound has been obtained for a fixed t ∈ R \ { 0 } . To extend the latter uniformly on R \ { 0 } , we proceed similarly to Theorem 4.1. Shortly, as proved in Lemma 5.5, the map t → dim H ,t ( Gr( H ) ; /rho1 H t ( ) ) is upper semi-continuous. Hence, there exists a countable set E such that for any t , there is sequence ( t n n ) ∈ N ∈ E satisfying
$$\dim _ { H, t } \left ( \text{Gr} ( H ) \, ; \varrho _ { H ( t ) } \right ) = \lim _ { n \rightarrow \infty } \dim _ { H, t _ { n } } \left ( \text{Gr} ( H ) \, ; \varrho _ { H ( t _ { n } ) } \right ).$$
The equality on the dimension of the graph holds almost surely for all s ∈ E . Therefore, with probability one, for all t ∈ \{ 0 } ,
$$\dim _ { H, t } \text{Gr} ( X ) & \geq \limsup _ { n \in \mathbb { N } } \dim _ { H, t _ { n } } \text{Gr} ( X ) \\ & = 1 + \limsup _ { n \in \mathbb { N } } \, H ( t _ { n } ) \left ( \dim _ { H, t _ { n } } \left ( \text{Gr} ( H ) \, ; \varrho _ { H ( t _ { n } ) } \right ) - 1 \right ) \\ & = 1 + H ( t ) \left ( \dim _ { H, t } \left ( \text{Gr} ( H ) \, ; \varrho _ { H ( t ) } \right ) - 1 \right ).$$
This last inequality concludes the proof, since the upper bound is already uniform on R .
Note that Theorem 4.2 does not contradict the result obtained by Peltier and Lévy Véhel [29] on H 0 -mBms. Indeed, under this assumption H 0 and due to the form of the parabolic metric /rho1 H t ( ) , covering the graph of H in the neighbourhood of t is equivalent to covering a straight line, implying that the local dimension satisfies: dim H ,t ( Gr( H ) ; /rho1 H t ( ) ) = H t ( ) -1 . Hence, the application of Theorem 4.2 proves that with probability one
$$\forall t \in \mathbf R \, \sum \{ 0 \} ; \ \dim _ { H, t } \text{Gr} \, X = 2 - H ( t ),$$
where X is an H 0 -multifractional Brownian motion.
It is interesting to observe that the Hausdorff and Box dimensions of the graph of mBm and fBm with variable drift have a similar form despite the rather different Gaussian structures
of these two processes. Indeed, the first one is a centered process whose incremental variance satisfies
$$\mathbb { E } [ X _ { u } - X _ { v } ] ^ { 2 } \asymp | u - v | ^ { 2 H ( t ) } + | H ( u ) - H ( v ) | ^ { 2 },$$
whereas the latter still has the covariance of fBm, but a mean equal to the drift f ( ). t
As a consequence of these similarities, the examples constructed by Peres and Sousi [31] can also be applied to the mBm, proving that the graph of the latter may have different Hausdorff and Box dimensions.
↦
Example 3. Peres and Sousi [31] have computed the parabolic Hausdorff dimension of the celebrated deterministic construction of McMullen [23]. Hence, there exists a continuous function H : R → [1 / 2 1) such that for all , t ∈ R , ˜ α X,t = log(2) / log(3) := θ ,
$$\dim _ { H, t } ( \text{Gr} ( H ) \, ; \, \varrho _ { 1 / 2 } ) = \log _ { 2 } ( 5 ^ { 2 \theta } + 1 ) \ \text{ and } \ \dim _ { B, t } \text{Gr} ( H ) = 2 - \theta.$$
Then, considering the mBm X with Hurst function H , we first observe that with probability one, for all t > 0, ˜ α X,t = θ . Furthermore, as a consequence of the properties of H , at every t > 0 such that H t ( ) = 1 / 2,
$$\dim _ { H, t } G r ( X ) = \frac { 1 + \log _ { 2 } \left ( 5 ^ { 2 \theta } + 1 \right ) } { 2 } < \dim _ { B, t } G r ( X ) = 2 - \theta.$$
Hence, at these particular times, the local Hausdorff dimension is strictly smaller that the Box one. In addition, as shown by Peres and Sousi [31], the local Hausdorff dimension is such that dim H ,t Gr( X > ) max dim { H ,t Gr( H , ) 3 / 2 } , meaning the simple and straightforward lower bound on the Hausdorff dimension is not optimal.
This example illustrates the kind of unusual fractal geometry the mBm can present. It clearly shows that the parabolic Hausdorff dimension is the relevant concept, as in this case, dim H ,t Gr( H ) happens to be insufficient to characterise entirely the Hausdorff dimension of the graph. We also note that in contrary to fBm, the fractal dimension of the graph can be completely unrelated to the local Hölder exponent.
As presented in the following second example, more classic sample path properties also occur.
Example 4. Suppose H is a sample path of a fractional Brownian motion with Hurst exponent α . Up to a rescaling, it is assumed to be valued in the interval (0 , 1). Then, using some classic estimates (see e.g. [13]), we easily obtain that
$$\forall t \in \mathbf R ; \ \dim _ { H, t } \left ( \mathbf G r ( H ) \, ; \, \varrho _ { H ( t ) } \right ) = \max \left \{ \frac { 1 } { H ( t ) }, 1 + \frac { 1 - \alpha } { H ( t ) } \right \}.$$
Then, owing to Theorem 4.2, a mBm X with Hurst function H satisfies with probability one:
$$\forall t \in { \mathbf R } \, \left \{ 0 \right \} ; \quad \dim _ { H, t } { \mathbf G } ( X ) = \dim _ { B, t } { \mathbf G } ( X ) = \max \left \{ 2 - \alpha, 2 - H ( t ) \right \} = 2 - \widetilde { \alpha } _ { X, t }.$$
This example displays a more common fractal geometry, where the Hausdorff and Box dimensions are both equal to 2 -˜ α X,t . Nevertheless, the local Hölder exponent may still have an irregular behaviour, depending on the value of H t ( ).
In the second part of this section, we are interested in studying the fractal dimension of an image X F ( ), where F can be any fractal set. A large literature has already investigated the
question of the geometry of images of Gaussian processes. It is a well-known result (see e.g. [19]) that for any fixed Borel set F ∈ (0 , ∞ ),
$$\dim _ { H } B ^ { H } ( F ) = \min \left \{ 1, \frac { \dim _ { H } F } { H } \right \} \text{ a.s.}$$
Mountford [27], Khoshnevisan et al. [21], Wu and Xiao [38] have generalized the previous result to ( N,d )-(fractional) Brownian sheet, and in particular, obtained uniform results on F under some assumption on the dimensions N and d .
Adapting classic arguments from Kahane [19], we can easily extend Equation (4.3) to an H 0 -mBm, i.e. for any fixed Borel set F ∈ (0 , ∞ ) and all t ∈ (0 , ∞ )
$$\dim _ { H, X ( t ) } X ( F ) = \min \left \{ 1, \frac { \dim _ { H, t } F } { H ( t ) } \right \} \text{ a.s.}$$
In the irregular case, the statement is not as simple and straightforward. Furthermore, the parabolic Hausdorff dimension again happens to be the relevant tool to describe the geometry of X F ( ).
Theorem 4.3. Suppose X is a multifractional Brownian motion. Then, with probability one,
$$\forall t \in \mathbf R \, \sum \{ 0 \} ; \ \dim _ { H, X ( t ) } X ( F ) = \min \left \{ 1, \dim _ { H, t } \left ( \mathbf G r ( H, F ) \, ; \varrho _ { t } \right ) \right \}.$$
Proof. The proof of this result is similar to the work of Peres and Sousi [31] on the drifted fBm. Let us first remark that
$$\dim _ { H, X ( t ) } X ( F ) \leq \dim _ { H, t } \text{Gr} ( X, F ).$$
↦
This inequality is a simple consequence of the property of the Hausdorff under the application of Lipschitz function (here, the projection p : ( t, X ( )) t → X t ( )). Then, using Lemma 2.1 from [31], we obtain dim H ,t Gr( X,F ) ≤ dim H ,t ( Gr( H,F ) ; /rho1 t ) , thus proving the upper bound.
To obtain the lower bound, we use a classic potential argument. Owing to Frostman's lemma, for any ρ > 0 and any γ < dim H ( Gr( H,F ∩ B t, ρ ( )) ; /rho1 H t ( ) ) , there exists a probability measure µ on B t, ρ ( ) such that
$$\iint _ { B ( t, \rho ) ^ { 2 } } \frac { \mu ( \mathrm d u ) \, \mu ( \mathrm d v ) } { \left ( | u - v | ^ { H ( t ) } + | H ( u ) - H ( v ) | \right ) ^ { \gamma } } < \infty.$$
Recall, as proved by Herbin [15] the covariance of the mBm has the following form
$$\sigma _ { u, v } ^ { 2 } \coloneqq \mathbb { E } [ X _ { u } - X _ { v } ] ^ { 2 } \asymp | u - v | ^ { 2 H ( t ) } + | H ( u ) - H ( v ) | ^ { 2 }$$
↦
Suppose µ X denotes the image of the measure µ under the map u → X u . Then, for any s ∈ (0 , 1), the expectation of its s -energy is equal to
$$\mathbb { * } _ { 1 } & \sim \mathbb { * } _ { 2 } \\ \mathbb { E } \left [ \iint _ { \mathbf R ^ { 2 } } \frac { \mu _ { X } ( \mathrm d x ) \, \mu _ { X } ( \mathrm d y ) } { \| x - y \| ^ { s } } \right ] & = \iint _ { \mathbf R ^ { 2 } } \mathbb { E } [ | X _ { u } - X _ { v } | ^ { - s } ] \mu ( \mathrm d u ) \, \mu ( \mathrm d v ) \\ & = c _ { 2 } \iint _ { \mathbf R ^ { 2 } } \sigma _ { u, v } ^ { - 2 s } \, \mu ( \mathrm d u ) \, \mu ( \mathrm d v )$$
Therefore, owing to the form of the covariance, the integral is finite when s ≤ γ . Hence, considering the limit ρ → 0, we obtain dim H ,X t ( ) X F ( ) ≥ dim H ,t ( Gr( H,F ) ; /rho1 t ) , which concludes the proof.
The local parabolic Hausdorff dimension dim H ,t ( Gr( H,F ) ; /rho1 t ) might not seem to be a very intuitive notion at first. To have a better intuition of the result described in Theorem 4.3, we may first note, similarly to Theorem 4.2, it does not contradict Equation (4.4) in the case of an H 0 -multifractional Brownian motion. Indeed, when H t ( ) < α ˜ H,t , due to the property of the parabolic metric /rho1 H t ( ) , it is completely equivalent to cover F and Gr( H,F ) in the neighbourhood of t . Hence, in this case
$$\dim _ { H, t } \left ( \text{Gr} ( H, F ) \, ; \, \varrho _ { t } \right ) & = \lim _ { \rho \rightarrow 0 } \dim _ { H } \left ( \text{Gr} ( H, F \cap B ( t, \rho ) ) \, ; \, \varrho _ { t } \right ) \\ & = \lim _ { \rho \rightarrow 0 } H ( t ) ^ { - 1 } \dim _ { H } \left ( F \cap B ( t, \rho ) \right ) = H ( t ) ^ { - 1 } \dim _ { H, t } F.$$
The next example shows that different behaviours may occur when the mBm is irregular.
Example 5. Let H be a trajectory of fractional Brownian motion defined on the interval [0 , 1], with Hurst index α ∈ (0 1). , Up to a rescaling and a translation, the sample path satisfies H ([0 , 1]) ⊂ (0 , 1).
We first assume that α < H t ( ). Owing to Equation (4.3), there exists a Borel set F such that dim H ,H t ( ) H F ( ) = α -1 dim H ,t F . Furthermore, since the proof of this statement is based on Frostman's lemma, there exists a probability measure on F ∩ B t, ρ ( ) such that ∫∫ R 2 | H u ( ) -H v ( ) | -γ µ (d u µ ) (d v ) < ∞ where γ < dim H ,H t ( ) H F ( ). Therefore, still using Frostman's lemma, we obtain that α -1 dim H ,t F ≤ dim H ,t ( Gr( H,F ) ; /rho1 t ) . The other side inequality is simply obtained by observing that ˜ α X,t = α , inducing the upper bound using classic covering arguments. Hence, we obtain
$$\dim _ { H, X ( t ) } X ( F ) = \dim _ { H, H ( t ) } H ( F ) = \frac { 1 } { \alpha } \dim _ { H, t } F \text{ \ a.s. }$$
For the second example, we consider the set F = H -1 { 1 2 } , which might be assumed to be non-empty. Monrad and Pitt [26] have proved that dim H F = 1 -α . In this case, as the Hurst function H is constant on the set F , we easily prove that the dimension of Gr( H,F ) is characterised by the dimension on F . Therefore,
$$\dim _ { H, X ( t ) } X ( F ) = \dim _ { H, t } ( \text{Gr} ( H, F ) \, ; \, \varrho _ { t } ) = \frac { 1 } { H ( t ) } \dim _ { H, t } F = \frac { 1 - \alpha } { H ( t ) } \text{ \ a.s. } \text{ \quad (4.6)}$$
The two previous examples show that the estimate obtained in Theorem 4.3 can lead to very different results, depending on the properties of the set F chosen. In particular, we observe that the Hausdorff geometry of images displays a richer structure than the local Hölder regularity. Indeed, in the irregular case, we know that the later only depends on H ( ) · sample path properties, whereas Equations (4.5) and (4.6) show a more complex geometry of the image X F ( ). In the first case, we observe that the formula depends only on the local exponent of H and the Hausdorff dimension of F , whereas in the second one, Equation (4.6) presents a case where the Hausdorff dimension follows the classic formula (4.3), even though the mBm is irregular at t .
To conclude this section, we discuss the question of obtaining the Hausdorff dimension of the level sets X -1 ( { x } ). This problem has already been investigated by Boufoussi et al. [9] for the H 0 -multifractional Brownian motion, proving in particular that for all t ∈ R \ { 0 } ,
$$\dim _ { H, t } X ^ { - 1 } ( \{ X _ { t } \} ) = 1 - H ( t ) \text{ a.s.}$$
It is quite natural to wonder if the previous result can be extended to the irregular mBm. The upper bound can easily extended, since it is a direct consequence of the Hölder regularity of the process at t . In addition, we note that the proof from [9] of the lower bound stands as well in the irregular case. As a consequence, we easily obtain bounds for all t ∈ R \ { 0 } ,
$$1 - H ( t ) \leq \dim _ { H, t } X ^ { - 1 } ( \{ X _ { t } \} ) \leq 1 - \min ( H ( t ), \widetilde { \alpha } _ { H, t } ) \ \text{ a.s.}$$
Lower and upper bounds are not equal in the irregular case, and the remaining question is whether the lower one can be improved. The sketch of the proof of this inequality is rather classic and relies on the study of the Hölder continuity of the local time (see e.g. the seminal works of Berman [7] and Geman and Horowitz [14] on the subject). The key point to prove the Hölder regularity is a property called the local nondeterminism (LND) . In the case of the mBm, it states that for every n ∈ N , there exists c n > 0 such that for all ( t 1 , . . . , t n , t ) ∈ ( ε, ∞ ) n +1
$$\text{Var} ( X _ { t } | X _ { t _ { 1 } }, \dots, X _ { t _ { n } } ) \geq c _ { n } \min _ { 1 \leq k \leq n } | t - t _ { k } | ^ { 2 H ( t ) }.$$
To obtain a better lower bound for the local dimension of the level set, we would need to improve the exponent in the right term. Unfortunately, the next lemma proves that it is optimal under a mild assumption on the Hurst function.
Lemma 4.1. Let t ∈ R and suppose there exists a sequence s i → t such that H s ( i ) = H t ( ) . Then, for any n ∈ N , any ε > 0 and any c n > 0 , there exists ( t 1 , . . . , t n ) ∈ R n such that
$$\text{Var} ( X _ { t } \left | X _ { t _ { 1 } }, \dots, X _ { t _ { n } } ) \leq c _ { n } \min _ { 1 \leq k \leq n } \left | t - t _ { k } \right | ^ { 2 H ( t ) - \varepsilon }.$$
Proof. For all ( t 1 , . . . , t n ) ∈ R n and any k ∈ { 1 , . . . , n } , we note that Var( X t | X ,...,X t 1 t n ) ≤ Var( X t | X t k ) ≤ E [ ( X t -X t k ) 2 ] . Using elements t k ∈ { s i } i ∈ N and the estimate of the variance of the increments, we obtain that
$$\text{Var} ( X _ { t } | X _ { t _ { 1 } }, \dots, X _ { t _ { n } } ) \leq C \min _ { 1 \leq k \leq n } | t - t _ { k } | ^ { 2 H ( t ) }.$$
Let set ρ > 0. We may assume that t k ∈ B t, ρ ( ) for every k ∈ { 1 , . . . , n } , implying that
$$\text{Var} ( X _ { t } | X _ { t _ { 1 } }, \dots, X _ { t _ { n } } ) \leq C \rho ^ { \varepsilon } \min _ { 1 \leq k \leq n } | t - t _ { k } | ^ { 2 H ( t ) - \varepsilon }.$$
Using ρ sufficiently small, it proves that the property of local nondeterminism does not stand with the exponent 2 H t ( ) -ε , for any ε > 0.
Hence, the classic exponent 2 H t ( ) in the LND property of the mBm is most of the time optimal. Intuitively, it can be understood as following: in the neighbourhood of t , the Hurst function can influence the Hölder regularity, but is still a deterministic component and thus does
not bring any randomness which could increase the conditional variance. As the consequence, the latter is only a result of the local fBm form of the multifractional Brownian motion.
Therefore, the estimate of the Hausdorff dimension of the level sets can not be improved using the classic methods based on the LND property, and it remains an open question to determine the precise form of the fractal dimension in terms of the geometric properties of the Hurst function H .
## 5. Appendix: a few deterministic properties
We gather in the appendix a few deterministic properties related to 2-microlocal analysis and local graph dimensions and which have been used through the paper. We begin with a simple upper bound on the 2-microlocal which extends the classic property of lower semi-continuity on the local Hölder exponent.
Lemma 5.1. Suppose f is a continuous function and t ∈ R such that lim inf u → t ˜ α f,t < 1 . Then, the 2-microlocal frontier of f at t satisfies
$$\forall s ^ { \prime } \in \mathbf R ; \ \ \sigma _ { f, t } ( s ^ { \prime } ) \leq \liminf _ { u \to t } \widetilde { \alpha } _ { f, u }.$$
Proof. Let us set ρ > 0 and ε > 0. There exists s ∈ B t, ρ ( ) such that ˜ α f,s ≤ lim inf u → t ˜ α f,u + . ε Then, we can find u , v n n → s such that
$$| f ( u _ { n } ) - f ( v _ { n } ) | \geq n \, | u _ { n } - v _ { n } | ^ { \tilde { \alpha } _ { f, s } + \varepsilon }.$$
Therefore, for any s ′ ∈ R , we have
$$\frac { | f ( u _ { n } ) - f ( v _ { n } ) | } { | u _ { n } - v _ { n } | ^ { \widetilde { \alpha } _ { f, s } + \varepsilon } \left ( | u _ { n } - t | + | v _ { n } - t | \right ) ^ { - s ^ { \prime } } } \geq \frac { n } { \left ( | u _ { n } - t | + | v _ { n } - t | \right ) ^ { - s ^ { \prime } } } \longrightarrow _ { n \to + \infty } + \infty,$$
$$\iota \, \sigma _ { f, t } ( s ^ { \prime } ) \leq \liminf _ { u \to t } \widetilde { \alpha } _ { f, u } + \varepsilon, \, \text{for any } \varepsilon > 0.$$
proving that σ f,t ( s ) ≤ lim inf u → t ˜ α f,u + , for any ε ε >
In the next two lemmas, we present an extension of a well-known property connecting the local dimension of the graph to the local Hölder exponent: dim B ,t Gr( f ) ≤ 2 -˜ α f,t .
Lemma 5.2. Suppose f is a continuous function. Then, for all t ∈ R ,
$$\underline { \dim } _ { B, t } \, \text{Gr} ( f ) \leq \overline { \dim } _ { B, t } \, \text{Gr} ( f ) \leq 2 - \left ( \sigma _ { f, t } ( 1 ) \wedge 1 \right ).$$
Proof. We need to construct a local cover of the graph which is sufficiently efficient. Let first set t ∈ R and σ < σ f,t (1) ∧ 1. Then, there exists ρ > 0 and C > 0 such that for all u, v ∈ B t, ρ ( ),
$$| f ( u ) - f ( v ) | \leq C | u - v | ^ { \sigma } \left ( | u - t | + | v - t | \right ) ^ { - 1 }.$$
The cover is constructed similarly to the original proof. Hence, if N δ denotes the number of
δ -boxes necessary to cover Gr( f, [ t -ρ, t + ]), we have ρ
$$N _ { \delta } \leq C \delta ^ { - 1 } + \sum _ { k = - \lceil \rceil \rho / \delta \rceil } ^ { \lceil \rho / \delta \rceil } \sup _ { u, v \in t + [ k \delta, ( k + 1 ) \delta \rceil } | f ( u ) - f ( v ) | \cdot \delta ^ { - 1 }$$
$$N _ { \delta } & \leq C \delta ^ { - 1 } + \sum _ { k = - \lceil \rho / \delta \rceil } ^ { \lceil \rho / \delta \rceil } \sup _ { u, v \in t + \lfloor k \delta, ( k + 1 ) \delta \rceil } | f ( u ) - f ( v ) | \cdot \delta ^ { - 1 } \\ & \leq C \delta ^ { - 1 } + \sum _ { k = - \lceil \rho / \delta \rceil } ^ { \lceil \rho / \delta \rceil } C \delta ^ { \sigma } ( k \delta ) ^ { - 1 } \cdot \delta ^ { - 1 } \\ & = C \delta ^ { - 1 } + 2 C \delta ^ { \sigma - 2 } \sum _ { k = 1 } ^ { \lceil \rho / \delta \rceil } k ^ { - 1 } \leq 2 C \delta ^ { \sigma - 2 } ( 1 + \log \lceil \rho / \delta \rceil ). \\ \cdot 1. \text{ Then}.$$
since 2 -σ > 1. Then,
$$\overline { \dim } _ { B } \, \text{Gr} ( f, [ t - \rho, t + \rho ] ) = \lim \sup _ { \delta \to 0 } \frac { N _ { \delta } } { - \log \delta } \leq 2 - \sigma,$$
for all σ < σ f,t 0 (1) ∧ 1, which proves to the expected inequality.
Lemma 5.3. Suppose f is a continuous function. Then, for all t ∈ R
$$\dim _ { H, t } \text{Gr} ( f ) \leq 2 - \left ( \sigma _ { f, t } ( + \infty ) \wedge 1 \right ).$$
Proof. As the estimate we want to obtain differs from the previous lemma, we need to adopt a constructing procedure which is slightly different. Let t ∈ R and σ < σ f,t (+ ∞ ∧ ) 1. Then, there exist s ′ ≥ 0, ρ > 0 and C > 0 such that for all u, v ∈ B t, ρ ( )
$$| f ( u ) - f ( v ) | \leq C | u - v | ^ { \sigma } \left ( | u - t | + | v - t | \right ) ^ { - s ^ { \prime } }.$$
Let us set δ > 0 and γ > 1. The δ -cover of Gr( f, [ t -ρ, t + ]) is constructed as following. There ρ exists n ∈ N such that for all k ≥ n , k -γ < δ . To cover the set [ t -ρ, t + ], we first choose ρ successive intervals ( I k ) k ≥ n of size k -γ in the neighbourhood of t . The rest is simply covered by intervals of size δ . The first type of intervals cover a set whose length is equal to:
$$\sum _ { k = n } ^ { + \infty } k ^ { - \gamma } \geq \int _ { n } ^ { + \infty } x ^ { - \gamma } \mathrm d x = C _ { \gamma } n ^ { 1 - \gamma }.$$
Hence, the number M δ of intervals of size δ which are necessary is upper bounded by
$$M _ { \delta } \leq \delta ^ { - 1 } ( 2 \rho - C _ { \gamma } n ^ { 1 - \gamma } ).$$
Let now consider the cover ( O i ) i ∈ N of Gr( f, [ t -ρ, t + ]) which can be defined from the previous ρ intervals (still using the same idea of covering by columns). It satisfies for any s > 0
$$\sum _ { i \in \mathbb { N } } | O _ { i } | ^ { s } \leq \sum _ { k = n } ^ { + \infty } ( k ^ { - \gamma } ) ^ { s } \cdot k ^ { \gamma } \sup _ { u, v \in I _ { k } } | f ( u ) - f ( v ) | + \sum _ { k = 1 } ^ { M _ { \delta } } \delta ^ { s } \cdot \delta ^ { - 1 } \sup _ { u, v \in \widetilde { I } _ { k } } | f ( u ) - f ( v ) |.$$
Let us estimate the first term,
$$\sum _ { k = n } ^ { + \infty } ( k ^ { - \gamma } ) ^ { s } \cdot k ^ { \gamma } \sup _ { u, v \in I _ { k } } | f ( u ) - f ( v ) | \leq \sum _ { k = n } ^ { + \infty } ( k ^ { - \gamma } ) ^ { \sigma + s - 1 } | x _ { k } - t | ^ { - s ^ { \prime } }$$
$$\sum _ { k = n } ^ { \sum _ { k } ^ { \prime } } ( \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \mu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^{\ 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2} \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nu ^ { 2 } \nolimits _ { k = n } ^ { \sum _ { k = n } ^ { + \infty } ( k ^ { - 1 } ) ^ { \gamma ( \sigma + s - 1 ) - s ^ { \prime } ( \gamma - 1 ) }.$$
where we have used that d t, I ( k ) ∼ k 1 -γ . The previous series converges if s satisfies γ σ ( + s -1) -s ′ ( γ -1) > 1, i.e. if s > 1 -σ + s ′ (1 -γ -1 ) + γ -1 .
Let now consider the second term,
$$\sum _ { k = 1 } ^ { M _ { \delta } } \delta ^ { s } \cdot \delta ^ { - 1 } \sup _ { u, v \in \widetilde { I } _ { k } } | f ( u ) - f ( v ) | \leq \sum _ { k = 1 } ^ { M _ { \delta } } \delta ^ { \sigma + s - 1 } \, | \widetilde { x } _ { k } - t | ^ { - s ^ { \prime } }$$
$$\leq \delta ^ { \sigma + s - 1 } \sum _ { \L } ^ { M _ { \delta } } \left ( C _ { \gamma } n ^ { 1 - \gamma } + k \delta \right ) ^ { - s ^ { \prime } }.$$
$$\sum _ { k = 1 } ^ { \mu } \mu ^ { \mu } \sum _ { \substack { \mu \mu \nu \\ u, v \in \widetilde { I } _ { k } } } \nu \sum _ { \substack { \nu \nu \\ k = 1 } } ^ { \mu } \mu ^ { \mu \nu } \mu ^ { \mu } \\ \leq \delta ^ { \sigma + s - 1 } \sum _ { k = 1 } ^ { M _ { \delta } } ( C _ { \gamma } n ^ { 1 - \mu }$$
The sum can upper bounded by
$$\int _ { 0 } ^ { M _ { \delta } } \left ( C _ { \gamma } n ^ { 1 - \gamma } + x \delta \right ) ^ { - s ^ { \prime } } \mathrm d x & \leq \delta ^ { - 1 } \int _ { 0 } ^ { \rho } \left ( C _ { \gamma } n ^ { 1 - \gamma } + v \right ) ^ { - s ^ { \prime } } \mathrm d v \\ & \leq c _ { 2 } \, \delta ^ { - 1 } \cdot n ^ { ( \gamma - 1 ) ( s ^ { \prime } - 1 ) } \\ & \leq c _ { 3 } \, \delta ^ { - 1 } \cdot \delta ^ { - ( \gamma - 1 ) ( s ^ { \prime } - 1 ) / \gamma }.$$
Combining the previous estimate, we obtain
$$\sum _ { k = 1 } ^ { M _ { \delta } } \delta ^ { s } \cdot \delta ^ { - 1 } \sup _ { u, v \in \widetilde { I } _ { k } } | f ( u ) - f ( v ) | \leq c _ { 4 } \, \delta ^ { \sigma + s - 2 - ( \gamma - 1 ) ( s ^ { \prime } - 1 ) / \gamma }.$$
Hence, the latter converges as δ → 0 when s > 2 -σ +( γ -1)( s ′ -1) /γ . Therefore, combining the two conditions we have obtained, the Hausdorff dimension satisfies
$$\dim _ { H } \text{Gr} ( f, [ t - \rho, t + \rho ] ) \leq \min \{ 1 - \sigma + s ^ { \prime } ( 1 - \gamma ^ { - 1 } ) + \gamma ^ { - 1 }, 2 - \sigma + ( \gamma - 1 ) ( s ^ { \prime } - 1 ) / \gamma \},$$
for all γ > 1 and s ′ ∈ R . Considering the limit γ → 1, we obtain dim H Gr( f, [ t -ρ, t + ]) ρ ≤ 2 -σ , which implies the expected inequality.
↦
$$\forall s ^ { \prime } \in \mathbf R ; \ \ \sigma _ { f, 0 } ( s ^ { \prime } ) = \frac { s ^ { \prime } + \alpha } { 1 + \beta }.$$
We note that the two new upper bounds presented in the previous Lemmas are consistent with the original one, since ˜ α f,t = σ f,t (0) ≤ σ f,t (1) ≤ σ f,t (+ ∞ ). Furthermore, it proves to be the optimal ones for a classic deterministic function called the chirp function : f : x → | x | α sin( | x | -β ), with α, β > 0. As proved by Echelard [12], its 2-microlocal frontier at 0 is equal to
It is an easy calculus to check that the local Hausdorff and Box dimensions at zero are given by
$$\dim _ { H, 0 } \, \text{Gr} ( f ) = 1 \quad \text{and} \quad \dim _ { B, 0 } \, \text{Gr} ( f ) = 2 - \left ( \frac { 1 + \alpha } { 1 + \beta } \right ) \wedge 1,$$
Hence, they are equal to the upper bounds presented in Lemmas 5.2 and 5.3, on the contrary to the usual Hölder estimate 2 -α/ (1 + β ) which is not optimal.
To end this section, we prove two useful lemmas related to the parabolic Hausdorff dimension. The first one constitutes an extension of an inequality proved by Peres and Sousi [31] (Lemma 2 1). .
Lemma 5.4. For every A ⊂ R 2 and all H > H 1 2 > 0 ,
$$\dim _ { H } ( A \, ; \, \varrho _ { H _ { 2 } } ) + \frac { 1 } { H _ { 2 } } - \frac { 1 } { H _ { 1 } } \ \leq \ \dim _ { H } ( A \, ; \, \varrho _ { H _ { 1 } } ) \ \leq \ 1 + \frac { H _ { 2 } } { H _ { 1 } } ( \dim _ { H } ( A \, ; \, \varrho _ { H _ { 2 } } ) - 1 ).$$
↦
As a consequence, if H ( ) · is a positive continuous function, the map t → dim H ( A /rho1 ; H t ( ) ) is also continuous.
Proof. Suppose A ⊂ R 2 and H > H 1 2 . Let us first prove the upper bound on dim H ( A /rho1 ; H 1 ) . For any γ > dim H ( A /rho1 ; H 2 ) , there exists a /rho1 H 2 -cover ( O i ) i ∈ N of A such that
$$\sum _ { i = 0 } ^ { \infty } \delta _ { i } ^ { \gamma } < \infty \ \text{ where } \ \delta _ { i } \coloneqq \text{diam} ( O _ { i } \, ; \, \varrho _ { H _ { 2 } } ) \ \forall i \in \mathbf N.$$
We denote by ρ i the quantity δ H /H 1 2 i . Without any loss of generality, we may assume that for every i ∈ N , O i is a rectangle of size δ 1 /H 2 i × δ i (as a simple consequence of the definition of /rho1 H 2 ). We want to construct a /rho1 H 1 -cover of A from ( O i ) i ∈ N . Since ρ 1 /H 1 i = δ 1 /H 2 i , we need to split O i along the space axis to obtain /rho1 H 1 -balls. More precisely, there will be at most 2 δ /ρ i i = 2 δ 1 -H /H 1 2 i resulting /rho1 H 1 -balls of diameter ρ i . Hence, we obtain a /rho1 H 1 -cover ( V k ) k ∈ N of A such that
$$\sum _ { k = 0 } ^ { \infty } \text{diam} \left ( V _ { k } \, ; \varrho _ { H _ { 1 } } \right ) ^ { s } \leq 2 \sum _ { i = 0 } ^ { \infty } \delta _ { i } ^ { 1 - H _ { 1 } / H _ { 2 } } \cdot \delta _ { i } ^ { s H _ { 1 } / H _ { 2 } }.$$
The last sum is finite when 1 + H /H 1 2 ( s -1) > γ , i.e. when s > 1+( γ -1) H /H 2 1 . Considering γ → dim H ( A /rho1 ; H 2 ) , we obtain the expected upper bound.
The lower bound is obtained using a similar reasoning. For any s > dim H ( A /rho1 ; H 1 ) , there exists a /rho1 H 1 -cover ( V k ) k ∈ N of A such that
$$\sum _ { k = 0 } ^ { \infty } \rho _ { k } ^ { s } < \infty \ \text{ where } \ \rho _ { k } \coloneqq \text{diam} \left ( V _ { k } \, ; \, \varrho _ { H _ { 1 } } \right ) \ \forall k \in \mathbf N.$$
Similarly, we may assume that V k is a rectangle of size ρ 1 /H 1 k × ρ k for every k ∈ N . As H > H 1 2 , we note that ρ 1 /H 2 < ρ 1 /H 1 k . Hence, to obtain a /rho1 H 2 -cover of A , we split the previous balls along the time axis, in at most 2 ρ 1 /H 1 -1 /H 2 k sub-balls. Therefore, we obtain a /rho1 H 2 -cover ( O i ) i ∈ N of A such that
$$\sum _ { i = 0 } ^ { \infty } \text{diam} \left ( O _ { i } \, ; \, \varrho _ { H _ { 2 } } \right ) ^ { \gamma } \leq 2 \sum _ { k = 0 } ^ { \infty } \rho _ { k } ^ { 1 / H _ { 1 } - 1 / H _ { 2 } } \cdot \rho _ { k } ^ { \gamma }.$$
↦
This sum is finite when γ +1 /H 1 -1 /H 2 > s , inducing the lower bound on dim H ( A /rho1 ; H 1 ) . The continuity of the map t → dim H ( A /rho1 ; H t ( ) ) is straightforward as we observe that lower and upper bounds converge to dim H ( A /rho1 ; H 1 ) when H 2 → H 1 .
The previous lemma helps us to extend the property of upper semi-continuity to the local parabolic Hausdorff dimension.
↦
Lemma 5.5. Suppose A ⊂ R 2 and H ( ) · is a positive continuous function. Then, the map t → dim H ,t ( A /rho1 ; H t ( ) ) is upper semi-continuous, i.e. for all t ∈ R
$$\dim _ { H, t } ( A \, ; \, \varrho _ { H ( t ) } ) \geq \lim _ { s \to t } \sup _ { s \to t } \dim _ { H, s } ( A \, ; \, \varrho _ { H ( s ) } ).$$
Proof. For any t ∈ R , the local parabolic Hausdorff dimension satisfies
$$\dim _ { H, t } ( A \, ; \varrho _ { H ( t ) } ) \geq \lim _ { s \to t } \sup _ { s \to t } \dim _ { H, s } ( A \, ; \varrho _ { H ( t ) } ).$$
This inequality is straightforward extension of the usual one on the Hausdorff dimension. Let δ > 0. Owing to the uniform bounds obtained in Lemma 5.4, there exists c δ > 0 such that for all s ∈ B t, δ ( ) and all ρ > 0,
$$\left | \dim _ { H } ( A \cap B ( s, \rho ) \, ; \, \varrho _ { H ( t ) } ) - \dim _ { H } ( A \cap B ( s, \rho ) \, ; \, \varrho _ { H ( s ) } ) \right | \leq c _ { \delta } | H ( s ) - H ( t ) |.$$
↦
Hence, considering the limit ρ → 0, we obtain | dim H ,s ( A /rho1 ; H t ( ) ) -dim H ,s ( A /rho1 ; H s ( ) ) | ≤ c δ | H s ( ) -H t ( ) , | proving, jointly with the first inequality, the upper semi-continuity of the map t → dim H ,t ( A /rho1 ; H t ( ) ) .
## References
- [1] A. Ayache. Continuous Gaussian multifractional processes with random pointwise Hölder regularity. J. Theoret. Probab. , 26(1):72-93, 2013.
- [2] A. Ayache and M. S. Taqqu. Multifractional processes with random exponent. Publ. Mat. , 49(2):459-486, 2005.
- [3] A. Ayache, S. Cohen, and J. Lévy Véhel. The covariance structure of multifractional brownian motion, with application to long range dependence. In Proceedings of the Acoustics, Speech, and Signal Processing, 2000 , volume 6, pages 3810-3813, Washington, DC, USA, 2000. IEEE Computer Society.
- [4] A. Ayache, N.-R. Shieh, and Y. Xiao. Multiparameter multifractional brownian motion: local nondeterminism and joint continuity of the local times. Ann. Inst. H. Poincaré Probab. Statist , 2011.
- [5] P. Balança. Fine regularity of Lévy processes and linear (multi)fractional stable motion. Preprint , 2013. arXiv:1302.3140 .
- [6] A. Benassi, S. Jaffard, and D. Roux. Elliptic Gaussian random processes. Rev. Mat. Iberoamericana , 13(1):19-90, 1997.
- [7] S. M. Berman. Local nondeterminism and local times of Gaussian processes. Indiana Univ. Math. J. , 23:69-94, 1973.
- [8] J.-M. Bony. Second microlocalization and propagation of singularities for semilinear hyperbolic equations. In Hyperbolic equations and related topics (Katata/Kyoto, 1984) , pages 11-49. Academic Press, Boston, MA, 1986.
- [9] B. Boufoussi, M. Dozzi, and R. Guerbaz. On the local time of multifractional Brownian motion. Stochastics , 78(1):33-49, 2006.
- [10] B. Boufoussi, M. Dozzi, and R. Guerbaz. Sample path properties of the local time of multifractional Brownian motion. Bernoulli , 13(3):849-867, 2007.
- [11] P. H. A. Charmoy, Y. Peres, and P. Sousi. Minkowski dimension of brownian motion with drift. 2012.
- [12] A. Echelard. Analyse 2-microlocale et application au débruitage . PhD thesis, Université de Nantes, 2007. http://tel.archives-ouvertes.fr/tel-00283008/fr/ .
- [13] K. Falconer. Fractal geometry . John Wiley & Sons Inc., Hoboken, NJ, second edition, 2003. Mathematical foundations and applications.
- [14] D. Geman and J. Horowitz. Occupation densities. Ann. Probab. , 8(1):1-67, 1980.
- [15] E. Herbin. From N parameter fractional Brownian motions to N parameter multifractional Brownian motions. Rocky Mountain J. Math. , 36(4):1249-1284, 2006.
- [16] E. Herbin and J. Lévy Véhel. Stochastic 2-microlocal analysis. Stochastic Process. Appl. , 119(7):2277-2311, 2009.
- [17] E. Herbin, B. Arras, and G. Barruel. From almost sure local regularity to almost sure Hausdorff dimension for Gaussian fields. Preprint , 2012.
- [18] S. Jaffard. Pointwise smoothness, two-microlocalization and wavelet coefficients. Publ. Mat. , 35(1):155-168, 1991. Conference on Mathematical Analysis (El Escorial, 1989).
- [19] J.-P. Kahane. Some random series of functions , volume 5 of Cambridge Studies in Advanced Mathematics . Cambridge University Press, Cambridge, second edition, 1985.
- [20] D. Khoshnevisan and Y. Xiao. Brownian motion and thermal capacity. Preprint , 2012.
- [21] D. Khoshnevisan, D. Wu, and Y. Xiao. Sectorial local non-determinism and the geometry of the Brownian sheet. Electron. J. Probab. , 11:no. 32, 817-843, 2006.
- [22] K. M. Kolwankar and J. Lévy Véhel. A time domain characterization of the fine local regularity of functions. J. Fourier Anal. Appl. , 8(4):319-334, 2002.
- [23] C. McMullen. The Hausdorff dimension of general Sierpiński carpets. Nagoya Math. J. , 96:1-9, 1984.
- [24] M. Meerschaert, D. Wu, and Y. Xiao. Local times of multifractional Brownian sheets. Bernoulli , 14(3):865-898, 2008.
- [25] Y. Meyer. Wavelets, vibrations and scalings , volume 9 of CRM Monograph Series . American Mathematical Society, Providence, RI, 1998.
- [26] D. Monrad and L. D. Pitt. Local nondeterminism and Hausdorff dimension. In Seminar on stochastic processes, 1986 (Charlottesville, Va., 1986) , volume 13 of Progr. Probab. Statist. , pages 163-189. Birkhäuser Boston, Boston, MA, 1987.
- [27] T. S. Mountford. Uniform dimension results for the Brownian sheet. Ann. Probab. , 17(4): 1454-1462, 1989.
- [28] G. C. Papanicolaou and K. Sølna. Wavelet based estimation of local Kolmogorov turbulence. In Theory and applications of long-range dependence , pages 473-505. Birkhäuser Boston, Boston, MA, 2003.
- [29] R. F. Peltier and J. Lévy Véhel. Multifractional brownian motion: Definition and preliminary results. Rapport de recherche INRIA , (2645):1-39, 1995.
- [30] Y. Peres and P. Sousi. Brownian motion with variable drift: 0-1 laws, hitting probabilities and Hausdorff dimension. Math. Proc. Cambridge Philos. Soc. , 153(2):215-234, 2012.
- [31] Y. Peres and P. Sousi. Dimension of fractional brownian motion with variable drift. 2013.
- [32] J. Picard. Representation formulae for the fractional Brownian motion. Séminaire de
Probabilités , XLIII:3-70, 2011.
- [33] S. G. Samko, A. A. Kilbas, and O. I. Marichev. Fractional integrals and derivatives . Gordon and Breach Science Publishers, Yverdon, 1993.
- [34] S. Seuret and J. Lévy Véhel. A time domain characterization of 2-microlocal spaces. J. Fourier Anal. Appl. , 9(5):473-495, 2003.
- [35] S. A. Stoev and M. S. Taqqu. How rich is the class of multifractional Brownian motions? Stochastic Process. Appl. , 116(2):200-221, 2006.
- [36] K. Takashima. Sample path properties of ergodic self-similar processes. Osaka J. Math. , 26(1):159-189, 1989.
- [37] S. J. Taylor and N. A. Watson. A Hausdorff measure classification of polar sets for the heat equation. Math. Proc. Cambridge Philos. Soc. , 97(2):325-344, 1985.
- [38] D. S. Wu and Y. M. Xiao. Dimensional properties of fractional Brownian motion. Acta Math. Sin. (Engl. Ser.) , 23(4):613-622, 2007. | null | [
"Paul Balança"
] | 2014-08-01T23:01:09+00:00 | 2014-08-01T23:01:09+00:00 | [
"math.PR",
"60G07, 60G17, 60G22, 60G44"
] | Some sample path properties of multifractional Brownian motion | The geometry of the multifractional Brownian motion (mBm) is known to present
a complex and surprising form when the Hurst function is greatly irregular.
Nevertheless, most of the literature devoted to the subject considers
sufficiently smooth cases which lead to sample paths locally similar to a
fractional Brownian motion (fBm). The main goal of this paper is therefore to
extend these results to a more general frame and consider any type of
continuous Hurst function. More specifically, we mainly focus on obtaining a
complete characterization of the pointwise H\"older regularity of the sample
paths, and the Box and Hausdorff dimensions of the graph. These results, which
are somehow unusual for a Gaussian process, are illustrated by several
examples, presenting in this way different aspects of the geometry of the mBm
with irregular Hurst functions |
1408.0318v1 | ## Jointly Sparse Global SIMPLS Regression
## Tzu-Yu Liu
University of California, Berkeley, CA, USA. e-mail: [email protected]
## Laura Trinchera
NEOMA Business School, Mont-Saint-Aignan, France. e-mail: [email protected]
## Arthur Tenenhaus
Sup´ elec, Gif-sur-Yvette, France. e-mail: [email protected]
## Dennis Wei
IBM T. J. Watson Research Center, NY, USA. e-mail: [email protected]
## Alfred O. Hero ∗
University of Michigan, Ann Arbor, MI, USA.
e-mail:
[email protected]
Abstract: Partial least squares (PLS) regression combines dimensionality reduction and prediction using a latent variable model. Since partial least squares regression (PLS-R) does not require matrix inversion or diagonalization, it can be applied to problems with large numbers of variables. As predictor dimension increases, variable selection becomes essential to avoid over-fitting, to provide more accurate predictors and to yield more interpretable parameters. We propose a global variable selection approach that penalizes the total number of variables across all PLS components. Put another way, the proposed global penalty encourages the selected variables to be shared among the PLS components. We formulate PLS-R with joint sparsity as a variational optimization problem with objective function equal to a novel global SIMPLS criterion plus a mixed norm sparsity penalty on the weight matrix. The mixed norm sparsity penalty is the glyph[lscript] 1 norm of the glyph[lscript] 2 norm on the weights corresponding to the same variable used over all the PLS components. A novel augmented Lagrangian method is proposed to solve the optimization problem and soft thresholding for sparsity occurs naturally as part of the iterative solution. Experiments show that the modified PLS-R attains better or as good performance with many fewer selected predictor variables.
Keywords and phrases: PLS, variable selection, dimension reduction, augmented Lagrangian optimization.
∗ to whom correspondence should be addressed.
## Contents
| 1 | Introduction . . . . . . . | . . . . . . . . . . . . . . . 2 |
|--------------|------------------------------------------------------------------------------------------------------------------------------------------------------|------------------------------------------------------------------------------------------------------------------------------------------------------|
| 2 | Partial Least Squares Regression . . . . . . . . . . . . . . . . . . . . . 3 | Partial Least Squares Regression . . . . . . . . . . . . . . . . . . . . . 3 |
| | 2.1 | . . . . . . . . . . . . . . . 4 |
| | 2.2 | . . . . . . . . . . . . . . . 5 |
| 3 | Sparse Partial Least Squares Regression . . . . . . . . . . . . . . . . . 6 | Sparse Partial Least Squares Regression . . . . . . . . . . . . . . . . . 6 |
| | 3.1 | . . . . . . . . . . . . . . . 6 |
| | 3.2 | Regression . . . . . . . . . . . . . 7 |
| 4 | Algorithmic Implementation for Jointly Sparse Global SIMPLS Re- gression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 | Algorithmic Implementation for Jointly Sparse Global SIMPLS Re- gression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 |
| 5 | Simulation Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . 12 | Simulation Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . 12 |
| 6 | Application 1: Chemometrics Study . . . . . . . . . . . . . . . . . . . 15 | Application 1: Chemometrics Study . . . . . . . . . . . . . . . . . . . 15 |
| 7 | Application 2: Sparse Prediction of Disease Symptoms from Gene Ex- pression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 | Application 2: Sparse Prediction of Disease Symptoms from Gene Ex- pression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 |
| 8 | Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 | Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 |
| References . | References . | . . . . . . . . . . . . . . . 20 |
## 1. Introduction
With advancing technology comes the need to extract information from increasingly high-dimensional data, whereas the number of samples is often limited. Dimension reduction techniques and models incorporating sparsity become important solution strategies. Partial least squares regression (PLS-R) combines dimensionality reduction and prediction using a latent variable model. It was first developed for regression analysis in chemometrics [1, 2], and has been successfully applied to many different areas, including sensory science and, more recently, genetics [3, 4, 5, 6, 7].
Moreover, PLS-R algorithm is designed precisely to operate with high dimensional data. Since the first proposed algorithm does not require matrix inversion nor diagonalization but deflation to find the latent components, it can be applied to problems with large numbers of variables. The latent components reduce the dimension by constructing linear combinations of the predictors, which successfully solved the collinearity problems in chemometrics [8]. However, the linear combinations are built on all the predictors. The resulting PLS-R model tends to overfit when the number of predictors increases for a fixed number of samples. Therefore, variable selection becomes essential for PLS-R in high-dimensional sample-limited problems. It not only avoids over-fitting, but also provides more accurate predictors and yields more interpretable estimates. For this reason sparse PLS-R was developed by H. Chun and S. Keles [9]. The sparse PLS-R algorithm performs variable selection and dimension reduction simultaneously using an glyph[lscript] 1 type variable selection penalty. However, the glyph[lscript] 1 penalty used in [9] penalizes each variable in each component independently and this can result in different sets of variables being selected for each PLS component leading to an excessively large number of variables.
In this work we propose a global criterion for PLS that changes the sequential optimization for a K component model in Statistically Inspired Modification of PLS (SIMPLS) [10] into a unified optimization formulation, which we refer to as global SIMPLS. This enables us to perform global variable selection, which penalizes the total number of variables across all PLS components. We formulate PLS-R with global sparsity as a variational optimization problem with the objective function equal to the global SIMPLS criterion plus a mixed norm sparsity penalty on the weight matrix. The mixed norm sparsity penalty is the glyph[lscript] 1 norm of the glyph[lscript] 2 norm on the weights corresponding to the same variable used over all the PLS components. The proposed global penalty encourages the selected variables to be shared among all the K PLS components. A novel augmented Lagrangian method is proposed to solve the optimization problem, which enables us to obtain the global SIMPLS components and to perform joint variable selection simultaneously. A greedy algorithm is proposed to overcome the computation difficulties in the iterations, and soft thresholding for sparsity occurs naturally as part of the iterative solution. Experiments show that our approach to PLS regression attains better or as good performance (lower mean squared error, MSE) with many fewer selected predictor variables. These experiments include a chemometric data set, and a human viral challenge study dataset, in addition to numerical simulations.
We review the developments in PLS-R for both univariate and multivariate responses in Section 2, in which we discuss different objective functions that have been proposed for PLS-R, particularly the Statistically Inspired Modification of PLS (SIMPLS) proposed by de Jong [10]. In Section 3, we formulate the Jointly Sparse Global SIMPLS-R by proposing a new criterion that jointly optimizes over K weight vectors (components) and imposing a mixed norm sparsity penalty to select variables jointly. The algorithmic implementation is discussed in Section 4 with simulation experiments presented in Section 5. The proposed Jointly Sparse Global SIMPLS Regression is applied to two applications: (1) Chemometrics in Section 6 and (2) Predictive health studies in Section 7. Section 8 concludes the paper.
## 2. Partial Least Squares Regression
Partial Least Squares (PLS) methods embrace a suite of data analysis techniques based on algorithms belonging to the PLS family. These algorithms consist of various extensions of the Nonlinear estimation by Iterative PArtial Least Squares (NIPALS) algorithm that was proposed by Herman Wold [11] as an alternative algorithm for implementing Principal Component Analysis (PCA) [12]. The NIPALS approach was slightly modified by Svante Wold, and Harald Martens, in order to obtain a regularized component based regression tool, known as PLS Regression (PLS-R) [1, 8].
Suppose that the data consists of n samples of p independent variables X ∈ R n × p and q dependent variables (responses) Y ∈ R n × q . In standard PLS Regression the aim is to define orthogonal latent components in R p , and then use
such latent components as predictors for Y in an ordinary least squares framework. The X weights used to compute the latent components can be specified by using iterative algorithms belonging to the NIPALS family or by a sequence of eigen-decompositions. The general underlying model is X = TP ′ + E and Y = TQ ′ + F , where T ∈ R n × K is the latent component matrix, P ∈ R p × K and Q ∈ R q × K are the loading matrices, K is the number of components, E and F are the residual terms. The latent components in T = [ t 1 t 2 ... t K ] are linear combinations of the independent variables, hence PLS can be viewed as a dimensional reduction technique, reducing the dimension from p to K . The latent components should be orthogonal to each other either by construction as in NIPALS [1, 8] or via constrained optimizations as in SIMPLS [10]. This allows PLS to build a parsimonious model for high dimensional data with collinearity [8].
## 2.1. Univariate Response
We assume, without loss of generality, that all the variables have been centered in a pre-processing step. For univariate Y , i.e q = 1, PLS Regression, also often denoted as PLS1, successively finds X weights R = [ r 1 r 2 ... r K ] as the solution to the constrained optimization
$$\mathbf r _ { k } = \arg \max _ { \mathbf r } \{ \mathbf r ^ { \prime } X ^ { \prime } _ { ( k - 1 ) } Y Y ^ { \prime } X _ { ( k - 1 ) } \mathbf r \} \ s. t. \ \mathbf r ^ { \prime } \mathbf r = 1. \quad \quad ( 2. 1 )$$
where X ( k -1) is the matrix of the residuals (i.e., the deflated matrix) from the regression of the X -variables on the first k -1 latent components, and X (0) = X . These weights are then used to find the latent components T = [ X (0) r 1 X (1) r 2 ... X ( K -1) r K ]. Such components can be also expressed in terms of original variables (instead of deflated variables), i.e., as T = XW , where W is the matrix containing the weights to be applied to the original variables in order to exactly obtain the latent components [13].
For a fixed number of components, the response variable Y is predicted in an ordinary least squares regression model, where the latent components play the role of the exogenous variables,
$$\hat { Q } = \arg \min _ { Q } \{ | | Y - T Q ^ { \prime } | | _ { 2 } \} = ( T ^ { \prime } T ) ^ { - 1 } T ^ { \prime } Y.$$
This provides the regression coefficients ˆ β PLS = WQ ˆ ′ for the model Y = Xβ PLS + F .
Depending on the number of selected latent components the length ‖ ˆ β PLS ‖ 2 of the vector of PLS coefficients changes. In particular, de Jong [14] had shown that the sequence of these coefficient vectors has lengths that are strictly increasing as the number of components increases. This sequence converges to the ordinary least squares coefficient vector and the maximum number of latent components obtainable equals the rank of the X matrix. Thus, by using a
number of latent components K < rank X ( ), PLS-R performs a dimension reduction by shrinking the X matrix. Hence, PLS-R is a suitable tool for problems for which the data contains many more variables p than observations n .
The objective function in (2.1) can be interpreted as maximizing the squared covariance between Y and the latent component: corr 2 ( Y, X k -1 r k )var( X k -1 r k ). Because the response Y has been taken into account to formulate the latent matrix, PLS usually has better performance in prediction problems than principle component analysis (PCA) does. This is one of the main differences between PLS-R and PCA [15].
## 2.2. Multivariate Response
Similarly to univariate response PLS-R, multivariate response PLS-R selects latent components in R p and R q , i.e., t k and v k , such that the covariance between t k and v k is maximized. For a specific component, the sets of weights r k ∈ R p and c k ∈ R q are obtained by solving
$$& \max \{ \mathfrak { t } ^ { \prime } \mathbf v \} = \max \{ \mathfrak { r } ^ { \prime } X ^ { \prime } _ { ( k - 1 ) } Y _ { ( k - 1 ) } \mathfrak { c } \} \\ & s. t. \ \, \mathfrak { r } ^ { \prime } \mathbf r = \mathfrak { c } ^ { \prime } \mathbf c = 1$$
where t k = X ( k -1) r k , v k = Y ( k -1) c k , and X ( k -1) and Y ( k -1) are the deflated matrices associated with X and Y . Notice that the optimal solution c k should be proportional to Y ′ k -1 X k -1 r k . Therefore, the optimization in (2.2) is equivalent to
$$& \max _ { r } \left \{ { \mathbf r } ^ { \prime } X ^ { \prime } _ { k - 1 } Y _ { k - 1 } Y ^ { \prime } _ { k - 1 } X _ { k - 1 } { \mathbf r } \right \} \\ & s. t. \ r ^ { \prime } { \mathbf r } = 1.$$
For each component, the solution to this criterion can be obtained by using a so called PLS2 algorithm. A detailed description of the iterative algorithm as presented by H¨skuldsson [16] is in Algorithm 1 . o
In 1993 de Jong proposed a variant of the PLS2 algorithm, called Statistically Inspired Modification of PLS (SIMPLS), which calculates the PLS latent components directly as linear combinations of the original variables [10]. The SIMPLS was first developed as the solution to an optimization problem
$$& \text{w} _ { k } = \underset { w } { \arg \max } ( \text{w} ^ { \prime } X ^ { \prime } Y Y ^ { \prime } X \text{w} ) & & ( 2. 4 ) \\ & s. t. \quad \text{w} ^ { \prime } \text{w} = 1, \quad \text{w} ^ { \prime } X ^ { \prime } X \text{w} _ { j } = 0 \ f o r \ j = 1, \dots, k - 1.$$
Ter Braak and de Jong [17] provided a detailed comparison between the objective functions for PLS2 in (2.3) and SIMPLS in (2.4) and showed that the successive weight vectors w k can be derived either from the deflated data matrices or the original variables in PLS2 and SIMPLS respectively. Let W + be the Moore-Penrose inverse of W = [ w w 1 2 ... w k -1 ]. The PLS2 algorithm (Algorithm 1) is equivalent to solving the optimization
$$w _ { k } = \underset { w } { \arg \max } ( w ^ { \prime } X ^ { \prime } Y Y ^ { \prime } X w )$$
## Algorithm 1: PLS2 algorithm
```
Algorithm 1: PLS2 algorithm
for k=1:K do
| initialize r
X = X.new
Y = Y.new
while solution has not converged do
| t = Xr
| c = Y't
| Scale c to length 1
| v = Yc
| r = X'v
| Scale r to length 1
loading vector p = X't/(t't)
deflate X.new = X - tp'
regression b = Y't/(t't)
deflate Y.new = Y - tb'
| r.k = r
```
$$s. t. \, { \mathbf w } ^ { \prime } ( I - W W ^ { + } ) { \mathbf w } = 1, { \mathbf w } ^ { \prime } X ^ { \prime } X { \mathbf w } _ { i } = 0 \ f o r \ i = 1, \dots, k - 1.$$
Both NIPALS and SIMPLS have the same objective function but each is maximized under a different normalization constraint. NIPALS and SIMPLS are equivalent when Y is univariate, but provide slightly different weight vectors in multivariate scenarios. The performance depends on the nature of the data, but SIMPLS appears easier to interpret since it does not involve deflation of the data set [10]. We develop our globally sparse PLS-R based on the SIMPLS optimization formulation.
## 3. Sparse Partial Least Squares Regression
## 3.1. glyph[lscript] 1 Penalized Sparse PLS Regression
One approach to sparse PLS-R is to add the glyph[lscript] 1 norm of the weight vector, a sparsity inducing penalty, to (2.4). The solution for the first component would be obtained by solving
$$& w _ { 1 } = \underset { w } { \arg \max } ( \mathbf w ^ { \prime } X ^ { \prime } Y Y ^ { \prime } X \mathbf w ) & & ( 3. 1 ) \\ & s. t. \ \mathbf w ^ { \prime } \mathbf w = 1, \ | | \mathbf w | | _ { 1 } \leq \lambda.$$
The addition of the glyph[lscript] 1 norm is similar to SCOTLASS (simplified component lasso technique), the sparse PCA proposed by Jolliffe [18]. However, the solution of SCOTLASS is not sufficiently sparse, and the same issue remains in (3.1). Chun and Keles [9] reformulated the problem, promoting the exact zero property by imposing the glyph[lscript] 1 penalty on a surrogate of the weight vector instead of the original weight vector [9]. For the first component, they solve the following optimization by alternating between updating w and updating z (block
coordinate descent).
$$w _ { 1 }, z _ { 1 } = \underset { w, z } { \arg \min } \{ - \kappa w ^ { \prime } X ^ { \prime } Y Y ^ { \prime } X w + ( 1 - \kappa ) ( z - w ) ^ { \prime } X ^ { \prime } Y Y ^ { \prime } X ( z - w ) + \lambda _ { 1 } \| z \| _ { 1 } + \lambda _ { 2 } \| z \| _ { 2 } ^ { 2 } \}$$
$$s. t. \ w ^ { \prime } w = 1$$
Allen et. al proposed a general framework for regularized PLS-R [20].
$$\max _ { w, v } \, w ^ { \prime } M \mathbf v - \lambda P ( \mathbf w ) \ s. t. \ w ^ { \prime } \mathbf w \leq 1, \mathbf v ^ { \prime } \mathbf v = 1$$
in which M is the cross-product matrix X Y ′ , and the regularization function P is a convex penalty function. The formulation is a relaxation of SIMPLS with penalties being applied to the weight vectors, and can be viewed as a generalization of [9].
As mentioned in the Introduction, these formulations ([9, 20]) penalize the variables in each PLS component independently. This paper proposes an alternative in which variables are penalized simultaneously over all components. First, we define the global weight matrix, consisting of the K weight vectors, as
$$W = \left [ \begin{array} { c c c c } | & | & | & & | \\ w _ { 1 } & w _ { 2 } & \cdots & w _ { K } \\ | & | & | & & | \end{array} \right ] = \left [ \begin{array} { c c c$$
Notice that the elements in a particular row of W, i.e., w ′ ( j ) , are all associated with the same predictor variable x j . Therefore, rows of zeros correspond to variables that are not selected. To illustrate the drawbacks of penalizing each variable in each component independently, as in [9], suppose that each entry in W is selected independently with probability p 1 . The probability that the ( j ) th variable is not selected becomes (1 -p 1 ) K , and the probability that all the variables are selected by at least one weight vector is [1 -(1 -p 1 ) K p ] , which increases as the number of weight vectors K increases. This suggests that for large K the local variable selection approach of [9] may not lead to an overall sparse and parsimonious PLS-R model. In such cases a group sparsity constraint can be employed to limit the overall number of selected variables.
## 3.2. Jointly Sparse Global SIMPLS Regression
The Jointly Sparse Global SIMPLS Regression variable selection problem is to find the top K weight vectors that best relate X to Y , while using limited number of variables. This is a subset selection problem that is equivalent to adding a constraint on the glyph[lscript] 0 norm of the vector consisting of the norms of the rows of W , i.e, the number of nonzero rows in W . For concreteness we use the glyph[lscript] 2 norm for the rows. This leads to the optimization problem
$$W = \underset { W } { \arg \min } - \frac { 1 } { n ^ { 2 } } \sum _ { k = 1 } ^ { K } w _ { k } ^ { \prime } X ^ { \prime } Y Y ^ { \prime } X w _ { k } \quad \quad \quad$$
glyph[negationslash]
s.t. || glyph[pi1] || 0 ≤ t, w w ′ k k = 1 ∀ k, and w ′ k X X ′ w i = 0 ∀ i = k in which
$$\varpi = \left [ \begin{array} { c } | | \mathbf w _ { ( 1 ) } | | _ { 2 } \\ | | \mathbf w _ { ( 2 ) } | | _ { 2 } \\ \vdots \\ | | \mathbf w _ { ( p ) } | | _ { 2 }$$
The objective function (3.2), which we refer to as global SIMPLS, is the sum of the objective functions (2.4) in the first K iterations of SIMPLS. Instead of the sequential greedy solution in PLS2 algorithm, the proposed jointly sparse global SIMPLS Regression solves for the K weight vectors simultaneously. We introduce the 1 n 2 factor to the objective function to interpret it as an empirical covariance. Given the complexity of this combinatorial problem, as in standard optimization practice, we relax the glyph[lscript] 0 norm optimization to a mixed norm structured sparsity penalty [21].
glyph[negationslash]
$$W = & \underset { W } { \arg } \min - \frac { 1 } {$$
The glyph[lscript] 2 norm of each row of W promotes grouping entries in W that relate to the same predictor variable, whereas the glyph[lscript] 1 norm promotes a small number of groups, as in (3.1).
## 4. Algorithmic Implementation for Jointly Sparse Global SIMPLS Regression
Constrained eigen-decomposition and group variable selection are each wellstudied problems for which efficient algorithms have been developed. We propose to solve the optimization (3.3) by augmented Lagrangian methods, which allows one to solve (3.3) by variable splitting iterations. Augmented Lagrangian methods introduce a new variable M , constrained such that M = W , such that the row vectors m ( j ) of M obey the same structural pattern as the rows of W :
glyph[negationslash]
$$& \min _ { W, M } - \frac { 1 } { n ^ { 2 } }$$
The optimization (4.1) can be solved by replacing the constrained problem by an unconstrained one with an additional penalty on the Frobenius norm of the
difference M -W . This penalized optimization can be iteratively solved by an alternating direction method of multipliers (ADMM) algorithm [25, 26, 27, 28, 29, 30, 31, 32], a block coordinate descent method that alternates between optimizing over W and over M (See Algorithm 2). We initialize Algorithm 2 with M (0) equal to the solution of SIMPLS, and D (0) equal to the zero matrix. Setting the parameter µ is nontrivial [31], and is hand-tuned for fastest convergence in some applications [27]. Once the algorithm converges, the final PLS regression coefficients are obtained by applying SIMPLS regression on the selected variables keeping the same number of components K . The optimization over W can be further simplified to a secular equation problem, whereas the optimization over M can be shown to reduce to a soft thresholding operation. The algorithm iterates until the stopping criterion based on the norm of the residuals || W ( τ ) -M ( τ ) || F < glyph[epsilon1] is satisfied, for some given tolerance glyph[epsilon1] . As described later in the experimental comparisons section, the parameters λ and K are decided by cross validation.
Algorithm 2: Algorithm for solving the global SIMPLS with global variable selection problem using the augmented Lagrangian method.
glyph[negationslash]
$$\begin{array} {$$
Optimization over W The following optimization in Algorithm 2 is a nonconvex quadratically constrained quadratic program (QCQP).
$$W ^ { ( \tau + 1 ) } = \underset { W } { \arg \$$
glyph[negationslash]
$$s. t. \, { \mathbf w } ^ { \prime } _ { k } {$$
We propose solving for the K vectors in W successively by a greedy approach. Let m k and d k be the columns of the matrices M and D , and ω k = m k + d k . The optimization over W becomes
$$\mathbf w _ { k } ^ { ( \tau + 1 ) } \quad = \quad & \underset { \mathbf w } { \arg \min } - \frac { 1 } { n ^ { 2 } } \mathbf w ^ { \prime } X ^ { \prime } Y Y ^ { \prime } X \mathbf w + \frac { \mu } { 2 } \| \mathbf w - \omega _ { k } \| _ { 2 } ^ { 2 } } \\ s. t. \quad & \ \mathbf w ^ { \prime } \mathbf w = 1, \ \mathbf w ^ { \prime } X ^ { \prime } X \mathbf w _ { i } = 0 \ \forall \ i < k.$$
Lemma 4.1. Let N be an orthonormal basis for the orthogonal complement of { X X ′ w i } , i < k . The optimization (4.2) can be solved by the method of Lagrange multipliers. The solution is w k = N A ( -αI ) -1 b , in which A = -1 n 2 N X Y Y XN ′ ′ ′ , b = µ 2 N ′ ω k and α is the minimum solution that satisfies b ′ ( A -αI ) -2 b = 1 .
Proof. Let w = N ˜ w , then the optimization (4.2) can be written as
$$\min \tilde { w } ^ { \prime } A \tilde { w } - 2 b ^ { \prime } \tilde { w } \ s. t. \ \tilde { w } ^ { \prime } \tilde { w } = 1.$$
Since we assume that w is a linear combination of the basis vectors in N , the orthogonality conditions in (4.2) are automatically satisfied. Hence these conditions have been dropped in the new formulation. Then using Lagrange multipliers, we can show that the solution takes the form as stated above.
Suppose there are two solutions of α that satisfy b ′ ( A -αI ) -2 b = 1, corresponding to two pairs of solutions to the optimization, ( ˜ w 1 , α 1 ) and ( ˜ w 2 , α 2 ). Since ˜ = ( w A -αI ) -1 b ,
$$A \tilde { w } _ { 1 } = \alpha _ { 1 } \tilde { w } _ { 1 } + b$$
$$A \tilde { w } _ { 2 } = \alpha _ { 2 } \tilde { w } _ { 2 } + b.$$
By multiplying (4.3) by ˜ w 1 , and (4.4) by ˜ w 2 , then subtracting the two new equations, we have
$$\tilde { w } _ { 1 } ^ { \prime } A \tilde { w } _ { 1 } - \tilde { w } _ { 2 } ^ { \prime } A \tilde { w } _ { 2 } = \alpha _ { 1 } - \alpha _ { 2 } + ( \tilde { w } _ { 1 } ^ { \prime } - \tilde { w } _ { 2 } ^ { \prime } ) b.$$
On the other hand, by multiplying (4.3) by ˜ w 2 , and (4.4) by ˜ w 1 , and subtracting the two new equations, we have
$$( \tilde { w } _ { 1 } ^ { \prime } - \tilde { w } _ { 2 } ^ { \prime } ) b = ( \alpha _ { 1 } - \alpha _ { 2 } ) \tilde { w } _ { 1 } ^ { \prime } \tilde { w } _ { 2 }.$$
Given (4.5) and (4.6), it can be shown that
$$( \tilde { w } _ { 1 } ^ { \prime } A \tilde { w } _ { 1 } - 2 b ^ { \prime } \tilde { w } _ { 1 } ) - ( \tilde { w } _ { 2 } ^ { \prime } A \tilde { w } _ { 2 } - 2 b ^ { \prime } \tilde { w } _ { 2 } ) = ( \alpha _ { 1 } - \alpha _ { 2 } ) - ( \tilde { w } _ { 1 } ^ { \prime } - \tilde { w } _ { 2 } ^ { \prime } ) b = \frac { \alpha _ { 1 } - \alpha _ { 2 } } { 2 } | | \tilde { w } _ { 1 } - \tilde { w } _ { 2 } | | _ { 2 } ^ { 2 }.$$
Hence, one should select the minimum among all the feasible α
's.
The equation b ′ ( A -αI ) -2 b = 1 is a secular equation, a well studied problem in constrained eigenvalue decomposition [23, 24]. The more general problem of least squares with a quadratic constraint was discussed in [22]. We can diagonalize the matrix A as A = UDU ′ , in which D is diagonal with eigenvalues d , d 1 2 , ..., d p in decreasing order on the diagonal, and the columns of U are the corresponding eigenvectors. Define
$$g ( \alpha ) = b ^ { \prime } ( A - \alpha I ) ^ { - 2 } \mathbf b = b ^ { \prime } U \left [ \begin{array} { c c c c } \frac { 1 } { ( d _ { 1 } - \alpha ) ^ { 2 } } & & & \\ & \frac { 1 } { ( d _ { 2 } - \alpha ) ^ { 2 } } & \\ & & \ddots & \\ & & & \frac { 1 } { ( d _ { p } - \alpha ) ^ { 2 } } \end{array} \right ] U ^ { \prime } \mathbf b.$$
Let ˜ b = U ′ b , then g α ( ) = b ′ ( A -αI ) -2 b = ∑ i ˜ b 2 i ( d i -α ) 2 , and hence g α ( ) = 1 is a secular equation. g α ( ) increases strictly as α increases from -∞ to d p , since
$$g ^ { \prime } ( \alpha ) = \sum _ { i } \frac { 2 \tilde { b } _ { i } ^ { 2 } } { ( d _ { i } - \alpha ) ^ { 3 } }$$
is positive for -∞ < α < d p . Moreover, given the limits
$$\lim _ { \mu \mu \mu \nu } g ( \alpha ) = 0$$
$$\lim _ { \alpha \to d _ { p } ^ { - } } g ( \alpha ) = \infty$$
$$\underset { \lambda - 1 } { \min } g ( \alpha ) = \upsilon \\ \lim _ { \lambda - 1 } g ( \alpha ) = \infty$$
we can conclude that there is exactly one solution α < d p to the equation g α ( ) = 1, [23]. An iterative algorithm (Algorithm 3 ) is used to solve g α ( ) = 1 starting from a point to the left of the smallest eigenvalue d p [24].
Notice that calculating g α ( ) = b ′ ( A -αI ) -2 b involves inverting a ( p -k +1) × ( p -k +1) matrix, but A has rank at most q . We can reduce the computational burden by the use of Woodbury matrix identity,
$$( A - \alpha I ) ^ { - 1 } = \frac { - 1 } { \alpha } ( I - \frac { 1 } { \alpha n ^ { 2 } } N ^ { \prime } X ^ { \prime } Y ( I + \frac { 1 } { \alpha n ^ { 2 } } Y ^ { \prime } X N N ^ { \prime } X ^ { \prime } Y ) ^ { - 1 } Y ^ { \prime } X N ).$$
The new format only requires inverting a q × q matrix, and in most applications, the number of responses q is much less than the number of predictors p . Furthermore, N is involved in g α ( ) in the form of NN ′ = I -HH ′ , in which H is an orthonormal basis for { X X ′ w i } , i < k . H can be constructed by Gram-Schmidt process as the algorithm successively finds the weight vectors w ′ i s .
## Algorithm 3: Iteration for solving secular equation.
$$\begin{matrix} \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \ \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \ & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad &\quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad \\ \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \null & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad &
& \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \\ \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad &
\quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \ } & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \ \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \ \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad &
\quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \ \quad & \quad & \quad & \quad & \quad & \quad & \quad & \ \quad & \quad & \quad & \ \quad & \quad & \quad & \ \quad & \quad & \quad & \quad & \ \quad & \quad & \quad & \ \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \ \quad & \quad & \quad & \quad & \quad & \ \quad & \quad & \ \quad & \quad & \ \quad & \quad & \ \quad & \ \quad & \ \quad & \ \quad & \ \quad & \ \quad & \quad & \quad & \quad & \ \quad & \ \quad & \ \quad & \ \quad & \ \quad & \ \quad & \ \quad & \ \quad & \ \quad & \ \ & \quad & \quad & \quad & \ \quad & \ \quad & \ \quad & \quad & \ \quad & \ \quad & \quad & \quad & \quad & \ \quad & \ \quad & \ \quad & \ \quad & \quad & \ \quad & \quad & \ \quad & \ \ } & \quad & \quad & \ \quad & \ \quad & \ \quad & \ \quad & \ \quad & \ \ } & \quad & \ \quad & \ \ } & \ \ \ \\ & \quad & \ \quad & \ \quad & \ \quad & \ \quad & \ \ } & \quad & \ \quad & \ \quad & \ \quad & \ \quad & \ \ \\ & \quad & \quad & \ \ \ & \quad & \ \ & \quad & \ \ \ & & \quad & \ \ & & \ \ & & \ \ & \ \ & & \ \ & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & \ \ & & \ \ & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & \ & \ \ & & \ \ & \ & \ \ & & \ \ & & \ \ & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & \ \ & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & \ & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & \ \ & & \ \ & \ & & \ \ & & \ \ & & \ \ & \ \ & & \ \ & \ \ & & \ \ & & \ \ & & \ \ & \ \ & \ & \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & \ \ & & \ \ & \ & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & \ & & \ \ & \ \ & & \ \ & \ & \ \ & \ \ & & \ \ & & \ \ & & \ \ & \ \ & & \ \ & \ \ & & \ \ & \ \ & & \ \ & & \ \ & \ \ & \ & & \ \ & \ \ & \ \ & & \ \ & \ & \ \ & & \ \ & \ & \ \ & & \ \ & \ & \ & \ \ & \ & \ \ & & \ \ & & \ \ & & \ \ & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & \ \ & & \ \ & \ \ & & \ \ & & \ \ & & \ \ & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & \ \ & & \ \ & \ \ & & \ \ & & \ \ & \ \ & & \ \ & & \ \ & & \ \ & \ \ & & \ \ & & \ \ & \ & \ \ & & \ \ & \ \ & \ \ & & \ \ & \ \ & \ & & \ \ & \ & \ & & \ \ & & \ \ & \ & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & \ \ & & \ \ & & \ \ & & \ \ & \ \ & & \ \ & \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & \ & \ & \ & \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ & \ \ & & \ & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ \ & & \ & & \ & & \ & \ & & \ \$$
Optimization over M The optimization over M has a closed form solution. Let ∆ = W ( τ +1) -D ( τ ) , and δ ( j ) denote the j th row of ∆, then each row of M is given as m ( j ) = [ || δ ( j ) || -λ µ ] + δ ( j ) || δ ( j ) || , in which [ z ] + = max z, { 0 } .
Convergence analysis of ADMM can be found in [25, 29, 30, 32]. In particular, it has been shown that ADMM converges linearly for strongly convex objective functions [30]. Although the convergence is based on strong convexity assumptions, ADMM has been widely applied in practice [26, 27], even to nonconvex problems [28]. The proposed Jointly Sparse Global SIMPLS Regression is one example of these nonconvex applications.
## 5. Simulation Experiments
We implement the simulation models in [33, 9]. There are four models all following Y = Xβ + F , in which the number of observation is n = 100, and the dimension is p = 5000. Full details of the models are given in Table 1. We compare five different methods: the standard PLS regression (denoted as PLS in the following comparison tables), PLS generalized linear regression proposed by Bastien et al. [34], glyph[lscript] 1 penalized PLS regression [9] (denoted as glyph[lscript] 1 SPLS), Lasso [35] and the Jointly Sparse Global SIMPLS regression ( denoted as glyph[lscript] 1 /glyph[lscript] 2 SPLS). All the methods select their parameters by ten fold cross-validation, except for the PLS generalized linear regression, which stops including an additional component if the new component is not significant. The parameter µ in the Jointly Sparse Global SIMPLS-R is fixed to 2000, and updated in each iteration by a scaling factor 1 01 . µ . The experiments on real data in Section 6 and 7 are using the same setting of µ . Two i.i.d sets are generated for each trial: one as the training set and one as the test set. Ten trials are conducted for each model, and the averaged results are listed in Table 2.
In most of the simulations, we observe that the proposed Jointly Sparse Global SIMPLS-R performs as good or better than other methods in terms of the prediction MSE. In particular, the number of variables and the number of components chosen in Jointly Sparse Global SIMPLS-R are usually less than the glyph[lscript] 1 penalized PLS-R. We also calculate the R 2 for each method on the training data to measure the variation explained. The standard PLS regression and the PLS generalized linear regression proposed by Bastien et al. [34] both have R 2 close to 1, but the performance in terms of MSE is not ideal in the first three models for these methods. This may suggest that these models overfit the data. In addition to the averaged performance, the p-values of one sided paired t-test suggest that Jointly Sparse Global SIMPLS-R reduces model complexity significantly from those in the standard PLS regression and the PLS generalized linear regression. Lasso achieves low complexity in terms of the number of variables, but the MSE is high compared to Jointly Sparse Global SIMPLS-R and the glyph[lscript] 1 penalized PLS-R. The cross validation time for Jointly Sparse Global SIMPLS Regression is long, searching over a two-dimensional grid of the number of components K and the regularization parameter λ to minimize MSE. However, the performance in terms of prediction MSE improves, and the model complexity in terms of the number of variables and the number of components both decreases compared with other methods in most simulations.
## Table 1
Simulation models in [33, 9]: These models were originally proposed to test supervised principal components. Model 1 and Model 2 are designed such that one latent variable dominates the multi-collinearity. Model 3 has multiple latent variables, and Model 4 has a correlation structure from an autoregressive process. In these model descriptions, I is used to denote the indicator function. We use i ( 1 ≤ i ≤ n ) to index the n samples, j ( 1 ≤ j ≤ p ) to index the p variables, and k to index the hidden components.
## Model 1
H 1 ( ) = 3 i I ( i ≤ 50) + 4 I ( i > 50) , 1 ≤ i ≤ n
H 2 ( ) = 3 5 i .
X j = H k + ε , j p k - 1 < j ≤ p k , k = 1 2 , , ( p , ..., p 0 2 ) = (0 , 50 , p )
β
j
=
{
1
25
1
≤
j
≤
50
0
51
≤
j
≤
p
ε j is N (0 , I n ) distributed, and F is N (0 , 1 5 . 2 I n ) distributed.
## Model 2
H 1 ( ) = 2 5 i . I ( i ≤ 50) + 4 I ( i > 50)
H 2 ( ) = 3 5 + 1 5 i . . I ( u 1 i ≤ 0 4) .
H 3 ( ) = 3 5 + 0 5 i . . I ( u 2 i ≤ 0 7) .
H 4 ( ) = 3 5 i . - 1 5 . I ( u 3 i ≤ 0 3) .
H 5 ( ) = 3 5 i .
u 1 i , u 2 i , u 3 i are i.i.d from Unif (0 , 1)
X j = H k + ε , j p k - 1 < j ≤ p k , k = 1 , ..., 5 ( , p , ..., p 0 5 ) = (0 , 50 100 200 300 , , , , p )
β
j
=
{
1
25
1
≤
j
≤
50
0
51
≤
j
≤
p
ε j is N (0 , I n ) distributed, and F is N (0 , I n ) distributed.
## Model 3
H 1 ( ) = 2 5 i . I ( i ≤ 50) + 4 I ( i > 50)
H 2 ( ) = 2 5 i . I (1 ≤ i ≤ 25 , or 51 ≤ i ≤ 75) + 4 I (26 ≤ i ≤ 50 , or 76 ≤ i ≤ 100)
H 3 ( ) = 3 5 + 1 5 i . . I ( u 1 i ≤ 0 4) .
H 4 ( ) = 3 5 + 0 5 i . . I ( u 2 i ≤ 0 7) .
H 5 ( ) = 3 5 i . - 1 5 . I ( u 3 i ≤ 0 3) .
H 6 ( ) = 3 5 i .
u 1 i , u 2 i , u 3 i are i.i.d from Unif (0 , 1)
X j = H k + ε , j p k - 1 < j ≤ p k , k = 1 , ..., 5 ( , p , ..., p 0 6 ) = (0 , 25 50 100 200 300 , , , , , p )
β j = 25 0 51 ≤ j ≤ p
{
1
1
≤
j
≤
50
ε j is N (0 , I n ) distributed, and F is N (0 , I n ) distributed.
## Model 4
H 1 ( ) = i I ( i ≤ 50) + 6 I ( i > 50)
H 2 ( ) = 3 5 + 1 5 i . . I ( u 1 i ≤ 0 4) .
H
3
( ) = 3 5 + 0 5
i
.
.
I
(
u
2
i
≤
0 7)
.
H 4 ( ) = 3 5 i . - 1 5 . I ( u 3 i ≤ 0 3) .
H 5 ( ) = 3 5 i .
u 1 i , u 2 i , u 3 i are i.i.d from Unif (0 , 1)
X = ( X (1) , X (2) )
X (1) is generated from N (0 , Σ 50 × 50 ) , Σ is from AR (1) with ρ = 0 9 . .
X (2) j = H k + ε , j p k - 1 < i ≤ p k , k = 1 , ..., 5 ( , p , ..., p 0 5 ) = (0 , 50 100 200 300 , , , , p - 50)
β i = r m for p m - 1 < i ≤ p m , m = 1 , ..., 6 , where
( p , ..., p 0 6 ) = (0 , 10 20 30 40 50 , , , , , p ) , ( r 1 , ..., r 6 ) = (8 , 6 4 2 1 0) , , , , / 25
ε i is N (0 , I n ) distributed, and F is N (0 , 1 5 . 2 I n ) distributed.
Table 2
Performance comparison table for the 4 simulation models. We compare five different methods: the standard PLS regression (PLS), PLS generalized linear regression proposed by Bastien et al. (Bastien), glyph[lscript] 1 penalized PLS regression ( glyph[lscript] 1 SPLS), Lasso and the Jointly Sparse Global SIMPLS regression ( glyph[lscript] 1 /glyph[lscript] 2 SPLS). These results are obtained by averaging over 10 trails.
| (5,4) | NA 0 . 053 0 . 006 | NA 0 . 0119 2 . 70 × 10 - 4 | NA 0 . 1726 0 . 0054 | NA 0 . 0485 0 . 0056 |
|-----------------------------|------------------------------------------------------------------------------------------------------|------------------------------------------------------------------------------------------------------|------------------------------------------------------------------------------------------------------|------------------------------------------------------------------------------------------------------|
| paired t-test (5,3) | 0 . 19 0 . 35 0 . 075 | 0 . 0184 0 . 1041 0 . 0344 | 0 . 3632 0 . 1697 0 . 3430 | 0 . 2201 0 . 3918 0 . 3812 |
| p-values of one sided (5,2) | 4 . 49 × 10 - 7 5 . 46 × 10 - 5 0 . 043 | 1 . 19 × 10 - 11 6 . 91 × 10 - 9 0 . 0046 | 5 . 53 × 10 - 6 1 . 39 × 10 - 12 0 . 0104 | 6 . 82 × 10 - 5 0 . 4618 0 . 1874 |
| (5,1) | 0 . 5 1 . 84 × 10 - 11 0 . 01 | 0 . 0671 1 . 13 × 10 - 14 8 . 69 × 10 - 4 | 0 . 4201 2 . 32 × 10 - 19 1 . 18 × 10 - 4 | 0 . 4057 9 . 62 × 10 - 5 0 . 0087 |
| 1 /glyph[lscript] 2 SPLS | 1.4 276.1 2.82 0.83 11414 6.40 0.002 11420 | 1.1 171.7 2.69 0.75 11420 5.72 0.001 11426 | 1.5 60.5 1.25 0.73 11295 5.48 0.001 11300 | 2.1 1180.5 2.36 0.91 12051 7.97 0.004 12059 |
| 4. Lasso | NA 40.7 3.23 0.59 51.59 0.04 0.04 51.67 | NA 15.8 3.09 0.39 53.97 0.04 0.039 54.05 | NA 41.3 1.48 0.75 49.45 0.04 0.039 49.53 | NA 9.4 2.14 0.19 50.52 0.04 0.042 50.60 |
| 3. glyph[lscript] 1 SPLS | 1.9 246.5 3.00 0.71 43.49 0.05 0.002 43.54 | 2.3 273.4 2.93 0.79 43.03 0.06 0.002 43.09 | 1.4 89.2 1.27 0.77 43.84 0.04 0.001 43.88 | 2.6 1260.8 2.41 0.78 44.31 0.10 0.007 44.41 |
| 2. Bastien | 5 1129.4 2.98 1 0 121.91 0.011 121.92 | 5 1158.4 2.99 1 0 122.69 0.011 122.70 | 5 1156.4 1.48 1 0 126.08 0.01 126.09 | 5 1118.8 2.29 1 0 123.79 0.011 123.8 |
| 1. PLS | 1.4 5000 3.14 0.98 101.47 0.89 0.010 102.37 | 2 5000 3.18 0.98 100.51 1.28 0.010 101.80 | 1.4 5000 1.82 0.98 102.61 1.03 0.01 103.65 | 2 5000 2.15 1 98.16 1.55 0.010 99.73 |
| | Model 1 number of comp. number of variables MSE R 2 Time CV Time analysis Time prediction Total time | Model 2 number of comp. number of variables MSE R 2 Time CV Time analysis Time prediction Total time | Model 3 number of comp. number of variables MSE R 2 Time CV Time analysis Time prediction Total time | Model 4 number of comp. number of variables MSE R 2 Time CV Time analysis Time prediction Total time |
Table 3 Performance comparison table for the Octane data.
| | | | | p-values of one sided paired t-test | p-values of one sided paired t-test |
|--------------------------------|---------|--------------------------|--------------------------------------------|-----------------------------------------------|---------------------------------------|
| | 1. PLS | 2. glyph[lscript] 1 SPLS | 3. glyph[lscript] 1 /glyph[lscript] 2 SPLS | (3,1) | (3,2) |
| number of comp. number of var. | 5.5 225 | 4.5 87.3 | 3.8 38.5 | 5 . 1888 × 10 - 16 1 . 1176 × 10 - 121 0.0032 | 0.0027 1 . 0967 × 10 - 15 0.1575 |
| MSE | 0.0564 | 0.0509 | 0.0481 | | |
## 6. Application 1: Chemometrics Study
In this section we show experimental results obtained by comparing standard PLS-R, glyph[lscript] 1 penalized PLS-R [9] (denoted as glyph[lscript] 1 SPLS in the performance table), and our proposed Jointly Sparse Global SIMPLS-R (denoted as glyph[lscript] 1 /glyph[lscript] 2 SPLS in the performance table). All the methods have been applied on the Octane data set (see [13]). The Octane data is a real data set consisting of 39 gasoline samples for which the digitized Octane spectra have been recorded at 225 wavelengths (in nm). The aim is to predict the Octane number, a key measurement of the physical properties of gasoline, using the spectra as predictors. This is of major interest in real applications, because the conventional procedure to calculate the Octane number is time consuming and involves expensive and maintenanceintensive equipment as well as skilled labor.
The experiments are composed of 150 trials. In each trial we randomly split the 39 samples into 26 training samples and 13 test samples. The regularization parameter λ and number of components K are selected by 2-fold cross validation on the training set. The averaged results over the 150 trials are shown in Table 3. We further show the variable selection frequencies for the sparse PLS methods over the 150 trials superimposed on the octane data in Fig. 1 (B) and (C). In chemometrics, the rule of thumb is to look for variables that have large amplitudes in first derivatives with respect to wavelength. Notice that both glyph[lscript] 1 penalized PLS-R and Jointly Sparse Global SIMPLS-R have selected variables around 1200 and 1350 nm, and the selected region in the latter case is more confined. Box and Whisker plots for comparing the MSE, number of selected variables, and number of components of these three PLS formulations are shown in Fig. 1 (A). Comparing our proposed Jointly Sparse Global SIMPLS Regression with standard PLS-R and glyph[lscript] 1 penalized PLS-R [9], we show that Jointly Sparse Global SIMPLS-R attains better performance in terms of MSE, the number of predictors, and the number of components. Besides, the model complexity in Jointly Sparse Global SIMPLS-R is significantly lower than both standard PLS-R and glyph[lscript] 1 penalized PLS-R, given the p-values of one sided paired t-test.
(C)
Fig 1: (A) Box and Whisker plots for comparing the MSE, number of selected variables, and number of components of three PLS formulations: standard PLSR, glyph[lscript] 1 penalized PLS-R ( glyph[lscript] 1 SPLS), our proposed Jointly Sparse Global SIMPLSR ( glyph[lscript] 1 /glyph[lscript] 2 SPLS). (B) Variable selection frequency of glyph[lscript] 1 SPLS superimposed on the octane data and its first derivative: The height of the surfaces represents the exact value of the data over 225 variables for the 39 samples. The color of the surface shows the selection frequency of the variables as depicted on the colorbar. (C)Variable selection frequency of glyph[lscript] 1 /glyph[lscript] 2 SPLS superimposed on the octane data and its first derivative.
## 7. Application 2: Sparse Prediction of Disease Symptoms from Gene Expression
In this section we apply the Jointly Sparse Global SIMPLS Regression to 4 types of predictive health challenge studies involving the H3N2, the H1N1, the HRV, and the RSV viruses. In these challenge studies, publicly available from the NCBI-GEO website, serial peripheral blood samples were acquired from a population of subjects inoculated with live flu viruses [38, 39, 40, 41]. The prediction task in these experiments is to predict the symptom scores based on gene expression of 12023 genes. There were 10 symptom scores, i.e., runny nose, stuffy nose, sneezing, sore throat, earache, malaise, cough, shortness of breath, headache, and myalgia, documented over time. The symptoms are self-reported scores, ranging from 0 to 3. We linearly interpolate the gene expressions to match them with the sampling time of the symptom reports. We compare the Jointly Sparse Global SIMPLS-R with standard PLS-R and glyph[lscript] 1 penalized PLS-R by leaving one subject out as the test set, and the rest as the training set. The process is repeated until all subjects have been treated as the test set. The number of components for all methods and the regularization parameter in Jointly Sparse Global SIMPLS-R are selected by 2-fold cross validation to minimize the sum of the MSE of the responses. Since each subject has multiple samples, we perform the cross validation by splitting by subjects, i.e., no samples from the same subject will appear in both training and tuning sets. We restrict the responses to the first 3 symptoms, which are the upper respiratory symptoms, and the results are shown in Table 4. In most of the cases, the proposed Jointly Sparse Global SIMPLS-R method outperforms the standard PLS-R and glyph[lscript] 1 penalized PLS-R in terms of prediction MSE, number of components, number of genes. As can be seen in Table 4, the number of selected variables decreases significantly by applying the glyph[lscript] 1 /glyph[lscript] 2 mixed norm sparsity penalty to the PLSR objective function. Thus the proposed PLS-R method is able to construct a more parsimonious predictor relative to the other PLS-R methods having similar accuracy.
The PLS-R method can also be viewed as an exploratory data analysis tool for constructing low dimensional descriptors of the independent variables and response variables. Specifically, the general underlying matrix factorization model X = TP ′ + E and Y = TQ ′ + F , with latent component T = XW , provides a factor analysis model for the independent and response variables X and Y . T , P and Q can be interpreted in a similar manner as the singular vectors of PCA. However, different from PCA that does not account for the response variables, T , P and Q contain information about both the independent variables and the response variables. The factor analysis interpretation of the underlying PLS model is that T is a latent score matrix and P , Q are latent factor loading matrices that associate T with the independent variables and the response variables via the approximate matrix factorizations X ≈ TP ′ and Y ≈ TQ ′ , respectively. The correlations between the latent component T and the sum of the 3 upper respiratory symptoms are reported in Table 5, which also shows results for classic matrix factorization methods including non-negative matrix
Table 4
Performance comparison table for the predictive health study. We apply the Jointly Sparse Global SIMPLS Regression ( glyph[lscript] 1 /glyph[lscript] 2 SPLS) for sparse prediction of disease symptoms from gene expression. The performance is compared with standard PLS and glyph[lscript] 1 sparse PLS. glyph[lscript] 1 /glyph[lscript] 2 SPLS achieves lower MSE with significantly fewer variables in most of the studies.
| | | | | p-values of one sided paired t-test | p-values of one sided paired t-test |
|----------------------|--------|--------------------------|--------------------------------------------|---------------------------------------|---------------------------------------|
| | 1. PLS | 2. glyph[lscript] 1 SPLS | 3. glyph[lscript] 1 /glyph[lscript] 2 SPLS | (3,1) | (3,2) |
| H1N1 number of comp. | 2.8 | 2.3 | 2.4 | 0.1163 | 0.3322 |
| number of genes | 12023 | 3624.1 | 3575.8 | 1 . 4451 × 10 - 10 | 0.4842 |
| Overall MSE | 0.599 | 0.603 | 0.591 | 0.2890 | 0.1094 |
| Runny nose MSE | 0.167 | 0.165 | 0.167 | | |
| Stuffy nose MSE | 0.281 | 0.282 | 0.269 | | |
| Sneezing MSE | 0.151 | 0.157 | 0.155 | | |
| H3N2 number of comp. | 3.2 | 2.5 | 1.9 | 0.0030 | 0.0863 |
| number of genes | 12023 | 3944.5 | 1721.5 | 1 . 7547 × 10 - 9 | 0.0601 |
| Overall MSE | 0.623 | 0.622 | 0.609 | 0.3073 | 0.2530 |
| Runny nose MSE | 0.186 | 0.174 | 0.173 | | |
| Stuffy nose MSE | 0.277 | 0.284 | 0.272 | | |
| Sneezing MSE | 0.160 | 0.164 | 0.165 | | |
| HRV number of comp. | 2.8 | 2.3 | 2.2 | 0.0484 | 0.3773 |
| number of genes | 12023 | 2193.2 | 1779.1 | 5 . 5038 × 10 - 13 | 0.3522 |
| Overall MSE | 0.628 | 0.607 | 0.603 | 0.2020 | 0.4490 |
| Runny nose MSE | 0.243 | 0.226 | 0.232 | | |
| Stuffy nose MSE | 0.324 | 0.323 | 0.314 | | |
| Sneezing MSE | 0.062 | 0.058 | 0.057 | | |
| RSV number of comp. | 3.2 | 2.3 | 2.4 | 0.0198 | 0.4103 |
| number of genes | 12023 | 2445.4 | 3889.8 | 1 . 1584 × 10 - 9 | 0.1472 |
| Overall MSE | 0.866 | 0.920 | 0.855 | 0.3318 | 0.0567 |
| Runny nose MSE | 0.312 | 0.327 | 0.312 | | |
| Stuffy nose MSE | 0.412 | 0.448 | 0.397 | | |
| Sneezing MSE | 0.143 | 0.145 | 0.145 | | |
factorization (NMF) [42] and Bayesian linear unmixing (BLU) [43, 44], previously applied to this dataset, for comparison. Notice the sparse PLS-R methods achieve higher correlation, as expected. Remarkably, the proposed Jointly Sparse global SIMPLS-R achieves this higher degree of correlation with many fewer components and variables than the NMF and BLU methods. This experiment demonstrates that Jointly Sparse Global SIMPLS-R can be used as a factor analysis method to find the hidden molecular factors that best relate to the response.
## 8. Conclusion
The formulation of the global SIMPLS objective function with an added group sparsity penalty greatly reduces the number of variables used to predict the response. This suggests that when multiple components are desired, the variable selection technique should take into account the sparsity structure for the same variables among all the components. Our proposed Jointly Sparse Global SIMPLS Regression algorithm is able to achieve as good or better performance with fewer predictor variables and fewer components as compared to competing methods. It is thus useful for performing dimension reduction and variable
Table 5
Matrix factorization. We use the cross-validated parameters reported in Table 4 for standard PLS-R and glyph[lscript] 1 penalized PLS-R, and the Jointly Sparse Global SIMPLS-R to decide the number of components, and search over the grid { 1 2 , , ..., 10 } to find the number of factors that achieves the highest correlation for NMF and BLU. The correlation between each factor and the sum of responses is listed for each method. The first 3 methods, PLS, glyph[lscript] 1 SPLS, and glyph[lscript] 1 /glyph[lscript] 2 SPLS are supervised matrix factorizations, where as NMF and BLU are unsupervised. The unsupervised methods require many more factors to achieve comparable correlation.
| factor | correlation of 1 2 | | each factor 3 | with 4 | the sum 5 | of upper 6 | respiratory 7 8 | | symptoms 9 |
|-----------------------------------------|----------------------|------|-----------------|----------|-------------|--------------|-------------------|------|--------------|
| H1N1 | | | | | | | | | |
| PLS | 0.47 | 0.38 | 0.38 | | | | | | |
| glyph[lscript] 1 SPLS | 0.52 | 0.37 | | | | | | | |
| glyph[lscript] 1 /glyph[lscript] 2 SPLS | 0.57 | 0.35 | | | | | | | |
| NMF | 0.32 | 0.45 | 0.04 | | | | | | |
| BLU | 0.27 | 0.19 | 0.02 | 0.51 | 0.16 | 0.06 | | | |
| H3N2 | | | | | | | | | |
| PLS | 0.67 | 0.42 | 0.33 | | | | | | |
| glyph[lscript] 1 SPLS | 0.73 | 0.33 | 0.33 | | | | | | |
| glyph[lscript] 1 /L 2 SPLS | 0.71 | 0.33 | | | | | | | |
| NMF | 0.62 | 0.70 | 0.10 | | | | | | |
| BLU | 0.54 | 0.73 | 0.26 | 0.33 | 0.01 | 0.02 | 0.28 | 0.05 | 0.00 |
| HRV | | | | | | | | | |
| PLS | 0.45 | 0.43 | 0.35 | | | | | | |
| glyph[lscript] 1 SPLS | 0.52 | 0.38 | | | | | | | |
| glyph[lscript] 1 /glyph[lscript] 2 SPLS | 0.53 | 0.42 | | | | | | | |
| NMF | 0.02 | 0.22 | 0.19 | | | | | | |
| BLU | 0.11 | 0.04 | 0.18 | 0.33 | 0.01 | 0.02 | 0.28 | 0.05 | 0.00 |
| RSV | | | | | | | | | |
| PLS | 0.66 | 0.35 | 0.35 | | | | | | |
| glyph[lscript] 1 SPLS | 0.70 | 0.34 | | | | | | | |
| glyph[lscript] 1 /glyph[lscript] 2 SPLS | 0.69 | 0.39 | | | | | | | |
| NMF | 0.41 | 0.13 | 0.16 | 0.01 | 0.31 | 0.02 | 0.11 | 0.11 | 0.01 |
| BLU | 0.01 | 0.02 | 0.23 | 0.20 | 0.03 | 0.68 | 0.12 | | |
selection simultaneously in applications with large dimensional data but comparatively few samples ( n < p ).
The Jointly Sparse Global SIMPLS Regression objective function is minimized using augmented Lagrangian techniques and, in particular, the ADMM algorithm. The ADMM algorithm splits the optimization into an eigen-decomposition problem and a soft-thresholding that enforces sparsity constraints. The general framework is extendable to more complicated regularization and can thus be tailored for other PLS-type applications, e.g., positivity constraints or smoothness penalties. For example, in the chemometric application, the data is smooth over the wavelengths and we can apply wavelet shrinkage on the data or include a total variation regularization to encourage smoothness. The sparsity constraints can be imposed on the wavelet coefficients if wavelet shrinkage is applied, or together with total variation regularization. The equivalence of soft wavelet shrinkage and total variation regularization was discussed in [45]. One can also consider imposing sparsity structures on the weights corresponding to the same components, adding glyph[lscript] 1 penalty within the groups, or total variation regularization, depending on the applications. The decoupling property of the ADMM algorithm allows one to extend the Jointly Sparse Global SIMPLS Regression to these various regularizations.
## References
- [1] Wold, S., Martens, H., and Wold, H. (1983). The multivariate calibration problem in chemistry solved by the PLS method. Proceedings of the Conference on Matrix Pencils. Lectures Notes in Mathematics , 286-293.
- [2] Sj¨str¨m, M., and Wold, S., and Lindberg, W., and Persson, J., and Martens, o o H. (1983). A multivariate calibration problem in analytical chemistry solved by partial least-squares models in latent variables. Analytica Chimica Acta 150 , 61-70.
- [3] Martens, H., and Martens, M. (1999). Validation of PLS Regression models in sensory science by extended cross-validation. PLS'99 .
- [4] Rossouw, D., Robert-Grani´, C., and Besse, P. (2008). A sparse PLS for varie able selection when integrating omics data. Genetics and Molecular Biology , 7 (1), 35.
- [5] Chun, H., and Kele¸ s, S. (2009). Expression quantitative trait loci mapping with multivariate sparse partial least squares regression. Genetics , 182 (1), 79-90.
- [6] Chung, D., and Kele¸ s, S. (2010). Sparse partial least squares classification for high dimensional data. Statistical applications in genetics and molecular biology , 9 (1), 17.
- [7] Chun, H., Ballard, D. H., Cho, J., and Zhao, H. (2011). Identification of association between disease and multiple markers via sparse partial least squares regression. Genetic epidemiology , 35 (6), 479-486.
- [8] Wold, S., Ruhe, A., Wold, H., and Dunn III, W. J. (1984). The collinearity problem in linear regression. The partial least squares (PLS) approach to
- generalized inverses. SIAM Journal on Scientific and Statistical Computing , 5 (3), 735-743.
- [9] Chun, H., and Kele¸ s, S. (2010). Sparse partial least squares regression for simultaneous dimension reduction and variable selection. Journal of the Royal Statistical Society: Series B (Statistical Methodology) , 72 , 3-25.
- [10] de Jong, S. (1993). SIMPLS: an alternative approach to partial least squares regression. Chemometrics and Intelligent Laboratory Systems , 18 (3), 251263.
- [11] Wold, H. (1966). Nonlinear estimation by iterative least squares procedures. Research papers in statistics , 411-444.
- [12] Hotelling, H. (1933). Analysis of a complex of statistical variables into principal components. Journal of educational psychology , 24 (6), 417.
- [13] Tenenhaus, M. (1998). La R´ egression PLS: th´ eorie et pratique . Editions Technip.
- [14] de Jong, S. (1995). PLS shrinks. Journal of Chemometrics , 9 (4), 323-326.
- [15] Boulesteix, A. L., and Strimmer, K. (2007). Partial least squares: a versatile tool for the analysis of high-dimensional genomic data. Briefings in bioinformatics , 8 (1), 32-44.
- [16] H¨ oskuldsson, A. (1988). PLS regression methods. Journal of Chemometrics , 2 (3), 211-228.
- [17] ter Braak, C. J., and de Jong, S. (1998). The objective function of partial least squares regression. Journal of chemometrics , 12 (1), 41-54.
- [18] Jolliffe, I. T., Trendafilov, N. T., and Uddin, M. (2003). A modified principal component technique based on the LASSO. Journal of Computational and Graphical Statistics , 12 (3), 531-547.
- [19] Wold, H. (1975). Soft modelling by latent variables: the non-linear iterative partial least squares (NIPALS) approach. Perspectives in Probability and Statistics, In Honor of MS Bartlett , 117-144.
- [20] Allen, G. I., Peterson, C., Vannucci, M., and Maleti´-Savati´, M. (2013). c c Regularized partial least squares with an application to NMR spectroscopy. Statistical Analysis and Data Mining , 6 (4), 302-314.
- [21] Bach, F. R. (2008). Consistency of the group Lasso and multiple kernel learning. The Journal of Machine Learning Research , 9 , 1179-1225.
- [22] Gander, W. (1980). Least squares with a quadratic constraint. Numerische Mathematik , 36 (3), 291-307.
- [23] Gander, W., Golub, G. H., and von Matt, U. (1989). A constrained eigenvalue problem. Linear Algebra and its applications , 114 , 815-839.
- [24] Beck, A., Ben-Tal, A., and Teboulle, M. (2006). Finding a global optimal solution for a quadratically constrained fractional quadratic problem with applications to the regularized total least squares. SIAM Journal on Matrix Analysis and Applications , 28 (2), 425-445.
- [25] Eckstein, J., and Bertsekas, D. P. (1992). On the Douglas-Rachford splitting method and the proximal point algorithm for maximal monotone operators. Mathematical Programming , 55 (1-3), 293-318.
- [26] Goldstein, T., and Osher, S. (2009). The split Bregman method for L1regularized problems. SIAM Journal on Imaging Sciences , 2 (2), 323-343.
- [27] Afonso, M. V., Bioucas-Dias, J. M., and Figueiredo, M. A. (2011). An augmented Lagrangian approach to the constrained optimization formulation of imaging inverse problems. Image Processing, IEEE Transactions on , 20 (3), 681-695.
- [28] Boyd, S., Parikh, N., Chu, E., Peleato, B., and Eckstein, J. (2011). Distributed optimization and statistical learning via the alternating direction method of multipliers. Foundations and Trends R © in Machine Learning , 3 (1), 1-122.
- [29] Hong, M., and Luo, Z. Q. (2012). On the linear convergence of the alternating direction method of multipliers. arXiv preprint arXiv:1208.3922.
- [30] Goldstein, T., O'Donoghue, B., and Setzer, S. (2012). Fast alternating direction optimization methods. CAM report, 12-35.
- [31] Ramani, S., and Fessler, J. A. (2012) A splitting-based iterative algorithm for accelerated statistical X-ray CT reconstruction. Medical Imaging, IEEE Transactions on , 31 (3), 677-688.
- [32] Nien, H., and Fessler, J. A. (2014). A convergence proof of the split Bregman method for regularized least-squares problems. arXiv preprint arXiv:1402.4371.
- [33] Bair, E., Hastie, T., Paul, D., and Tibshirani, R. (2006). Prediction by supervised principal components. Journal of the American Statistical Association , 101 (473).
- [34] Bastien, P., Vinzi, V. E., and Tenenhaus, M. (2005). PLS generalised linear regression. Computational Statistics & Data Analysis , 48 (1), 17-46.
- [35] Tibshirani, R. (1996). Regression shrinkage and selection via the lasso. Journal of the Royal Statistical Society. Series B (Methodological) , 267-288.
- [36] Martens, H., and Naes, T. (1989). Multivariate calibration . Wiley.
- [37] Masoum, S., Bouveresse, D. J. R., Vercauteren, J., Jalali-Heravi, M., and Rutledge, D. N. (2006). Discrimination of wines based on 2D NMR spectra using learning vector quantization neural networks and partial least squares discriminant analysis. Analytica chimica acta , 558 (1), 144-149.
- [38] Zaas, A. K., Chen, M., Varkey, J., Veldman, T., Hero, A. O., Lucas, J., Huang, Y., Turner, R., Gilbert, A., Lambkin-Williams, R., Øien, N. C., Nicholson, B., Kingsmore, S., Carin, L., Woods, C. W., and Ginsburg, G. S. (2009). Gene Expression Signatures Diagnose Influenza and Other Symptomatic Respiratory Viral Infections in Humans. Cell Host and Microbe , 6 (3), 207-217.
- [39] Huang, Y., Zaas, A. K., Rao, A., Dobigeon, N., Woolf, P. J., Veldman, T., Øien, N. C., McClain, M. T., Varkey, J. B., Nicholson, B., Carin, L., Kingsmore, S., Woods, C. W., Ginsburg, G. S., Hero, A. O. (2011). Temporal dynamics of host molecular responses differentiate symptomatic and asymptomatic influenza a infection. PLoS genetics , 7 (8), e1002234.
- [40] Woods., C. W., McClain, M. T., Chen, M., Zaas, A. K., Nicholson, B. P., Varkey, J., Veldman, T., Kingsmore, S. F., Huang, Y., Lambkin-Williams, R., Gilbert, A. G., Hero, A. O., Ramsburg, E., Glickman, S., Lucas1, J. E., Carin, L., and Ginsburg, G. S. (2013). A host transcriptional signature for presymptomatic detection of infection in humans exposed to influenza H1N1
or H3N2. PloS one 8.1: e52198.
- [41] Zaas, A. K., Burke, T., Chen, M., McClain, M., Nicholson, B., Veldman, T., Tsalik, E. L., Fowler, V., Rivers, E. P., Otero, R., Kingsmore, S. F., Voora, D., Lucas, J., Hero, A. O., Carin, L., Woods, C. W., and Ginsburg, G. S. (2013). A Host-Based RT-PCR Gene Expression Signature to Identify Acute Respiratory Viral Infection. Science translational medicine , 5 , 203ra126.
- [42] Paatero, P., Tapper, U. (1994). Positive matrix factorization: A nonnegative factor model with optimal utilization of error estimates of data values. Environmetrics , 5 (2), 111-126.
- [43] Dobigeon, N., Moussaoui, S., Coulon, M., Tourneret, J. Y., and Hero, A. O. (2009). Joint Bayesian endmember extraction and linear unmixing for hyperspectral imagery. Signal Processing, IEEE Transactions on , 57 (11), 4355-4368.
- [44] Bazot, C., Dobigeon, N., Tourneret, J.-Y., Zaas, A. K., Ginsburg, G. S. , Hero, A. O. (2013). Unsupervised Bayesian linear unmixing of gene expression microarrays. BMC Bioinformatics , 14 (1), 99.
- [45] Steidl, G., Weickert, J., Brox, T., Mrzek, P., and Welk, M. (2004). On the equivalence of soft wavelet shrinkage, total variation diffusion, total variation regularization, and SIDEs. SIAM Journal on Numerical Analysis , 42 (2), 686-713. | null | [
"Tzu-Yu Liu",
"Laura Trinchera",
"Arthur Tenenhaus",
"Dennis Wei",
"Alfred O. Hero"
] | 2014-08-01T23:19:49+00:00 | 2014-08-01T23:19:49+00:00 | [
"stat.ME"
] | Jointly Sparse Global SIMPLS Regression | Partial least squares (PLS) regression combines dimensionality reduction and
prediction using a latent variable model. Since partial least squares
regression (PLS-R) does not require matrix inversion or diagonalization, it can
be applied to problems with large numbers of variables. As predictor dimension
increases, variable selection becomes essential to avoid over-fitting, to
provide more accurate predictors and to yield more interpretable parameters. We
propose a global variable selection approach that penalizes the total number of
variables across all PLS components. Put another way, the proposed global
penalty encourages the selected variables to be shared among the PLS
components. We formulate PLS-R with joint sparsity as a variational
optimization problem with objective function equal to a novel global SIMPLS
criterion plus a mixed norm sparsity penalty on the weight matrix. The mixed
norm sparsity penalty is the $\ell_1$ norm of the $\ell_2$ norm on the weights
corresponding to the same variable used over all the PLS components. A novel
augmented Lagrangian method is proposed to solve the optimization problem and
soft thresholding for sparsity occurs naturally as part of the iterative
solution. Experiments show that the modified PLS-R attains better or as good
performance with many fewer selected predictor variables. |
1408.0319v2 | ## Metal Transport and Chemical Heterogeneity in Early Star Forming Systems
Jeremy S. Ritter, Alan Sluder, Chalence Safranek-Shrader, Miloˇ s Milosavljevi´ c and Volker Bromm
Department of Astronomy, The University of Texas at Austin, Austin, TX 78712, USA
4 May 2015
## ABSTRACT
To constrain the properties of the first stars with the chemical abundance patterns observed in metal-poor stars, one must identify any non-trivial effects that the hydrodynamics of metal dispersal can imprint on the abundances. We use realistic cosmological hydrodynamic simulations to quantify the distribution of metals resulting from one Population III supernova and from a small number of such supernovae exploding in close succession. Overall, supernova ejecta are highly inhomogeneously dispersed throughout the simulations. When the supernova bubbles collapse, quasi-virialized metal-enriched clouds, fed by fallback from the bubbles and by streaming of metal-free gas from the cosmic web, grow in the centers of the dark matter halos. Partial turbulent homogenization on scales resolved in the simulation is observed only in the densest clouds where the vortical time scales are short enough to ensure true homogenization on subgrid scales. However, the abundances in the clouds differ from the gross yields of the supernovae. Continuing the simulations until the cloud have gone into gravitational collapse, we predict that the abundances in second-generation stars will be deficient in the innermost mass shells of the supernova (if only one has exploded) or in the ejecta of the latest supernovae (when multiple have exploded). This indicates that hydrodynamics gives rise to biases complicating the identification of nucleosynthetic sources in the chemical abundance spaces of the surviving stars.
Key words: dark ages, reionization, first stars - galaxies: dwarf - galaxies: formation methods: numerical - stars: abundances - stars: Population II
## 1 INTRODUCTION
vestiges of the complexity to persist, requiring us to model it as much as possible from first principles.
The chemical abundance patterns of metal-poor and ancient stellar populations and intergalactic absorption systems provide information about the earliest stages of chemical enrichment (for reviews, see Beers & Christlieb 2005; Frebel & Norris 2013; Karlsson et al. 2013; Frebel & Norris 2015). Because the first stellar systems were likely enriched by only a few discrete sources, one must interpret the inherent complexity of enrichment in relating the abundance patterns measured in metal-poor stars to theoretical models (Audouze & Silk 1995; Shigeyama & Tsujimoto 1998). Cosmic star formation, the driver of enrichment, is already stochastic thanks to the randomness of cosmic primordial density fluctuations. Significant additional complexity enters chemical enrichment through the turbulent hydrodynamics of the intergalactic, circumgalactic, and interstellar medium, the stochasticity of the star formation process, and the mechanics of the dispersal of nucleosynthetic products from their sources-supernovae and stellar mass loss. The effects of complexity are partially, though perhaps not completely erased through the stirring and mixing in mature star forming systems like the Milky Way's disk. But in ancient stellar populations, we expect
The recent years have seen a class of high-resolution simulations of small numbers of primordial star forming systems (e.g., Wise & Abel 2008; Greif et al. 2010; Wise et al. 2012, 2014; Ritter et al. 2012; Jeon et al. 2014) as well as of moderate-resolution simulations (somewhat farther removed from the first-principles philosophy) that allow the authors to track large samples of such systems (e.g., Ricotti et al. 2008; Bovill & Ricotti 2009; Tassis et al. 2012; Muratov et al. 2013; O'Shea et al. 2015). Without exception, these works study the local enrichment amplitude as quantified by the metallicity but do not attempt to investigate potential complex effects that could influence differential chemical abundance patterns.
A critical source of uncertainty entering chemical evolution models is the chemical yields of supernovae. The yields could vary drastically with the progenitor mass, rotation rate, the presence of a binary companion, and initial metal content (e.g., Heger & Woosley 2002, 2010). In the case of core-collapse supernovae, firstprinciples theoretical modeling is still not sufficiently predictive to permit direct synthesis of the observed stellar chemical abundance
patterns. What is worse, the yields may not be deterministic and could be sensitive to instabilities taking place immediately prior to and in the course of the explosion (e.g., Arnett & Meakin 2011; Ellinger et al. 2012; Wongwathanarat et al. 2013; Smith & Arnett 2014; Couch et al. 2015). It has therefore been a longstanding hope that the process of nucleosynthesis will be reverse engineered from the abundance patterns (e.g., Talbot & Arnett 1973; Nomoto et al. 2013). Principal component analysis (PCA) can be applied in the stellar chemical abundance space to discern the contributing classes of nucleosynthetic sources, be they individual stars or star clusters (e.g., Ting et al. 2012), if one posits that abundance patterns of the sources define monolithic basis vectors of the resulting stellar chemical abundance space. This is becoming a particularly promising direction with the arrival of large spectroscopic surveys such as HERMES-GALAH (Zucker et al. 2012; De Silva et al. 2015) and Gaia-ESO (Gilmore et al. 2012) surveys. However, the applicability of PCA becomes less clear if the hydrodynamics of metal dispersal skews the patterns, introducing biases deviating the patterns from linear superpositions of the patterns of well-defined nucleosynthetic source classes. Properly characterizing such biases would place the recovery of the properties of nucleosynthetic sources from stellar abundance data on much firmer footing.
The most primitive and metal-poor stellar populations known are the ultra-faint dwarf spheroidal satellites of the Milky Way (UFDs; Brown et al. 2012; Frebel & Bromm 2012; Vargas et al. 2013; Frebel et al. 2014), the Milky Way's stellar halo which is thought to have formed (at least in part) through the disruption of dwarf satellite galaxies (Kirby et al. 2008, Norris et al. 2010, Lai et al. 2011, but see, e.g., Lee et al. 2013) and possibly an old, metalpoor sub-population within the Milky Way bulge. 1 The prominent chemical heterogeneity in UFDs is often invoked as a potential indicator of poor mixing of supernova ejecta or an enrichment by a small number of contributing supernovae (e.g., Norris et al. 2010; Simon et al. 2015). The primitive populations could be particularly sensitive to hydrodynamical biases, given the low masses and gravitational potential well depths of the progenitor star-forming systems and the low numbers of contributing supernovae and AGB stars.
Here we make an initial attempt to identify these biases with ultra-high-resolution cosmological hydrodynamic simulations. We track the degree of chemical heterogeneity in the enrichment by a single supernova, and by a cluster of seven consecutive supernovae. The supernovae explode in a previously metal-free cosmic minihalo, a plausible UFD progenitor (Ricotti et al. 2008; Salvadori & Ferrara 2009). The two simulations are initialized from Gaussian cosmological fluctuations, ensuring that the cosmic environment being enriched is realistic.
The finite resolution of the simulations implies that we are able to directly place only a lower limit on the degree of coarsegrained chemical heterogeneity. Perfect chemical homogeneity, down to molecular scales, is expected only if the gas has been stirred by such processes as gravitational infall, the mechanical feedback from star formation, and thermal instability, to the extent that would facilitate microscopic diffusion across chemically heterogeneous sheets produced by turbulent folding. Efficient stirring generally requires that the vortical time ∼ |∇× v | -1 be much
1 We can define a stellar population as 'primitive' if the number of times the nucleosynthetic output of individual supernovae and AGB stars has been recycled is small.
shorter than the lifetime of a cloud before it collapses to form new stars (see, e.g., Pan & Scannapieco 2010, and references therein). This condition can be fulfilled in specific sites, but certainly not in general. 2
We study the transport of supernova ejecta resolved by the mass coordinate inside the progenitor star (in the simulation with one supernova) and by the temporal order of the explosion (in the simulation with multiple supernovae). We follow the transport until gas clouds have assembled in which second-generation stars can be expected to form. This allows us to determine if these nucleosynthetic sources contribute monolithically, defining invariant basis vectors of the chemical abundance space, or if, perhaps, the hydrodynamics of supernova remnant evolution favors the reprocessing into new stars of only a biased subsample of the gross nucleosynthetic yields, skewing toward the reprocessing of some elements but not others.
## 2 NUMERICAL METHODOLOGY
The simulations were initialized from the same realization of cosmological initial conditions as in Safranek-Shrader et al. (2012) and Ritter et al. (2012). We simulated the gravitational collapse of collisionless dark matter and baryonic fluid in a box of size 1 Mpc (comoving) starting at redshift z = 146 . The initial density and velocity perturbations were generated with the multiscale cosmological initial conditions package GRAFIC2 (Bertschinger 2001) with Wilkinson Microwave Anisotropy Probe 7-year cosmological parameters (Komatsu et al. 2011). We utilized two levels of nested refinement to achieve an effective resolution of 512 3 , corresponding to 230 M glyph[circledot] per dark matter particle, in a patch encompassing the density maximum on mass scales 10 8 M glyph[circledot] . Time integration was carried out with the adaptive mesh refinement (AMR) code FLASH (Fryxell et al. 2000) as described in Safranek-Shrader et al. (2012). In what follows, we use physical units as opposed to comoving units. Metallicities are quoted in absolute metal mass fractions unless specified otherwise.
Gas cooled by H 2 rovibrational emission collapsed to form a ≈ 10 6 M glyph[circledot] minihalo at redshift z ∼ 19 6 . . We inserted a single collisionless particle at the location of the gas density maximum within the minihalo to represent a Population III star (simulation 1SN) or a small cluster of such stars (simulation 7SN). The equivalent gas mass was removed by reducing the nearby gas density to a constant maximum level. In view of the recent realization that protostellar disk fragmentation (Stacy et al. 2010; Clark et al. 2011a,b; Greif et al. 2011, 2012) and evaporation by protostellar radiation (Hosokawa et al. 2011; Stacy et al. 2012) can limit the masses of Population III stars in the few tens of solar masses, we picked their masses to be in the range 20 -80 M glyph[circledot] with typical values ∼ 40 M glyph[circledot] and assumed for simplicity that they all exploded with energies 10 51 erg. Preceding the first explosion, we let the collisionless particle emit ionizing radiation for 3 Myr with an ionizing luminosity Q ion and create an H II region. Hydrodynamic expansion of the ionized gas reduced the central gas density to n ∼ 0 3 . cm -3 . After 3 Myr, we either inserted a single supernova remnant (1SN; Q ion = 6 × 10 49 photons s -1 ) or initiated a sequence of 7 consecutive supernovae (7SN; Q ion = 2 2 . × 10 50 photons s -1 ), all centered on the location of the collisionless particle.
2 Microscopic chemical mixing is a sufficient, but not a necessary condition for chemical homogeneity, since any residual heterogeneity on substellar mass scales is erased in the stars themselves.
Figure 1. Simulation 7SN 200 Myr after the first supernova explosion: density-weighted density projection ( ∫ ρ dz/ 2 ∫ ρdz , left), metal-density-weighted metal density projection ( ∫ ρ 2 Z dz/ ∫ ρ Z dz , middle), and metallicity slice through the cell with maximum density (right). The metal density is derived from the advected passive scalar tracing supernova ejecta. The cosmic web filament extends diagonally from top left. Note the relatively low metallicity near the center, resulting from a dilution of supernova ejecta with metal-free gas streaming down the cosmic web filament.
In 1SN we excised from the cosmological simulation a 1 kpc region centered on the collisionless particle, replacing the dark matter particles with a simple, parametric, spherically symmetric, time-dependent analytical dark matter density profile. The analytical profile was centered in the initial, instantaneous local rest frame of the host dark matter halo. The baryonic density and velocity field from the cosmological box was mapped directly onto the excised region, allowing us to continue the simulation in the interior of the excised region at high spatial resolution. The supernova ejecta mass was set to 40 M glyph[circledot] . In 7SN, the seven supernova delay times 3 1 . -7 7 . Myr (measured after collisionless particle insertion) were selected to represent the lifetimes of stars with masses decreasing from 80 to 20 M glyph[circledot] , following a stellar IMF with a flat dN/d ln M . Each of the supernovae was inserted in the free expansion phase, with E SN = 10 51 erg in kinetic energy and an initial radius smaller than one-tenth of the radius containing gas mass equal to that of the ejecta. The ejecta masses were in the range 25 18 5 -. M glyph[circledot] , decreasing with the order of the explosion.
tion of the ionization parameter proportional to the local hydrogenionizing flux divided by the total hydrogen number density. 3
We terminated each simulation when sufficient gas returned to the halo center to form a self-gravitating cloud with density > 10 3 cm -3 , which happened 56 and 198 Myr after collisionless particle insertion in simulations 1SN and 7SN, respectively.
## 2.1 Thermodynamic Evolution
Preceding the insertion of the collisionless star particle we integrated the full set of coupled rate equations of the nonequilibrium chemical network for the primordial chemical species H, He, and D, their common ions, and the molecules H 2 and HD (SafranekShrader et al. 2012). After insertion of the star particle we turned on an ionizing point source. We mapped the gas density onto a sourcecentered spherical grid partitioned using the HEALPIX (G´rski et al. o 2005) algorithm in the angular coordinate (3000 pixels) and logarithmically in the radial coordinate (100 radial bins). The flux in any spherical cell was diminished by the cumulative extinction in the cells at smaller radii. We then mapped the flux from the spherical grid back onto the Cartesian simulation grid. On the basis of the mapped flux, we determined if a cell was expected to be ionized. We assumed photoionization equilibrium in ionized cells and interpolated the local thermodynamic state of the gas from data tabulated with the code CLOUDY (Ferland et al. 2013) as a func-
When the first supernova remnant was inserted, we switched off the photoionizing source and began computing the cooling rates assuming that the species' abundances were in collisional ionization equilibrium. Here, the cooling rate and the ionization state were again interpolated from tables pre-computed with CLOUDY, now as a function of density, temperature, and metallicity. For the interpolation we used the metallicity as defined by the passive mass scalar metallicity on the Cartesian grid. Artificial diffusion (or, better, 'numerical teleportation') of the passive scalar metallicity is a problem hitherto unsolved in Eulerian codes (see, e.g., Footnote 6 in Ritter et al. 2012). It produces pseudo-exponential lowmetallicity tails ahead of advancing metallicity fronts. In the hot, shocked shell preceding the supernova ejecta, the teleportation tail is present, but the metallicity in the tail is too low to contribute to gas cooling. In gas with temperature glyph[greatermuch] 10 4 K , metal cooling begins to dominate only at metallicities glyph[greaterorsimilar] 0 1 . Z glyph[circledot] , higher than in the artificial tail. On the other hand, at low temperatures glyph[lessmuch] 10 4 K , metallicities even as low as glyph[lessorsimilar] 10 -3 Z glyph[circledot] (and ∼ 10 -6 Z glyph[circledot] in the presence of dust) can define the cooling rate.
Similar to the numerical teleportation of chemical abundance scalar tracers, there is the potential of insidious mass teleportation at the contact discontinuity separating the hot, shock-heated gas of the supernova bubble from the much colder gas inside the enveloping dense cooling shell. Our calculations do not include the conduction of heat from the hot to the cold side of the interface. The net effect of thermal conduction should either be a cooling flow condensing the hot into the cold phase, or a reverse, evaporation flow, depending on the structure of the interface. For the specific temperature gradients and densities in our simulations, one would expect the heat conduction to produce an evaporation rather than a cooling flow (e.g., Cowie & McKee 1977; Draine 2011). The simulations exhibit a pressure dip at the interface, implying a narrow cooling layer. This arises from the artificial numerical smearing of the contact discontinuity. Dense gas from the cold side smears over onto the hot side thus artificially lowering the cooling time in a layer one cell thick. The cell-thick pressure dip is an under-resolved hy-
3 While not photon-conserving, this method is sufficient to simulate the anisotropic H II region expansion in the relatively dense gas of our minihalo.
drodynamic feature. The numerical artifacts that it may produce should be sensitive to the particulars of the numerical scheme such as the choice of slope limiter and approximate Riemann solver. We will perform a systematic study of numerical behavior at contact discontinuities in radiating flows in a separate, technical paper.
Since accurate tracking of molecular processes requires integration of stiff and computationally expensive chemical rate equations, to accelerate the computation, we did not track the H 2 abundance and did not include molecular cooling after the first supernova remnant insertion. This is a potential oversimplification as H 2 should form as the photo- and supernova-ionized gas recombines. Absent a high dust abundance that would drive molecule formation on grains (we do not track dust but its abundance is limited from above by the metallicity that remains low), H 2 abundance in the recombined gas should be n H 2 /n H ∼ 10 -3 -10 -2 (e.g., Mac Low & Shull 1986; Shapiro & Kang 1987; Kang & Shapiro 1992). Molecule abundance reaches maximum within about a million years after gas cools below 10 4 K. In our simulation, this gas has metallicity glyph[lessmuch] 10 -2 Z glyph[circledot] and cooling by H 2 should dominate (Glover & Clark 2014), at least in a relatively narrow density range n H ∼ 10 -100 cm -3 (see, e.g., Figure 6 in Safranek-Shrader et al. 2014). Again barring a full atomic-to-molecular transition that would require abundant dust, at still higher densities glyph[greaterorsimilar] 100 cm -3 , metal line cooling dominates even before molecular transitions thermalize at ∼ 10 4 cm -3 .
The artifact of our neglecting molecular cooling at densities ∼ 10 -100 cm -3 is that the supernova-enriched, recollapsed gas will linger longer under quasi-hydrostatic conditions; a longer interval will pass before gravitational potential well becomes deep enough to compress the quasi-isothermal gas to the threshold of runaway gravitational collapse. The results that follow should be interpreted as placing an upper limit on the cooling and recollapse time in a supernova-enriched dark matter minihalo following a single supernova or a cluster of supernovae. The limit is strictly valid only in the fine-tuned regime in which molecule formation is completely suppressed by a dissociating UV background. The true recollapse time could be factor of a few shorter than in our simulations if the molecular abundance is significant. We will specifically address molecular cooling in our next work on this topic.
## 2.2 Adaptive Mesh Refinement Control
Throughout the simulation, the AMR resolution was dynamically and adaptively adjusted to resolve the structures of interest. The refinement level glyph[lscript] of AMR blocks relates the size of the computational box L to size of an individual AMR grid cell by ∆ x = 2 -glyph[lscript] +1 -3 L , where the factor of 2 -3 arises from the subdivision of AMR blocks into 8 cells along each axis ( 8 3 = 512 cells per block). During the initial gravitational collapse of dark matter and gas leading to the formation of a minihalo we refined based on the gas density ρ . Specifically we raised the refinement to level glyph[lscript] to satisfy the condition ρ < 3¯2 ρ 3(1+ φ )( glyph[lscript] -glyph[lscript] base ) , where glyph[lscript] base = 5 is the initial refinement level at the start of the simulation, ¯ ρ is the mean density in the box, and φ = -0 3 . (see, e.g., Safranek-Shrader et al. 2012, and references therein). After the insertion of a star particle, which takes place in the most massive dark matter minihalo in the box, we discontinued enforcing density-based refinement in other halos. At the evolving location of the star particle we maintained the maximum refinement level attained when refining based on gas density even after the H II region broke out and gas density dropped.
In order the resolve the free expansion phase of a supernova remnant, it is necessary to resolve scales much smaller than the radius at which the expanding blast wave sweeps up a gas mass similar to the ejecta mass. The grid resolution at each supernova insertion was forced to be glyph[lessorequalslant] 0 03 . pc and somewhat coarser for the later supernovae in 7SN that explode in the bubble containing gas shocked by early supernovae. We additionally refined the grid up to the maximum resolution anywhere in the box using the standard second-derivative refinement criterion in FLASH tuned to aggressively refine ahead of compositional discontinuities where metallicity jumps. This ensured that the forward shock wave, contact discontinuity, and reverse shock were maintained at the highest available grid refinement level. As each remnant expanded, we degraded the resolution while ensuring that the diameter of the remnant (or superbubble) was resolved by at least 256 cells. In the late stages of the simulations focused on the recollapse in the halo, we degraded the resolution in regions causally disconnected from the halo center, outside the virial radius of the halo.
In all supernovae we assumed that 10% of the ejecta mass was in metals, with α -enhanced solar abundances ( [ α/ Fe ] = 0 5 . ). In the thermodynamic calculations (but not in tracking metal dispersal in 1SN), we assumed that the metal abundances were homogeneous within each supernova's ejecta. The tracking of ejecta was carried out with Lagrangian passive tracer particles, N trace = 10 7 in 1SN and (2 -3) × 10 6 per supernova or a total of N trace ≈ 2 × 10 7 in 7SN. The tracers allowed us to connect ejecta fluid elements to their origin in the explosions. In 1SN they allowed us to distinguish between the ejecta originating in distinct mass shells within the explosion, and in 7SN between the ejecta originating in different supernovae. It is worth noting that the numerical limitations affecting tracers, e.g., associated with the finiteness of the order of the velocity field representation on the computational mesh, are distinct from the numerical teleportation problem mentioned in Section 2.1. Thus, the simulated metallicity is generally to be trusted in cells in which the Eulerian and Lagrangian tracers indicate consistent values of the metallicity.
## 3 HYDRODYNAMIC EVOLUTION
The supernova remnants evolved through the free expansion, Sedov-Taylor, pressure-driven and momentum-conserving snowplow phases, and eventually, underwent partial collapse along the direction of the cosmic gas inflow, parallel to the filaments of the cosmic web. In 7SN, the later exploding supernovae expanded in the hot, low-density bubble evacuated by earlier ones, with blastwaves colliding with the dense radiative shell before the reverse shocks have completed traversal of the ejecta.
For the hydrodynamic and chemical evolution of the entire remnant or bubble, it is important that the photoionization by the supernova progenitor stars did not completely photoevaporate the densest primordial clouds inside the halo (Abel et al. 2007; BlandHawthorn et al. 2011; Ritter et al. 2012). While the photoevaporation flows are difficult to adequately resolve in cosmological simulations, survival of neutral clouds inside the primordial H II region is expected from the analytical evaporation solutions of Bertoldi & McKee (1990). The densest clouds with central densities n glyph[lessorsimilar] 10 cm -3 and distances ∼ 50 -100 pc from the center of the halo are associated with the filamentary inflow from the cosmic web. Supernova blastwaves swept past these clouds, partially ablating them and depositing some ejecta material at the perimeters of the clouds. The blastwave-cloud interaction drove turbulence inside the bubble. The ablated primordial gas found itself inside the hot,
Figure 2. A slice of vorticity magnitude |∇ × v | in the center of 1SN (left) and 7SN (right) overplotting ejecta tracer particles in a 5 pc thick slab containing the slice (black dots). Note that metal ejecta are unmixed outside the central high-vorticity core, tens of parsecs in radius.
turbulent interior, where it appeared to be susceptible to turbulentstirring-aided mixing, especially in the multi-supernova simulation. This intra-bubble mixing resulted in a modest, factor of glyph[lessorsimilar] 10 dilution of the ejecta by the primordial gas.
After ∼ 0 13 . and ∼ 1 Myr from the (first) explosion in simulation 1SN and 7SN, respectively, the ejecta material started to accumulate in the pressure-driven snowplow shell. The shell was thin < 10 pc and not adequately resolved at grid resolution ∼ 0 5 . -1 pc. The insufficient resolution blurred the contact discontinuity and its associated compositional gradient. As a result, the metallicity derived from the passive mass scalar was diluted to Z ∼ 10 -4 -10 -3 in the thin shell. Rayleigh-Taylor (RT) fingers first became prominent in the shell after ∼ 3 Myr and then became extended, with length scales comparable to the radius of the bubble, at ∼ 20 Myr. The long-term hydrodynamical evolution of the supernova bubble was highly anisotropic, with a fraction of the ejecta and swept-up primordial medium traveling many halo virial radii perpendicular to the cosmic web filaments, and another ∼ 50% of the ejecta remaining within the virial radius. 4
lead to the formation of second-generation stars. Outside the quasivirialized cloud, the vortical time scales are longer, 10 -100 Myr, precluding mixing. Figure 3 shows the joint distribution of gas density and metallicity in the two simulations. Metallicity spread in low-density gas is high, indicating a high degree of inhomogeneity. The spread decreases with increasing density, becoming narrow for n glyph[greaterorsimilar] 10 -100 cm -3 . The narrowing of the metallicity spread is a consequence of rapid turbulent homogenization. The densest gas has negligible metallicity spread with Z ≈ 5 × 10 -6 in 1SN and 5 × 10 -7 in 7SN. We can conclude that the very first generation of metal-enriched stars will be chemically homogeneous on the scale of a stellar group or cluster. Recently, Feng & Krumholz (2014) observed the same in simulations of star clusters forming in a very different environment, the Milky Way disk (see, also, BlandHawthorn et al. 2010). We shall see in the following section, however, that stellar abundance patterns in the first metal-enriched star forming systems will not be simple superpositions of the yield patterns of the contributing supernovae.
After the supernova bubbles started to collapse, the bubble interiors cooled to ∼ 10 4 K and began intermixing with the ambient unshocked gas. Dual inflows fed toward the halo center: from the infall of unenriched, primordial clouds (including from merging halos in 7SN), and from the fragments of the buckling thin shell. The terminus of the inflows was a turbulent quasi-virialized (or quasihydrostatic) cloud in which turbulence was stirred by gravitational infall. Figure 1 illustrates the overall geometry of the metal distribution at this stage. It shows a homogenized low-metallicity interior surrounded by more metal rich, inhomogeneous clouds.
Figure 2, plotting the amplitude of the fluid vorticity, shows that vortical time scales in the quasi-virialized cloud are |∇ × v | -1 ∼ 0 1 . -1 Myr, short enough to facilitate turbulent-cascadeaided fluid mixing in ∼ 10 Myr . Vorticity is the highest near the center of each plot, where the metal-enriched gas has gone into runaway gravitational collapse. We expect the collapse to ultimately
4 The morphological evolution of the metal-enriched volume bears similarities to both the centered and off-center idealized non-cosmological simulations of Webster et al. (2014).
## 4 MONOLITHIC AND BIASED ENRICHMENT
We procede to analyze how hydrodynamic dispersal depends on the nucleosynthetic site. In analyzing simulation 1SN, we split the ejecta at supernova insertion into N bin = 7 radial bins containing equal ejecta masses and treat the bin index i = 1 , ..., N bin as a crude proxy for the isotopic group synthesized in the corresponding bin. For example, the innermost bins could contain the explosively synthesized Fe peak and α elements, the intermediate bins could be rich in light hydrostatic elements (C and O), and the outermost bins would contain H and He. This idealizes the explosion as preserving spherical symmetry, which is certainly not the case, as symmetry is strongly broken by convection preceding and during the collapse and by RT fingering during the explosion (our simulations can capture the RT instability in the remnant but not in the explosion). However, we expect that the radial 'dredging' of elements by instabilities (and rotation-driven mixing, see, e.g., Maeder &Meynet 2012) is incomplete and that some radial stratification is preserved until the ejecta enter free expansion. In analyzing simulation 7SN, we ignore the stratification inside each explosion and take
Figure 3. Metallicity as a function of gas density at the end of the simulation 1SN at 56 Myr after the first explosion (left plot) and 7SN at 200 Myr after the first explosion (right plot). From red to blue, the color scales with the logarithm of the fluid mass in the density and metallicity bin. Solid curve is the mean metallicity and colored curves are the fractional contributions from the seven radial mass bins (1SN) and seven supernovae (7SN).
the bin index i = 1 , ..., N bin to range over the N bin = N SN = 7 supernovae.
We first examine the degree to which the ejecta in different bins are stirred with each other (to be further discussed in a forthcoming follow-up paper). In 1SN, the outer, high-velocity mass shells of the ejecta are well-stirred between themselves, but inner mass shells remain highly inhomogeneous. In 7SN, the ejecta from supernovae separated by short time intervals are well-stirred with each other. The ejecta from later supernovae, separated from the earlier supernovae by the longest time intervals, remain inhomogeneous.
We call the enrichment by the source (a single supernova in 1SN and a cluster of supernovae in 7SN) monolithic if the density of isotope A at location x is proportional to the sum of the isotopic yields of individual bins Y i,A over the bin index. Specifically,
$$\rho _ { A } ( x ) = Z ( x ) \rho ( x ) \frac { Y _ { A } } { \sum _ { A } Y _ { A } }, \ \ ( \text{monolithic} ), \quad \ ( 1 ) \quad \text{ even}$$
where
$$Y _ { A } & \equiv \sum _ { i = 1 } ^ { N _ { \text{bin} } } Y _ { i, A } & ( 2 )$$
and ρ ( x ) and Z ( x ) are the total mass density and total metallicity from the enrichment event, respectively, such that ∫ Zρd x 3 = ∑ A Y A . It is standard to assume (e.g., Ting et al. 2012) that enrichment is indeed monolithic so that each enrichment event E defines a unique chemical space vector Y ( E ) ≡ ( Y ( E ) A 1 , Y ( E ) A 2 , ... ) ∼ Y ( C ) representative of a class C of nucleosynthetic sources (e.g., core collapse events with or without r-process).
If star formation takes place at space-time points ( x k , t k ) , then stellar abundances are given by 5
$$Z _ { k, A } = \frac { 1 } { \rho ( { \mathbf x } _ { k }, t _ { k } ) } \sum _ { \mathcal { E } } \rho _ { A } ^ { ( \mathcal { E } ) } ( { \mathbf x } _ { k }, t _ { k } ), \ 1 \leqslant k \leqslant N _ { \text{star} } \quad \ \ ( 3 ) \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
5 Here and in what follows, coarse graining of ρ A on the spatial scales of star-forming clumps is implied.
The goal of chemical abundance analysis is then to isolate nucleosynthetic source classes by recovering their yield vectors Y ( C ) from stellar abundance data Z k,A . The dimensionality of the chemical abundance space corresponds to the number of different nucleosynthetic classes.
The metallicities resulting from individual events Z ( E ) ( x , t ) are random variables determined by the hydrodynamics of metal dispersal. They endow stellar metallicities with scatter. If a sufficient number of events contributes and the events are statistically independent, the central limit theorem implies that the scatter is Gaussian, justifying the PCA approach.
It cannot be taken for granted, however, that the source contributions are monolithic (Eqn. 1), because nucleosynthetic products are injected into the hydrodynamic environment with different velocities and at different times and are transported differently to their star formation sites. In perfect generality, for each nucleosynthetic event,
$$\lhd \nu \quad & \cdots \cdots, \\ \rho _ { A } ( x ) \quad & = \quad \sum _ { i = 1 } ^ { N _ { \text{bin} } } Z _ { i } ( x ) \rho ( x ) \frac { Y _ { i, A } } { \sum _ { A } Y _ { i, A } } \\ ( 2 ) \quad & = \quad Z ( x ) \rho ( x ) \frac { \sum _ { i = 1 } ^ { N _ { \text{bin} } } w _ { i } ( x ) Y _ { i, A } } { \sum _ { A } Y _ { A } }, \\ \text{all-} \quad & \text{when $Z_{\text{$\prime}\prime} \vee$ is the total multi-attempted at the time, $k$} \, $k$} \, $i$}$$
where Z i ( x ) is the total metallicity at x due to enrichment by bin i such that ∫ Z ρd x i 3 = ∑ A Y i,A , and in the second line, we introduced the weights
$$\underset { \text{core} } { \overset { \dots } { Y ^ { ( \mathcal { C } ) } } } \quad w _ { i } ( \mathbf x ) = \frac { Z _ { i } ( \mathbf x ) } { Z ( \mathbf x ) } \, \frac { \sum _ { A } Y _ { A } } { \sum _ { A } Y _ { i, A } }$$
for straightforward comparison with the monolithic case (Eq. 1).
If source contributions are monolithic then w i ≡ 1 /N bin , but in general, the enrichment at a specific location could be biased toward some bins, giving them higher weights. The dimensionality of the chemical abundance space can now be much larger than the number of nucleosynthetic source classes. The weights encapsulate the biases that hydrodynamics introduces into abundance patterns. If the weights are completely random and possess unknown statistics, this introduces uncontrolled biases frustrating the recov-
ery of nucleosynthetic source classes from stellar abundance data. Here, we make the first step toward characterizing the nature of the biases, aspiring to detect regularities that can be factored into chemical abundance analysis. Figure 3 shows bin-specific metallicities Z i spherically averaged around the center of gravitational collapse. In both simulations, in dense gas n glyph[greaterorsimilar] 1 cm -3 , departures from monolithic enrichment are evident. In 1SN, the innermost radial bin, which is expected to carry explosive elements, is deficient by a factor of ∼ 3 relative to the other bins, 〈 w 1 〉 ∼ 1 3 〈 w 2 〉 and 〈 w 2 〉 ∼ ... ∼ 〈 w 7 〉 , where the averages refer to gas-mass-weighted averages of w i in the dense gas. In 7SN, ejecta from the first two supernovae are ∼ 3 times as abundant as the ejecta from the last two supernovae, 〈 w 1 〉 ∼ 〈 w 2 〉 ∼ 3 〈 w 6 〉 ∼ 3 〈 w 7 〉 .
We propose the following physical interpretation. In the case of the solitary supernova, the reverse shock raises the inner ejecta shells, which it sweeps later and at lower densities, to a higher entropy than the outer shells, which it sweeps earlier and at higher densities. This can be seen by considering the Sedov-Taylor point explosion, in which the pressure asymptotes to a constant value near the center, but density decreases toward the center as ρ ∝ r 3 / γ ( -1) . Therefore, with γ = 5 3 / , the entropy s = ln( P/ρ γ ) + const rises toward the center of the ejecta as s ∼ -27 10 ln r + const. This is significant because as the ejecta become quasi-isobaric inside the remnant, the radiative cooling time is a steeply increasing function of entropy. The outer ejecta shells cool first and are incorporated into the snowplow shell, while the innermost ejecta avoid cooling. As the remnant stalls and collapses, the innermost ejecta, having higher entropy, are outward buoyant and rise to larger radii, thus avoiding collapse and allowing low-entropy outer ejecta mass shells to fall in to enrich the quasi-virialized cloud. In 1SN, the ejecta tracer particle radii containing 25% of the ejecta in bins 1 (the innermost bin) and 2 cross at ∼ 0 2 . Myr after the explosion and then bin 1 interchanges with bins glyph[greaterorequalslant] 3 at ∼ 0 3 . Myr.
In the case of clustered supernovae, the situation is similar, but now since the density inside the supernova bubble keeps dropping as the bubble expands, the ejecta from later supernovae are on average raised to higher entropies than those of earlier ones and are less susceptible to cooling. Since the ejecta of the latest supernovae remain hot, they are outward buoyant. Upon the collapse of the bubble, the hot ejecta of the later supernovae interchange with the cooled ejecta of the earlier supernovae. The latter fall in to enrich the central cloud.
## 5 CONFRONTING THE EMPIRICAL RECORD
Our results have important implications for understanding the early stages of cosmic chemical evolution. More precisely, the hydrodynamic biases in the transport of individual elements, introduced by the post-explosion evolution, need to be taken into account when confronting the empirical abundance trends. The ultimate goal here is to achieve a robust mapping from the observed abundance pattern in metal-poor stars or systems thereof to the individual sources of those metals. In the absence of any monolithic mapping between sources and fossil record (see Section 4), the hydrodynamic transport process constitutes the missing link in our current understanding. Simulations along the lines of our exploratory work here promise to bridge this crucial gap. However, we can already now address a long-standing problem in Galactic chemical evolution in a new light. This concerns the prevalence of peculiar abundance ratios in low-metallicity stars and systems.
To briefly summarize the main phenomenology, observations of metal-poor stars in the Galactic halo, assembled over more than two decades, have provided intriguing constraints on the nature of early chemical evolution (reviewed in Beers & Christlieb 2005; Frebel & Norris 2013; Karlsson et al. 2013). The current state of the art is defined by a large sample of halo red giant stars, where key lines are sufficiently strong to enable high signal-tonoise spectroscopy (Cayrel et al. 2004; Franc ¸ois et al. 2007). The main lessons are two-fold: both α -elements (Mg, Ca, Si, Ti), and iron-peak elements (V to Zn) exhibit extremely small scatter, down to [Fe/H] ∼ -3 5 . . The neutron-capture elements, comprising elements beyond Zn, on the other hand, exhibit equally small scatter down to [Fe/H] ∼ -3 , but show extremely large scatter, up to 5 dex, below this (Qian & Wasserburg 2002; Truran et al. 2002; Sneden et al. 2008). Finally, the lighter elements (C, N, O) again show large abundance variations at [Fe/H] < -4 , and approach well-defined trends for less metal-poor stars. 6 The most dramatic manifestation of this transition to well-behaved abundance trends, once a threshold metallicity is reached, is provided by the huge scatter in r-process abundances, seen in Galactic halo stars with [Fe/H] glyph[lessorsimilar] -3 (reviewed in Sneden et al. 2008). The origin of rprocess nucleosynthesis, such as the mass and properties of the supernova progenitor star, is still highly uncertain. There is, however, tentative evidence that the r-process might operate in progenitor stars with a very narrow mass range, possibly close to the lowermass limit for core-collapse supernovae (e.g., Qian & Wasserburg 2008).
It has been challenging to explain all of these trends within one comprehensive framework, but our work suggests a promising Ansatz to do so. Basically, our results show that early enrichment is differential, non-monolithic in nature. This specifically implies that second-generation star formation does not sample the enrichment from the full IMF, and possibly not even that from a single explosion (see Fig. 3). Our simulations build on earlier analytical work that had postulated a minimum number of supernovae, N SN glyph[greaterorsimilar] 20 , needed to average out any yield inhomogeneity from individual explosion sites (see, e.g., Tsujimoto & Shigeyama 1998; Tsujimoto et al. 1999). This absence of effective source-averaging, then, such that only a small number of explosion sites contribute, is the key requirement to preserve peculiar abundance ratios. Next to the classical r-process elements, a similar explanation may pertain to the strong odd-even pattern that is predicted for pair-instability supernova enrichment, but has not been detected so far (Heger & Woosley 2002; Karlsson et al. 2008). We note that our explanation of the r-process record, understanding huge scatter as a result of sparse IMF-sampling, where only a very narrow progenitor mass range gives rise to the r-process, does not require any stochastic contribution from neutron star mergers, as has recently been suggested (Shen et al. 2014; van de Voort et al. 2014).
Once cosmological structure formation advances to more massive systems, with deeper potential wells to facilitate the nearuniform mixing of the ejecta from a large number of supernovae, convergence towards well-defined, smooth abundance trends will set in. Extragalactic observations targeting systems of greatly different virial mass are in agreement with this overall picture. Specifically, recent medium- and high-resolution spectroscopy of red giant stars in Milky Way dwarf satellites has established that their abun-
6 The skewing of abundance patterns by post-supernova hydrodynamics may also be the explanation of abundance anomalies (e.g., Feltzing et al. 2009; Cohen et al. 2013; Yong et al. 2013) found in primitive stellar systems in the local universe.
dance properties, including the degree of scatter, are indistinguishable from the metal-poor tail of the Galactic halo stars (Frebel et al. 2010, 2014). A complementary view into early metal enrichment is provided by the abundances measured in damped Lymanα (DLA) systems. Here, the evidence points towards extremely low scatter, at least for the prominent α -elements (Cooke et al. 2011; Becker et al. 2012). This may be indicative of the onset of efficient gasphase mixing in the deep potential wells of the DLA dark matter host halos. An intriguing question for future simulation work is to test whether the empirical threshold metallicity, roughly measured by [Fe/H], for the disappearance of such anomalous abundance signatures can be reproduced.
Of special interest is the observed dichotomy of carbonenhanced and carbon-normal metal-poor stars (e.g., Beers & Christlieb 2005; Gilmore et al. 2013). Cooke & Madau (2014, hereafter CM14) have recently suggested that the strength of supernova driven outflows may be responsible for this bimodality. These authors invoke two classes of supernovae with greatly different times for recovery from the supernova explosions with different explosion energies. More energetic explosions E SN glyph[greatermuch] 10 51 erg imply longer recovery times (Jeon et al. 2014). According to CM14, rapid recovery is connected to weak explosions with large carbon overabundances, and slow recovery with strong explosions with normal carbon yields. Our simulation 1SN, where we find the return of supernova ejecta into the halo center deficient in the ejecta originating from the innermost 10% of the ejecta, suggests another mechanism for carbon-enhanced metal poor (CEMP) stars. A core-collapse explosion ejects a normal carbon-to-iron ratio [C / Fe] < 1 , but the iron-bearing ejecta are raised to a higher entropy upon reverse shock traversal than the carbon-bearing ejecta, and subsequently do not cool and settle into the halo center to form second-generation stars.
## 6 SUMMARY AND CONCLUSIONS
We have carried out two complementary very-high-resolution cosmological simulations of how the metals produced in the first supernova explosions are transported into the cold, dense gas out of which the second-generation of (Population II) stars is formed. The first simulation followed the ejecta from a single explosion, whereas the second traced the metal dispersal from seven sources. We arrived at two main conclusions. The re-condensed Population II star forming material exhibits strong turbulent vorticity, implying a likely fine-grained turbulent mixing of gas down to very small, unresolved, scales. Stellar clusters or groups forming out of this material are thus predicted to be chemically uniform, unless any self-enrichment process may operate during the later stages of stellar evolution. Our results also indicate that the hydrodynamic metal transport proceeds differentially, such that the monolithic mapping of source abundances into the fossil record is broken.
The hydrodynamically biased nature of early metal enrichment, as demonstrated in our pathfinder simulations, has multiple implications, requiring that we rethink a number of our traditional assumptions and methodologies. On the theory side, a common approach to chemical enrichment is to assume that homogenization of the chemical composition is instantaneous and complete in certain 'mixing volumes,' normally centered on nucleosynthetic sources, but that the volumes themselves occur stochastically and intermittently, tracing star formation. This approach provides a rudimentary model of inhomogeneous chemical evolution, but since the choice of mixing volumes is ad hoc, its predictive power is lim- ited. It is standard to motivate the choice of mixing volumes by considering the spatial extent of supernova remnants and galactic superbubbles, assuming that the medium is chemically homogeneous within these structures (Argast et al. 2000; Oey 2000, 2003; Karlsson 2005; Karlsson & Gustafsson 2005; Karlsson et al. 2008; Bland-Hawthorn et al. 2010; Corlies et al. 2013). In the absence of monolithic source mapping, it is not obvious how to adjust this technique.
Our results also present a challenge to the standard interpretational framework of near-field cosmology. The hydrodynamic biases need to be quantified as a function of the star forming environment by carrying out a dedicated program of simulations. Once these chemical transport 'maps' are in hand, the full power of stellar archaeology can be unleashed. Such a program is very timely, given the advent of large survey projects, including the dedicated efforts connected to Gaia-ESO, which promise a record of early chemical evolution in unprecedented detail.
## ACKNOWLEDGMENTS
We acknowledge helpful conversations with Anna Frebel, John Scalo, and Craig Wheeler. The FLASH code was in part developed by the DOE-supported Flash Center for Computational Science at the University of Chicago. The authors acknowledge the Texas Advanced Computing Center at The University of Texas at Austin for providing HPC resources under XSEDE allocation TGAST120024. CSS is grateful for support provided by the NASA Earth and Space Science Fellowship (NESSF) program. This study was supported by the NSF grant AST-1009928 and by the NASA grant NNX09AJ33G. Many of the visualizations were made with the YT package (Turk et al. 2011).
## REFERENCES
Abel, T., Wise, J. H., & Bryan, G. L. 2007, ApJ, 659, L87
Argast, D., Samland, M., Gerhard, O. E., & Thielemann, F.-K. 2000, A&A, 356, 873
Arnett, W. D., & Meakin, C. 2011, ApJ, 733, 78
Audouze, J., & Silk, J. 1995, ApJ, 451, L49
Becker, G. D., Sargent, W. L. W., Rauch, M., & Carswell, R. F. 2012, ApJ, 744, 91
Beers, T. C., & Christlieb, N. 2005, ARA&A, 43, 531
Bertoldi, F., & McKee, C. F. 1990, ApJ, 354, 529
Bertschinger, E. 2001, ApJS, 137, 1
Bland-Hawthorn, J., Karlsson, T., Sharma, S., Krumholz, M., & Silk, J. 2010, ApJ, 721, 582
Bland-Hawthorn, J., Sutherland, R., & Karlsson, T. 2011, EAS Publications Series, 48, 397
Bovill, M. S., & Ricotti, M. 2009, ApJ, 693, 1859
Brown, T. M., Tumlinson, J., Geha, M., et al. 2012, ApJ, 753, L21
Cayrel, R., Depagne, E., Spite, M., et al. 2004, A&A, 416, 1117
Clark, P. C., Glover, S. C. O., Smith, R. J., et al. 2011a, Science, 331, 1040
- Clark, P. C., Glover, S. C. O., Klessen, R. S., & Bromm, V. 2011b, ApJ, 727, 110
Cohen, J. G., Christlieb, N., Thompson, I., et al. 2013, ApJ, 778, 56
Cooke, R., Pettini, M., Steidel, C. C., Rudie, G. C., & Nissen, P. E. 2011, MNRAS, 417, 1534
Cooke, R., & Madau, P. 2014, arXiv:1405.7369
Corlies, L., Johnston, K. V., Tumlinson, J., & Bryan, G. 2013, ApJ, 773, 105
Couch, S. M., Chatzopoulos, E., Arnett, W. D., & Timmes, F. X. 2015, arXiv:1503.02199
Cowie, L. L., & McKee, C. F. 1977, ApJ, 211, 135
- De Silva, G. M., Freeman, K. C., Bland-Hawthorn, J., et al. 2015, arXiv:1502.04767
- Draine, B. T. 2011, Physics of the Interstellar and Intergalactic Medium by Bruce T. Draine. Princeton University Press, 2011. ISBN: 978-0-69112214-4,
- Ellinger, C. I., Young, P. A., Fryer, C. L., & Rockefeller, G. 2012, ApJ, 755, 160
- Feltzing, S., Eriksson, K., Kleyna, J., & Wilkinson, M. I. 2009, A&A, 508, L1
- Feng, Y., & Krumholz, M. R. 2014, Nature, 513, 523
- Ferland, G. J., Porter, R. L., van Hoof, P. A. M., et al. 2013, RMxAA, 49, 137
- Franc ¸ois, P., Depagne, E., Hill, V ., et al. 2007, A&A, 476, 935
- Frebel, A., Simon, J. D., Geha, M., & Willman, B. 2010, ApJ, 708, 560
- Frebel, A., & Bromm, V. 2012, ApJ, 759, 115
- Frebel, A., & Norris, J. E. 2013, Planets, Stars and Stellar Systems. Volume 5: Galactic Structure and Stellar Populations, 55
- Frebel, A., Simon, J. D., & Kirby, E. N. 2014, ApJ, 786, 74
- Frebel, A., & Norris, J. E. 2015, arXiv:1501.06921
- Fryxell, B., Olson, K., Ricker, P., et al. 2000, ApJS, 131, 273
- Gilmore, G., Randich, S., Asplund, M., et al. 2012, The Messenger, 147, 25
- Gilmore, G., Norris, J. E., Monaco, L., et al. 2013, ApJ, 763, 61 Greif, T. H., Glover, S. C. O., Bromm, V., & Klessen, R. S. 2010, ApJ, 716, 510
- Glover, S. C. O., & Clark, P. C. 2014, MNRAS, 437, 9
- G´ orski, K. M., Hivon, E., Banday, A. J., et al. 2005, ApJ, 622, 759
- Greif, T. H., Springel, V ., White, S. D. M., et al. 2011, ApJ, 737, 75
- Greif, T. H., Bromm, V., Clark, P. C., et al. 2012, MNRAS, 424, 399
- Heger, A., & Woosley, S. E. 2002, ApJ, 567, 532
- Heger, A., & Woosley, S. E. 2010, ApJ, 724, 341
- Hosokawa, T., Omukai, K., Yoshida, N., & Yorke, H. W. 2011, Science, 334, 1250
- Jeon, M., Pawlik, A. H., Bromm, V., & Milosavljevic, M. 2014, arXiv:1407.0034
- Kang, H., & Shapiro, P. R. 1992, ApJ, 386, 432
- Karlsson, T. 2005, A&A, 439, 93
- Karlsson, T., & Gustafsson, B. 2005, A&A, 436, 879
- Karlsson, T., Johnson, J. L., & Bromm, V. 2008, ApJ, 679, 6
- Karlsson, T., Bromm, V., & Bland-Hawthorn, J. 2013, Reviews of Modern Physics, 85, 809
- Kirby, E. N., Simon, J. D., Geha, M., Guhathakurta, P., & Frebel, A. 2008, ApJ, 685, L43
- Komatsu, E., Smith, K. M., Dunkley, J., et al. 2011, ApJS, 192, 18
- Lai, D. K., Lee, Y. S., Bolte, M., et al. 2011, ApJ, 738, 51
- Lee, D. M., Johnston, K. V., Tumlinson, J., Sen, B., & Simon, J. D. 2013, ApJ, 774, 103
- Mac Low, M.-M., & Shull, J. M. 1986, ApJ, 302, 585
- Maeder, A., & Meynet, G. 2012, Reviews of Modern Physics, 84, 25
- Muratov, A. L., Gnedin, O. Y., Gnedin, N. Y., & Zemp, M. 2013, ApJ, 773, 19
- Nomoto, K., Kobayashi, C., & Tominaga, N. 2013, ARA&A, 51, 457
- Norris, J. E., Gilmore, G., Wyse, R. F. G., Yong, D., & Frebel, A. 2010, ApJ, 722, L104
- Norris, J. E., Yong, D., Gilmore, G., & Wyse, R. F. G. 2010, ApJ, 711, 350
- Oey, M. S. 2000, ApJ, 542, L25
- Oey, M. S. 2003, MNRAS, 339, 849
- Oh, S. P., & Haiman, Z. 2002, ApJ, 569, 558
- O'Shea, B. W., Wise, J. H., Xu, H., & Norman, M. L. 2015, arXiv:1503.01110
- Pan, L., & Scannapieco, E. 2010, ApJ, 721, 1765
Qian, Y.-Z., & Wasserburg, G. J. 2002, ApJ, 567, 515
- Qian, Y.-Z., & Wasserburg, G. J. 2008, ApJ, 687, 272
- Ricotti, M., Gnedin, N. Y., & Shull, J. M. 2008, ApJ, 685, 21
- Ritter, J. S., Safranek-Shrader, C., Gnat, O., Milosavljevi´, M., & Bromm, c V. 2012, ApJ, 761, 56
- Safranek-Shrader, C., Agarwal, M., Federrath, C., et al. 2012, MNRAS, 426, 1159
Safranek-Shrader, C., Milosavljevi´ c, M., & Bromm, V. 2014, MNRAS,
438, 1669 Salvadori, S., & Ferrara, A. 2009, MNRAS, 395, L6 Shapiro, P. R., & Kang, H. 1987, ApJ, 318, 32 Shen, S., Cooke, R., Ramirez-Ruiz, E., et al. 2014, arXiv:1407.3796 Shigeyama, T., & Tsujimoto, T. 1998, ApJ, 507, L135 Simon, J. D., Jacobson, H. R., Frebel, A., et al. 2015, ApJ, 802, 93 Smith, N., & Arnett, W. D. 2014, ApJ, 785, 82 Sneden, C., Cowan, J. J., & Gallino, R. 2008, ARA&A, 46, 241 Stacy, A., Greif, T. H., & Bromm, V. 2010, MNRAS, 403, 45 Stacy, A., Greif, T. H., & Bromm, V. 2012, MNRAS, 422, 290 Talbot, R. J., Jr., & Arnett, W. D. 1973, ApJ, 186, 51 Tassis, K., Gnedin, N. Y., & Kravtsov, A. V. 2012, ApJ, 745, 68 Ting, Y.-S., Freeman, K. C., Kobayashi, C., De Silva, G. M., & BlandHawthorn, J. 2012, MNRAS, 421, 1231 Truran, J. W., Cowan, J. J., Pilachowski, C. A., & Sneden, C. 2002, PASP, 114, 1293 Tsujimoto, T., & Shigeyama, T. 1998, ApJ, 508, L151 Tsujimoto, T., Shigeyama, T., & Yoshii, Y. 1999, ApJ, 519, L63 Turk, M. J., Smith, B. D., Oishi, J. S., et al. 2011, ApJS, 192, 9 Vargas, L. C., Geha, M., Kirby, E. N., & Simon, J. D. 2013, ApJ, 767, 134 van de Voort, F., Quataert, E., Hopkins, P. F., Keres, D., & FaucherGiguere, C.-A. 2014, arXiv:1407.7039 Webster, D., Sutherland, R., & Bland-Hawthorn, J. 2014, ApJ, 796, 11 Wise, J. H., & Abel, T. 2008, ApJ, 685, 40 Wise, J. H., Turk, M. J., Norman, M. L., & Abel, T. 2012, ApJ, 745, 50 Wise, J. H., Demchenko, V. G., Halicek, M. T., et al. 2014, MNRAS, 442, 2560 Wongwathanarat, A., Janka, H.-T., M¨ uller, E. 2013, A&A, 552, A126 Yong, D., Norris, J. E., Bessell, M. S., et al. 2013, ApJ, 762, 26
- Zucker, D. B., de Silva, G., Freeman, K., Bland-Hawthorn, J., & Hermes Team 2012, Galactic Archaeology: Near-Field Cosmology and the Formation of the Milky Way, 458, 421 | 10.1093/mnras/stv982 | [
"Jeremy S. Ritter",
"Alan Sluder",
"Chalence Safranek-Shrader",
"Milos Milosavljevic",
"Volker Bromm"
] | 2014-08-01T23:22:26+00:00 | 2015-05-01T15:06:35+00:00 | [
"astro-ph.GA",
"astro-ph.CO"
] | Metal Transport and Chemical Heterogeneity in Early Star Forming Systems | To constrain the properties of the first stars with the chemical abundance
patterns observed in metal-poor stars, one must identify any non-trivial
effects that the hydrodynamics of metal dispersal can imprint on the
abundances. We use realistic cosmological hydrodynamic simulations to quantify
the distribution of metals resulting from one Population III supernova and from
a small number of such supernovae exploding in close succession. Overall,
supernova ejecta are highly inhomogeneously dispersed throughout the
simulations. When the supernova bubbles collapse, quasi-virialized
metal-enriched clouds, fed by fallback from the bubbles and by streaming of
metal-free gas from the cosmic web, grow in the centers of the dark matter
halos. Partial turbulent homogenization on scales resolved in the simulation is
observed only in the densest clouds where the vortical time scales are short
enough to ensure true homogenization on subgrid scales. However, the abundances
in the clouds differ from the gross yields of the supernovae. Continuing the
simulations until the cloud have gone into gravitational collapse, we predict
that the abundances in second-generation stars will be deficient in the
innermost mass shells of the supernova (if only one has exploded) or in the
ejecta of the latest supernovae (when multiple have exploded). This indicates
that hydrodynamics gives rise to biases complicating the identification of
nucleosynthetic sources in the chemical abundance spaces of the surviving
stars. |
1408.0322v1 | ## MINIMAL ALMOST CONVEXITY
MURRAY ELDER AND SUSAN HERMILLER 1
Abstract. In this article we show that the Baumslag-Solitar group BS (1 , 2) is minimally almost convex, or MAC . We also show that BS (1 , 2) does not satisfy Po´ enaru's almost convexity condition P (2), and hence the condition P (2) is strictly stronger than MAC . Finally, we show that the groups BS (1 , q ) for q ≥ 7 and Stallings' nonFP 3 group do not satisfy MAC . As a consequence, the condition MAC is not a commensurability invariant.
## 1. Introduction
Let G be a group with finite generating set A , let Γ be the corresponding Cayley graph with the path metric d , and let S r ( ) and B r ( ) denote the sphere and ball, respectively, of radius r in Γ. The pair ( G,A ) satisfies the almost convexity condition AC f,r 0 for a function f : N → R + and a natural number r 0 ∈ N if for every natural number r ≥ r 0 and every pair of vertices a, b ∈ S r ( ) with d a, b ( ) ≤ 2, there is a path inside B r ( ) from a to b of length at most f ( r ). Note that every group satisfies the condition AC f, 1 for the function f ( r ) = 2 r . A group is minimally almost convex , or MAC (called K (2) in [10]), if the condition AC f,r 0 holds for the function f ( r ) = 2 r -1 and some number r 0 ; that is, the least restriction possible is imposed on the function f . If the next least minimal restriction is imposed, i.e. if G is AC f,r 0 with the function f ( r ) = 2 r -2, then the group is said to be M AC ′ ( K ′ (2) in [10]). Increasing the restriction on the function f further, the group satisfies Poenaru's P (2) condition [7, 13] if AC f,r 0 holds for a sublinear function f : N → R + , i.e. f satisfies the property that for every number C > 0 the limit lim r →∞ ( r -Cf r ( )) = ∞ . All of these definitions are generalizations of the original concept of almost convexity given by Cannon in [3], in which the greatest restriction is placed on the function f , namely that a group is almost convex or AC if there is a constant function f ( r ) ≡ C for which the group satisfies the condition AC C, 1 . Results of [3, 10, 14] show that the condition MAC , and hence each of the other almost convexity conditions, implies finite presentation of the group and solvability of the word problem.
Date : August 5, 2014.
2000
Mathematics Subject Classification.
20F65.
Key words and phrases. (Minimally) almost convex, Baumslag-Solitar group, Stallings group.
1 Supported under NSF grant no. DMS-0071037
The successive strengthenings of the restrictions in the definitions above give the implications AC ⇒ P (2) ⇒ M AC ′ ⇒ MAC . For this series of implications, a natural question to ask is, which of the implications can be reversed? A natural family of groups to consider are the Baumslag-Solitar groups BS (1 , q ) := 〈 a, t | tat -1 = a q 〉 with | q | > 1, which Miller and Shapiro [12] proved are not almost convex with respect to any generating set.
In the present paper, the structure of geodesics in the Cayley graph of BS (1 , q ) is analyzed in greater detail, in Section 2. In Section 3, we use these results to show that the group BS (1 2) satisfies the property , M AC ′ . In Section 4 we show that the group BS (1 2) , does not satisfy the P (2) condition, and hence the implication P (2) ⇒ M AC ′ cannot be reversed.
In section 4 we also show that the groups BS (1 , q ) = 〈 a, t | tat -1 = a q 〉 for q ≥ 7 are not MAC . Since the group BS (1 8) , is a finite index subgroup of BS (1 2), an immediate consequence of this result is that both , MAC and M AC ′ are not commensurability invariants, and hence not quasiisometry invariants. The related property AC is also known to vary under quasi-isometry; in particular, Thiel [16] has shown that AC depends on the generating set.
Finally, in Section 5 we consider Stallings' nonFP 3 group [15], which was shown by the first author [4, 5] not to be almost convex with respect to two different finite generating sets. In Theorem 5.3, we prove the stronger result that this group also is not MAC , with respect to one of the generating sets. Combining this with a result of Bridson [2] that this group has a quadratic isoperimetric function, we obtain an example of a group with quadratic isoperimetric function that is not MAC . During the writing of this paper, Belk and Bux [1] have shown another such example; namely, they have shown that Thomson's group F , which also has a quadratic isoperimetric function function [9], does not satisfy MAC .
## 2. Background on Baumslag-Solitar groups
Let G := BS (1 , q ) = 〈 a, t | tat -1 = a q 〉 with generators A := { a, a -1 , t, t -1 } for any natural number q > 1. Let Γ denote the corresponding Cayley graph with path metric d , and let C denote the corresponding Cayley 2-complex.
The complex C can be built from 'bricks' homeomorphic to [0 1] , × [0 , 1], with both vertical sides labeled by a ' ' upward, the top horizontal side t labeled by an ' a ' to the right, and the bottom horizontal side split into q edges each labeled by an ' a ' to the right. These bricks can be stacked horizontally into 'strips'. For each strip, q other strips can be attached at the top, and one on the bottom. For any set of successive choices upward, then, the strips of bricks can be stacked vertically to fill the plane. The Cayley complex then is homeomorphic to the Cartesian product of the real line with a regular tree T of valence q + 1; see Figure 1. Let π : C → T be the horizontal projection map. For an edge e of T , e inherits an upward
a
Figure 1. A brick in a plane, and a side-on view of the Cayley graph Γ for BS (1 4). ,
direction from the upward labels on the vertical edges of C that project onto e . More details can be found in [6] (pages 154-160).
For any word w ∈ A ∗ , let w denote the image of w in BS (1 , q ). For words v, w ∈ A ∗ , denote v = w if v and w are the same words in A ∗ , and v = G w if v = w . Let l ( w ) denote the word length of w and let w i ( ) denote the prefix of the word w containing i letters. Then ( w -1 ( )) i -1 is the suffix of w of length i . Define σ t ( v ) to be the exponent sum of all occurrences of t and t -1 in v . Note that relator tat -1 a -q in the presentation of G satisfies σ t ( tat -1 a -q ) = 0; hence whenever v = G w , then σ t ( v ) = σ t ( w ).
The following Lemma is well-known; a proof can be found in [11].
Lemma 2.1 (Commutation) . If v, w ∈ A ∗ and σ t ( v ) = 0 = σ t ( w ) , then vw = G wv .
Let E denote the set of words in { a, a -1 } ∗ , P the words in { a, a -1 , t } ∗ containing at least one t letter, and N the words in { a, a -1 , t -1 } ∗ containing at least one t -1 letter. A word w = w w 1 2 with w 1 ∈ N and w 2 ∈ P , then, will be referred to as a word in NP . Finally, let X denote the subset of the words in PN with t -exponent sum equal to 0. Letters in parentheses denote subwords that may or may not be present; for example, P X ( ) := P ∪ PX . The following statement is proved in [8].
Lemma 2.2 (Classes of geodesics) . A word w ∈ A ∗ that is a geodesic in Γ must fall into one of four classes:
- (1) E or X,
- (2) N or XN,
- (3) P or PX,
- (4) NP, or NPX with σ t ( w ) ≥ 0 , or XNP with σ t ( w ) ≤ 0 .
Analyzing the geodesics more carefully, we find a normal form for geodesics in the following Proposition.
Proposition 2.3 (Normal form) . If w ∈ A ∗ is a geodesic in G = BS (1 , q ) , then there is another geodesic ˆ w ∈ A ∗ with ˆ = w w such that for w in each class, ˆ w has the following form (respectively):
- (1) ˆ = w a i for | i | ≤ C q where C q := /floorleft q 2 + 1 /floorright if q > 2 and C 2 := 3 , or ˆ = w w 0 ∈ X ,
- (3) ˆ = w a n 0 t · · · a n f -1 tw 0 with | n j | ≤ /floorleft q 2 /floorright for all j , f ≥ 1 , and either w 0 = a i for | i | ≤ C q or w 0 ∈ X ,
- (2) ˆ = w w t 0 -1 a m 1 · · · t -1 a m e with | m j | ≤ /floorleft q 2 /floorright for all j , e ≥ 1 , and either w 0 = a i for | i | ≤ C q or w 0 ∈ X ,
- (4) Either ˆ w = t -e a m f ta m f -1 · · · a m 1 tw 0 with 1 ≤ e ≤ f , or ˆ w = w t 0 -1 a m 1 · · · t -1 a m e t f with 1 ≤ f ≤ e , such that | m j | ≤ /floorleft q 2 /floorright for all j , and either w 0 = a i for | i | ≤ C q or w 0 ∈ X . Note that if σ t ( w ) = 0 then e = f and either expression is valid.
In every class the word w 0 ∈ X can be chosen to be either of the form w 0 = t h a t s -1 a k h -1 · · · a k 1 t -1 a k 0 or w 0 = a k 0 ta k 1 · · · ta k h -1 ta t s -h with | k j | ≤ /floorleft q 2 /floorright for all j , 1 ≤ | s | ≤ q -1 if q > 2 , 2 ≤ | s | ≤ 3 if q = 2 , and h ≥ 1 .
Proof. Note that the natural number q = /floorleft q 2 +1 + /floorright /ceilingleft q 2 - /ceilingright 1 .
Next suppose that w is a geodesic in class (2). Then w ∈ ( X N ) , so w = w t ′ 0 -1 a l 1 t -1 a l 2 · · · t -1 a l e for some word w ′ 0 in class (1), e ≥ 1, and integers l i . Again we reformulate the defining relation of G , in this case to t -1 a ±/floorleft q 2 +1 /floorright = G a ± 1 t -1 a ∓ /ceilingleft ( q 2 - /ceilingright 1 ) . If q is odd, then we may (repeatedly) replace any occurrence of t -1 a ±/floorleft q 2 +1 /floorright by a ± 1 t -1 a ∓ /ceilingleft ( q 2 - /ceilingright 1 ) . If q is even, t -1 a ±/floorleft q 2 +1 /floorright is not geodesic, so | l j | ≤ /floorleft q 2 /floorright for all j and replacements are not needed. In both cases, then, we obtain a geodesic word of the form w t ′′ 0 -1 a m 1 t -1 a m 2 · · · t -1 a m e with each | m j | ≤ /floorleft q 2 /floorright and w ′′ 0 in class (1); to form the normal form ˆ, then, replace w w ′′ 0 by its normal form.
For a geodesic w in class (1), if w ∈ E , then w = a i for some i . If q = 2, then a ± 6 = ta ± 3 t -1 so | i | ≤ 6, and the words a ± (4+ k ) have normal form ta ± 2 t -1 a ± k ∈ X for k = 0 and k = 1. If q > 2, then the relation tat -1 = G a q can be reformulated as a ±/floorleft q 2 +1 +1 /floorright = G ta ± 1 t -1 a ∓ /ceilingleft ( q 2 - /ceilingright1 1) . If q is even, then a ± q 2 +2 is not geodesic, so | i | ≤ /floorleft q 2 + 1 . /floorright On the other hand, if q is odd, then a ± q +1 2 +2 = G ta ± 1 t -1 a ∓ ( q -1 2 -2) so a ± q +1 2 +2 is not geodesic; hence | i | ≤ /floorleft q 2 +1 +1, and the words /floorright a ±/floorleft q 2 +1 +1 /floorright have a normal form in X .
The proof of the normal form for geodesics in class (3) is very similar, using the relation a ±/floorleft q 2 +1 /floorright t = G a ∓ /ceilingleft ( q 2 - /ceilingright 1 ) ta ± 1 .
Suppose next that w is a geodesic in class (4) with σ t ( w ) ≥ 0. Then w = t -1 a k 1 · · · a k e -1 t -1 a l f ta l f -1 · · · a l 1 tw ′ 0 with w ′ 0 in class (1), 1 ≤ e < f , and each k , l j i ∈ Z . First use Lemma 2.1 to replace w by the geodesic word
$$t ^ { - e } a ^ { l _ { f } } t a ^ { k _ { e - 1 } + l _ { f - 1 } } \cdots t a ^ { k _ { 1 } + l _ { f - e + 1 } } t a ^ { l _ { f - e } } \tilde { w } _ { 0 } t a ^ { l _ { f - e - 1 } } \cdots a ^ { l _ { 1 } } t w _ { 0 } ^ { \prime }.$$
To complete construction of the normal form ˆ from this word, replace the w subword a l f t · · · a l 1 tw ′ 0 by its normal form from class (3).
The constructions for the normal forms of geodesics w in class (4) with σ t ( w ) ≤ 0, and of geodesics w 0 ∈ X , are analogous. /square
## 3. The group BS(1,2) satisfies M AC ′
Let G := BS (1 2) = , 〈 a, t | tat -1 = a 2 〉 with generators A := { a, a -1 , t, t -1 } . In this section we prove, in Theorem 3.5, that this group is M AC ′ . We begin with a further analysis of the geodesics in G , via several lemmas which are utilized in many of the cases in the proof of Theorem 3.5.
Lemma 3.1 (Large geodesic) . If w is a geodesic of length r > 200 in one of the classes (1), (2), or (3) of Proposition 2.3 and | σ t ( w ) | ≤ 2 , then w is in either X XN , , or PX , respectively. Moreover, the X subword of w must have the form w w 1 2 with w 1 ∈ P and w 2 ∈ N such that σ t ( w 1 ) = -σ t ( w 2 ) > 10 .
Proof. Suppose that w is a geodesic in either E N , , or P of length r > 200, and | σ t ( w ) | ≤ 2. Then w contains at most two occurrences of the letters t and t -1 . As mentioned in the proof of Proposition 2.3, a ± 6 = ta ± 3 t -1 so a j is not geodesic for | j | ≥ 6. Hence w contains at most 15 occurrences of the letters a and a -1 interspersed among the t ± 1 letters. Then l ( w ) ≤ 17, giving a contradiction. /square
Given a word w 0 ∈ X , there is a natural number k ∈ N with w 0 = G a k ; denote ˜ w 0 := a k . If w is a geodesic word in E ∪ N ∪ P ∪ NP , then let ˜ := w w . Combining these, for any geodesic word w = w w 0 1 (or w = w w 1 0 ) with w 0 ∈ X and w 1 ∈ N ∪ P ∪ NP , define ˜ := w ˜ w w 0 1 = a w k 1 (or ˜ := w w w 1 ˜ 0 = w a 1 k , respectively). Then ˜ w ∈ N ∪ P ∪ NP , and the subword w 1 is geodesic.
Lemma 3.2. If w is a word in NP , NPX or XNP , and ˜ w contains a subword of the form t -1 a 2 i t with i ∈ Z , then w is not geodesic.
Proof. The word w can be written as w = w w w 0 1 2 with w 1 ∈ NP and each of w 0 and w 2 either in X or E . Since ˜ = w ˜ w w w 0 1 ˜ 2 contains the subword t -1 a 2 i t ∈ NP , the word t -1 a 2 i t must be a subword of w 1 , and hence also of w . Since t -1 a 2 i t = G a i , this subword is not geodesic, and hence w also is not geodesic. /square
Lemma 3.3. If w is any word in NP or NPN and w = G 1 , then w must contain a subword of the form t -1 a 2 i t for some i ∈ Z .
Proof. Since G = BS (1 2) is , an HNN extension, Britton's Lemma states that if w ∈ NP N ( ) and w = G 1, then w must contain a subword of the form ta t i -1 or t -1 a 2 i t for some i ∈ Z . If w ∈ NP then w must contain the second type of subword.
If w ∈ NPN , then w = w w w 1 2 3 with w , w 1 3 ∈ N and w 2 ∈ P . Since σ t ( w 1 ) < 0 and 0 = σ t (1) = σ t ( w 1 ) + σ t ( w 2 ) + σ t ( w 3 ), σ t ( w 2 ) > σ t ( w 3 ) and the word w w 2 3 ∈ PX . Then w w 2 3 = w w 4 5 with w 4 ∈ P and w 5 ∈ X , and
the word w w w 1 4 ˜ 5 ∈ NP with w w w 1 4 ˜ 5 = G w = G 1. Then Britton's Lemma applies again to show that the prefix w w 1 4 of w must contain a subword of the form t -1 a 2 i t for i ∈ Z . /square
Lemma 3.4. If w and u are geodesics, w ∈ NP ∪ XNP ∪ NPX σ w , t ( ) ≤ σ t ( u ) , and 1 ≤ d w, u ( ) ≤ 2 , then u ∈ NP ∪ XNP ∪ NPX and for some w , u 1 1 ∈ N and w , u 2 2 ∈ P with σ t ( w 1 ) = σ t ( u 1 ) , ˜ = w w w 1 2 and ˜ = u u u 1 2 .
Proof. The definition of ˜ shows that we can write w ˜ = w w w 1 2 with w 1 ∈ N and w 2 ∈ P . Let γ label a path of length 1 or 2 from w to u ; since σ t ( w ) ≤ σ t ( u ), then γ ∈ E ∪ P . Proposition 2.3 says that ˜ u ∈ E ∪ P ∪ N ∪ NP . Since w is a geodesic, Lemma 3.2 implies that ˜ cannot contain a subword w of the form t -1 a 2 i t for any integer i . Then Lemma 3.3 says that the word ˜ wγu ˜ -1 , which represents the trivial element 1 in G , cannot be in NP N ( ). Therefore ˜ u ∈ E ∪ P ∪ N , so ˜ u ∈ NP . Hence u ∈ NP ∪ XNP ∪ NPX .
/negationslash
-1
We can now write ˜ = u u u 1 2 with u 1 ∈ N and u 2 ∈ P . The word ˜ u ˜ wγ = u -1 2 u -1 1 w w γ 1 2 is another representative of 1. Repeatedly reducing subwords ta t j -1 to t 2 j in the subword u -1 1 w 1 ∈ PN results in a word ˜ u -1 1 w 1 ∈ E ∪ P ∪ N . Then 1 = G u -1 2 ˜ u -1 1 w w γ 1 2 ∈ NP N ( ), so this word must contain a subword of the form t -1 a 2 i t for some integer i . Since w and u are geodesics, w w 1 2 and u -1 2 u -1 1 cannot contain such a subword. Therefore we must have ˜ u -1 1 w 1 ∈ E . Hence σ t ( w 1 ) = σ t ( u 1 ). /square
We split the proof of Theorem 3.5 into 10 cases, depending on the classes from Proposition 2.3 to which the two geodesics w and u belong. In overview, we begin by showing that the first three cases cannot occur; that is, for a pair of length r geodesics w and u in the respective classes in these three cases, it is not possible for d w, u ( ) to be less than three. In cases 4-6, we show that a path δ can be found that travels from w along the path w -1 to within a distance 2 of the identity vertex, and, after possibly traversing an intermediate edge, δ then travels along a suffix of u to u . In case 7 we show that the path δ can be chosen to have length at most six, traveling around at most two bricks in the Cayley complex. In case 8 there are subcases in which each of the two descriptions above occur, as well as a subcase in which the path δ initially follows the inverse of a suffix of w from w , then travels along a path that 'fellow-travels' this initial subpath, and then repeats this procedure by traversing a fellow-traveler of a suffix of u , and then traveling along the suffix itself to u . In cases 9 and 10, the paths δ constructed in each of the subcases follow one of these three patterns.
Theorem 3.5. The group G = BS (1 2) = , 〈 a, t | tat -1 = a 2 〉 is M AC ′ with respect to the generating set A = { a, a -1 , t, t -1 } . In particular, if w and u are geodesics of length r > 200 with 1 ≤ d w, u ( ) ≤ 2 , then there is a path δ inside B r ( ) from w to u of length at most 2 r -2 .
Proof. Suppose that w and u are geodesics of length r > 200 with 1 ≤ d w, u ( ) ≤ 2. Using Proposition 2.3, by replacing w and u by ˆ w and ˆ u
respectively, we may assume that each of w and u are in one of the normal forms listed in that Proposition. Using Lemma 3.1, we may assume that neither w nor u is in E .
Let γ be the word labeling a geodesic path of length at most 2 from w to u , so that wγu -1 = G 1. Since d w, u ( ) ≥ 1,
$$\gamma \in \{ a ^ { \pm 1 }, t ^ { \pm 1 }, a ^ { \pm 2 }, a t ^ { \pm 1 }, a ^ { - 1 } t ^ { \pm 1 }, t a ^ { \pm 1 }, t ^ { - 1 } a ^ { \pm 1 }, t ^ { \pm 2 } \}.$$
Then γ is in one of the sets E P , , or N .
We divide the argument into ten cases, depending on the class of the normal forms w and u from Proposition 2.3, which we summarize in the following table.
| Case | Class of w | Class of u | Case | Class of w | Class of u |
|---------|--------------|--------------|----------|--------------|--------------|
| Case 1: | (4) | (1) | Case 6: | (3) | (3) |
| Case 2: | (4) | (3) | Case 7: | (2) | (2) |
| Case 3: | (2) | (3) | Case 8: | (1) | (3) |
| Case 4: | (1) | (1) | Case 9: | (2) | (4) |
| Case 5: | (1) | (2) | Case 10: | (4) | (4) |
This table represents a complete list of the cases to be checked. In particular, if w is in class (2) and u in class (1), then the inverse of the path in Case 5 will provide the necessary path δ , and similarly for the remainder of the cases.
Case 1: If w is in class (4) and u is in class (1) : Then w is in either NP , NPX or XNP , and u ∈ X . Since ˜ w ∈ NP u , ˜ ∈ E , and the path γ is either in E N , or P , then 1 = G wγu -1 = G ˜ wγu ˜ -1 ∈ NP N ( ) (that is, replacing the X subwords of w and u by powers of a ). By Lemma 3.3, ˜ wγu ˜ -1 contains a subword of the form t -1 a 2 s t ∈ NP , which therefore must occur within ˜. w Then Lemma 3.2 says that w is not a geodesic, which is a contradiction. Hence Case 1 cannot hold.
Case 2: If w is in class (4) and u is in class (3) : Then w is in either NP , XNP or NPX , and u ∈ P X ( ). In this case 1 = G ˜ wγu ˜ -1 ∈ NPN , and the same proof as in Case 1 shows that Case 2 cannot occur.
Case 3: If w is in class (2) and u is in class (3) : Then w ∈ ( X N ) and u ∈ P X ( ). Since σ t ( w ) < 0 and σ t ( u ) > 0 then we must have σ t ( w ) = -1, σ t ( u ) = 1, and γ = t 2 . Lemma 3.1 says that w ∈ XN , and u ∈ PX . Since w is in normal form, w = ˆ = w w t 0 -1 a i with | i | ≤ 1 and w 0 ∈ X , and similarly u = a tu j 0 with | j | ≤ 1 and u 0 ∈ X . Then 1 = G ˜ wγu ˜ -1 = G ˜ w t 0 -1 a t i 2 ˜ u -1 0 t -1 a -j ∈ NPN . Lemma 3.3 then says that w t 0 -1 a t i 2 ˜ u -1 0 t -1 a -j contains a subword of the form t -1 a 2 s t for some s ∈ Z , so i must be a multiple of 2, and hence i = 0. Using the last part of Proposition 2.3, we can further write the normal form for w 0 ∈ X as w 0 = w t 1 -1 ,
Figure 2. Case 6: w = a tw w i 1 0 , u = a -i tu u 1 0 .
so w = w t i -2 . Then u = G wγ = G w r ( -2), contradicting the hypothesis that u is a geodesic word of length r . Thus Case 3 does not hold.
Case 4: If both w and u are in class (1) : Then w and u are both in X . From Proposition 2.3 the normal forms w = ˆ and w u = ˆ can be chosen of the u form ˆ = w t h w 1 and ˆ = u t u i 1 with h, i > 0 and w , u 1 1 ∈ N . Then w and u have a common prefix t = w (1) = u (1), and the path δ := w -1 1 t -( h -1) t i -1 u 1 from w through w (1) to u has length 2 r -2 and stays inside B r ( ).
Case 5: If w is in class (1) and u is in class (2) : Then w ∈ X and u ∈ ( X N ) . In this case σ t ( w ) = 0, σ t ( γ ) = σ t ( w -1 u ) = σ t ( w ) + σ t ( u ) = σ t ( u ), and σ t ( u ) < 0, so σ t ( u ) is either -1 or -2. The hypothesis that r > 200 and Lemma 3.1 imply that u ∈ XN . Then both of the normal forms ˆ and ˆ w u can be chosen to begin with t , and the same proof as in Case 4 gives the path δ .
Case 6: If both w and u are in class (3) : In this case both w and u are in P X ( ). Without loss of generality assume that σ t ( w ) ≤ σ t ( u ), so σ t ( γ ) ≥ 0 and γ ∈ E ∪ P . Since both w and u are in normal form, w = a tw w i 1 0 and u = a tu u j 1 0 with w , u 1 1 ∈ P ∪ E ; w , u 0 0 ∈ X ∪ E ; and | i , | | j | ≤ 1. Then 1 = G u -1 wγ = G ˜ u -1 0 u -1 1 t -1 a i -j tw w γ 1 ˜ 0 ∈ NP . By Lemma 3.3, u -1 0 u -1 1 t -1 a i -j tw w γ 1 ˜ 0 has a subword of the form t -1 a 2 s t , so i -j is a multiple of 2 and hence either i = j with 0 ≤ | | ≤ i 1 or i = -j with | i | = 1.
If i = j then w and u have a common prefix a t i = w (1 + | i | ) = u (1 + | i | ). The path δ := w -1 0 w -1 1 u u 1 0 from w follows the suffix w w 1 0 of w backward to w (1 + | i | ) and then follows the suffix u u 1 0 of u to u , remaining in B r ( ). If i = -j with | i | = 1, define the path
δ := w -1 0 w -1 1 a -i u u 1 0 = G w -1 0 w -1 1 t -1 a -i a -i tu u 1 0 = w -1 u = G γ.
Then δ labels a path of length 2 r -3, traveling along w -1 from w to wδ r ( -2) = w (2), then along a single edge to wδ r ( -1) = u (2), and finally along a suffix of u to u , thus remaining in B r ( ). (See Figure 2.)
Figure 3. Case 7.3: w = w w t 0 1 -1 , u = u u t 0 1 -1 a j .
Case 7: If both w and u are in class (2) : In this case, both w and u are in ( X N ) . We can assume without loss of generality that σ t ( w ) ≤ σ t ( u ), so again σ t ( γ ) ≥ 0 and γ ∈ E ∪ P .
From Proposition 2.3 we have w = w w t 0 1 -1 a i and u = u u t 0 1 -1 a j with w , u 0 0 ∈ X ∪ E ; w , u 1 1 ∈ N ∪ E ; and | i , | | j | ≤ 1. Thus 1 = G ˜ wγu ˜ -1 = ˜ w w t 0 1 -1 a γa i -j tu -1 1 ˜ u -1 0 ∈ NP . By Lemma 3.3 the latter contains a subword of the form t -1 a 2 s t , and so t -1 a γa i -j t must contain this subword.
Since γ ∈ E ∪ P , then γ ∈ { t, t 2 , ta ± 1 , a ± 1 t, a ± 1 , a ± 2 } , and we may divide the argument into four subcases.
Case 7.1: If γ ∈ { t, a ± 1 t } : Then t -1 a 2 s t must be a subword of t -1 a γ i . If γ = t , then since | i | ≤ 1 we have i = 0 and w = w w t 0 1 -1 , so u = G wγ = G w r ( -1). If γ = a ± 1 t , then | i | = 1, and γ = a ± i t . If γ = a t i , then u = G wγ = w r ( -2) t -1 a a t i i = G w r ( -2) a i . Finally, if γ = a -i t , then u = G wγ = w r ( -2) t -1 a a i -i t = G w r ( -2). All three of these options result in a contradiction of the fact that u is a geodesic of length r , so subcase 7.1 can't occur.
Case 7.2: If γ ∈ { t 2 , ta ± 1 } : In this subcase, t -1 a 2 s t must be a subword of t -1 a t i again, so i = 0 and w = w w t 0 1 -1 . Note that γ (1) = t and wγ (1) = G w r ( -1). Then γ is a path of length 2 inside B r ( ) from w to u . In this subcase, we may define the path δ := γ .
Case 7.3: If γ ∈ { a ± 1 } : Write γ = a k with | k | = 1. Recall that 0 ≤ | | ≤ i 1. If i = 0, then t -1 a 2 s t must be a subword of t -1 a a k -j t , so 2 ( | k -j ) and | j | = 1. Then γ = a ± j . If γ = a j , then w = G uγ -1 = u r ( -2) t -1 a a j -j = G u r ( -2) t -1 , contradicting the length r of the geodesic w . Thus γ = a -j . The word δ := ta -j t -1 a j = G a -j labels a path from w to u of length 4. Since wδ (1) = G w r ( -1) and wδ (2) = G u r ( -2), the path δ stays inside B r ( ), and hence satisfies the required properties. (See Figure 3.)
If | i | = 1, then we can write γ = a ± i . If γ = a -i , then u = G wγ = G w r ( -1), again giving a contradiction; hence γ = a i . Note that the word
t -1 a 2 s t must be a subword of t -1 a 2 i a -j t , so 2 (2 | i -j ) and j = 0. Defining δ := a -i ta t i -1 = G a i , then δ labels a path of length 4 from w to u , with wδ (2) = G w r ( -2) and wδ (3) = G u r ( -1), so the path remains in B r ( ) as required.
Case 7.4: If γ ∈ { a ± 2 } : Write γ = a 2 k with | k | = 1. As in the previous subcase, we consider the options i = 0 and | i | = 1 in separate paragraphs.
If i = 0, then t -1 a 2 s t must be a subword of t -1 a 2 k a -j t , so j = 0. Then the length 3 path labeled by δ := ta k t -1 from w to u traverses the vertices represented by wδ (1) = G w r ( -1) and wδ (2) = G u r ( -1), hence remaining in B r ( ).
If | i | = 1, then γ = a ± 2 i . If γ = a -2 i , then wγ (1) = G w r ( -1), so we may define δ := γ .
If | i | = 1 and γ = a 2 i , then t -1 a 2 s t must be a subword of t -1 a 3 i a -j t . Thus | j | = 1, so j = ± i . If j = , then i wγ (1) = G u r ( -1), so again the path δ := γ has the required properties. If j = -i , then the path of length 6 labeled by δ := a -i ta 2 i t -1 a -i = G a 2 i starting at w ends at u . Since wδ (2) = G w r ( -2) and wδ (4) = G u r ( -2), this path also remains within B r ( ).
Case 8: If w is in class (1) and u is in class (3) : Then w ∈ X and u ∈ P X ( ). In this case, σ t ( w ) = 0, σ t ( u ) > 0, and σ t ( u ) = σ t ( w ) + σ t ( γ ) = σ t ( γ ), so 0 < σ t ( u ) = σ t ( γ ) ≤ 2. Thus γ ∈ P , so γ ∈ { t, ta ± 1 , t 2 , a ± 1 t } .
/negationslash
Suppose γ ∈ { t, ta ± 1 , t 2 } . By Proposition 2.3 and Lemma 3.1, the normal form w can be chosen in the form w = w t 1 -h with w 1 ∈ P and h > 10. Then the length r geodesic u cannot represent wt = G w r ( -1) or wt 2 = g w r ( -2), so γ = t and γ = t 2 . For γ = ta ± 1 , since wγ (1) = G w r ( -1), we may define δ := γ .
/negationslash
Suppose for the rest of Case 8 that γ = a ± 1 t and write γ = a m t with | m | = 1. Proposition 2.3 says that the normal form w can also be chosen in the form w = tw t 0 -1 a i with w 0 ∈ X and 0 ≤ | | ≤ i 1. If i = m , then u = G wγ = tw t 0 -1 a m m a t = G w r ( -2) a m , and if i = -m , then u = G tw t 0 -1 a -m m a t = w r ( -2), both contradicting the geodesic length r of u . Then i = 0 and w = tw t 0 -1 with w 0 in X . We also have σ t ( u ) = σ t ( γ ) = 1, and Lemma 3.1 implies that u ∈ PX , so the normal form u = a tu j 0 with u 0 in X and | j | ≤ 1.
If j = 0 then w and u have a common t prefix, and the path δ := tw -1 0 u 0 has the required properties.
Suppose for the remainder of Case 8 that | j | = 1. Then either γ = a t j or γ = a -j t ; we consider these two subcases separately.
Case 8.1: If γ = a t j : Applying Lemma 2.1 to commute the subwords in parentheses with zero t -exponent-sum,
$$1 = _ { G } w \gamma u ^ { - 1 } = t w _ { 0 } t ^ { - 1 } ( a ^ { j } ) ( t u _ { 0 } ^ { - 1 } t ^ { - 1 } ) a ^ { - j } = _ { G } t w _ { 0 } u _ { 0 } ^ { - 1 } t ^ { - 1 },$$
which yields w 0 = G u 0 . By Proposition 2.3 we can replace each subword with a normal form w 0 = u 0 = vta t k -p such that 2 ≤ | k | ≤ 3 and v ∈ P
Figure 4. Case 8.1: w = tvta k t - -p 1 , u = a tvta j k t -p , γ = a t j .
with σ t ( v ) = p -1. Since r > 200, Lemma 3.1 implies that p > 10. Let s := sign( k ). Then w = tvta | k s | t -( p +1) and u = a tvta j | k s | t -p .
Consider the path δ := t p a -2 s t -p a t a j p 2 s t 1 -p starting at w . Using Lemma 2.1,
$$\delta = ( t ^ { p } a ^ { - 2 s } t ^ { - p } ) ( a ^ { j } ) t ^ { p } a ^ { 2 s } t ^ { - ( p - 1 ) } = _ { G } ( a ^ { j } ) ( t ^ { p } a ^ { - 2 s } t ^ { - p } ) t ^ { p } a ^ { 2 s } t ^ { - ( p - 1 ) } = _ { G } a ^ { j } t = \gamma,$$
so δ labels a path from w to u . This path δ both follows and 'fellow travels' suffixes of w and u ; see Figure 4 for a view of this path, shown in shading, when k and j have the same sign.
In order to check that δ remains in the ball B r ( ), we analyze the distances from 1 of several vertices along the path δ , and together with the lengths of the subpaths between the vertices. The prefix t p of δ is the inverse of a suffix of w , so starting from w the path δ follows the path w backward. Then d (1 , wδ ( )) = i r -i for 0 ≤ i ≤ p and wδ p ( ) = G w r ( -p ). The point wδ p ( +1) must then also lie in the ball B r ( -( p -1)). Since wδ p ( +2) = wt a p -2 s = G w r ( -p a ) -2 s = G w r ( -p ta ) -s t -1 = G w r ( -( p +2)) t -1 , the point C := wδ p ( +2) must lie in the ball B r ( -( p +1)). Then the initial segment of δ of length p +2 from w to C lies inside B r ( ).
Similarly, the suffix t -( p -1) of δ is also a suffix of u , so d (1 , wδ (3 p +5+ )) = i r -( p -1) + i for 0 ≤ i ≤ p -1 and wδ (3 p +5) = G u r ( -( p -1)). The point wδ (3 p +4) ∈ B r ( -( p -2)). Since wδ (3 p +3) = G wδ (3 p +5) a -2 s = G u r ( -( p -1)) ta -s t -1 = G a tvta j | k s | t -1 ta -s t -1 = G u r ( -( p +1)) t -1 , the point D := wδ (3 p +3) must lie in the ball B r ( -p ). So the final segment of δ of length p +1 from D to u also lies in B r ( ).
Finally, the central section labeled t -p a t j p of the path δ from C ∈ B r ( -( p +1)) to D ∈ B r ( -p ) has length 2 p +1, and hence never leaves the ball B r ( ). The entire path δ has length 4 p +4, whereas r = ( ) + l v | k | + p +3 ≥ ( p -1) + 2 + p + 3 = 2 p + 4, so l ( δ ) ≤ 2 r -4. Thus the path δ has the required properties in this subcase.
Case 8.2: If γ = a -j t : Applying Lemma 2.1 again yields 1 = G wγu -1 = tw t 0 -1 ( a -j )( tu -1 0 t -1 ) a -j = G tw t 0 -1 tu -1 0 t -1 a -j a -j = G tw u 0 -1 0 a -j t -1 , implying that u 0 = G a -j w 0 . Plugging this into the expression for u gives u = a tu j 0 = G a ta j -j w 0 = G a -j tw 0 . Note that r = ( l w ) = ( l w 0 ) + 2, so a -j tw 0 is another geodesic from 1 to u . Replacing u with a -j tw 0 , we can now find the path δ using subcase 8.1.
/negationslash
Case 9: If w is in class (2) and u is in class (4) : Then w ∈ ( X N ) and u is either in NP , XNP , or NPX . Since w ∈ ( X NP ) ∪ NPX , Lemma 3.4 says that σ t ( w < σ ) t ( u ). Therefore σ t ( γ ) > 0, so γ ∈ { t, t 2 , a ± 1 t, ta ± 1 } . We divide this case into two subcases, depending on the t -exponent sum of u .
Case 9.1: If σ t ( u ) ≤ 0 : By Proposition 2.3, the geodesic normal form u = u u t 0 1 -1 a m e t f with u 0 ∈ X ∪ E u , 1 ∈ N , | m e | = 1, and 1 ≤ f ≤ e = | σ t ( u 1 ) | +1.
/negationslash
Case 9.1.1: If γ ∈ { t, t 2 , a ± 1 t } : Then γ and u share a suffix t , so u r ( -1) = uγ -1 (1). The geodesic w = u r ( -1), so γ = . For t γ ∈ { t 2 , a ± 1 t } , the path δ := γ has the required properties.
/negationslash
Case 9.1.2: If γ ∈ { ta ± 1 } : Write γ = ta k with | k | = 1. Also write w = w w 0 1 with w 0 ∈ X ∪ E and w 1 ∈ N . Then
= G ˜ u u t 0 1 -1 a m e t f -1 a -2 k w -1 1 ˜ w -1 0 ∈ NP .
1 = G ˜ uγ -1 ˜ w -1 = ˜ u u t 0 1 -1 a m e t f a -k t -1 w -1 1 ˜ w -1 0
/negationslash
Lemma 3.3 implies that the latter word must contain a non-geodesic t -1 a 2 s t subword. Then the first occurrence of a t must be in w -1 1 , so f = 1. Write u = u u t 0 1 -1 a m e t = u u t 0 ′ 1 with u ′ 1 := u t 1 -1 a m e ∈ N . Note that w = G uγ -1 = G u u ta 0 ′ 1 -k t -1 = G u r ( -1) a -2 k . Let v := u u a 0 ′ 1 -k = u r ( -1) a -k = G wa k . The vertex v ∈ B r ( ). If v ∈ B r ( -1), then the path δ := a 2 k t from w to u satisfies wδ (1) = G v and wδ (2) = G u r ( -1), so δ is a path of length 3 inside B r ( ) from w to u . On the other hand, if v ∈ B r ( -1), then v is a length r geodesic in ( X N ) , and w = G va -k so d v, u ( ) = 1. Applying case 7.3 to the geodesics v and w in class (2), there is a path δ ′ of length 4 inside B r ( ) from w to v . Let δ := δ a ′ k t . Then δ is a path of length 6 from w to u inside B r ( ).
/negationslash
Case 9.2: If σ t ( u ) > 0 : Since σ t ( w ) < 0 and σ t ( w ) + σ t ( γ ) = σ t ( u ) then we must have σ t ( w ) = -1, σ t ( u ) = 1, and γ = t 2 . By Proposition 2.3 and Lemma 3.1, w = w t 0 -1 a i with w 0 ∈ X , | i | ≤ 1, w 0 = vta k t - -p 1 with v ∈ P , σ t ( v ) = p > 9, and 2 ≤ | k | ≤ 3. Since u = G wγ = vta k t -p t -2 a t i 2 and u has geodesic length r = ( l w ) = l ( v ) + | k | + | i | + p +3, then i = 0. Using Lemma 2.1, u = G ( t -1 a t i )( vta k t - -p 1 ) . t The word x := t -1 a tvta i k t -p is another geodesic labeling a path from the identity to u .
Let s :=sign( k ), and define the path
$$\delta \coloneqq a ^ { - i } t ^ { p + 1 } a ^ { - 2 s } t ^ { - ( p - 1 ) } t ^ { - 2 } a ^ { i } t ^ { 2 } t ^ { p - 1 } a ^ { 2 s } t ^ { - ( p - 1 ) }$$
starting at w . Using Lemma 2.1, so δ labels a path from w to u = x . The length l ( δ ) = 4 p +8, and the length r l w l v k p p l δ r
δ = G a -i t p +1 a -2 s t -( p -1) ( t p -1 a 2 s t -( p -1))( t -2 a t i 2 ) = G t 2 = γ, = ( ) = ( ) + | | + +4 ≥ 2 +6, so ( ) ≤ 2 -4.
The proof that δ remains in B r ( ) is similar to case 8.1. In particular, note that wδ p ( +2) = w r ( -( p +2)) ∈ B r ( -( p +2)). Since wδ p ( +4) = ( vta k t -p t -2 a i )( a -i t p +1 a -2 s ) = G vta k -s t -1 = G w r ( -( p +4)) t -1 , the point wδ p ( +4) ∈ B r ( -( p +3)). Also wδ (3 p +9) = x r ( -( p -1)) ∈ B r ( -( p -1)). Finally, since wδ (3 p + 7) = G x r ( -( p -1)) a -2 s = G t -1 a tvta i k t -1 a -2 s = G t -1 a tvta i k -s t -1 = G x r ( -( p +1)) t -1 , then the point wδ (3 p +7) ∈ B r ( -p ). Then the five successive intermediate subpaths of δ between w , these four points, and u are too short to allow δ to leave B r ( ).
Case 10: If both w and u are in class (4) : In this case both w and u are in ( X NP ) ∪ NPX . We may assume without loss of generality that σ t ( w ) ≤ σ t ( u ). It follows that σ t ( γ ) ≥ 0, so γ ∈ { a ± 1 , a ± 2 , a ± 1 t, ta ± 1 , t, t 2 } .
We divide this case into three subcases, depending on the t -exponents of w and u .
Case 10.1: If σ t ( w ) ≥ 0 and σ t ( u ) ≥ 0 : In this case Proposition 2.3 says that we have geodesic normal forms u, w ∈ NP X ( ), and moreover w = t -p 1 w ′ and u = t -p 2 u ′ with p 1 > 0, p 2 > 0, and w , u ′ ′ ∈ P X ( ). Thus w (1) = t -1 = u (1), and we may define δ := w ′-1 t p 1 -1 t -( p 2 -1) u ′ .
Case 10.2: If σ t ( w ) < 0 and σ t ( u ) ≤ 0 : In this subcase, we have normal forms w,u ∈ ( X NP ) , and we can write
$$w = w _ { 0 } w _ { 1 } t ^ { - 1 } a ^ { i _ { 1 } } t ^ { f _ { 1 } } \quad \text{ and } \quad u = u _ { 0 } u _ { 1 } t ^ { - 1 } a ^ { i _ { 2 } } t ^ { f _ { 2 } }$$
with w , u 0 0 ∈ X ∪ E w , 1 ∈ N u , 1 ∈ N ∪ E f , 1 ≥ 1, f 2 ≥ 1, σ t ( w 1 ) ≤ -f 1 , σ t ( u 1 ) ≤ -( f 2 -1), and | i 1 | = | i 2 | = 1. Lemma 3.4 implies that σ t ( w 1 ) = σ t ( u 1 ). Then σ t ( wγ ) = σ t ( w 1 ) -1 + f 1 + σ t ( γ ) = σ t ( u ) = σ t ( u 1 ) -1 + f 2 , so f 2 = f 1 + σ t ( γ ) ≥ f 1 .
Case 10.2.1: If γ ∈ { t, t 2 , a ± 1 t } : Since the last letter of u is t , the proof of Case 9.1.1 shows that γ = , and for t γ ∈ { t 2 , a ± 1 t } , we may define δ := γ .
/negationslash
Case 10.2.2: If γ ∈ { a ± 1 } : Write γ = a k with | k | =1. The word δ := t -1 a 2 k t labels a path of length 4 from w to u . Since both words w and u end with a t , then wδ (1) = G w r ( 1 ) and wδ (3) = G u r ( -1), hence δ lies in B r ( ).
Case 10.2.3: If γ ∈ { ta ± 1 } : Write γ = ta k with | k | = 1. In this subcase, f 2 = f 1 + σ t ( γ ) = f 1 +1 ≥ 2. Then the word w ends with a t and u ends with t 2 . Now u = u r ( -1) t = G wγ = wta k = G wa 2 k t . Let v := u r ( -1) a -k = G wa k . Then v ∈ ( X NP ) and v ∈ B r ( ). The remainder of the proof in this subcase is similar to Case 9.1.2. If v ∈ B r ( -1), then δ := a 2 k t has the required properties. If v ∈ B r ( -1), then Case 10.2.2 provides a path
/negationslash
δ ′ = t -1 a 2 k t inside B r ( ) from w to v , and the path δ := δ a ′ k t = t -1 a 2 k ta k t from w to u satisfies the required conditions.
Case 10.2.4: If γ ∈ { a ± 2 } : Write γ = a 2 k with | k | = 1. In this case we have σ t ( u ) = σ t ( w < ) 0 and f 2 = f 1 + σ t ( γ ) = f 1 .
/negationslash
/negationslash
/negationslash
Define δ := t -f 1 ( a -i 1 )( t f 1 a 2 k t -f 1 ) a i 1 t f 1 . Using Lemma 2.1 to commute the subwords in parentheses, and freely reducing the resulting word, shows that δ = G a 2 k = γ , so δ labels a path from w to u . The length l ( γ ) = 4 f 1 + 4 = 2(2 f 1 +3) -2 ≤ 2 r -2. The prefix δ f ( 1 +2) = w -1 ( f 1 +2) is the inverse of a suffix of w , so wδ f ( 1 +2) = w r ( -( f 1 +2)) ∈ B r ( -( f 1 + 2)). The word t f i is a suffix of both δ and u , so wδ (3 f 1 +4) = u r ( -f 1 ) ∈ B r ( -f 1 ). The three subpaths of δ between w , the two points above, and u are again too short to allow δ to leave B r ( ).
The radius r = l ( w ) = l ( w 0 ) + l ( w 1 ) + 2 + f 1 ≥ σ t ( w 1 ) + 2 + f 1 ≥ 2 f 1 + 2. If r = 2 f 1 + 2, then w = t -f 1 -1 a i 1 t f 1 and u = t -f 1 -1 a i 2 t f 1 . Since w = u , then i 1 = i 2 , so i 2 = -i 1 . Since r > 200, then f 1 > 1. Now u -1 wγ = G ( t -f 1 -1 a i 2 t f 1 ) -1 ( t -f 1 -1 a i 1 t f 1 ) a 2 k = G t -( f 1 -1) a i 1 t f 1 -1 ) a 2 k ; according to Britton's Lemma, this last expression cannot equal the trivial element 1 in G . Thus the radius r = 2 f 1 +2, so r ≥ 2 f 1 +3.
Case 10.3: If σ t ( w ) < 0 and σ t ( u ) > 0 : Then γ = t 2 , σ t ( w ) = -1, and σ t ( u ) = 1. From Proposition 2.3, the normal form
$$w = w _ { 0 } t ^ { - 1 } a ^ { m _ { 1 } } t ^ { - 1 } \cdots t ^ { - 1 } a ^ { m _ { p } } t ^ { p - 1 }$$
with either w 0 ∈ X or w 0 = a k ∈ E for some | k | ≤ 3, p ≥ 2, and | m p | = 1. Using Lemma 2.1, then the word ˇ := w w t 0 -p a m p ta m p -1 · · · ta m 1 is another geodesic representative of w . The normal form for u ∈ NP X ( ) has the form u = t -e a tu u j 1 0 with e ≥ 1, | j | = 1, u 1 ∈ P with σ t ( u 1 ) = p , and u 0 ∈ X ∪ E . Lemma 3.4 shows that p = . Replacing e w by the alternate normal form ˇ, w then we can write
$$w = w _ { 0 } t ^ { - p } a ^ { i } t w _ { 2 } \quad \text{ and } \quad u = t ^ { - p } a ^ { j } t u _ { 1 } u _ { 0 }$$
such that either w 0 ∈ X or w 0 = a k for | k | ≤ 3, p ≥ 2, | i | = 1, w 2 ∈ P ∪ E with σ t ( w 2 ) = p -2, | j | = 1, u 1 ∈ P with σ t ( u 1 ) = p , and u 0 ∈ X ∪ E .
We will divide case 10.3 into further subcases, depending on the form of w 0 and the length of w 2 .
Case 10.3.1: If w 0 ∈ X : In this case Proposition 2.3 says that we can write w 0 = w ta t 3 k -l with l ≥ 1, w 3 ∈ P ∪ E σ , t ( w 3 ) = l -1, and 2 ≤ | m | ≤ 3. Let s = ± 1 be the sign of m , so that m = | ms | . Then
$$w = w _ { 3 } t a ^ { | m | s } t ^ { - l } t ^ { - p } a ^ { i } t w _ { 2 } \.$$
The radius r = ( l w 3 ) + l ( w 2 ) + | m | + + l p +3 ≥ σ t ( w 3 ) + l ( w 2 ) + l + p +5 = l ( w 2 ) + 2 l + p +4.
Applying Lemma 2.1, we obtain u = G ˇ wt 2 = ( w ta 3 | ms | t -l )( t -p a tw t t i 2 ) = G t -p a tw tw ta i 2 3 | ms | t - -( l 1) . Then
$$\check { u } \coloneqq t ^ { - p } a ^ { i } t w _ { 2 } t w _ { 3 } t a ^ { | m | s } t ^ { - ( l - 1 ) } \.$$
is another geodesic representative of u .
$$C a s e \, 1 0. 3. 1. 1 \colon I f \, l \geq 2 \colon Define$$
$$\delta \coloneqq ( w _ { 2 } ^ { - 1 } t ^ { - 1 } a ^ { - i } t ^ { p - 1 } ) ( t ^ { l } a ^ { - 2 m } t ^ { - l } ) t ^ { - ( p - 1 ) } a ^ { i } t w _ { 2 } t ^ { l } a ^ { 2 m } t ^ { - ( l - 2 ) } \.$$
Applying Lemma 2.1 to the subwords in parentheses shows that δ = G t 2 = γ , so δ labels a path from w to ˇ = u u . The length l ( δ ) = 2 ( l w 2 ) + 4 l +2 +4 p ≤ 2 r -4.
The vertex wδ l w ( ( 2 ) + p + +1) = l w r ( -( ( l w 2 ) + p + +1)) l ∈ B r ( -( ( l w 2 ) + p + l + 1)). Now wδ l w ( ( 2 ) + p + l + 3) = G w ta 3 | ms | t -1 a -2 s = G w ta 3 ( | m |-1) s t -1 w r ( -( ( l w 2 )+ + +3)) p l t -1 , implying wδ l w ( ( 2 ) + p + +3)) l ∈ B r ( -( ( l w 2 ) + p + l + 2)). The suffix t - -( l 2) of δ is also a suffix of ˇ, u so wδ (2 ( l w 2 ) + 3 l +2 +6) = ˇ( p u r -( l -2)) ∈ B r ( -( l -2)). Finally, wδ (2 ( l w 2 )+3 +2 +4) = l p G ˇ( u r - -( l 2)) a -2 s = G t -p a tw tw ta i 2 3 | ms | t -1 a -2 s = G t -p a tw tw ta i 2 3 ( | m |-1) s t -1 , so wδ (2 ( l w 2 ) + 3 l +2 +4) = ˇ( p u r -l ) t -1 ∈ B r ( -( l -1)). The five subpaths of δ between w , these four points, and u are each too short to leave B r ( ).
Case 10.3.1.2: If l = 1 : In this case define
$$\delta \coloneqq ( w _ { 2 } ^ { - 1 } t ^ { - 1 } a ^ { - i } t ^ { p - 1 } ) ( t a ^ { - 2 s } t ^ { - 1 } ) t ^ { - ( p - 1 ) } a ^ { i } t w _ { 2 } t ^ { 2 } a ^ { s } \.$$
Commuting the subwords in parentheses, then δ = G t 2 = γ and δ labels a path from w to u . The length l ( δ ) = 2 ( l w 2 ) +2 p +9 = 2 ( l w 2 ) +4 +2 l p +5 ≤ 2 r -3.
The proof that δ remains in B r ( ) is similar to Case 10.3.1.1. In particular, wδ l w ( ( 2 ) + p +2) = w r ( -( ( l w 2 ) + p +2)) ∈ B r ( -( r -( ( l w 2 ) + p + 2)), wδ l w ( ( 2 ) + p +4) = w r ( -( ( l w 2 ) + p +4)) t -1 ∈ B r ( -( r -( ( l w 2 ) + p +3)), wδ (2 ( l w 2 ) + 2 p +8) = ˇ( u r -1) ∈ B r ( -1), and the four successive subpaths between w , these three points, and u are too short for δ to leave B r ( ).
Case 10.3.2: If w 0 = a k with | k | ≤ 3 , and l ( w 2 ) = p -2 : Since σ t ( w 2 ) = p -2, then w 2 = t p -2 and w = a t k -p a t i p -1 with p ≥ 2 and | i | = 1. Recall that u = t -p a tu u j 1 0 with | j | = 1 and σ t ( u 1 ) = p . The radius r = | k | + 2 p = p +2+ ( l u u 1 0 ), so | k | = ( l u u 1 0 ) -p +2 ≥ 2.
/negationslash
/negationslash
/negationslash
If | k | = 2, then l ( u u 1 0 ) = p and u = t -p a t j p +1 . The trivial element 1 = G wt u 2 -1 = a t k -p a t i p -1 t 2 t -( p +1) a -j t p = G a t k -p a a i -j t p . Since a k = G 1, i = . But j a t k -p a a t i i p = G a t k -( p -1) a t i ( p -1) , and Britton's Lemma says that the latter word cannot represent the trivial element 1, so i = -j . Therefore we cannot have | k | = 2.
If | k | = 3, then l ( u u 1 0 ) = p +1 and σ t ( u 1 ) = p , so the word u u 1 0 contains one occurrence of a or a -1 . Write u u 1 0 = t b a t l p -b for some 0 ≤ b ≤ p and
| l | = 1. Then wt u 2 -1 freely reduces to a t k -p a t t a i ( b -l t -b )( a -j ) t p . Commuting the parenthetical subwords and reducing again gives 1 = G wt u 2 -1 = G a t k -p a ta i -j t b a -l t p -b . This word is in NP , and doesn't contain a subword of the form t -1 a 2 m t , so Britton's Lemma (or Lemma 3.3) implies a contradiction again. Therefore case 10.3.2 cannot occur.
Case 10.3.3: If w 0 = a k with | k | ≤ 3 and l ( w 2 ) ≥ p -1 : In this case we have the geodesic normal forms w = a t k -p a tw i 2 and u = t -p a tu u j 1 0 . The radius r = ( l w 2 ) + | k | + p +2 = ( l u u 1 0 ) + p +2. We have two subcases, depending on whether i = j or i = . j
/negationslash
Case 10.3.3.1: If i = j : In this subcase note that γ = t 2 = G w -1 u = w -1 2 t -1 ( a -j )( t p a -k t -p ) a tu u j 1 0 . Applying Lemma 2.1 and reducing shows that the word δ := w -1 2 t p -1 a -k t -( p -1) u u 1 0 = G γ , and so δ labels a path from w to u . The length l ( δ ) = 2 r -6.
As usual we analyze the distances from 1 of several vertices along the path δ . wδ l w ( ( 2 )) = w r ( -l ( w 2 )) ∈ B r ( -l ( w 2 )), wδ l w ( ( 2 ) + 2 p -2 + | k | ) = u r ( -l ( u u 1 0 )) ∈ B r ( -l ( u u 1 0 )), and the three intervening subpaths are each too short to allow δ to leave B r ( ).
Case 10.3.3.2: If i = -j : Similar to Case 10.3.3.1, after commuting and reduction we have γ = G w -1 u = G w -1 2 t p -1 a -k t -p a a tu u j j 1 0 . Then the word δ := w -1 2 t p -1 w -1 0 t -( p -1) a u u j 1 0 = G γ labels a path from w to u and has length l ( δ ) = 2 r -5.
As in Case 10.3.3.1, we have wδ l w ( ( 2 )) = w r ( -l ( w 2 )) ∈ B r ( -l ( w 2 )) and wδ l w ( ( 2 ) + 2 p -1 + | k | ) = u r ( -l ( u u 1 0 )) ∈ B r ( -l ( u u 1 0 )). Now wδ l w ( ( 2 )+ 2 p -2 + | k | ) = G ( t -p a t a j ) -j = G t -p a -j t , so wδ l w ( ( 2 ) + 2 p -2 + | k | ) = u r ( -( ( l u u 1 0 ) + 2)) a -j t ∈ B r ( -l ( u u 1 0 )) as well. The four successive subpaths of δ between w , these three vertices, and u have lengths too short to allow δ to leave B r ( ). Therefore δ has the required properties.
Therefore in all of cases 1-10, either the case cannot occur or the path δ with the required properties can be constructed, completing the proof of Theorem 3.5. /square
## 4. Non-convexity properties for BS (1 , q )
In the first half of this section we show, in Theorem 4.3, that the group G := BS (1 2) = , 〈 a, t | tat -1 = a 2 〉 with generators A := { a, a -1 , t, t -1 } does not satisfy Poenaru's P (2) almost convexity condition. We start by defining some notation. Let n be an arbitrary natural number with n > 100 and let w := t n a t 2 -n and u := at n a t 2 -( n -1) (see Figure 5). Then w and u are words of length R := 2 n +2. Moreover, using Lemma 2.1, wat = ( a )( t n a t 2 -n ) t = G u , so d w, u ( ) = 2 and the word γ := at labels a path from w to u .
Lemma 4.1. If m ∈ Z and a m is in the ball B R ( ) = B (2 n + 2) in the Cayley graph of G , then either m = 2 n +1 or m ≤ 2 n +2 n -1 +2 n -2 .
Figure 5. w = t n a t 2 -n , u = at n a t 2 -( n -1)
Proof. For a m ∈ B R ( ), Proposition 2.3 says that there is a geodesic word V in the normal form v = t h a t s -1 a k h -1 t -1 · · · t -1 a k 0 with 2 ≤ | s | ≤ 3 and each | k i | ≤ 1, such that v = G a m . This word contains 2 h letters of the form t ± 1 and l ( v ) ≤ 2 n +2, so h ≤ n . Also, v = G a 2 h s +2 h -1 k h -1 + ··· + k 0 = G a m , so m = 2 h s +2 h -1 k h -1 + · · · + k 0 .
Finally, if h ≤ n -2, then m ≤ 2 n -2 (3) + 2 n -3 (1) + · · · +(1) = 2 n -1 2 -1 , so m< 2 n +2 n -1 +2 n -2 . /square
If h = n , then v = t n a ± 2 t -n , so m = ± 2 n +1 . If h = n -1, then there are at most 4 occurrences of a ± 1 in the expression for v ; that is, | s | + ∑ | k i | ≤ 4. The value of m will be maximized if s = +3, k h -1 = +1, and k i = 0 for all i ≤ h -2; in this case, m = (3)2 n -1 +2 n -2 = 2 n +2 n -1 +2 n -2 .
Lemma 4.2. The words w = t n a t 2 -n , wa wa , -1 , wt -1 , and u = at n a t 2 -( n -1) label geodesics in the Cayley graph of G .
Proof. As a consequence of Lemma 4.1, the vertices a 2 n +1 +1 and a 2 n +1 -1 are not in the ball B R ( ). The words t n a t 2 -n a = wa and aw both label paths from the identity to a 2 n +1 +1 , and the word wa -1 labels a path from 1 to a 2 n +1 -1 . Each of these words has length 2 n +3 = R +1, so all three paths must be geodesic. As a consequence, the subwords w of wa and u of aw are also geodesic.
The element wt -1 has a geodesic normal form from Proposition 2.3 given by v = t h a t s -1 a k h -1 t -1 · · · t -1 a k 0 t -1 a l with | l | ≤ 1. Since a 2 n +1 t -1 v -1 = G 1, Lemma 3.3 shows that a l = a 2 i with i ∈ Z , so l = 0. Hence wt -1 is also geodesic. /square
Theorem 4.3. The group G = BS (1 2) = , 〈 a, t | tat -1 = a 2 〉 is not P (2) with respect to the generating set A = { a, a -1 , t, t -1 } .
Proof. Let n ∈ N with n > 100, w = t n a t 2 -n , u = at n a t 2 -( n -1) , and R = 2 n +2. Then w and u lie in the sphere S R ( ) and d w, u ( ) = 2. Let δ be a path inside the ball B R ( ) from w to u that has minimal possible length. In particular, δ does not have any subpaths that traverse a single vertex more than once.
From Lemma 4.2, wa ± 1 and wt -1 are not in B R ( ), so the first letter of the path δ must be t . Let π : C → T denote the horizontal projection map from the Cayley complex of G to the regular tree T of valence 3, as described at the beginning of Section 2. The vertices π wδ ( (1)) = π t ( ) and π u ( ) = π at ( ) are the terminal vertices of the two distinct edges of T with initial vertex π w ( ) = π (1). Since the projection of the path δ begins at π (1), goes to π t ( ), and eventually ends at π at ( ), there must be another point P := wδ j ( ) along the path δ with π P ( ) = π wδ j ( ( )) = π (1) and 1 < j < l δ ( ). Let δ 1 be the subpath of δ from w to P .
/negationslash
Our assumption that δ has minimal possible length implies that P = w . Since π P ( ) = π (1), P = G a m for some m ∈ Z . Then Lemma 4.1 shows that m ≤ 2 n +2 n -1 +2 n -2 . Since δ 1 labels a path from a 2 n +1 to a m , then δ -1 1 = G a 2 n +1 -m = a k with k = 2 n +1 -m ≥ 2 n -2 > 2 ( n -4)+1 . Applying the contrapositive of Lemma 4.1, a k is not in the ball B (2( n -4) + 2), so the length l ( δ -1 1 ) > 2( n -4) + 2. Therefore l ( δ ) > R -8, so this length cannot be bounded above by a sublinear function of R . /square
For the remainder of this section, let G q := BS (1 , q ) = 〈 a, t | tat -1 = a q 〉 with q ≥ 7 and with generators A := { a, a -1 , t, t -1 } . We will apply methods very similar to those developed above, to show that these groups are not MAC .
Lemma 4.4. If m ∈ Z and a m is in the ball B R ( ) = B (2 n + 1) in the Cayley graph of G q , then either m = q n or m ≤ 3 q n -1 .
Proof. For a m ∈ B R ( ), Proposition 2.3 says that there is a geodesic word V in the normal form v = t h a t s -1 a k h -1 t -1 · · · t -1 a k 0 with 1 ≤ | s | ≤ q -1 and each | k i | ≤ /floorleft q 2 /floorright , such that v = G a m . Then h ≤ n and m = q h s + q h -1 k h -1 + · · · + k 0 .
If h = n , then v = t n a ± 1 t -n , and m = ± q n . If h = n -1, then there are at most 3 occurrences of a ± 1 in the expression for v . The value of m will be maximized if s = +3, in which case, m = (3) q n -1 . Finally, if h ≤ n -2, then m ≤ q n -2 ( q -1) + q n -3 /floorleft q 2 /floorright + · · · +( /floorleft q 2 /floorright ) = q n -1 -q n -2 + q n -2 -1 2 -1 /floorleft q 2 /floorright , so m< q 3 n -1 . /square
Theorem 4.5. The group G q = BS (1 , q ) = 〈 a, t | tat -1 = a q 〉 with q ≥ 7 is not MAC with respect to the generating set A = { a, a -1 , t, t -1 } .
Proof. Let n be an arbitrary natural number with n > 100. Let w ′ := t n at -n and u ′ := at n at -( n -1) . Then w ′ and u ′ are words of length R := 2 n +1. As a consequence of Lemma 4.4, an argument similar to the proof of Lemma 4.2 shows that the words w ′ , u ′ , w a ′ ± 1 , and w t ′ -1 are geodesics. Hence w ′ and u ′ lie in S R ( ). Lemma 2.1 says that w at ′ -1 = ( a )( t n at -n ) t 1 = G u ′ , so d w , u ( ′ ′ ) = 2 and the word γ := at 1 labels a path from w ′ to u ′ . Let δ be a path inside the ball B R ( ) from w ′ to u ′ that has minimal possible length.
Using the information on geodesics from the previous paragraph, the first letter of the path δ must be t . Let π : C → T denote the horizontal projection
map from the Cayley complex of G q to the regular tree T of valence q +1. The vertices π w δ ( ′ (1)) = π t ( ) and π u ( ′ ) = π at ( ) are the terminal vertices of two distinct edges of T with initial vertex π w ( ′ ) = π (1). Consequently, there must be a point P := w δ j ′ ( ) along the path δ with π P ( ) = π w δ j ( ′ ( )) = π (1) and 1 < j < l δ ( ). Write δ = δ δ 1 2 where δ 1 is the subpath of δ from w ′ to P .
/negationslash
The vertex P = w ′ = a q n , and P = G q a m for some m ∈ Z , so Lemma 4.4 shows that m ≤ 3 q n -1 . The word δ 1 labels a path from a q n to a m , so δ 1 = G q a k with k = q n -m ≥ ( q -3) q n -1 > q 3 n -1 since q ≥ 7. Then Lemma 4.4 says that either k = q n or a k is not in the ball B (2 n +1).
If a k is not in the ball B (2 n +1), then the length l ( δ 1 ) > n 2 +1 = R . Note that u t ′ -1 = aw ′ = G q a q n +1 . The word δ t 2 -1 labels a path from P = a m to a q n +1 , so δ t 2 -1 = G q a k +1 with k +1 > q 3 n -1 . Then the length l ( δ t 2 -1 ) > R as well. Thus l ( δ 2 ) ≥ R and the length l ( δ ) ≥ R +1+ R = 2 R +1. Since there is a path w ′-1 u ′ of length 2 R inside B R ( ) from w ′ to u ′ , this contradicts our choice of δ with minimal length.
Then the path δ satisfies k = q n , so P = 1. Therefore the path δ reaches the vertex corresponding to the identity, and the length of the path δ is 2 R . /square
Corollary 4.6. The properties MAC and M AC ′ are not commensurability invariant, and hence also not quasi-isometry invariant.
Proof. The index 3 subgroup of BS (1 2) , = 〈 a, t | tat -1 = a 2 〉 generated by a and t 3 is isomorphic to BS (1 8). , Theorem 3.5 shows that BS (1 2) is , M AC ′ and hence MAC , and Theorem 4.5 proves that BS (1 8) has neither , property. /square
## 5. Stallings' group is not MAC
In [15], Stallings showed that the group with finite presentation
$$S \ \coloneqq \ \langle a, b, c, d, s \ | \ [ a, c ] = [ a, d ] = [ b, c ] = [ b, d ] = 1, \\ ( a ^ { - 1 } b ) ^ { s } = a ^ { - 1 } b, ( a ^ { - 1 } c ) ^ { s } = a ^ { - 1 } c, ( a ^ { - 1 } d ) ^ { s } = a ^ { - 1 } d \rangle$$
does not have homological type FP 3 . In our notation, [ a, c ] := aca -1 c -1 and ( a -1 b ) s := sa -1 bs . Let X := { a, b, c, d, s, a -1 , b -1 , c -1 , d -1 , s -1 } be the inverse closed generating set, and let Γ be the corresponding Cayley graph of S .
Let G be the subgroup of S generated by Y := { a, b, c, d, a -1 , b -1 , c -1 , d -1 } , and let Λ be the corresponding Cayley graph of G . Then G is the direct product of the nonabelian free groups 〈 a, b 〉 and 〈 c, d 〉 . Let H be the finitely generated subgroup of G given by H = 〈 a -1 b, a -1 c, a -1 d 〉 .
Lemma 5.1. The group H consists of all elements of G that can be represented by a word over Y of exponent sum zero. Moreover, every word, over X or Y , representing an element of H must have exponent sum zero.
Proof. Using the fact that a commutes with both c and d , we have that ca -1 , da -1 ∈ H , and so their inverses ac -1 , ad -1 ∈ H . Using the fact that b and c commute,
$$\ a b ^ { - 1 } \ = _ { S } \ a ( c ^ { - 1 } c ) b ^ { - 1 } = _ { S } ( a c ^ { - 1 } ) b ^ { - 1 } c = _ { S } ( a c ^ { - 1 } ) b ^ { - 1 } ( a a ^ { - 1 } ) c \\ = _ { S } \ ( a c ^ { - 1 } ) ( a ^ { - 1 } b ) ^ { - 1 } ( a ^ { - 1 } c ) \in H$$
as well. Taking products of the form h -1 1 h 2 with
$$h _ { 1 }, h _ { 2 } \in \{ a ^ { - 1 } b, a ^ { - 1 } c, a ^ { - 1 } d, a b ^ { - 1 }, a c ^ { - 1 }, a d ^ { - 1 } \}$$
shows that l -1 1 l 2 , l 1 2 l -1 ∈ H for all positive letters l 1 , l 2 ∈ { a, b, c, d } . Finally, consider an arbitrary word w = l /epsilon1 1 1 · · · l /epsilon1 n n with l i ∈ { a, b, c, d } , /epsilon1 i = ± 1, and ∑ i /epsilon1 i = 0. For each i , there is a letter m ∈ Y which commutes with both l /epsilon1 i i and l /epsilon1 i +1 i +1 . Repeating the technique above of inserting the inverse pair mm -1 between l /epsilon1 i i and l /epsilon1 i +1 i +1 and applying the commutation relations as needed, then w can be written as a product of elements of exponent sum zero of the form m m 1 2 with m i ∈ Y . Then w ∈ H . The second sentence of this lemma follows from the fact that the exponent sum for each of generators of H and each of the relators in the presentation for S is zero. /square
Let φ : H → H be the identity function. Then S is the HNN extension extension S = G/star φ with stable letter s , and s commutes with all of the elements of H .
## Lemma 5.2. Let w ∈ X ∗ .
- (1) If w is a geodesic in Γ , then the word w cannot contain a subword of the form sus -1 or s -1 us with u ∈ H .
- (2) If w ∈ G and w is a geodesic in Γ , then w ∈ Y ∗ and w is a geodesic in Λ .
- (3) If w ∈ Y ∗ and w is a geodesic in Λ , then w , sw and ws are all geodesics in Γ .
Proof. Part (1) follows directly from the fact that for u ∈ H , sus -1 = S s -1 us = S u . In parts (2) and (3), suppose that g ∈ G v , is a geodesic word over X in Γ representing g , and w ∈ Y ∗ is a geodesic in Λ with w = g also. Then vw -1 = S 1. Britton's Lemma applied to the HNN extension S says that if either s or s -1 occurs in vw -1 , then vw -1 , and hence v , must contain a subword of the form sus -1 or s -1 us with u ∈ H , contradicting part (1). Therefore v ∈ Y ∗ . Since v, w ∈ Y ∗ and v is is a geodesic in Λ, then l ( v ) ≤ l ( w ) Similarly since v, w ∈ X ∗ and w is a geodesic in Γ, l ( w ) ≤ l ( v ). Thus v is also a geodesic in Λ, and w is a geodesic in Γ.
For the remainder of part (3), suppose that µ is a geodesic representative of sw in Γ. Then w -1 s -1 µ = S 1. Britton's Lemma then says that w -1 s -1 µ must contain a subword of the form sus -1 or s -1 us with u ∈ H . Since µ is a geodesic, part (1) says that the sus -1 or s -1 us cannot be completely contained in µ , so we can write µ = µ sµ 1 2 with µ 1 ∈ H . Since µ 1 and s commute, sµ µ 1 2 = S µ = S sw , so µ µ 1 2 = S w and both µ µ 1 2 and w are
s
Figure 6. Paths in the Cayley graph of Stallings' group.
geodesics representing the same element of G . Hence l ( µ µ 1 2 ) = l ( w ). Then l ( µ ) = l ( µ 1 ) + 1 + l ( µ 2 ) = l ( w ) + 1 = l ( sw ), so sw is a geodesic in Γ. The proof that ws is also a geodesic in Γ is similar. /square
The proof of the following theorem relies further on the HNN extension structure of Stallings' group S . In particular, we utilize an ' s -corridor' to show that the path δ in the definition of MAC cannot exist.
Theorem 5.3. ( S, X ) is not MAC with respect to the generating set X .
Proof. Let α := b -( n +1) a n +1 and β := sb -( n +1) a n , and let χ = b -( n +1) a n be their maximal common subword. The word α ∈ Y ∗ , so α ∈ G ; in particular, the exponent sum of α is zero, so Lemma 5.1 says α ∈ H also. Since α is a geodesic in the Cayley graph Λ of the group G = F 2 × F 2 , Lemma 5.2(3) says that α is a geodesic in Γ. Similarly, χ is a geodesic in Λ, so Lemma 5.2(3) says that β = sχ is also geodesic in Γ. Thus α and β lie in the sphere of radius 2 n +2 in Γ. Since
$$\alpha ^ { - 1 } \beta = a ^ { - ( n + 1 ) } b ^ { n + 1 } s b ^ { - ( n + 1 ) } a ^ { n } = _ { S } s a ^ { - ( n + 1 ) } b ^ { n + 1 } b ^ { - ( n + 1 ) } a ^ { n } = _ { S } s a ^ { - 1 },$$
the distance d α, β ( ) = 2, for all natural numbers n .
Suppose there is a path δ of length at most 2(2 n +2) -1 inside the ball of radius 2 n +2 between α and β . Since the relators in the presentation of S have even length, the word δ must have length at most 2(2 n +2) -2 = 4 n +2.
Applying Britton's Lemma to the product δas -1 = S 1 implies that δ = w sw 1 2 with w , w a 1 2 ∈ H . Then w , w 1 2 ∈ G = F 2 × F 2 . Lemma 5.2(2) and the direct product structure imply that there are geodesic representatives q 1 and q 2 of w 1 and w 2 , respectively, that have the form q 1 = q 1 a,b q 1 c,d and q 2 = q 2 c,d q 2 a,b with q 1 a,b , q 2 a,b ∈ { a, b, a -1 , b -1 } ∗ and q 1 c,d , q 2 c,d ∈ { c, d, c -1 , d -1 } ∗ .
Since α and q 1 = w 1 are both elements of H αq , 1 ∈ H as well. From the direct product structure, there is a geodesic representative σ ∈ Y ∗ of αq 1 of the form σ = σ a,b σ c,d with σ a,b ∈ { a, b, a -1 , b -1 } ∗ and σ c,d ∈ { c, d, c -1 , d -1 } ∗ . (see Figure 6).
The edge in Γ labeled by s connecting σ and σs is part of the path δ , so this edge must lie in the ball of radius 2 n +2 in Γ. Lemma 5.2(3) says that
σs is a geodesic, so d (1 , σ ) + 1 = l ( σ ) + 1 = l ( σs ) = d (1 , σs ) ≤ 2 n +2. Then the vertex σ ∈ B (2 n +1) and l ( σ ) ≤ 2 n +1.
Now l ( q 1 ) + 1 + l ( q 2 ) ≤ l ( δ ) ≤ 4 n + 2, and thus either l ( q 1 ) ≤ 2 n or l ( q 2 ) ≤ 2 n (or both).
Case A: If l ( q 1 ) ≤ 2 n : Note that αq σ 1 -1 = b -( n +1) a n +1 q 1 a,b q 1 c,d σ -1 c,d σ -1 a,b = F 2 × F 2 1. Hence q 1 c,d = F 2 σ c,d and αq 1 a,b = F 2 σ a,b . Since geodesics in free groups are unique, we also have q 1 c,d = σ c,d .
/negationslash
There is an integer 0 ≤ i 1 ≤ 2 n such that q 1 a,b ( i 1 ) = α -1 ( i 1 ) but q 1 a,b ( i 1 + 1) = α -1 ( i 1 + 1), where we denote q 1 a,b (0) := 1 and q 1 a,b ( k ) := q for all k > l q ( 1 a,b ). Write q 1 a,b = α -1 ( i 1 ) r with r ∈ F 2 = 〈 a, b 〉 . The words α q , 1 a,b , and σ a,b are all geodesic representatives of elements of the free group F 2 , and hence these are freely reduced words that define non-backtracking edge paths in the tree given by the Cayley graph for this group. By definition of i 1 , the product αq 1 a,b freely reduces to α (2 n +2 -i 1 ) r , with no further free reduction possible. Then α (2 n +2 -i 1 ) r is the unique geodesic representative in F 2 = 〈 a, b 〉 of αq 1 a,b , and hence α (2 n +2 -i 1 ) r = σ a,b .
Case A.1: If i 1 ≤ n +1 : In this case, q 1 a,b = a -i 1 r . Now q 1 = q 1 a,b q 1 c,d = a -i 1 rq 1 c,d represents an element of H , and so Lemma 5.1 says q 1 has exponent sum zero. Then l ( rq 1 c,d ) ≥ i 1 . We also have σ = σ a,b σ c,d = α (2 n + 2 -i 1 ) rq 1 c,d . Then the length l ( σ ) ≥ (2 n + 2 -i 1 ) + i 1 = 2 n + 2, contradicting the result above that l ( σ ) ≤ 2 n +1. Thus this subcase cannot occur.
Case A.2: If i 1 > n + 1 : In this case, σ a,b = b -(2 n +2 -i 1 ) r . Since σ = σ a,b σ c,d = b -(2 n +2 -i 1 ) rσ c,d represents an element of H , this word has exponent sum zero, so l ( rσ c,d ) ≥ 2 n + 2 -i 1 in this subcase. The word q 1 = q 1 a,b q 1 c,d = α -1 ( i 1 ) rσ c,d then has length l ( q 1 ) ≥ i 1 +(2 n +2 -i 1 ) = 2 n +2, contradicting the fact that we are in Case A.
Case B: If l ( q 2 ) ≤ 2 n : Since σ ∈ H σ , commutes with s . Then
$$\sigma = _ { S } s ^ { - 1 } \sigma s = _ { S } s ^ { - 1 } \alpha q _ { 1 } s = _ { S } s ^ { - 1 } \alpha q _ { 1 } s q _ { 2 } q _ { 2 } ^ { - 1 } = _ { S } s ^ { - 1 } \beta q _ { 2 } ^ { - 1 } = _ { S } \chi q _ { 2 } ^ { - 1 } \.$$
/negationslash
In this case σ a,b σ c,d = F 2 × F 2 b -( n +1) a n +1 q -1 2 a,b q 2 c,d -1 , so q 2 c,d -1 = F 2 σ c,d and χq -1 2 a,b = F 2 σ a,b . Uniqueness of geodesics in F 2 = 〈 c, d 〉 implies q 2 c,d -1 = σ c,d . There is an integer 0 ≤ i 2 ≤ 2 n such that q -1 2 a,b ( i 2 ) = χ -1 ( i 2 ) but q -1 2 a,b ( i 2 + 1) = χ -1 ( i 2 +1). Write q 2 a,b = ( r χ -1 ( i 2 )) -1 with r ∈ F 2 = 〈 a, b 〉 . The words χ , q 2 a,b , and σ a,b are all geodesics, and hence freely reduced words, in F 2 . By definition of i 2 , the product χq -1 2 a,b freely reduces to χ (2 n +1 -i 2 ) r -1 , with no further reduction possible. Then χ (2 n +1 -i 2 ) r -1 = σ a,b .
Case B.1: If i 2 ≤ n : In this case, q 2 a,b = ra i 2 . Now q 2 = q 2 c,d q 2 a,b = q 2 c,d ra i 2 . Recall that q 2 was chosen as a geodesic representative of an element
w 2 ∈ G for which w a 2 ∈ H . Then q a 2 represents an element of H , and so (by Lemma 5.1) has exponent sum zero. Therefore the exponent sum of q 2 is -1. Then l ( q 2 c,d r ) ≥ i 2 +1. We also have σ = σ a,b σ c,d = χ (2 n +1 -i 2 ) r -1 q -1 2 c,d . Then the length l ( σ ) ≥ (2 n +1 -i 2 ) + ( i 2 +1) = 2 n +2, again contradicting the result above that l ( σ ) ≤ 2 n +1.
Case B.2: If i 2 > n : In this case, σ a,b = b -(2 n +1 -i 2 ) r -1 . Since σ = σ a,b σ c,d = b -(2 n +1 -i 2 ) r -1 σ c,d represents an element of H , this word has exponent sum zero, so l ( r -1 σ c,d ) ≥ 2 n +1 -i 2 in this subcase. Therefore the word q 2 = q 2 c,d q 2 a,b = σ -1 c,d r χ ( -1 ( i 1 )) -1 has length l ( q 2 ) ≥ (2 n +1 -i 2 ) + i 2 = 2 n +1, contradicting the fact that we are in Case B.
Therefore every possible subcase results in a contradiction implying that the subcase cannot occur. Then the path δ cannot exist, so S is not MAC with respect to the generating set X . /square
## References
- [1] James Belk and Kai-Uwe Bux. Thompson's group F is maximally nonconvex. In Geometric methods in group theory , volume 372 of Contemp. Math. , pages 131-146. Amer. Math. Soc., Providence, RI, 2005.
- [2] Martin R. Bridson. Doubles, finiteness properties of groups, and quadratic isoperimetric inequalities. J. Algebra , 214(2):652-667, 1999.
- [3] James W. Cannon. Almost convex groups. Geom. Dedicata , 22(2):197-210, 1987.
- [4] Murray Elder. Automaticity, almost convexity and falsification by fellow traveler properties of some finitely generated groups. PhD Dissertation, University of Melbourne, 2000.
- [5] Murray Elder. The loop shortening property and almost convexity. Geom. Dedicata , 102(1):1-18, 2003.
- [6] David B. A. Epstein, James W. Cannon, Derek F. Holt, Silvio V. F. Levy, Michael S. Paterson, and William P. Thurston. Word processing in groups . Jones and Bartlett Publishers, Boston, MA, 1992.
- [7] Louis Funar. On discrete solvgroups and Po´naru's condition. e Arch. Math. (Basel) , 72(2):81-85, 1999.
- [8] J. R. J. Groves. Minimal length normal forms for some soluble groups. J. Pure Appl. Algebra , 114(1):51-58, 1996.
- [9] V. S. Guba. The Dehn function of Richard Thompson's group F is quadratic. Invent. Math. , 163(2):313-342, 2006.
- [10] Ilya Kapovich. A note on the Po´ enaru condition. J. Group Theory , 5(1):119-127, 2002.
- [11] Charles F. Miller, III. Normal forms for some Baumslag-Solitar groups. Preprint 1997.
- [12] Charles F. Miller, III and Michael Shapiro. Solvable Baumslag-Solitar groups are not almost convex. Geom. Dedicata , 72(2):123-127, 1998.
- [13] V. Po´ enaru. Almost convex groups, Lipschitz combing, and π ∞ 1 for universal covering spaces of closed 3-manifolds. J. Differential Geom. , 35(1):103-130, 1992.
- [14] Tim R. Riley. The geometry of groups satisfying weak almost-convexity or weak geodesic-combability conditions. J. Group Theory , 5(4):513-525, 2002.
- [15] John Stallings. A finitely presented group whose 3-dimensional integral homology is not finitely generated. Amer. J. Math. , 85:541-543, 1963.
- [16] Carsten Thiel. Zur fast-Konvexit¨t a einiger nilpotenter Gruppen . Universit¨t a Bonn Mathematisches Institut, Bonn, 1992. Dissertation, Rheinische Friedrich-WilhelmsUniversit¨t Bonn, Bonn, 1991. a
School of Mathematics and Statistics, University of St. Andrews, North Haugh, St. Andrews, Fife, KY16 9SS, Scotland
E-mail address :
[email protected]
Department of Mathematics, University of Nebraska, Lincoln NE 685880323, USA
E-mail address : [email protected] | 10.1515/jgth.2005.8.2.239 | [
"Murray Elder",
"Susan Hermiller"
] | 2014-08-01T23:59:23+00:00 | 2014-08-01T23:59:23+00:00 | [
"math.GR",
"20F65"
] | Minimal Almost Convexity | In this article we show that the Baumslag-Solitar group $BS(1,2)$ is
minimally almost convex, or $MAC$. We also show that $BS(1,2)$ does not satisfy
Po\'enaru's almost convexity condition $P(2)$, and hence the condition $P(2)$
is strictly stronger than $MAC$. Finally, we show that the groups $BS(1,q)$ for
$q \geq 7$ and Stallings' non-$FP_3$ group do not satisfy $MAC$. As a
consequence, the condition $MAC$ is not a commensurability invariant. |
1408.0323v3 | ## Laboratory plasma physics experiments using merging supersonic plasma jets
S. C. Hsu, 1 † A. L. Moser, 1 ‡ E. C. Merritt, 1 2 , C. S. Adams, 1 2 , J. P. Dunn, 1 S. Brockington, 3 A. Case, 3 M. Gilmore, 2 A. G. Lynn, 2 S. J. Messer, 3 ¶ and F. D. Witherspoon 3
1
Los Alamos National Laboratory, Los Alamos, NM 87545, USA 2 University of New Mexico, Albuquerque, NM 87131, USA 3 HyperV Technologies Corp., Chantilly, VA 20151, USA
(Received ?; revised ?; accepted ?. - To be entered by editorial office)
We describe a laboratory plasma physics experiment at Los Alamos National Laboratory that uses two merging supersonic plasma jets formed and launched by pulsed-powerdriven railguns. The jets can be formed using any atomic species or mixture available in a compressed-gas bottle and have the following nominal initial parameters at the railgun nozzle exit: n e ≈ n i ∼ 10 16 cm -3 , T e ≈ T i ≈ 1 4 eV, . V jet ≈ 30-100 km/s, mean charge ¯ Z ≈ 1, sonic Mach number M s ≡ V jet /C s > 10, jet diameter = 5 cm, and jet length ≈ 20 cm. Experiments to date have focused on the study of merging-jet dynamics and the shocks that form as a result of the interaction, in both collisional and collisionless regimes with respect to the inter-jet classical ion mean free path, and with and without an applied magnetic field. However, many other studies are also possible, as discussed in this paper.
## PACS codes:
## 1. Introduction
The Plasma Liner Experiment (PLX) is a laboratory plasma physics facility at Los Alamos National Laboratory. The primary purpose and goal of PLX is to form and study spherically imploding plasma liners via multiple merging plasma jets, motivated by high-energy-density (HED) physics and magneto-inertial fusion (MIF) applications (Thio et al. 1999, 2001; Hsu et al. 2012 a b , ; Santarius 2012; Knapp & Kirkpatrick 2014). For fundamental HED-physics studies (Drake 2006), PLX is envisioned to provide an economic means for forming inertially confined cm-, µ s-, and Mbar-scale plasmas (Awe et al. 2011; Davis et al. 2012; Cassibry et al. 2013) upon the stagnation of up to 60 merged pulsed-power-driven, supersonic plasma jets. Such an experiment would allow for good diagnostic measurements on larger-spatial and longer-temporal scales and a much higher shot rate than those of typical laser- or Z-pinch-driven HED experiments. For the MIF application (Lindemuth & Kirkpatrick 1983; Kirkpatrick et al. 1995; Lindemuth & Siemon 2009), the use of a spherically converging plasma liner for compressing a magnetized target plasma would avoid the repetitive hardware destruction inherent in solid-liner approaches (e.g., Intrator et al. 2004; Slutz et al. 2010; Degnan et al. 2013), and it would
†
Email: [email protected]
- ‡ Now at General Atomics, San Diego, CA.
- ¶ Now at Foreground Security, Herndon, VA.
allow for a higher repetition rate ( ∼ 1 Hz), which is desirable for an economic, pulsed fusion-energy system.
The PLX facility was built in 2010-2011, and the first plasma jet was fired in September, 2011. Initial research characterized in detail the plasma properties and evolution of a single jet (Hsu et al. 2012 ). A second railgun was installed in mid-2012, and ensuing b experiments focused on the oblique merging of two plasma jets, resulting in detailed observations shown to be consistent with collisional plasma shocks (Merritt et al. 2013, 2014). Due to loss of funding, we were not able to build the facility up to 30 railguns as originally planned. Instead, ongoing experiments are focused on the head-on merging of two plasma jets to form and study unmagnetized and magnetized plasma shocks. An in-chamber pulsed Helmholtz coil was installed in late 2013 to enable the study of magnetized shocks. The facility now enables a unique experimental research program on the detailed study of plasma shocks, from collisional-to-collisionless and unmagnetized-tomagnetized regimes. Preliminary designs for a 36- or 60-jet experiment (with ∼ 1 5 MJ . of capacitive stored energy) to form spherically imploding plasma liners are also in hand.
Plasma jet and shock studies on PLX complement many other related, contemporary experimental efforts. Examples include studies of plasma jet dynamics, interactions, and/or shock formation in astrophysically or fusion-relevant contexts using a variety of plasma formation methods, ranging from coaxial plasma guns (e.g., Bellan et al. 2005; Hsu & Bellan 2005; Moser & Bellan 2012) to wire-array-driven pinches (e.g., Haas et al. 2011; Gourdain & Seyler 2013; Swadling et al. 2014) to laser-driven experiments (e.g., Ross et al. 2013; Li et al. 2013; Fox et al. 2013; Fiksel et al. 2014). In general, PLX plasmas are colder and more collisional than the other experiments, but have the benefits of larger spatial size for ease of diagnostic measurements, more choices for the working plasma species, and the option of both unmagnetized and magnetized experiments. A brief overview of many plasma-jet experiments and formation methods is given in Hsu (2009).
The rest of this paper is organized as follows. Section 2 describes the PLX experimental device and parameters. Section 3 discusses the physics questions that can be addressed on PLX. Section 4 summarizes the major results obtained thus far, and Sec. 5 describes future opportunities.
## 2. Description of experimental device and parameters
2.1. Device description
PLX consists of a 9-ft. (2.74-m) diameter spherical vacuum chamber (Fig. 1), two pulsedpower-driven plasma railguns (Sec. 2.1.1), a diagnostic suite (Sec. 2.1.2), and an inchamber Helmholtz coil (Sec. 2.1.4) to generate a pulsed magnetic field up to 0.22 T. The facility is housed in a 3000-ft. 2 (279-m ) high-bay space with a 10-ton (9072-kg) 2 bridge crane, roll-up door with 18.5-ft. (5.6-m) tall and 16-ft. (4.9-m) wide clearance, 208- and 480-VAC power, and building chilled water and compressed air.
## 2.1.1. Plasma railguns and pulsed-power systems
The plasma railguns [Fig. 2(a) and (b)] and their high-current, spark-gap switches [Fig. 2(c)] were designed and built by collaborator HyperV Technologies Corp. The railgun has two parallel-plate electrodes made of tungsten alloy (HD-17BB), separated by insulators made of zirconium-toughened alumina (ZTA). The railgun bore is 1 in. (2.54 cm) × 1 in. (2.54 cm), and the body is a clamshell design consisting of two Noryl halves bolted together [Figs. 2(a) and (b)]. A custom-built gas-puff valve injects neutral gas into a preionization (PI) chamber at the rear of the railgun. A cylindrical acrylic nozzle with 5-cm
Figure 1. Photograph of the PLX vacuum chamber (2.74-m diameter).
Figure 2. (a) Photograph of railgun. (b) Schematic of railgun (courtesy of HyperV Technologies). (c) Photograph of two linear, spark-gap switches (1 main switch, 1 crowbar) installed on top of a railgun capacitor bank.
| | capacitor model | rated bank voltage (kV) | total bank capacitance ( µ F) | max. energy (kJ) |
|-------------------------------------|-------------------------------------|-------------------------------------|-------------------------------------|-------------------------------------|
| Railguns (one bank per gun) | Maxwell 32184 | 40 | 12 or 36 | 9.6 or 28.8 |
| Pre-ionization (for both guns) | Maxwell 32814 | 40 | 12 | 9.6 |
| Gas valve (for both guns) | Maxwell 32567 | 20 | 24 | 1.7 |
| Helmholtz coil | GA 32934 | 5 | 4000 | 50 |
| Table 1. Capacitor bank parameters. | Table 1. Capacitor bank parameters. | Table 1. Capacitor bank parameters. | Table 1. Capacitor bank parameters. | Table 1. Capacitor bank parameters. |
inner diameter is mounted at the exit of the railgun bore. The rails, PI, and gas valves (GV) are driven by four separate capacitor banks, which are summarized in Table 1. The GVs and PIs of both guns are connected in parallel and driven by a single GV and PI bank, respectively.
## 2.1.2. Diagnostics
Operational diagnostics include a multi-chord visible interferometer, a visible-to-nearinfrared survey spectrometer, an array of three photodiodes, a charge-coupled-device (CCD) camera, a schlieren imaging system, an electrostatic triple probe, and magnetic probe arrays. We briefly summarize the diagnostic capabilities here; more technical details are given elsewhere (Hsu et al. 2012 ; Lynn b et al. 2010).
The 561-nm multi-chord interferometer (Merritt et al. 2012 a b , ) is fiber-coupled between the launch laser optics and the vacuum chamber, allowing for relatively easy (a few days work) modification of chord positioning for different experiments and different plasma gun geometries. Several different chord arrangements have been used for singlejet propagation (Hsu et al. 2012 ), two-jet oblique-merging experiments (Merritt b et al. 2013, 2014), and two-jet head-on merging experiments (Moser & Hsu 2014). The timeresolution of the continuous interferometer phase signals is 50 ns, limited by the 40-MHz digitization rate. Spatial resolution is determined by the chord spacing (typically a few cm) and the laser spot size ( ≈ 3-mm diameter at the position of the plasma).
The survey spectrometer system consists of a 0.275-m spectrometer and a gated (0.45µ s duration), 1024-pixel, multi-channel-plate array covering the wavelength range 300-900 nm. The best spectral resolution, corresponding to a 600 line/mm grating, is 0.152 nm/pixel. Plasma light is collected at a vacuum chamber window into a 5-mm diameter collimating lens and brought to the spectrometer with a optical fiber bundle. The diameter of the spectrometer viewing chord is ≈ 7 cm at the position of the plasma. The lens and fiber can be moved easily to different viewing positions.
A three-channel photodiode (PD) array collects chord-integrated and time-resolved broadband visible (300-850 nm) plasma emission for determining jet propagation speed, based on the viewing-chord positions and time-of-flight estimates of features in the PD data traces. The peak PD responsivity is 0.65 A/W at 970 nm, and the frequency response decreases with increasing gain (2.1 MHz and 0.1 MHz at 20 and 50 dB, respectively). The light is collected through an adjustable aperture that is typically set at 1 cm, which constitutes the nominal spatial resolution.
An intensified CCD camera (DiCam Pro ICCD), with spectral sensitivity from the UV to near-IR, is used to capture time-gated, visible images of the experiment. The camera records 1280 × 1024-pixel images with 12-bit dynamic range. The typical gate time used is 20 ns. The camera is triggered remotely via an optical fiber, housed inside a metal shielding box, and mounted next to a large rectangular borosilicate window on the vacuum chamber (although it can be moved to other ports as well).
A schlieren system (Settles 2001) for obtaining 2D images (4 cm × 16 cm) sensitive to plasma-density gradients has been constructed. The PLX schlieren system (Adams et al. 2012) uses a pulsed 1.064µ mNd:YAG laser (3-ns pulse, 30-80 mJ) expanded to a 20-cm diameter beam that is then collimated and passed through the plasma between 25-cm diameter, f /8 mirrors. A tunable knife edge blocks light refracted by density gradients in the plasma. The image is captured by an IR camera (Apogee Alta U1109) which contains a 2048 × 512 CCD detector with 8% quantum efficiency at 1.064 µ m. The system is designed to give images with > 5% contrast for electron-density gradients /greaterorsimilar 10 15 cm -4 , i.e., for 1-cm-scale gradients in the presence of 10 14 -cm -3 density plasmas. The intensity of the recorded image is proportional to ∇ N , where N is the refractive index of the plasma. The PLX schlieren diagnostic has yet to be tested on plasmas with sufficient density gradients to yield images with detectable contrast.
Magnetic probe arrays measure all three components of the magnetic field along the railgun nozzle and within the plasma-jet-interaction region. The nozzle probe array con-
sists of hand-wound coils (using 0.5-mm diameter Kapton-coated magnet wire) around a rectangular Delrin rod (0 37 . × 0 48 cm ). These probes have effective turns . 2 × area ( NA ) of ≈ 10 cm 2 in the frequency range 10-5000 kHz. The internal probe array is mounted inside an insertable probe shaft and contains two sets of 3-axis coils at two positions separated by 27.75 cm, for the simultaneous measurement of up to six B -dot signals simultaneously. These probe arrays use commercial inductor chips (Coilcraft 1008CS272XJLB, with dimensions 2 92 . × 2 79 . × 2 03 mm ); the probe construction is similar . 3 to that described by Romero-Talam´s a et al. (2004). The internal probe array coils have NA ≈ 0 7 . cm 2 between 10-1000 kHz. The B -dot signals are integrated by a passive integration circuit with RC = 100 µ s.
An electrostatic triple probe (e.g., Ji et al. 1991) measures the instantaneous floating potential V f , electron temperature T e , and ion saturation current I sat as a function of time at one position in space. The probe can be moved to different radial positions with respect to chamber center. Due to the lack of a reliable, nearby reference ground potential (due to large voltage excursions on the vacuum chamber during a shot), the triple probe is not yet producing reliable data. To improve the probe operation, we have built 4 battery-powered, analog optical channels (up to 1-MHz frequency response) to bring probe signals back to the data acquisition system.
Finally, there are Rogowski coils, high-voltage probes, voltage divider networks, and pulsed current monitors to record the pulsed electrical currents and voltages of all the capacitor banks.
## 2.1.3. Control and data acquisition systems
PLX has an FPGA-based control/trigger system and CAMAC-based digital-dataacquisition system, all integrated through a LabVIEW software interface and operated from a desktop computer. Both analog and optical channels are available for diagnostic triggering and operating relays for the various hardware sub-systems. The dataacquisition system consists of 64 channels with 12-bit resolution, up to 40-MHz sampling rate, and 32 kilosamples of memory per channel (4 Joerger TR modules). Shot information and diagnostic data are stored in an MDSplus tree (www.mdsplus.org).
## 2.1.4. In-chamber Helmholtz coils
We have installed inside the vacuum chamber a pair of magnetic coils arranged in a Helmholtz (HH) configuration (coil radius and separation ≈ 30 cm) inside the vacuum chamber. The coils and mounting structure were designed and fabricated by Woodruff Scientific, LLC, and are presently installed such that the HH field is perpendicular to the direction of jet propagation, but the HH field can be adjusted up to ± 30 ◦ about the perpendicular and parallel directions with respect to the jet propagation direction. The coils are driven by a capacitor bank (see Table 1) with a current-rise time of ≈ 1 3 ms. The maximum HH-bank operating voltage of 4 kV gives a coil current . ≈ 8 2 kA . and a peak magnetic field of 0.22 T. The minimum operating voltage is around 200 V (corresponding to about 0.44 kA and 0.012 T peak magnetic field), below which the HH-bank ignitron switch does not fire reliably. The magnetic-field profiles have been characterized experimentally using three concentric flux loops of different radii.
## 2.2. Parameters
Table 2 summarizes the experimentally measured or inferred physical parameters of our plasma jets, both at the gun-nozzle exit and after propagating ≈ 41 cm. For head-on jetmerging experiments, the jets propagate ≈ 1 1 m before colliding with each other near . chamber center, where the jet n tot ∼ 10 14 cm -3 and T e ≈ 1 eV just before the collision.
Figure 3. Photograph of the in-chamber Helmholtz coil, looking down the barrel of one railgun.
For the achievable range of n tot , T e , and V just before jets collide, the inter-jet counterstreaming ion mean free path λ i , ctr (mean free path of an ion in one jet interacting with all species in the opposing jet) can range from collisional, i.e., λ i , ctr < ∆ (Merritt et al. 2013, 2014), where ∆ is the gradient-scale-length of the shock that forms as a result of the colliding jets, to collisionless, i.e., λ i , ctr /greatermuch ∆ (the goal of ongoing work). Note that both ion and electron collisionality are very high within each jet, and that the inter-jet counter-streaming electron mean free path is very short, i.e., λ e ctr , /lessmuch ∆.
## 3. Physics questions that can be addressed
3.1. Plasma shock structure and dynamics
Highly supersonic plasmas, i.e., in which the sonic Mach number M ≡ V/C s /greatermuch 1 (with V the plasma flow speed and C s the ion sound speed), are needed and provide a natural opportunity to study plasma shock physics (e.g., Jaffrin & Probstein 1964; Casanova et al. 1991; Sagdeev & Kennel 1991). Detailed laboratory studies of the latter have been scarce for two main, coupled reasons: (i) the most readily available ways to form supersonic plasmas tended to be in very high-density regimes, e.g., using laser, z -pinch, or θ -pinch sources that (ii) presented significant diagnostic challenges. The latter arose due to a combination of small spatial scales, high plasma densities, and/or lack of diagnostic access. PLX, with its macroscopic plasma jets at moderate density and completely open diagnostic access, presents a unique experimental platform for studying detailed plasma shock structure and dynamics on multiple-cm spatial and severalµ s time scales. The experiment also offers the ability to span collisional to collisionless regimes for the counter-streaming ions and unmagnetized to magnetized jet interactions.
3.2. Equation-of-state in regimes intermediate between magnetic and inertial fusion Plasma equation of state (EOS), i.e., ionization state and the relationship between specific internal energies and pressure, can be complicated depending on the regime of plasma
temperature and density as well as the plasma species or mixture of species. In magnetic fusion plasmas, which typically are at low density and consist of fully ionized hydrogen isotopes, EOS is relatively straightforward, e.g., governed by the ideal gas law p = n kT e e + n kT i i with n e = n i . On the other hand, in inertial fusion and HED plasmas, which can have mixed-species plasmas with less than fully stripped ions, EOS is a leading order issue complicated by atomic physics, and significant effort has been made and sophisticated capabilities developed to model EOS.
The PLX railguns can generate low-to-moderate density plasmas using any species or mixture available in a compressed-gas bottle. The density and temperature range (see Table 2) are such that the mean charge state ¯ is rapidly varying, and local-thermodynamicZ equilibrium (LTE) and non-LTE models predict notably different EOS. PLX plasmas, whether single- or merging-jet, offer a platform for measuring and constraining EOS in a plasma regime that has not received much detailed attention.
## 3.3. Interaction of plasma flow with vacuum magnetic field or magnetized plasma
Studies of plasma flow across a magnetic field (e.g., Baker & Hammel 1965), or into another magnetized plasma (e.g., Liu & Hsu 2011) have a long history, and have been motivated by both basic plasma physics and applications such as tokamak fueling (e.g., Perkins et al. 1988). With our in-chamber HH coil, we can study supersonic plasma jet interactions with a vacuum magnetic field. In addition, because our plasma jets each consist of multiple 'blobs' due to the ringing railgun current, we can study the interaction of trailing blobs with stagnated magnetized plasma formed from the merged leading blobs.
## 4. Summary of major results
## 4.1. Characterization of single-jet parameters and evolution
The first set of experiments that we performed on PLX focused on the detailed characterization of the plasma parameters, profiles, and evolution over ≈ 1 m of propagation of a single argon plasma jet launched by a plasma railgun (Hsu et al. 2012 b ). These experiments were done to help us understand and interpret planned, subsequent experiments based on the merging of two or more plasma jets, and also to provide accurate initial conditions for numerical modeling of experiments using these plasma jets. These experiments also contributed uniquely to the plasma railgun literature, most of which has focused on in-bore characterization of the plasma armature as a pusher for launching solid projectiles to high velocities (see Batteh 1991, for example). Table 2 summarizes the measured (or experimentally inferred) argon plasma jet parameters at the exit of the railgun nozzle ( Z ≈ 2 cm) and downstream ( Z ≈ 41 cm), as well as the expected range of values for some of the parameters, although the latter were not systematically measured in the single-jet studies.
The plasma jet is essentially unmagnetized after a few tens of centimeters of propagation (at a typical speed of 3-7 cm/ µ s) due to fast resistive decay of the magnetic field that is embedded in the jet as it exits the railgun bore. Using magnetic pickup coils mounted on the exterior surface of the railgun nozzle, we have measured (Merritt et al. 2014) the decay time of the transverse (relative to the jet propagation) component of the magnetic field (of order 750 G near the railgun nozzle exit for peak railgun current of order 200 kA) to be ≈ 5 6 . µ s, which is consistent with the expected magnetic diffusion time estimated using the plasma jet parameters and radial scale lengths. Thus, for the jet-merging studies described in Secs. 4.2 and 4.3, the jet kinetic energy density ( ρv / 2 2) is much greater (by a factor of 50-10 ) than the magnetic energy density ( 4 B / µ 2 2 0 ) at the time of initial jet merging.
Table 2. Measured and achievable range of railgun-driven argon plasma jet parameters. Hydrogen and helium jets have similar values.
| parameter | Z ≈ 2 cm | Z ≈ 41 cm | range |
|-------------------------------------|------------|-------------|--------------|
| ion+neutral density n tot (cm - 3 ) | 2 × 10 16 | 2 × 10 15 | 10 14 -10 17 |
| electron temperature T e (eV) | 1.4 | 1.4 | 1-3 |
| jet velocity V (km/s) | 30 | 30 | 30-100 |
| mean charge ¯ Z | 0.96 | 0.94 | ∼ 1 |
| jet length L (cm) | 20 | 45 | |
| jet diameter D (cm) | 5 | 10-20 | |
## 4.2. Evidence for collisional shock formation between obliquely merging plasma jets
Following the single-jet experiments, we conducted experiments on the oblique merging of two plasma jets (Merritt et al. 2013, 2014), in order to gain better understanding of the essential underlying physical process for plasma liner formation via merging plasma jets. This work led to the identification and characterization of the stagnation layer between obliquely merging jets and the demonstration that the stagnation layer morphology is consistent with that of a collisional oblique plasma shock (Merritt et al. 2013, 2014).
We obtained temporally and spatially resolved survey-spectrometer and interferometer measurements of the stagnation layer, along with sequences of time-gated visible CCD images. Based on experimentally inferred plasma parameters, the inter-jet ion mean free path were less than the stagnation layer thickness, i.e., the oblique jet-merging was in a collisional regime. Detailed analyses of the data from all three diagnostics allowed us to arrive at the following key results (Merritt et al. 2013, 2014): demonstration that (a) the observed stagnation layer morphology is consistent with that expected of a collisional oblique plasma shock, (b) the peak density exceeds that of simple plasma interpenetration but is about a factor of 2 smaller than the peak density predicted by hydrodynamic shock theory, and (c) the observed gradient scale length at the postulated shock boundary is consistent with two-fluid plasma theory predictions of shock layer thickness.
## 4.3. Transition from collisionless interpenetration to collisional stagnation during head-on jet merging
In working toward a more collisionless regime for the counter-streaming ions between the merging jets, we performed a set of head-on jet-merging experiments in which the interaction between jets began with collisionless interpenetration but then transitioned to collisional stagnation due to a rising ¯ (Moser & Hsu 2014). This observation demonZ strates a clear way by which an initially collisionless interaction between colliding plasmas can evolve toward a collisional one, and provides data that can be used to validate physics models of plasma collisionality in the presence of complex EOS. The observation of unmagnetized, supersonic interpenetration also establishes a regime in which neither macronor micro-instabilities are present or strong enough to introduce collisionless shock effects.
## 4.4. Observation of Rayleigh-Taylor-instability growth and progression to longer wavelength
As mentioned earlier, our ringing railgun current produces several serial plasma 'blobs,' each separated in time by 20-30 µ s. The leading blob of each jet each compresses the applied magnetic field and stagnates in the center of the chamber upon merging with the opposing jet. We observed one trailing blob decelerating against the stagnated mag-
netized plasma in the center (formed by the merged leading blobs). Images from a multiple-frame CCD camera (Invisible Vision model UHSi 12/24) revealed the growth of Rayleigh-Taylor-like fingers at the leading edge of this trailing blob as well as the evolution toward longer wavelength of these fingers. Preliminary instability-growth-rate estimates and magnetohydrodynamic (MHD) simulations suggest that our observations are consistent with Rayleigh-Taylor instability of this decelerating interface. The progression toward longer mode wavelength could be due to a combination of magnetic stabilization and viscous damping. Detailed analysis is ongoing, and these results will be reported elsewhere.
## 4.5. Ongoing studies of magnetized head-on jet merging to study cosmically relevant collisionless shocks
Ongoing experiments on PLX are aimed at forming and studying 'cosmically relevant' collisionless shocks, as described by Drake (2000), who argues that such experiments must meet a number of competing constraints on magnetization, plasma β , Alfv´ en Mach number M A , and collisionality. Table 3 summarizes the original design parameters for collisionless shock studies on PLX using head-on-merging plasma jets. Hybrid particlein-cell (PIC) and fully kinetic PIC simulations of head-on merging jets in this parameter regime confirmed that we would be in a collisionless regime and that magnetized shocks should form (Thoma et al. 2013). However, the level of impurities in our railgun-driven jets was higher than expected (Hsu et al. 2012 b ; Merritt et al. 2014; Moser & Hsu 2014), leading to an unanticipated rise in ¯ (even in hydrogen experiments), causing a reduction Z in the counter-streaming ion-ion mean free path that scales as ¯ Z -4 , and a transition to a collisional regime (described in Sec. 4.3) at the originally chosen jet velocities. Thus, ongoing work is aimed at reaching collisionless regimes for the counter-streaming ions by achieving even higher counter-streaming jet velocities to overcome the rising- ¯ effect. Z Failing this, we may need to upgrade to coaxial guns (Witherspoon et al. 2009), which are expected to have far lower levels of impurities.
## 5. Future opportunities
Future opportunities on PLX with two plasma guns include further and more detailed studies of all the topics discussed in Secs. 3 and 4. Upgraded coaxial guns (e.g., Witherspoon et al. 2009) would reduce impurities and allow for higher velocities than railguns. In addition, the plasma shock and interpenetration data offer a rare opportunity to validate collisionality, EOS, and shock-handling physics models employed in MHD, multi-fluid, and kinetic plasma codes.
If PLX is upgraded to have many more guns, e.g., 36 or 60, then the merging and spherical convergence of all the jets could produce cm-, µ s-, and Mbar-scale plasmas, enabling unique fundamental physics studies of many HED topics (as described in the U.S. Dept. of Energy Report of the Workshop on High Energy Density Laboratory Physics Research Needs, Nov. 15-18, 2009) and the exploration of a standoff MIF driver for fusion energy (Hsu et al. 2012 a ).
## Acknowledgments
We thank Drs. Y. C. F. Thio and J. T. Cassibry for many useful conversations and Drs. Glen Wurden and Thomas Intrator for loaning numerous items of laboratory and diagnostic equipment, especially the multiple-frame CCD camera. This work was supported by the Office of Science (Office of Fusion Energy Sciences) and the LANL Laboratory
Table 3. Proposed reference experimental values, and evaluation of relevant physics criteria against those of Drake (2000) for a cosmically relevant collisionless shock experiment on PLX. Here, ρ i is the thermal ion gyro-radius, β the ratio of plasma thermal-to-magnetic pressure, M A the Alfv´n Mach number, e λ i the ion mean free path, R M the magnetic Reynold's number, ω ci the ion-cyclotron frequency, τ exp the shock-transit time, and c/ω pi the ion inertial length. All values given are for a quasi-perpendicular shock, and all speeds are in the laboratory frame.
| parameter | jets at collision | post-shock |
|--------------------|---------------------|---------------|
| species | H + | H + |
| density (cm - 3 ) | 3 × 10 14 | 1 . 2 × 10 15 |
| temperature (eV) | 1 | T i = 139 |
| speed (km/s) | V jet = 100 | V shock = 167 |
| magnetic field (G) | 300 (applied) | 1200 |
| length L (cm) | 50 | |
| radius R (cm) | 15 | |
| criterion | estimate for proposed experiment (post-shock values) |
|------------------------------|--------------------------------------------------------|
| 2 R/ρ i /greatermuch 1 | 30 |
| β > 1 | 4.7 |
| M A > 1 | 2.2 |
| λ i /ρ i /greatermuch 1 | 28 |
| R M /greatermuch 1 | ∼ 100 (for T e ∼ 10 eV) |
| ω ci τ exp /greatermuch 1 | 34 |
| L/ ( c/ω pi ) /greatermuch 1 | 76 |
Directed Research and Development (LDRD) Program under U.S. Department of Energy contract no. DE-AC52-06NA25396.
## REFERENCES
Adams, C. S., Lynn, A. G., Gilmore, M. A., Merritt, E. C., Moser, A. L. & Hsu, S. C. 2012 Schlieren imaging diagnostic for a collisionless shock experiment. Bull. Amer. Phys. Soc. 57 , 130.
- Awe, T. J., Adams, C. S., Davis, J. S., Hanna, D. S., Hsu, S. C. & Cassibry, J. T. 2011 One-dimensional radiation-hydrodynamic scaling studies of imploding spherical plasma liners. Phys. Plasmas 18 , 072705.
- Baker, D. A. & Hammel, J. E. 1965 Experimental studies of the penetration of a plasma stream into a transverse magnetic field. Phys. Fluids 8 , 713.
- Batteh, J. H. 1991 Review of armature research. IEEE Trans. Magn. 27 , 224.
- Bellan, P. M., You, S. & Hsu, S. C. 2005 Simulating astrophysical jets in laboratory experiments. Astrophys. Space Sci. 298 , 203.
Casanova, M., Larroche, O. & Matte, J.-P. 1991 Kinetic simulation of a collisional shock wave in a plasma. Phys. Rev. Lett. 67 , 2143.
Cassibry, J. T., Stanic, M. & Hsu, S. C. 2013 Ideal hydrodynamic scaling relations for a stagnated imploding spherical plasma liner formed by an array of merging plasma jets. Phys. Plasmas 20 , 032706.
Davis, J. S., Hsu, S. C., Golovkin, I. E., MacFarlane, J. J. & Cassibry, J. T. 2012 One-dimensional radiation-hydrodynamic simulations of imploding spherical plasma liners with detailed equation-of-state modeling. Phys. Plasmas 19 , 102701.
Degnan, J.H., Amdahl, D.J., Domonkos, M., Lehr, F.M., Grabowski, C., Robinson, P.R., Ruden, E.L., White, W.M., Wurden, G.A., Intrator, T.P., Sears, J., Weber, T., Waganaar, W.J., Frese, M.H., Frese, S.D., Camacho, J.F., Coffey, S.K., Makhin, V., Roderick, N.F., Gale, D.G., Kostora, M., Lerma, A., McCullough, J.L., Sommars, W., Kiuttu, G.F., Bauer, B., Fuelling, S.R., Siemon, R.E., Lynn,
Laboratory plasma physics experiments using merging supersonic plasma jets 11
A.G. & Turchi, P.J. 2013 Recent magneto-inertial fusion experiments on the field reversed configuration heating experiment. Nucl. Fusion 53 , 093003.
Drake, R. P. 2000 The design of laboratory experiments to produce collisionless shocks of cosmic relevance. Phys. Plasmas 7 , 4690.
Drake, R. P. 2006 High-Energy-Density-Physics . Berlin: Springer.
Fiksel, G., Fox, W., Bhattacharjee, A., Barnak, D. H., Chang, P.-Y., Germaschewski, K., Hu, S. X. & Nilson, P. M. 2014 Magnetic reconnection between colliding magnetized laser-produced plasma plumes. Phys. Rev. Lett. 113 , 105003.
Fox, W., Fiksel, G., Bhattacharjee, A., Chang, P.-Y., Germaschewski, K., Hu, S. X. & Nilson, P. M. 2013 Filamentation instability of counterstreaming laser-driven plasmas. Phys. Rev. Lett. 111 , 225002.
Gourdain, P.-A. & Seyler, C. E. 2013 Impact of the hall effect on high-energy-density plasma jets. Phys. Rev. Lett. 110 , 015002.
Haas, D. M., Bott, S. C., Kim, J., Mariscal, D. A., Madden, R. E., Eshaq, Y., Ueda, U., Collins IV, G., Gunasekera, K., Beg, F. N., Chittenden, J. P., Niasse, N. & Jennings, C. A. 2011 Supersonic jet formation and propagation in x-pinches. Astrophys. Space Sci. 336 , 33.
Hsu, S. C. 2009 Technical summary of the first U.S. plasma jet workshop. J. Fusion Energy 28 , 246.
Hsu, S. C., Awe, T. J., Brockington, S., Case, A., Cassibry, J. T., Kagan, G., Messer, S. J., Stanic, M., Tang, X., Welch, D. R. & Witherspoon, F. D. 2012 a Spherically imploding plasma liners as a standoff driver for magnetoinertial fusion. IEEE Trans. Plasma Sci. 40 , 1287.
Hsu, S. C. & Bellan, P. M. 2005 On the jets, kinks, and spheromaks formed by a planar magnetized coaxial gun. Phys. Plasmas 12 , 032103.
Hsu, S. C., Merritt, E. C., Moser, A. L., Awe, T. J., Brockington, S. J. E., Davis, J. S., Adams, C. S., Case, A., Cassibry, J. T., Dunn, J. P., Gilmore, M. A., Lynn, A. G., Messer, S. J. & Witherspoon, F. D. 2012 b Experimental characterization of railgundriven supersonic plasma jets motivated by high energy density physics applications. Phys. Plasmas 19 , 123514.
Intrator, T., Zhang, S. Y., Degnan, J. H., Furno, I., Grabowski, C., Hsu, S. C., Ruden, E. L., Sanchez, P. G., Taccetti, J. M., Tuszewski, M., Waganaar, W. J. & Wurden, G. A. 2004 A high density field reversed configuration (FRC) target for magnetized target fusion: First internal profile measurements of a high density FRC. Phys. Plasmas 11 , 2580-2585.
Jaffrin, M. Y. & Probstein, R. F. 1964 Structure of a plasma shock wave. Phys. Fluids 7 , 1658.
Ji, H., Toyama, H., Yamagishi, K., Shinohara, S., Fujisawa, A. & Miyamoto, K. 1991 Probe measurements in the REPUTE-1 reversed field pinch. Rev. Sci. Instrum. 62 , 2326. Kirkpatrick, R. C., Lindemuth, I. R. & Ward, M. S. 1995 Magnetized target fusion: An overview. Fusion Tech. 27 , 201.
Knapp, C. E. & Kirkpatrick, R. C. 2014 Possible energy gain for a plasma-liner-driven magneto-inertial fusion concept. Phys. Plasmas 21 , 070701.
Li, C. K., Ryutov, D. D., Hu, S. X., Rosenberg, M. J., Zylstra, A. B., S´guin, F. H., e Frenje, J. A., Casey, D. T., Gatu Johnson, M., Manuel, M. J.-E., Rinderknecht, H. G., Petrasso, R. D., Amendt, P. A., Park, H. S., Remington, B. A., Wilks, S. C., Betti, R., Froula, D. H., Knauer, J. P., Meyerhofer, D. D., Drake, R. P., Kuranz, C. C., Young, R. & Koenig, M. 2013 Structure and dynamics of colliding plasma jets. Phys. Rev. Lett. 111 , 235003.
Lindemuth, I. R. & Kirkpatrick, R. C. 1983 Parameter space for magnetized fuel targets in inertial confinement fusion. Nucl. Fusion 23 , 263.
Lindemuth, I. R. & Siemon, R. E. 2009 The fundamental parameter space of controlled thermonuclear fusion. Amer. J. Phys. 77 , 407.
Liu, W. & Hsu, S. C. 2011 Ideal magnetohydrodynamic simulations of unmagnetized dense plasma jet injection into a hot strongly magnetized plasma. Nucl. Fusion 51 , 073026.
Lynn, A. G., Merritt, E., Gilmore, M., Hsu, S. C., Witherspoon, F. D. & Cassibry,
- J. T. 2010 Diagnostics for the Plasma Liner Experiment. Rev. Sci. Instrum. 81 , 10E115.
Merritt, E. C., Lynn, A. G., Gilmore, M. A. & Hsu, S. C. 2012 a Multi-chord fiber-coupled interferometer with a long coherence length laser. Rev. Sci. Instrum. 83 , 033506.
Merritt, E. C., Lynn, A. G., Gilmore, M. A., Thoma, C., Loverich, J. & Hsu, S. C. 2012 b Multi-chord fiber-coupled interferometry of supersonic plasma jets. Rev. Sci. Instrum. 83 , 10D523.
Merritt, E. C., Moser, A. L., Hsu, S. C., Adams, C. S., Dunn, J. P., Holgado, A. M. & Gilmore, M. 2014 Experimental evidence for collisional shock formation via two obliquely merging supersonic plasma jets. Phys. Plasmas 21 , 055703.
Merritt, E. C., Moser, A. L., Hsu, S. C., Loverich, J. & Gilmore, M. 2013 Experimental characterization of the stagnation layer between two obliquely merging supersonic plasma jets. Phys. Rev. Lett. 111 , 085003.
Moser, A. L. & Bellan, P. M. 2012 Magnetic reconnection from a multiscale instability cascade. Nature 482 , 379.
Moser, A. L. & Hsu, S. C. 2014 Observation of ionization-mediated transition from collisionless interpenetration to collisional stagnation during merging of two supersonic plasmas. submitted ; http://arxiv.org/abs/1405.2286.
Perkins, L. J., Ho, S. K. & Hammer, J. H. 1988 Deep penetration fuelling of reactor-grade tokamak plasmas with accelerated compact toroids. Nucl. Fus. 28 , 1365.
Romero-Talam´ as, C. A., Bellan, P. M. & Hsu, S. C. 2004 Multielement magnetic probe using commercial chip inductors. Rev. Sci. Instrum. 75 , 2664.
Ross, J. S., Park, H.-S., Berger, R., Divol, L., Kugland, N. L., Rozmus, W., Ryutov, D. & Glenzer, S. H. 2013 Collisionless coupling of ion and electron temperatures in counterstreaming plasma flows. Phys. Rev. Lett. 110 , 145005.
Sagdeev, R. Z. & Kennel, C. F. 1991 Collisionless shock waves. Sci. Amer. 264 , 106.
Santarius, J. F. 2012 Compression of a spherically symmetric deuterium-tritium plasma liner onto a magnetized deuterium-tritium target. Phys. Plasmas 19 , 072705.
Settles, G. S. 2001 Schlieren and Shadowgraph Techniques . New York: Springer.
Slutz, S. A., Herrmann, M. C., Vesey, R. A., Sefkow, A. B., Sinars, D. B., Rovang, D. C., Peterson, K. J. & Cuneo, M. E. 2010 Pulsed-power-driven cylindrical liner implosions of laser preheated fuel magnetized with an axial field. Phys. Plasmas 17 , 056303. Swadling, G. F., Lebedev, S. V., Harvey-Thompson, A. J., Rozmus, W., Burdiak, G. C., Suttle, L., Patankar, S., Smith, R. A., Bennett, M., Hall, G. N., Suzuki-Vidal, F. & Yuan, J. 2014 Interpenetration, deflection, and stagnation of cylindrically convergent magnetized supersonic tungsten plasma flows. Phys. Rev. Lett. 113 , 035003.
Thio, Y. C. F., Knapp, C. E., Kirkpatrick, R. C., Siemon, R. E. & Turchi, P. J. 2001 A physics exploratory experiment on plasma liner formation. J. Fusion Energy 20 , 1.
Thio, Y. C. F., Panarella, E., Kirkpatrick, R. C., Knapp, C. E., Wysocki, F., Parks, P. & Schmidt, G. 1999 Magnetized target fusion in a spheroidal geometry with standoff drivers. In Current Trends in International Fusion Research-Proceedings of the Second International Symposium (ed. E. Panarella), p. 113. Ottawa: National Research Council of Canada.
Thoma, C., Welch, D. R. & Hsu, S. C. 2013 Particle-in-cell simulations of collisionless shock formation via head-on merging of two laboratory supersonic plasma jets. Phys. Plasmas 20 , 082128.
Witherspoon, F. D., Case, A., Messer, S. J., Bomgardner II, R., Phillips, M. W., Brockington, S. & Elton, R. 2009 A contoured gap coaxial plasma gun with injected plasma armature. Rev. Sci. Instrum. 80 , 083506. | 10.1017/S0022377814001184 | [
"S. C. Hsu",
"A. L. Moser",
"E. C. Merritt",
"C. S. Adams",
"J. P. Dunn",
"S. Brockington",
"A. Case",
"M. Gilmore",
"A. G. Lynn",
"S. J. Messer",
"F. D. Witherspoon"
] | 2014-08-01T23:59:48+00:00 | 2015-04-21T21:03:24+00:00 | [
"physics.plasm-ph"
] | Laboratory plasma physics experiments using merging supersonic plasma jets | We describe a laboratory plasma physics experiment at Los Alamos National
Laboratory that uses two merging supersonic plasma jets formed and launched by
pulsed-power-driven rail guns. The jets can be formed using any atomic species
or mixture available in a compressed-gas bottle and have the following nominal
initial parameters at the railgun nozzle exit: $n_e\approx n_i \sim 10^{16}$
cm$^{-3}$, $T_e \approx T_i \approx 1.4$ eV, $V_{\rm jet}\approx 30$-100 km/s,
mean charge $\bar{Z}\approx 1$, sonic Mach number $M_s\equiv V_{\rm
jet}/C_s>10$, jet diameter $=5$ cm, and jet length $\approx 20$ cm. Experiments
to date have focused on the study of merging-jet dynamics and the shocks that
form as a result of the interaction, in both collisional and collisionless
regimes with respect to the inter-jet classical ion mean free path, and with
and without an applied magnetic field. However, many other studies are also
possible, as discussed in this paper. |
1408.0324v1 | ## To Adjust or Not to Adjust? Sensitivity Analysis of M -Bias and Butterfly-Bias
Peng Ding and Luke Miratrix Department of Statistics, Harvard University [email protected]
Emails: [email protected] and
## Abstract
' M -Bias,' as it is called in the epidemiologic literature, is the bias introduced by conditioning on a pretreatment covariate due to a particular ' M -Structure' between two latent factors, an observed treatment, an outcome, and a 'collider.' This potential source of bias, which can occur even when the treatment and the outcome are not confounded, has been a source of considerable controversy. We here present formulae for identifying under which circumstances biases are inflated or reduced. In particular, we show that the magnitude of M -Bias in linear structural equation models tends to be relatively small compared to confounding bias, suggesting that it is generally not a serious concern in many applied settings. These theoretical results are consistent with recent empirical findings from simulation studies. We also generalize the M -Bias setting (1) to allow for the correlation between the latent factors to be nonzero, and (2) to allow for the collider to be a confounder between the treatment and the outcome. These results demonstrate that mild deviations from the M -Structure tend to increase confounding bias more rapidly than M -Bias, suggesting that choosing to condition on any given covariate is generally the superior choice. As an application, we re-examine a controversial example between Professors Donald Rubin and Judea Pearl.
Key Words : Causality; Collider; Confounding; Controversy; Covariate.
## 1 Introduction
The hallmark of an observational study is selection bias (Heckman, 1979, Copas and Li, 1997, Hern´ an, Hernandez-Diaz, and Robins, 2004). Many statisticians believe that 'there is no reason to avoid adjustment for a variable describing subjects before treatment' in observational studies (Rosenbaum, 2002, pp 76), because 'typically, the more conditional an assumption, the more generally acceptable it is' (Rubin, 2009). This advice, recently dubbed the 'pretreatment criterion' (VanderWeele and Shpitser, 2011), is widely used in empirical studies, as more covariates generally seem to make the ignorability assumption, i.e., the assumption that conditionally on the observed pretreatment covariates, treatment assignment is independent of the potential outcomes (Rosenbaum and Rubin, 1983), more plausible. And, as the validity of causal inference in observational studies relies strongly on this (untestable) assumption (Rosenbaum and Rubin, 1983), it seems reasonable to make all efforts to render it plausible.
However, other researchers (Pearl, 2009b,c, Shrier, 2008, 2009, Sj¨lander, o 2009), mainly from the causal diagram community, do not accept this view because of the possibility of a so-called M -Structure, illustrated in Figure 1(c). In sharp contrast to Rubin and Rosenbaum's advice, Pearl (2009b) and Pearl (2009c) warn practitioners that spurious bias may arise due to adjusting for a collider M in an M -Structure, even if it is a pretreatment covariate . This form of bias, typically called M -bias, a special version of so-called 'collider bias,' has since generated considerable controversy and confusion.
We attempt to resolve some of these debates by an analysis of M -bias under the causal diagram or directed acyclic graph (DAG) framework. For readers unfamiliar with the terminologies from the DAG (or Bayesian Network) literature, more details can be found in Pearl (1995) or Pearl (2009a). We here use only a small part of this larger framework. Arguably the most important structure in the DAG, and certainly the one at root of almost all controversy, is the 'V-Structure' illustrated in Figure 1(b). Here, U and W are marginally independent with a common outcome M , which shapes a 'V' with the vertex M being called a 'collider.' From a data-generation viewpoint, one might imagine Nature generating data in two steps: She first picks independently two values for U and W from two distributions, and then she combines them (possibly along with some additional random variable) to create M . Given this, conditioning on M can cause a spurious correlation between U and W , which is known as the collider bias (Greenland, 2002), or, in epidemiology, Berkson's Paradox (Berkson, 1946). Conceptually, this correlation happens because if one cause of an observed outcome is known to have not occurred, the other cause becomes more likely. Consider an automatic-timer sprinkler system where the sprinkler being on is independent of whether it is raining. Here, the weather gives no information on the sprinkler. However, given wet grass, if one observes a sunny day, one will likely conclude that the sprinklers have recently run. Correlation has been induced.
Where things get interesting is when this collider is made into a pre-treatment variable. Consider Figure 1(c), an extension of Figure 1(b). Here U and W are now also causes of the treatment T and the outcome Y , respectively. Nature, as a last, third step generates T as a function of U and some randomness, and Y as a function of W and some randomness. This structure is typically used to represent a circumstance where a researcher observes T , Y , and M in nature and is attempting to derive the causal impact of T on Y . U and W are unobserved, or latent. Clearly, the causal effect of T on Y is zero, which is also equal to the marginal association between T and Y . If a researcher regressed Y on T , he or she would obtain a zero in expectation, which is correct for estimating the causal effect. But perhaps there is a concern that M , a pretreatment covariate, may be a confounder that is masking a treatment effect. Typically, one would then 'adjust' for M to take this possibility into account, e.g., by including M in a regression or by matching units on similar values of M . If we do this in this circumstance, however, then we will not find a zero causal effect, in expectation. This is the so-called ' M -Bias,' and this special structure is called the ' M -Structure' in the DAG literature.
Previous qualitative analysis for binary variables shows that collider bias generally tends to be small (Greenland, 2002), and simulation studies (Liu, Brookhart, Schneeweiss, Mi, and Setoguchi, 2012) again demonstrate that M -Bias is small in many realistic settings. While mathematically describing the magnitudes of M -Bias in general models is intractable, it is possible to derive exact formulae of the biases as functions of the correlation coefficients in linear structural equation models (LSEMs). The LSEM has a long history in statistics (Wright, 1921, 1934) to describe dependence among multiple random variables. Sprites (2002) uses linear models to illustrate M -Bias in observational studies, and Pearl (2013) also utilize the transparency of such linear models to examine various types of causal phenomena, biases, and paradoxes. We here extend these works and provide exact formulae for biases, allowing for a more detailed quantitative analysis of M -bias.
While M -Bias does exist when the true underlying data generating process (DGP) follows the exact M -Structure, it might be rather sensitive to various deviations from the exact M -Structure. Furthermore, some might argue that an exact M -Structure is unlikely to hold in practice. Gelman (2011), for example, doubts the exact independence assumption required for the M -Structure in the social sciences by arguing that there are '(almost) no true zeros' in this discipline. Indeed, since U and W are often latent characteristics of the same individual, the independence assumption U W is a rather strong structural assumption. Furthermore, it might be plausible that the pretreatment covariate M is also a confounder between, i.e., has some causal impact on both, the treatment and outcome. We extend our work by accounting for these departures from a pure M -Structure, and find that even slight departures from the M -Structure can dramatically change the forms of the biases.
This paper theoretically compares the bias from conditioning on an M to not under several scenarios and finds that M -Bias is indeed small relative to other concerns unless there is a strong correlation structure for the variables. We further show that these findings extend to a binary treatment regime as well. This argument proceeds in several stages. First, in Section 2, we examine a pure M -Structure and introduce our LSEM framework. We then discuss the cases when the latent variables U and W may be correlated and M may also be a confounder between the treatment T and the outcome Y . In Section 3, we generalize the results in Section 2 to a binary treatment. In Section 4, we illustrate the theoretical findings using a controversial example between Professors Donald Rubin and Judea Pearl (Rubin, 2007, Pearl, 2009c). Section 5 discusses the relevance of our findings by examining M -Bias in actual practice and by comparing asymptotic to finite sample properties. We conclude with a brief discussion and present all technical details in the Appendix.
## 2 M -Bias and Butterfly-Bias in LSEMs
We begin by examining pure M -Bias in a LSEM. As our primary focus is bias, we assume data are ample and that anything estimable is estimated with nearly perfect precision. In particular, when we say we obtain a result from a regression, we implicitly mean we obtain that result in expectation; in practice an estimator will be near the given quantities. We do not compare relative uncertainties of different estimators given the need to estimate more or fewer parameters. There are likely degrees-of-freedom issues that would implicitly advocate using estimators with fewer parameters, but in the circumstances considered here these concerns are likely to be minor as all the models have few parameters.
A causal DAG can be viewed as a hierarchical DGP. In particular, any variable on the graph can be viewed as a function of its parents and some additional noise, i.e., if R had parents A B , , and C , we would have
$$R = f ( A, B, C, \varepsilon _ { R } ) \text{ with } \varepsilon _ { R } \bigsqcup ( A, B, C ).$$
Generally noise terms such as e R are considered to be independent from each other, but they can also be given an unknown correlation structure corresponding to earlier variables not explicitly included in the diagram. This is typically represented by drawing the dependent noise terms jointly from some multivariate distribution. This framework is quite general; we can represent any distribution that can be factored as a product of conditional distributions corresponding to a DAG (which is one representation of the Markov Condition, a fundamental assumption for DAGs).
LSEMsare special cases of the above with additional linearity and additivity constraints. For simplicity, and without loss of generality, we also rescale all primary variables ( U W M T Y , , , , ) to have zero mean and unit variance. For example, consider this data generating process corresponding to Figure 1(a):
$$\begin{cases} \begin{array} { c c } M, \epsilon _ { T }, \epsilon _ { Y } & \text{ind} \\ T & = & a M + \sqrt { 1 - a ^ { 2 } } \epsilon _ { T }, \\ Y & = & b T + c M + \sqrt { 1 - b ^ { 2 } - c ^ { 2 } } \epsilon _ { Y }, \end{array} \end{cases}$$
where we use A ∼ [ 0 1 , ] to denote a random variable with mean zero and variance one.
In the causal DAG literature, we think about causality as reaching in and fixing a given node to a set value, but letting Nature take her course otherwise. For example, if we were able to set T at t , the above data generation process would be transformed to:
$$\begin{cases} \begin{array} { c c } M, \varepsilon _ { T }, \varepsilon _ { Y } & \text{ind} \\ T & = & t, \\ Y & = & b t + c M + \sqrt { 1 - b ^ { 2 } - c ^ { 2 } } \varepsilon _ { Y }. \end{array} \end{cases}$$
The previous cause, M , of T has been broken, but the impact of T on Y remains intact. This changes the distribution of Y but not M . More importantly, this results in a distribution distinct from that of conditioning on T = t . Consider the case of positive a b , , and c . If we observe a high T , we can infer a high M (as T and M are correlated) and a high Y due to both the bT and cM terms in Y 's equation. However, if we set T to a high value, M is unchanged. Thus, while we will still have the large bT term for Y , the cM term will be 0 in expectation. Thus, the expected value for Y will be less.
This setting as compared to conditioning is represented with the 'do' operator. Given the 'do' operator, we define a local causal effect of T on Y at T = t as:
$$\tau _ { t } = \frac { \partial \mathbb { E } \{ Y \ | \, \text{do} ( T ) = t \} } { \partial t }.$$
For linear models, the local causal effect is a constant, and thus we do not need to specify t . We use 'do' here purely to indicate the different distributions. For a more technical overview, see Pearl (1995) or Pearl (2009a). Our results, with more formality, can easily be expressed in this more technical notation.
Figure 2: M -Structure with Possibly Correlated Hidden Causes
If we extend the M -Structure in Figure 1(c) by allowing possible correlation between the two hidden causes U and W , we obtain the DAG in Figure 2. This in turn gives the following DGP:
$$\begin{cases} \begin{array} { c } \varepsilon _ { M }, \varepsilon _ { T }, \varepsilon _ { Y } & \text{ind} \\ ( U, W ) & \end{array} \begin{array} { c } \text{$\varepsilon_{\text{$M}$} } & [ 0, 1 ], \\ ( U, W ) & \end{array} \begin{array} { c } \text{$\varepsilon_{\text{$O}$}, 1, 1, \rho ], \\ M & = & b U + c W + \sqrt{1-b^{2}-c^{2}} \varepsilon _ { M }, \\ T & = & a U + \sqrt{1-a^{2}} \varepsilon _ { T }, \\ Y & = & d W + \sqrt{1-d^{2}} \varepsilon _ { Y }, \end{array} \begin{array} { c } \text{$\varepsilon_{\text{$O}$}, 1, 1, \rho ], \\ \end{array} \begin{array} { c } \text{$\varepsilon_{\text{$O}$}, 1, 1, \rho ], \\ \end{array} \begin{array} { c } \text{$\varepsilon_{\text{$O}$}, 1, 1, \rho ], \\ \end{array} \begin{array} { c } \text{$\varepsilon_{\text{$O}$}, 1, 1, \rho ], \\ \end{array} \begin{array} { c } \text{$\varepsilon_{\text{$O}$}, 1, 1, \rho ], \\ \end{array} \begin{array} { c } \text{$\varepsilon_{\text{$O}$} }. \end{array}.$$
where we use ( A B , ) ∼ [ 0 0;1 1 , , , r ] to denote a bivariate random vector with means zero, variances one and correlation coefficient r .
Here, the true causal effect of T on Y is zero, namely, t t = 0 for all t . The unadjusted estimator for the causal effect obtained by regressing Y onto T is the same as the covariance between T and Y :
$$\text{Bias} _ { u n a d j } = \text{Cov} ( T, Y ) = \text{Cov} ( a U, d W ) = a d \text{Cov} ( U, W ) = a d \rho.$$
The adjusted estimator (see Lemma 2 in Appendix A for a proof) obtained by regressing Y onto ( T M , ) is
$$\text{Bias} _ { a d j } = \frac { a d \rho ( 1 - b ^ { 2 } - c ^ { 2 } - b c \rho ) - a b c d } { 1 - ( a b + a c \rho ) ^ { 2 } }.$$
The results above and some of the results discussed later in this paper can be obtained directly from traditional path analysis (Wright, 1921, 1934). However, we provide elementary proofs, which can easily be extended to binary treatment, in the Appendix. If we allowed for a treatment effect, our results would remain essentially unchanged; the only difference would be due to restrictions on the correlation terms needed to maintain unit variance for all variables.
The above can also be expressed in the potential outcomes framework (Neyman, 1923/1990, Rubin, 1974). In particular, for a given unit let Nature draw e M , e T , e Y , U , and V as before. Let T be the 'natural treatment' for that unit, i.e., what treatment it would receive sans intervention. Then calculate Y t ( ) for any t of interest using the 'do' operator. These are what we would see if we set T = t . How Y t ( ) changes for a particular unit defines that unit's collection of potential outcomes. Then E { Y t ( ) } for some t is the expected potential outcome over the population for a particular t . We can examine the derivative of this function as above to get a local treatment effect. This connection is exact: the findings in this paper are the same as what one would find using this DGP and the potential outcomes framework. We here examine regression as the estimator. Note that matching would produce identical results as the amount of data grew (assuming the data generating process ensures common support, etc.).
Exact M -Bias. The M -Bias originally considered in the literature is the special case where the correlation coefficient between U and W is r = 0. In this case, the unadjusted estimator is unbiased and the absolute bias of the adjusted estimator is | abcd | / { 1 -( ab ) 2 } . With moderate correlation coefficients a b c d , , , the denominator 1 -( ab ) 2 is close to one, and the bias is close to -abcd . Since abcd is a product of four correlation coefficients, it can be viewed as a 'higher order bias.' For example, if a = b = c = d = 0 2, then 1 . -( ab ) 2 = 0 9984 . ≈ 1, and the bias of the adjusted estimator is -abcd / { 1 -( ab ) 2 } = -0 0016 . ≈ 0; if a = b = c = d = 0 3, then 1 . -( ab ) 2 = 0 9919 . ≈ 1, and the bias of the adjusted estimator is -abcd / { 1 -( ab ) 2 } = -0 0082 . ≈ 0 . Even moderate correlation results in little bias.
In Figure 3, we plot the bias of the adjusted estimator as a function of the correlation coefficients, and let these coefficients change to see how the bias changes. In the first subfigure, we assume all the correlation coefficients have the same magnitude ( a = b = c = d ), and we plot the absolute bias of the adjusted estimator versus a . The constraints on variance and correlation only allow for some combinations of values for a b c , , and d which limits the domain of the figures. In this case, for example, | a | ≤ √ 2 2 due to the requirement that / b 2 + c 2 = 2 a 2 ≤ 1. Other figures have limited domains due to similar constraints. In the second subfigure of Figure 3, we assume that M is more predictive to the treatment T than to the outcome Y , with a = b = 2 c = 2 d . In the third subfigure of Figure 3, we assume that M is more predictive to the outcome Y , with 2 a = 2 b = c = d . The biases are generally very small within wide ranges of the feasible regions of the correlation coefficients. However, the biases do blow up when the correlation coefficients are extremely large. Near the boundary of the feasible regions in Figure 3, the M -Structure is approximately deterministic, which is rare in social sciences. Pearl (2009b) does not exclude the worst cases, and thus he considers M -Bias as a severe problem.
Figure 3: M -Bias with Independent U and W . The three subfigures correspond to the cases when ( U W , ) are equally/more/less predictive to the treatment than to the outcome. In each subfigure, we show the proportions of the areas where the adjusted estimator has a bias smaller than 0 01. .
In Figure 4(a), we assume a = b and c = d and examine a broader range of relationships. Here, the grey area satisfies | Bias ad j | < min ( | a | , | c | ) / 20. For example, when the absolute values of the correlation coefficients are smaller than 0 5 . (the square with dashed boundary in Figure 4(a)), the corresponding area is almost grey, implying small bias.
Due to the four dimensional sensitivity parameters ( a b c d , , , ) , a full exploration and graphical illustration over all possible values of the sensitivity parameters is formidably hard. In the absence of prior knowledge about the DAG, our sensitivity analysis here is based on some simplifications (e.g., a = b and c = d ), which may reflect some real situations. Using the bias formulae in this paper, we can easily conduct sensitivity analysis for other parameter combinations, depending on our practical problem and background knowledge about the DAG.
As a side note, Pearl (2013) noticed a surprising fact: the stronger the correlation between T and M , the larger the absolute bias of the adjusted estimator, since the absolute bias is monotone increasing in | ab | . From the second and the third subfigure of Figure 3, we see that when M is more predictive of the treatment, the biases of the adjusted estimator indeed tends to be larger.
(a) Pure M -Bias. Within the grey region, the absolute bias of the adjusted estimator is less than 1 / 20 of the minimum of | a | (= | b | ) and | c | (= | d | ) .
(b) M -Bias with Correlated U and W . Within the grey region, the adjusted estimator is superior.
Figure 4: M -Bias under Different Scenarios
glyph[negationslash]
Correlated Latent Variables. When the latent variables U and W are correlated with r = 0, both the unadjusted and adjusted estimators may be biased. The ques- tion then becomes: which is worse? The ratio of the absolute biases is
$$\left | \frac { \text{Bias} _ { a d j } } { \text{Bias} _ { u n a d j } } \right | = \left | \frac { \rho ( 1 - b ^ { 2 } - c ^ { 2 } - b c \rho ) - b c } { \rho \left \{ 1 - ( a b + a c \rho ) ^ { 2 } \right \} } \right |,$$
which does not depend on d (the relationship between W and Y ). For example, if the correlation coefficients a b c , , , r all equal 0 2, the ratio above is 0 714; in this . . case the adjusted estimator is superior to the unadjusted one by a factor of 1 4. . Figure 4(b) compares this ratio to 1 for all combinations of r and a (= b = ) c . Generally, the adjusted estimator has smaller bias except when a b , , and c are quite large.
In Figure 5, we again assume a = b = c = d and investigate the absolute biases as functions of a for fixed r at 0 1 . , 0 2 . , and 0 4 . . When the correlation coefficients a (= b = c = ) d are not dramatically larger than r , the adjusted estimator has smaller bias than the unadjusted one.
glyph[negationslash]
Figure 5: M -Bias with correlated U and W when r = 0 with a = b = c = d . In each subfigure, we show the proportion of the areas where the adjusted estimator has a smaller bias than the unadjusted estimator.
glyph[negationslash]
The Disjunctive Cause Criterion. In order to remove biases in observational studies, VanderWeele and Shpitser (2011) propose a new 'disjunctive cause criterion' for selecting confounders, which requires controlling for all the covariates that are either causes of the treatment, causes of the outcome, or causes of both. According to the 'disjunctive cause criterion,' when r = 0, we should control for
( U W , ) if possible. Unfortunately, neither of ( U W , ) is observable. However, controlling the 'proxy variable' M for ( U W , ) may reduce bias when r is relatively large. In the special case with b = 0, the ratio of the absolute biases is
$$\left | \frac { \text{Bias} _ { a d j } } { \text{Bias} _ { u n a d j } } \right | = \frac { 1 - c ^ { 2 } } { 1 - ( a c \rho ) ^ { 2 } } \leq 1 ;$$
in another special case with c = 0, the ratio of the absolute biases is
$$\left | \frac { \text{Bias} _ { a d j } } { \text{Bias} _ { u n a d j } } \right | = \frac { 1 - b ^ { 2 } } { 1 - ( a b ) ^ { 2 } } \leq 1.$$
Therefore, if either U or W is not causative to M , the adjusted estimator is always better than the unadjusted one.
Butterfly-Bias: M -Bias with Confounding Bias. Models, especially in the social sciences, are approximations. They rarely hold exactly. In particular, for any covariate M of interest, there is likely to be some concern that M is indeed a confounder, even if it is also a possible source of M -Bias. If we let M both be a confounder as well as the middle of an M -Structure we obtain a 'Butterfly-Structure' (Pearl, 2013) as shown in Figure 6. In this circumstance, conditioning will help with confounding bias, but hurt with M -Bias. Ignoring M will not resolve any confounding, but will avoid M -Bias. The question then becomes that of determining which is the lesser of the two evils.
Figure 6: Butterfly-Structure
We can examine this trade-off for a LSEM corresponding to Figure 6. The DGP is given by the following equations:
$$\begin{cases} \begin{array} { c c c } U, W, \varepsilon _ { M }, \varepsilon _ { T }, \varepsilon _ { Y } & \underset { \sim } { \text{$ind} } \\ M & = & b U + c W + \sqrt { 1 - b ^ { 2 } - c ^ { 2 } } \varepsilon _ { M }, \\ T & = & a U + e M + \sqrt { 1 - a ^ { 2 } - e ^ { 2 } } \varepsilon _ { T }, \\ Y & = & d W + f M + \sqrt { 1 - d ^ { 2 } - f ^ { 2 } } \varepsilon _ { Y }. \end{array} \end{cases}$$
Again, the true causal effect of T on Y is zero. The unadjusted estimator obtained by regressing Y onto T is the covariance between T and Y :
$$\text{Bias} _ { u n a d j } = \text{Cov} ( T, Y ) = a b f + c d e + e f.$$
It is not, in general, zero, implying bias. The adjusted estimator (see Lemma 3 in Appendix for a proof) obtained by regressing Y onto ( T M , ) has bias
$$\text{Bias} _ { a d j } = - \frac { a b c d } { 1 - ( a b + e ) ^ { 2 } }.$$
If the values of e and f are relatively high (i.e., M has a strong effect on both T and Y ), the confounding bias is large and the unadjusted estimator will be severely biased. For example, if a b c d e , , , , , and f all equal 0 2, the bias of the unadjusted . estimator is 0 056, but the bias of the adjusted estimator is only . -0 0017, an order . of magnitude smaller. Generally, the largest term for the unadjusted bias is the second-order term of e f , while the adjusted bias only has, ignoring the denominator, a fourth-order term of abcd . This suggests adjustment is generally preferable and that M -bias is in some respect a 'higher order bias.'
Detailed comparison of the ratio of the biases is difficult, since we can vary six parameters ( a b c d e , , , , , f ) . In Figure 7(a), we assume all the correlation coefficients have the same magnitude, and plot bias for both estimators as a function of the correlation coefficient within the feasible region, defined by the restrictions -√ 2 2 / < a < ( -1 + √ 5 ) / 2, due to the restrictions
$$b ^ { 2 } + c ^ { 2 } < 1, \, a ^ { 2 } + e ^ { 2 } < 1, \, d ^ { 2 } + f ^ { 2 } < 1, \, \text{and} \, | a ^ { 2 } + e | < 1.$$
Within 74 9% of the feasible region, the adjusted estimator has smaller bias than the . unadjusted one. The unadjusted estimator only has smaller bias than the adjusted estimator when the correlation coefficients are extremely large. In Figure 7(b), we assume a = b = c = d and e = f , and compare | Bias ad j | and | Bias unad j | within the feasible region of ( a e , ) defined by (2). We can see that the adjusted estimator is superior to the unadjusted one for 71% (colored in grey in Figure 7(b)) of the feasible region. In the area satisfying | e | > | a | in Figure 7(b), where the connection between M to T and Y is stronger than the other connections, the area is almost entirely grey suggesting that the adjusted estimator is preferable. This is sensible because here the confounding bias has larger magnitude than the M -Bias. In the area satisfying | a | < | e | , where M -bias is stronger than confounding bias, the unadjusted estimator is superior for some values, but still tends to be inferior when the correlations are roughly the same size.
(a) Absolute biases of both estimators with a = b = c = d = e = f .
(b) Comparison of the absolute biases with a = b = c = d and e = f . Within 71 2% (in grey) of the feasible region, the . adjusted estimator has smaller bias than the unadjusted one.
Figure 7: Butterfly-Bias
## 3 Extensions to a Binary Treatment
One might worry that the conclusions in the previous section are not applicable for a binary treatment. It turns out, however, that they are. In this section, we extend the results in Section 2 to binary treatments by representing the treatment through a latent Gaussian variable as shown in Figure 8.
Correlated Latent Variables. We extend Figure 2 to Figure 8(a). Here, T ∗ is the old T . The generating equations for T and T ∗ become
$$T = I ( T ^ { * } \geq \alpha ), \text{ and } T ^ { * } = a U + \sqrt { 1 - a ^ { 2 } } \epsilon _ { T }.$$
Other variables and noise terms remain the same. Although it might be relaxed, we make reference to the Normally assumption of the error terms for mathematical simplicity. The intercept a determines the proportion of the individuals receiving the treatment: F ( -a ) = P T ( = ) 1 , where F ( ) · is the cumulative distribution function of a standard Normal distribution. When a = 0, the number of individuals
U
a
ρ
M
T
W
d
Y
T*
(a) Correlated Hidden Causes
Figure 8: DAGs for Binary Treatment
exposed to the treatment and control are balanced; when a < 0, more individuals are exposed to the treatment; when a > 0, the reverse.
The true causal effect of T on Y is again zero. Let f ( ) = · F ′ ( ) · and h a ( ) ≡ f a ( ) / { F ( a ) F ( -a ) } . Then Lemma 6 in Appendix shows that the unadjusted estimator has bias
$$\text{Bias} _ { u n a d j } = a d \rho \eta ( \alpha ),$$
and the adjusted estimator has bias
$$\text{Bias} _ { a d j } = \frac { a d \eta ( \alpha ) \{ \rho ( 1 - b ^ { 2 } - c ^ { 2 } - b c \rho ) - b c \} } { \rho \{ 1 - ( a b + a c \rho ) ^ { 2 } \phi ( \alpha ) \eta ( \alpha ) \} }.$$
When r = 0, the unadjusted estimator is unbiased, but the adjusted estimator has bias
$$- \frac { a b c d \eta ( \alpha ) } { 1 - ( a b ) ^ { 2 } \phi ( \alpha ) \eta ( \alpha ) }.$$
glyph[negationslash]
When r = 0, the ratio of the absolute biases is
$$\left | \frac { \text{Bias} _ { a d j } } { \text{Bias} _ { u n a d j } } \right | = \left | \frac { \rho ( 1 - b ^ { 2 } - c ^ { 2 } - b c \rho ) - b c } { \rho \{ 1 - ( a b + a c \rho ) ^ { 2 } \phi ( \alpha ) \eta ( \alpha ) \} } \right |.$$
The patterns for a binary treatment do not differ much from a continuous treatment. As before, if the correlation coefficient is moderately small, the M -Bias also tends to be small. As shown in Figure 9 (analogous to Figure 5), when | r | is comparable to | a | (= | b | = | c | = | d | ) , the adjusted estimator is less biased than the unadjusted estimator. Only when | a | is much larger than | r | is the unadjusted estimator superior.
b
c
Figure 9: M -Bias with correlated ( U W , ) and binary treatment. Compare to Figure 5
Butterfly-Bias with a Binary Treatment. We can extend the LSEM ButterflyBias setup to binary treatment just as we extended the M -Bias setup. Compare Figure 8(b) to Figure 6. T becomes T ∗ and T is built from T ∗ as above. The structural equations for T and T ∗ for butterfly bias in the binary case are then
$$T = I ( T ^ { * } \geq \alpha ), \text{ and } T ^ { * } = a U + e M + \sqrt { 1 - a ^ { 2 } - e ^ { 2 } } \varepsilon _ { T }.$$
The other equations and variables are the same as before.
Although the true causal effect of T on Y is zero, Lemma 7 in Appendix shows that the unadjusted estimator has bias
$$\text{Bias} _ { u n a d j } = ( c d e + a b f + e f ) \eta ( \alpha ),$$
and the adjusted estimator has bias
$$\text{Bias} _ { a d j } = - \frac { a b c d \eta ( \alpha ) } { 1 - ( a b + e ) ^ { 2 } \phi ( \alpha ) \eta ( \alpha ) }.$$
Therefore, the ratio of the absolute biases is
$$\left | \frac { \text{Bias} _ { a d j } } { \text{Bias} _ { u n a d j } } \right | = \left | \frac { a b c d \eta ( \alpha ) } { ( c d e + a b f + e f ) \{ 1 - ( a b + e ) ^ { 2 } \phi ( \alpha ) \eta ( \alpha ) \} } \right |.$$
Complete investigation of the ratio of the biases is intractable with seven varying parameters ( a b c d e , , , , , f , a ) . However, in the very common case with a = 0, which gives equal-sized treatment and control groups, we again find trends similar to the continuous treatment case. See Figure 10. As before, only in the cases with very small e (= f ) but large a (= b = c = ) d , does the unadjusted estimator tend to be superior. Within a reasonable region of a , these patterns are quite similar.
(a) Absolute biases with a = b = c = d = e = f .
(b) Comparison of the absolute bias with a = b = c = d and e = f . Within 74 4% . (in grey) of the feasible region, the adjusted estimator is better than the unadjusted estimator.
Figure 10: Butterfly-Bias with a Binary Treatment
## 4 Illustration: The Rubin-Pearl Controversy
Pearl (2009c) cites Rubin (2007)'s example about the causal effect of smoking habits ( T ) on lung cancer ( Y ), and argues that conditioning on the pretreatment covariate 'seat-belt usage' ( M ) would introduce spurious associations, since M could be reasonably thought of as an indicator of a person's attitudes toward societal norms ( U ) as well as safety and health related measures ( W ). Assuming all the analysis is already conditioned on other observed covariates, we focus our discussion on the five variables ( U W M T Y , , , , ) , of which the dependence structure is illustrated by Figure 11. Since the patterns with a continuous treatment and a binary treatment are similar, we focus our discussion on LSEMs.
As Pearl (2009c) points out,
If we have good reasons to believe that these two types of attitudes are marginally independent, we have a pure M-structure on our hand.
In the case with r = 0, conditioning on M will lead to spurious correlation between T and Y under the null, and will bias the estimation of the causal effect of T on Y . However, Pearl (2009c) also recognizes that the independence assumption seems very strong in this example, since U and W are both background variables about the habit and personality of a person. Pearl (2009c) further argues:
But even if marginal independence does not hold precisely, conditioning on 'seat-belt usage' is likely to introduce spurious associations, hence bias, and should be approached with caution.
Although we believe most things should be approached with caution, our work, above, suggests that even mild perturbations of an M -Structure can switch which of the two approaches, conditioning or not conditioning, is likely to remove more bias. In particular, Pearl (2009c) is correct in that the adjusted estimator indeed tends to introduce more bias than the unadjusted one when an exact M -Structure holds and thus the general advice 'to condition on all observed covariates' may not be sensible in this context. However, in the example of Rubin (2007), the exact independence between a person's attitude toward societal norms U and safety and health related measures W is questionable, since we have good reasons to believe that other hidden variables such as income and family background will affect both U and W simultaneously, and thus Pearl's fears may be unfounded.
To examine this further, we consider two possible deviations from the exact M -Structure, and investigate the biases of the unadjusted and adjusted estimators for each.
- (a) (Correlated U and W ) Assume the DGP follows the DAG in Figure 11, with an additional correlation between the attitudes U and W as shown in Figure 2. If we then assume that all the correlation coefficients have the same positive magnitude, earlier results demonstrate that the adjusted estimator is preferable as it strictly dominates the unadjusted estimator except for extremely large values of the correlation coefficients.
- Furthermore, in Rubin (2007)'s example, attitudes toward societal norms U are more likely to affect the 'seat-belt usage' variable M than safety and health
An Arrow from W to T
- (a) M -Bias, with possible deviations: correlated U and V , and an additional arrow from W to T
- (b) Biases of resulting unadjusted and adjusted estimators with a = b = c = d = g .
Figure 11: Sensitivity Analysis of Pearl (2009c)'s Critique on Rubin (2007)
related measures W , which further strengthens the case for adjustment. If we were willing to assume that c is zero but r is not, equation (1) in Section 2 again shows that the adjusted estimator is superior.
- (b) (An arrow from W to T ) Pearl's example seems a bit confusing on further inspection, even if we accept his independence assumption U W . In particular, one's 'attitudes towards safety and health related measures' likely impact one's decisions about smoking. Therefore, we might reasonably expect an arrow from W to T . In Figure 11(a), we remove the correlation between U and W , but we allow an arrow from W to T , i.e., the generating equation for T becomes T = aU + gW + √ 1 -a 2 -g 2 e T . Lemma 8 in Appendix gives the associated formulae for biases of the adjusted and unadjusted estimators. Figure 11(b) shows that, assuming a = b = c = d = g (i.e., equal correlations), the adjusted estimator is uniformly better.
## 5 Two Further Issues
One controversy about M -Bias is whether M -Structure is rare or not in practice, and we go through several examples to discuss this issue. In the second part of this section, we make a distinction between asymptotic and finite sample properties of M -Bias.
Is M -Structure Rare? Although Pearl (2009c) argues that M -Bias is a structural property, Rubin (2009) claims that M -bias is a rare phenomenon such as 'trying to balance a multidimensional cone on its point with no external supports in some visible directions.' As mentioned in the introduction, Gelman (2011) argues that, in social sciences, 'true zeros' are rare and consequently the independence structure in the exact M -Structure is also rare. Section 4 revisited the controversial example between Professors Pearl and Rubin, and Figure 11(a) illustrated two possible deviations from the exact M -Structure. Both deviations suggested conditioning is a superior choice. In the following, we review three other examples of M -Bias in the current literature, and investigate the plausibility of the exact M -Structure.
(a) Glymour (2006)
(b) Kelcey and Carlisle (2011)
(c) Liu et al. (2012)
Figure 12: Three M -Structures with Possible Deviations in Dashed Arrows
As shown in Figure 12(a), Glymour (2006) postulates a possible M -Structure with exposure 'low income,' outcome 'diabetes,' and M variable 'mother had diabetes,' where 'family income during childhood' affects both exposure and M , and 'mother's genetic diabetes risk' affect both outcome and M . However, this M -Structure is subject to several plausible deviations: 'mother's genetic diabetes risk' may affect 'family income during childhood; 'mother had diabetes' may affect 'low education;' and 'family income during childhood' may affect 'diabetes.'
Kelcey and Carlisle (2011) have ( T Y M U W , , , , ) as 'teacher's instructional reading practice,' 'student's reading comprehension achievement,' 'teacher's reading knowledge,' 'professional development in reading,' and 'teacher's general knowledge.' See Figure 12(b). However, this M -Structure is dubious because of the possible correlation between the latent ( U W , ) and the confounding effect of M on the relationship between T and Y .
Figure 12(c) is a possible M -Structure investigated by Liu et al. (2012), where ( T Y M U W , , , , ) are 'use of selective serotonin reuptake inhibitors (SSRI),'
'lung cancer,' 'coronary artery disease,' 'depression,' and 'ever smoker.' Although it is plausible that 'coronary artery disease' is not the confounder between 'use of SSRI' and 'lung cancer,' it is very likely that 'depression' affects both 'ever smoker' and 'lung cancer.'
In summary, while all the examples above were quite useful to illustrate M -Bias in theoretical research, it is unwise to believe that these M -Structures are exact based on our background knowledge. Therefore, we suggest researchers conduct sensitivity analysis, such as illustrated earlier, according to their scientific knowledge about the structure of the DAG and the associated parameters.
Asymptotic versus Finite Sample Properties. The discussion in the previous sections are mainly based on asymptotic theory assuming large samples. As argued by Pearl (2009b), this approach allows for investigating the existence of bias in a certain DAG, and asymptotic analysis helps reveal the structural property of a DAG. A referee pointed out that the asymptotic theory is quite different from the more practical finite sample theory. In finite sample data analysis, practitioners, often interested in interval estimation and hypothesis testing, are typically more interested in whether associated confidence intervals cover the true causal parameters at nominal rates, and whether tests for null hypotheses about the causal effect have valid size. These questions are related to the asymptotic property of the DAGs, but also depend on the sample size, the procedure for constructing confidence interval, and choice of test statistic. Theoretical discussion of the finite sample theory is unfortunately more difficult. Simulation study, however, is an alternative tool for these questions. Some studies exist. In particular, Liu et al. (2012) simulate large cohort studies under an M -Structure corresponding to their science question of interest, and find that the impact of M -Bias was small for most of their 178 scenarios unless the association between M and the unmeasured confounders is very large.
## 6 Discussion
For objective causal inference, Rubin and Rosenbaum suggest balancing all the pretreatment covariate in observational studies to parallel with the design of randomized experiments (Rubin, 2007, 2008, 2009, Rosenbaum, 2002), which is called the 'pretreatment criterion' (VanderWeele and Shpitser, 2011). However, Pearl and other researchers (Pearl, 2009b,c, Shrier, 2008, 2009, Sj¨lander, 2009) criticize the o 'pretreatment criterion' by pointing out that this criterion may lead to biased inference in presence of a possible M -Structure even if the treatment assignment is unconfounded. We investigate this controversy in detail for LSEMs, ideally providing a template for future research about more general DAGs (e.g., nonparametric and nonlinear models). While we agree that Pearl's warning is very insightful, our asymptotic theory shows that, at least for LSEMs, this conclusion is quite sensitive to various deviations from the exact M -Structure, e.g., to circumstances where latent causes may be correlated or the M variable may also be a confounder between the treatment and the outcome. We also go through several candidate M -Structures in the existing literature, and find that exact M -Structure is likely to be rare with various deviations typically being more plausible. Overall, this coupled with our asymptotic theory suggests that for linear systems, except in some extreme cases, adjusting for all the pretreatment covariates is in fact a reasonable choice.
## Acknowledgment
The authors thank all the participants in the 'Causal Graphs in Low and High Dimensions' seminar at Harvard Statistics Department in Fall, 2012, and thank Professor Peter Spirtes for sending us his slides (Sprites, 2002). Comments from the associate editor and two reviewers greatly improved the quality of our paper.
## Appendix: Lemmas and Proofs
Lemma 1 In the linear regression model Y = b 0 + b T T + b MM + e with e ( T M , ) and E ( e ) = 0 , we have
$$\beta _ { T } \ = \ \frac { C o v ( Y, T ) V a r ( M ) - C o v ( Y, M ) C o v ( M, T ) } { V a r ( T ) V a r ( M ) - C o v ^ { 2 } ( M, T ) }.$$
Proof. Solve for ( b T , b M ) using the following moment conditions
$$\begin{cases} \text{ Cov} ( Y, T ) & = & \beta _ { T } \text{Var} ( T ) + \beta _ { M } \text{Cov} ( M, T ), \\ \text{ Cov} ( Y, M ) & = & \beta _ { T } \text{Cov} ( M, T ) + \beta _ { M } \text{Var} ( M ). \end{cases}$$
Lemma 2 Under the model generated by Figure 2, the regression coefficient of T by regressing Y onto ( T M , ) is
$$\beta _ { T } = \frac { a d \rho ( 1 - b ^ { 2 } - c ^ { 2 } - b c \rho ) - a b c d } { 1 - ( a b + a c \rho ) ^ { 2 } }.$$
Proof. We apply Lemma 1, where all variance terms such as Var ( M ) are 1, and the covariance terms are easily calculated. For example, we have Cov ( Y T , ) = Cov ( dW aU , ) = ad r , and Cov ( Y M , ) = Cov ( dW bU , + cW ) = bd r + cd .
Lemma 3 Under the model generate by Figure 7, the regression coefficient of T from regressing Y onto ( T M , ) is
$$\beta _ { T } = - \frac { a b c d } { 1 - ( a b + e ) ^ { 2 } }.$$
Proof. Similar to the proof of Lemma 2, we apply Lemma 1.
Lemma 4 Assume that ( X 1 , X 2 ) follows a bivariate Normal distribution with means zero, variances one, and correlation coefficient r . Then E ( X 1 | X 2 ≥ z ) -E ( X 1 | X 2 < z ) = r h ( ) z , where h ( ) = z f ( ) z / { F ( ) z F ( - } z ) .
Proof. Since X 1 = rX 2 + √ 1 -r 2 Z with Z ∼ N ( 0 1 , ) and Z X 2, we have
$$\mathbb { E } ( X _ { 1 } \ | \ X _ { 2 } \geq z ) = r \mathbb { E } ( X _ { 2 } \ | \ X _ { 2 } \geq z ) = \frac { r } { \Phi ( - z ) } \int _ { z } ^ { \infty } x \phi ( x ) d x = - \frac { r } { \Phi ( - z ) } \int _ { z } ^ { \infty } d \phi ( x ) = r \frac { \phi ( z ) } { \Phi ( - z ) }.$$
Similarly, we have E ( X 1 | X 2 < z ) = E ( X 1 | -X 2 > -z ) = -r f ( -z ) / F ( ) = z -r f ( ) z / F ( ) z . Therefore, E ( X 1 | X 2 ≥ z ) -E ( X 1 | X 2 < z ) = r f ( ) z { 1 / F ( -z ) + 1 / F ( ) z } = r h ( ) z .
Lemma 5 The covariance between X and B ∼ Bernoulli ( p ) is
$$C o v ( X, B ) = p ( 1 - p ) \{ \mathbb { E } ( X \ | \ B = 1 ) - \mathbb { E } ( X \ | \ B = 0 ) \}.$$
Proof. It follows from the definition of the covariance.
Lemma 6 Under the model generated by Figure 8(a), the regression coefficient of T from regressing Y onto ( T M , ) is
$$\beta _ { T } = \frac { a d \eta ( \alpha ) \{ \rho ( 1 - b ^ { 2 } - c ^ { 2 } - b c \rho ) - b c \} } { \rho \{ 1 - ( a b + a c \rho ) ^ { 2 } \phi ( \alpha ) \eta ( \alpha ) \} }.$$
Proof. We have the following joint Normality of ( Y M T , , ∗ ) :
$$\begin{pmatrix} Y \\ M \\ T ^ { * } \end{pmatrix} \ \sim \ N _ { 3 } \left \{ \begin{pmatrix} 0 \\ 0 \\ 0 \end{pmatrix}, \begin{pmatrix} 1 & b d \rho + c d & a d \rho \\ b d \rho + c d & 1 & a b + a c \rho \\ a d \rho & a b + a c \rho & 1 \end{pmatrix} \right \}.$$
From Lemma 4, we have
$$\mathbb { E } ( M \, | \, T = 1 ) - \mathbb { E } ( M \, | \, T = 0 ) \ & = \ \mathbb { E } ( M \, | \, T ^ { * } \geq \alpha ) - \mathbb { E } ( M \, | \, T ^ { * } < \alpha ) = ( a b + a c \rho ) \eta ( \alpha ), \\ \mathbb { E } ( Y \, | \, T = 1 ) - \mathbb { E } ( Y \, | \, T = 0 ) \ & = \ \mathbb { E } ( Y \, | \, T ^ { * } \geq \alpha ) - \mathbb { E } ( Y \, | \, T ^ { * } < \alpha ) = a d \rho \eta ( \alpha ).$$
Therefore, from Lemma 5, the covariances are Cov ( M T , ) = F ( a ) F ( -a )( ab + ac r h a ) ( ) , and Cov ( Y T , ) = F ( a ) F ( -a ) ad rh a ( ) . According to Lemma 1, the regression coefficient b T is
$$\begin{array} { r c l } \supset & \cdot \cdot \\ \beta _ { T } & = & \frac { \Phi ( \alpha ) \Phi ( - \alpha ) a d \rho \eta ( \alpha ) - ( b d \rho + c d ) \Phi ( \alpha ) \Phi ( - \alpha ) ( a b + a c \rho ) \eta ( \alpha ) } { \Phi ( \alpha ) \Phi ( - \alpha ) - \Phi ^ { 2 } ( \alpha ) \Phi ^ { 2 } ( - \alpha ) ( a b + a c \rho ) ^ { 2 } \eta ^ { 2 } ( \alpha ) } \\ & = & \frac { a d \eta ( \alpha ) \{ \rho ( 1 - b ^ { 2 } - c ^ { 2 } - b c \rho ) - b c \} } { \rho \{ 1 - ( a b + a c \rho ) ^ { 2 } \phi ( \alpha ) \eta ( \alpha ) \} }. \end{array}$$
Lemma 7 Under the model generated by Figure 8(b), the regression coefficient of T from regressing Y onto ( T M , ) is
$$\beta _ { T } = - \frac { a b c d \eta ( \alpha ) } { 1 - ( a b + e ) \phi ( \alpha ) }.$$
Proof. We have the following joint Normality of ( Y M T , , ∗ ) :
$$\begin{pmatrix} Y \\ M \\ T ^ { * } \end{pmatrix} \ \sim \ N _ { 3 } \left \{ \begin{pmatrix} 0 \\ 0 \\ 0 \end{pmatrix}, \begin{pmatrix} 1 & c d + f & c d e + a b f + e f \\ c d + f & 1 & a b + e \end{pmatrix} \right \}.$$
From Lemma 4, we have
E ( M T | = ) 1 -E ( M T | = ) 0 = E ( M T | ≥ a ) -E ( M T | < a ) = ( ab + ) e h a ( )
$$\mathbb { E } ( Y \, | \, T = 1 ) - \mathbb { E } ( Y \, | \, T = 0 ) \ \ = \ \mathbb { E } ( Y \, | \, T ^ { * } \geq \alpha ) - \mathbb { E } ( Y \, | \, T ^ { * } < \alpha ) = ( c d e + a b f + e f ) \eta ( \alpha ).$$
$$& \mathbb { E } ( M \, | \, T ^ { * } \geq \alpha ) - \mathbb { E } ( M \, | \, T ^ { * } < \alpha ) = ( a b + e ) \eta ( \alpha ), \\ & \mathbb { E } ( Y \, | \, T ^ { * } \geq \alpha ) - \mathbb { E } ( Y \, | \, T ^ { * } < \alpha ) = ( c d e + a b f + e f ) \eta ( \alpha ).$$
From Lemma 5, we obtain their covariances Cov ( M T , ) = F ( a ) F ( -a )( ab + ) e h a ( ) , and Cov ( Y T , ) = F ( a ) F ( -a )( cde + abf + e f ) h a ( ) . According to Lemma 1, the regression coefficient b T is
$$\begin{array} { r c l } \Lambda & = & \cdot \cdot \\ \beta _ { T } & = & \frac { \Phi ( \alpha ) \Phi ( - \alpha ) ( c d + f ) \Phi ( \alpha ) \Phi ( - \alpha ) ( a b + e ) \eta ( \alpha ) } { \Phi ( \alpha ) \Phi ( - \alpha ) - \Phi ^ { 2 } ( \alpha ) \Phi ^ { 2 } ( - \alpha ) ( a b + e ) ^ { 2 } \eta ^ { 2 } ( \alpha ) } \\ & = & - \frac { a b c d \eta ( \alpha ) } { 1 - ( a b + e ) ^ { 2 } \phi ( \alpha ) \eta ( \alpha ) }. \end{array}$$
Lemma 8 Under the model generated by Figure 11(a) with an arrow from W to T, the unadjusted estimator has bias ad r + dg, and the adjusted estimator has bias
$$\frac { d g - ( c d ) ( a b + c g ) } { 1 - ( a b + c g ) ^ { 2 } }.$$
Proof. The unadjusted estimator is Cov ( T Y , ) = ad r + dg . Expanding Lemma 1 gives the above as the regression coefficient of T from regressing Y onto ( T M , ) .
## References
Berkson, J. (1946): 'Limitations of the application of fourfold table analysis to hospital data,' Biometrics Bulletin , 2, 47-53.
Copas, J. B. and H. G. Li (1997): 'Inference for non-random samples (with discussion),' Journal of the Royal Statistical Society: Series B , 59, 55-95.
- Gelman, A. (2011): 'Causality and statistical learning,' American Journal of Sociology , 117, 955-966.
- Glymour, M. M. (2006): 'Using causal diagrams to understand common problems in social epidemiology,' In Methods in Social Epidemiology, Oakes M, and Kaufman J, eds. Jossey-Bass: San Francisco, CA , 393-428.
Greenland, S. (2002): 'Quantifying biases in causal models: classical confounding vs collider-stratification bias,' Epidemiology , 14, 300-306.
Heckman, J. J. (1979): 'Sample selection bias as a specification error,' Econometrica , 153-161.
Hern´ an, M. A., S. Hernandez-Diaz, and J. M. Robins (2004): 'A structural approach to selection bias,' Epidemiology , 15, 615-625.
Kelcey, B. and J. Carlisle (2011): 'The threshold of embedded M collider bias and confounding bias.' Society for Research on Educational Effectiveness Conference , available at http://files.eric.ed.gov/fulltext/ED519118.pdf .
Liu, W., M. A. Brookhart, S. Schneeweiss, X. Mi, and S. Setoguchi (2012): 'Implications of M bias in epidemiologic studies: a simulation study,' American Journal of Epidemiology , 176, 938-948.
- Neyman, J. (1923/1990): 'On the application of probability theory to agricultural experiments. essay on principles. section 9,' Statistical Science , 5, 465-472.
- Pearl, J. (1995): 'Causal diagrams for empirical research,' Biometrika , 82, 669688.
Pearl, J. (2009a): Causality: Models, Reasoning and Inference, 2nd Edition , Cambridge University Press.
Pearl, J. (2009b): 'Letter to the editor,' Statistics in Medicine , 28, 1415-1416.
- Pearl, J. (2009c): 'Myth, confusion, and science in causal analysis,' Technical Report , available at http://ftp.cs.ucla.edu/pub/stat\_ser/r348.pdf .
- Pearl, J. (2013): 'Linear models: A useful microscope for causal analysis,' Journal
- of Causal Inference , 1, 155-170.
Rosenbaum, P. R. (2002): Observational Studies , New York: Springer.
Rosenbaum, P. R. and D. B. Rubin (1983): 'The central role of the propensity score in observational studies for causal effects,' Biometrika , 70, 41-55.
Rubin, D. B. (1974): 'Estimating causal effects of treatments in randomized and nonrandomized studies,' Journal of Educational Psychology , 66, 688-701.
Rubin, D. B. (2007): 'The design versus the analysis of observational studies for causal effects: parallels with the design of randomized trials,' Statistics in Medicine , 26, 20-36.
Rubin, D. B. (2008): 'For objective causal inference, design trumps analysis,' The Annals of Applied Statistics , 2, 808-840.
Rubin, D. B. (2009): 'Should observational studies be designed to allow lack of balance in covariate distributions across treatment groups?' Statistics in Medicine , 28, 1420-1423.
Shrier, I. (2008): 'Letter to the editior,' Statistics in Medicine , 27, 2740-2741.
- Shrier, I. (2009): 'Propensity scores,' Statistics in Medicine , 28, 1315-1318.
- Sj¨ olander, A. (2009): 'Propensity scores and M-structures,' Statistics in Medicine , 28, 1416-1420.
Sprites, P. (2002): 'Presented at: WNAR/IMS Meeting. Los Angeles, CA, June 2002,' .
VanderWeele, T. J. and I. Shpitser (2011): 'A new criterion for confounder selection,' Biometrics , 67, 1406-1413.
- Wright, S. (1921): 'Correlation and causation,' Journal of Agricultural Research , 20, 557-585.
- Wright, S. (1934): 'The method of path coefficients,' The Annals of Mathematical Statistics , 5, 161-215. | null | [
"Peng Ding",
"Luke Miratrix"
] | 2014-08-02T01:31:25+00:00 | 2014-08-02T01:31:25+00:00 | [
"math.ST",
"stat.TH"
] | To Adjust or Not to Adjust? Sensitivity Analysis of M-Bias and Butterfly-Bias | "M-Bias," as it is called in the epidemiologic literature, is the bias
introduced by conditioning on a pretreatment covariate due to a particular
"M-Structure" between two latent factors, an observed treatment, an outcome,
and a "collider." This potential source of bias, which can occur even when the
treatment and the outcome are not confounded, has been a source of considerable
controversy. We here present formulae for identifying under which circumstances
biases are inflated or reduced. In particular, we show that the magnitude of
M-Bias in linear structural equation models tends to be relatively small
compared to confounding bias, suggesting that it is generally not a serious
concern in many applied settings. These theoretical results are consistent with
recent empirical findings from simulation studies. We also generalize the
M-Bias setting (1) to allow for the correlation between the latent factors to
be nonzero, and (2) to allow for the collider to be a confounder between the
treatment and the outcome. These results demonstrate that mild deviations from
the M-Structure tend to increase confounding bias more rapidly than M-Bias,
suggesting that choosing to condition on any given covariate is generally the
superior choice. As an application, we re-examine a controversial example
between Professors Donald Rubin and Judea Pearl. |
1408.0326v1 | and spectropolarimetry
Proceedings IAU Symposium No. 307, 2014
G. Meynet, C. Georgy, J.H. Groh & Ph. Stee, eds.
c © 2014 International Astronomical Union DOI: 00.0000/X000000000000000X
## 3D and Some Other Things Missing from the Theory of Massive Star Evolution
## W. David Arnett 1
1
Steward Observatory University of Arizona Tucson AZ 85721, USA
email: [email protected]
Abstract. This is a sketch of a 321D approximation which is nonlocal, and thus has nonzero fluxes of KE (to be published in more detail elsewhere). We plan to add this as an option to MESA. Inclusion of KE fluxes seems to help resolve the solar abundance problem (Asplund et al. 2009). Smaller cores may ease the explosion problems with core collapse supernova simulations.
Keywords. convection, turbulence, Sun: abundances, stars: evolution, supernovae
## 1. Introduction
Stars are three dimensional (3D), turbulent plasma, and much more complex than the simplified one dimensional (1D) models we use for stellar evolution. Computer power is inadequate † at present for adequately resolved (i.e., turbulent) 3D simulations of whole stars. In his review talk, Meakin (2015) has illustrated this complexity and shown how it may be tamed by use of 3D simulations and Reynolds-Averaged Navier-Stokes (RANS) equations.
A minimalist, more approximate step may be easier to implement in stellar evolutionary codes, and instructive for the closure problem. Formally, the RANS equations are incomplete unless taken to infinite order; they must be closed by truncation at low order to be useful. This may be due to the nature of the Reynolds averaging, which allows all fluctuations rather than restriction only to dynamically consistent ones. Closure requires additional information to remove these new, extraneous solutions. As a complement to the RANS approach, approximations which focus on dynamics are emphasized here. In the cascade, turbulent kinetic energy and momentum are concentrated in the largest eddies. We introduce a simple dynamically self-consistent model which contains the largest eddies and the Kolmogorov cascade, and examine the consequences.
Here we will concentrate on the following issues:
- ( a ) the turbulent cascade (Kolmogorov 1941, 1962),
- ( b ) chaos (Lorenz 1963) and time dependence (Arnett & Meakin 2011b),
- ( c ) boundary layers (Prandtl & Tietjens 1934) and surfaces of separation (Landau & Lifshitz 1959),
- ( d ) combined mixing and burning, and
- ( e ) sensitivity of core collapse to progenitor structure
The important and related issues of coherent treatment of pulsations, eruptions and explosions, rotation and magnetic fields, and variable mass and angular momentum must be deferred.
† But perhaps close; see Herwig et al. (2013).
W. David Arnett
Table 1. A Few Examples of Three-dimensional Simulations of Convection in Stars.
| Attribute | 3D Atmospheres | Solar Convection Zone | Stellar Interiors |
|-------------------------------------------------------------------------------------------------------------------------------------|-------------------------------------------------------------------------|------------------------------------------------------|----------------------------------------------------------------------------|
| representative photosphere composition gradient nuclear burning magnetic field rotation driving geometry boundary inside grid hydro | Stein & Nordlund 1 Y(yes) N N Y N top box in star top only compressible | J. Toomre 2 N(no) N N Y Y top CZ in box No anelastic | Meakin & Arnett 3 N(no) Y Y N N top or bottom box in star Yes compressible |
Notes:
- 1 See Stein & Nordlund (1998), and Magic et al. (2013, 2014) which also refers to other recent work.
- 2 See Brun et al. (2011, 2004), and many papers in the IAUS271 proceesings (Brun 2011).
- 3 See Meakin & Arnett (2007); Viallet et al. (2013), and Arnett et al. (2014) for an overview.
Erika B¨hm-Vitense developed mixing-length theory in the 1950's (Vitense 1953; B¨hmo o Vitense 1958), prior to the publication in the west of Andrey Kolmogorov's theory of the turbulent cascade (Kolmogorov 1941, 1962). MLT might have been different had she been aware of the original work Kolmogorov had done in 1941. Edward Lorenz showed that a simple convective roll had chaotic behavior (a strange attractor, Lorenz (1963)). Ludwig Prandtl developed the theory of boundary layers (Prandtl & Tietjens 1934). All these ideas will be relevant to our discussion, which is based upon theory and 3D simulations.
## 1.1. 3D Simulations
Numerical approaches for fluid dynamics simulations include implicit large eddy simulations (ILES) and direct numerical simulation (DNS). DNS resolves the smallest relevant scales, and procedes to larger scales until limited by computer power available. At present this limit is far smaller than stellar scales, so DNS is important in stellar problems for points of principle (e.g., Wood et al. (2013)). ILES includes the largest scales in the system, and procedes to smaller scales until limited by computer power available. Implict is the assumption that sub-grid scale phenomena are correctly treated. This seems to be valid for 3D turbulence using state-of-the-art methods (Grinstein, F. F. et al. 2007).
In order to be useful in a 1D stellar evolution code, 3D information must be projected onto a 1D coordinate system, hence '321D' . Table 1 gives a brief comparison of fea-† tures of some 3D simulations which are relevant to designing 1D stellar algorithms. The simulations have different strengths and weaknesses, and tend to complement each other.
3D atmospheres. Simulations of stellar atmospheres in 3D were pioneered by Stein & Nordlund (1998), see also Magic et al. (2013) and references therein. These simulations are one of the great successes of radiation hydrodynamics, removing the necessity for the micro-turbulence and macro-turbulent fudge factors previously used to calculate stellar spectra. It is tempting to use MLT to fit such 3D simulations as a 'stellar engineering' exercise to connect atmospheres to interiors, because stellar evolution codes are almost always formulated in the language of MLT. Magic et al. (2014) give a clear discussion of this process, and show that MLT must be modified in at least one respect to make the identification: a ram pressure term must be added to MLT. It appears that such fits necessarily ignore a quantity important for stellar interiors which comes from the same
† The acronym '321D' for 'projection of three dimensions to one dimension' is due to John Lattanzio.
term but in the energy equation. This is the flux of turbulent kinetic energy, which is incorrectly defined as zero in MLT.
These 3D stellar atmospheres generally do not contain the whole convective zone, but use a lower boundary condition which has been shown to have little effect on the predicted spectra. By the same token, this means that these simulations are not a sensitive probe of the convection at the bottom of the convection zone.
Convection, rotation and MHD. Juri Toomre and his students and collaborators have pioneered anelastic simulations of the solar convection zone, focusing on global MHD and differential rotation. For 2D see (Hurlburt et al. 1984), and 3D (Brun et al. 2004). Unlike the 'box in a star' grids used in 3D atmospheres and stellar interiors, such global solutions make more serious demands on computer resources because of their larger extent, so that, other things being equal, their zoning tends to be coarser. Recent results suggest that the numerical viscosity is sufficiently low so that the flows are becoming realistically turbulent (Brun et al. 2011). Merging of the insights from these simulations with stellar evolution with rotation is a present challenge.
Stellar Interiors. Another option for use of limited computational resourses is to examine deep interiors in which composition gradients and nuclear burning occur. By using a 'box in a star' approach it is possible to include convective zone boundaries within the computational grid. The treatment of radiation flow is simpler (radiative diffusion). Due to neutrino cooling the thermal time scales become shorter, and make thermal relaxation easier to deal with; also higher luminosity as found in deep layers of red giants reduces the thermal time scale. Deep zones (large stratification) and pressure dilatation (Viallet et al. 2013) have been examined. The 'box in a star' approach truncates the lowest order modes, so that a natural complement is the pioneering work of Paul Woodward (e.g., Herwig et al. (2013)) to put a 'whole star in a box'.
## 2. A 321D Algorithm
This procedes in several steps:
- ( a ) add turbulent cascade,
- ( b ) add dynamic (acceleration) equation for integral scale,
- ( c ) balance between driving, damping, and the role of turbulent KE flux,
- ( d ) make quantitative connections to MLT and the Lorenz model,
- ( e ) use the steady-state Lorenz model to approximate average behavior,
- ( ) f use acceleration equation to define boundary behavior, and
- ( g ) add composition effects.
Because an acceleration equation is used, it is straight-forward (in principle at least) to add inertial forces (centrifugal and coriolis), Lorenz forces and differential rotation.
## 2.1. The Turbulent Cascade
Arnett et al. (2014) estimate the Reynolds number to be Re ∼ 10 18 at the base of the solar convection zone. Numerical simulations and laboratory experiments become turbulent for Re ∼ 10 , so fluid flows in stars are strongly turbulent if, as we assume 3 for the moment, rotational and magnetic field effects may be neglected. This special, simplier case is thought to be widely but not universally appropriate to stellar interiors.
For homogeneous, isotropic, and steady-state turbulence, the Kolmogorov relation between the dissipation rate of turbulent kinetic energy, velocity, and length scale is,
$$\epsilon & \sim v ^ { 3 } / \ell.$$
This is a global constraint, and applies to each length scale λ in the turbulent cascade,
for all scales λ , or,
$$\Delta v _ { \lambda } & \sim ( \epsilon \lambda ) ^ { \frac { 1 } { 3 } }.$$
so that the velocity variation across a scale λ is ∆ v λ , and increases as λ 1 3 . A description of the cascade needs both large and small scales; Eq. 2.3 implies that the largest (integral) scales have most of the KE and momentum while the smallest have the fastest relaxation times. Simulations confirm this (Arnett et al. 2009).
## 2.2. Dynamics
In MLT the buoyant force is approximately integrated over a mixing length to obtain an average velocity u (e.g., Smith & Arnett (2014)),
$$u ^ { 2 } = g \beta _ { T } \Delta \nabla \left ( \frac { \ell _ { M L T } ^ { 2 } } { 8 H _ { P } } \right ).$$
Working backward, this may be expressed as
$$d u / d t = g \beta _ { T } \Delta \nabla - u / \tau,$$
where glyph[lscript] d ≡ glyph[lscript] 2 MLT / 8 H P and τ = glyph[lscript] d / u | | , and for ∆ ∇ > 0. Multiplying by u gives a kinetic energy equation,
$$d ( u ^ { 2 } / 2 ) / d t = u \cdot g \beta _ { T } \Delta \nabla - u ^ { 2 } / \tau,$$
for which the steady-state solution † is Eq. 2.4 (only positive ∆ ∇ are allowed).
glyph[negationslash]
Had it been available, B¨hm-Vitense might have identified the damping term with the o Kolmogorov value (Eq. 2.1). However, Kolmogorov found the damping length glyph[lscript] d to be the depth of the turbulent region, so that it is not a free parameter, unlike MLT. There is a further issue; glyph[epsilon1] is the average dissipation rate, not the instantaneous value ( u /glyph[lscript] 3 d ) which fluctuates over time; that is, u = v except on average over τ (see Fig. 4 in Meakin & Arnett (2007)); we return to this below.
## 2.3. Kinetic Energy Flux
The flow is relative to the grid of the background stellar evolution code, so
$$d ( u ^ { 2 } / 2 ) / d t = \partial _ { t } ( \mathbf u \cdot \mathbf u ) / 2 + \nabla \cdot \mathbf F _ { \text{KE} },$$
where F KE = ρ u u ( · u ) / 2 is a flux of kinetic energy. The generation of the divergence of a kinetic energy flux in this way is robust for dynamic models; it occurs in the more precise RANS approach as well (Meakin & Arnett 2007). It does not occur in MLT, which assumes symmetry in velocity between upflows and downflows, a condition that is violated at even weak levels of stratification (Meakin & Arnett 2010; Viallet et al. 2013). Alternatively, MLT is equivalent to Eq. 2.5 with du/dt = 0, the local approximation.
Viallet et al. (2013); Moc´k et al. (2014) find a a global balance between buoyant driving (at the largest scales) and turbulent damping (at the smallest). Because of the large separation of these length scales, they are weakly coupled, giving rise to fluctuating behavior typical of turbulence. This balance (on average only) is a new condition beyond MLT, and allows the elimination of the free parameter in MLT, the mixing length, or α = glyph[lscript]/H P . The actual flow patterns that correspond to this global balance depend upon the details of the turbulent energy input; nuclear heating and photospheric cooling give different flows (Meakin & Arnett 2010), i.e., fluxes of kinetic energy.
- † Care must be taken with the sign of the transit time τ and the deceleration for negative u .
$$\epsilon \sim ( \Delta v _ { \lambda } ) ^ { 3 } / \lambda,$$
We choose a vector equation for the integral scale velocity u ,
$$\partial \mathbf u / \partial t + ( \mathbf u \cdot \nabla ) \mathbf u = \mathbf g \beta _ { T } \Delta \nabla - \mathbf u / \tau.$$
The damping term is chosen to give the Kolmogorov expression for glyph[epsilon1] , the turbulent dissipation, if Eq. 2.8 is dotted by u and averaged over a turnover time τ .
## 2.4. Connections to MLT and Lorenz
In the local, steady-state, limiting case, the left-hand side of Eq. 2.8 vanishes, and an equation similar to Eq. 2.4 results, but with the mixing length replaced by the turbulent damping length (essentially the lesser of the depth of the convective zone or 4 pressure scale heights, Arnett & Meakin (2011b)). With this change, the cubic equation of B¨hmo Vitense may be derived (Smith & Arnett 2014), and we recover a form of MLT.
If it is assumed that the integral scale motion is that of a convective roll, Eq. 2.8 may be reduced to the form of the classic Lorenz equations , but with a nonlinear damping term provided by the Kolmogorov cascade (Arnett & Meakin 2011b). Because of the time lag, the modified equations become even more unstable than the original ones.
For 321D the stellar evolution code must be supplied a smooth time-averaged value for the convective variables; we approximate that by the steady state. The weak coupling between large scale driving and dissipation at the small scale causes time dependent fluctuations of significant amplitude in luminosity and turbulent velocity; see Meakin & Arnett (2007), Fig. 4, for fluctuations in KE in an oxygen burning shell. The term ∂ u /∂t in Eq. 2.8 is needed for chaotic fluctuations, wave generation and large scale dynamic behavior. A stellar evolution code must step over the shorter turnover time scales (weather) to solve for the evolutionary times (climate); this requires an average over active cells, so that there is a cellular structure in both space and time. The steady-state limit of the Lorenz equation gives a reasonable approximation to its average behavior, filtering out the fluctuations (Arnett & Meakin 2011b); we apply the same approximation to Eq. 2.8 for slow stages of stellar evolution.
## 2.5. Boundary Layers
Fluid motion in a star may be separated into two fundamentally different flows: solenoidal flow (divergence free) and potential flow (curl free) (Landau & Lifshitz 1959). Fig. 1 shows the striking separation in flow velocities at such boundary layers (see Prandtl & Tietjens (1934). We do not claim that we have resolved these boundary layers yet , but their † structure and nature are important to the rate at which turbulent flow moves into or from non-turbulent regions-the entrainment rate.
Peter Eggleton took an early step in dealing with steep gradients in composition with the introduction of an admittedly ad-hoc diffusion operator (Eggleton 1973); this numerically advantageous procedure has been widely adopted for stellar evolution.
Amore physical picture results from consideration of the dynamics of the motion. At its most elemental level, the velocity vector must turn at boundaries to maintain solenoidal flow. Most of the momentum is at the largest scales, where dissipation is least and the flow almost adiabatic, so this velocity must be the integral scale velocity. The magnitude of the acceleration required is just the centrifugal value v /w 2 where w is the width of the turning region and v the relevant velocity. Using Eq. 2.8 in the steady state limit, and taking w ∼ ∆ r << glyph[lscript] , the radial component of the acceleration equation becomes
$$u _ { r } \partial u _ { r } / \partial r \sim \Delta ( \frac { 1 } { 2 } u _ { r } ^ { 2 } ) / \Delta r \sim g \beta _ { T } \Delta \nabla.$$
† We have fewer than 10 zones across the lower boundary layer; numerical viscosity may affect the computed entrainment rate.
Sorting out the signs we see that the buoyancy must be negative for the radial kinetic energy to decrease, so that negative buoyancy decelerates the motion, and gives the required turning. This defines a layer at which the velocity goes to zero on average. The 'overshoot' region has a width w ; this material is mixed because it moves back into the convective region after it turns.
There are fluctuations, so that the layer undulates, generating waves in the neighboring stable region. In a coordinate system moving with the undulating layer, there is a boundary between the mixed (overshoot) region and the overlying layer which is stably stratified. The overshoot region has solenoidal flow which flow in the stably stratified region is potential (wave) flow. By Kelvin's theorem (conservation of circulation), entropy change (e.g., dissipation, radiative diffusion) is required to move matter from one region to the other. Most of the turbulent momentum is in the largest scale convective motions, the integral scale. In simulations we see that these couple well with gravity waves † , whose speed increases with wavelength (Landau & Lifshitz (1959), § 12), so that the longer wavelengths carry most of the energy. The shorter wavelengths are more dissipative; they are generated by nonlinear interaction of the long wavelengths (wave breaking; think water waves). This is an example of a mechanism for changing entropy at the boundary, allowing the turbulent region to grow. If this example is representative, such entrainment rates are not universal but depend upon local conditions at the boundary as well as turbulent velocities. See also Viallet et al. (2013) discussion of radiative heating at bottom of a deep (strongly stratified) convection zone.
- † Sound waves couple well only for higher Mach number flow (Landau & Lifshitz (1959), § 64).
A snapshot of turbulent convection Viallet et al. 20/3
Figure 1. Boundary Layers enclosing Convection (after Viallet et al. (2013).
Grolsch (2015) approaches this issue from a quite different point of view, but comes to some similar conclusions.
## 2.6. Mixing and Burning
Change in composition due to nuclear burning is a fundamental feature of stellar evolution. In general stars do not have uniform composition. In stellar evolution, mixing is determined according to the criterion of Schwarzschild, or of Ledoux. The Schwarzschild criterion is defined as S = ∇-∇ e , and the Ledoux criterion is L = S -∇ Y , so ∆ ∇ = L which reverts to S if there is no composition gradient ( ∇ Y = 0).
Mixing motions (solenoidal flow) result from buoyancy, pressure perturbations, or differential rotation. We focus on the buoyant acceleration, -g ρ /ρ. ′ Composition gradients enter the buoyancy on an equal basis with entropy gradients; for the simplest case of an ideal gas, ρ /ρ ′ = T /T ′ + Y /Y ′ -P /P, ′ where Y = 1 /µ is the number of free particles per baryon, or inverse mean molecular weight. For a more general equation of state, there are mutiplicative factors of order unity on the r.h.s. Traditionally P /P ′ is taken to be zero even though this is incorrect for strongly stratified cconvecton (Viallet et al. 2013).
Brunt frequency, which is important for for asteroseismology, is N 2 = -L ( gβ T /H P ), indicating a fundamental connection with boundary conditions (Aerts, C. et al. 2010).
## 3. Solar Abundances
In the steady-state but non-local case, the ( u ·∇ ) u term gives a coupling between driving regions and damping regions in the form of a flux of turbulent kinetic energy (Meakin & Arnett 2007, 2010); this is assumed to be zero by symmetry in MLT (a major flaw). Even moderate stratification breaks the symmetry, and gives a finite flux of turbulent kinetic energy.
Any model of the Sun which uses MLT will neglect the effect of turbulent kinetic energy fluxes. To estimate the sign and size of the necessary changes, we scale from the simulations in Viallet et al. (2013) for a convection zone which like the Sun is highly stratified. The luminosity due to kinetic energy is L KE ∼ -0 35 . L near the base of the convection zone, which is significant. This negative luminosity must be balanced by increased positive radiative luminosity, to maintain a solar luminosity, so L new /L glyph[circledot] = 1 35. This may . obtained by a reduction in opacity. The opacity is well known, and depends upon the composition, primarily the metal abundance. If we simply assume that the opacity scales with metalicity, κ new /κ old ∼ 3 / 4, and the actual metalicity should shift from the stellar evolution value (from the 'standard solar model', SSM) to z new ∼ 0 02(3 . / 4) ∼ 0 015, . which is the value determined from 3D atmospheres. Asplund et al. (2009) conclude that the discrepancy between abundances determined from 3D atmospheres, and from stellar evolution is unidentified. The argument just given suggests that there is a significant flaw in the physics of SSM, correction for which tends to resolve the discrepancy. Until stellar models including kinetic energy flux are constructed, it seems reasonable to take the 3D atmospheric abundances as our best estimate, and attribute the disagreement to use of a local convection theory.
## 4. Sensitivity to Progenitor Structure
Simulations of core collapse may be sensitive to convection algorithms used to construct precollapse models. Couch & Ott (2013) get increased tendency toward explosion simply by adding a nonradial velocity component to the progenitor model (Arnett & Meakin
2011a), as turbulence requires. This also suggests changes to the initial models used by Kochanek (2014).
Fig. 2 shows two TYCHO models at the end of oxygen burning. The density structure is sensitive to the assumed entrainment physics. The two curves represent stars of 15 M glyph[circledot] , with only the entrainment physics changed. Not only are the sizes of the cores changed, but the carbon/oxygen ratio is affected by entrainment as well.
Core-collapse simulations often fail because of the high rate of infall of mantle matter onto the newly formed, nuclear density core. This matter must be photo-dissociated by the explosion shock if an explosion is to occur. Lower rates of infall aid the explosion. The rate of infall is
$$\dot { M } _ { i n } = u _ { i n } 4 \pi r _ { i n } ^ { 2 } \rho _ { i n }.$$
The infall velocity is essentially the local sound speed in the progenitor mantle; matter falls in as a rarefaction wave moves out. To make the rate of mass infall ˙ M in small, the initial density ρ in should be small. The critical time occurs during the infall of the mantle (labeled 'Infall'); the envelope falls in too slowly to affect the explosion shock. Clearly the curve in Fig. 2 with the larger core will have larger ρ in , and be harder to explode. This large core resulted from taking the maximum entrainment rate that was energetically allowed, throughout the evolution from the main sequence (for simplicity no mass loss or binary stripping were considered).
The small core case resulted from an entrainment rate of 0.01 of the maximum, so that the boundary dynamics was almost elastic, and consistent with analytic estimates. Smaller rates and smaller cores are possible. The highest resolution 3D simulations we
Figure 2. Changes in Density structure of 15 M glyph[circledot] with Entrainment.
have done were the most nearly elastic in the boundary layers, so we regard this as the more reasonable case. It has smaller cores that conventional progenitor models; ad-hoc diffusion smoothes gradients, leading to larger cores.
## 5. Conclusions
This is a sketch of a 321D approximation which is nonlocal, and thus has nonzero fluxes of KE (to be published in more detail elsewhere). We plan to add this as an option to MESA. Inclusion of KE fluxes seems to help resolve the solar abundance problem (Asplund et al. 2009). Smaller cores may ease the explosion problems with core collapse supernova simulations.
## References
Aerts, C., Christensen-Dalsgaard, J., & Kurtz, D. W. 2010, Asteroseismology , Springer
Arnett, D., Meakin, C., & Young, P. A. 2009, ApJ 690, 1715
Arnett, W. D. & Meakin, C. 2011a,
ApJ
733, 78
Arnett, W. D. & Meakin, C. 2011b, ApJ 741, 33
Arnett, W. D., Meakin, C., & Viallet, M. 2014, AIP Advances 4(4), 041010
Asplund, M., Grevesse, N., Sauval, A. J., & Scott, P. 2009, ARAA 47, 481
B¨ ohm-Vitense, E. 1958, ZfA 46, 108
Brun, A. S. (ed.) 2011, Astrophysical Dynamics: From Stars to Galaxies , Vol. S 271, International Astronomical Union, CUP
Brun, A. S., Miesch, M. S., & Toomre, J. 2004, ApJ 614, 1073
Brun, A. S., Miesch, M. S., & Toomre, J. 2011, ApJ 742, 79
Couch, S. M. & Ott, C. D. 2013, ApJ 778, L7
Eggleton, P. P. 1973, MNRAS 163, 279
Grinstein, F. F., Margolin, L. G., & Rider, W. J. (eds.) 2007, Implicit Large Eddy Simulations , Cambridge University Press
Grolsch, A. 2015, in Proceedings of this conference
Herwig, F., Woodward, P. R., Lin, P.-H., Knox, M., & Fryer, C. 2013, ArXiv e-prints
Hurlburt, N. E., Toomre, J., & Massaguer, J. M. 1984, ApJ 282, 557
Kochanek, C. S. 2014, MNRAS
Kolmogorov, A. 1941, Akademiia Nauk SSSR Doklady 30, 301
Kolmogorov, A. N. 1962, Journal of Fluid Mechanics 13, 82
Landau, L. D. & Lifshitz, E. M. 1959, Fluid mechanics , Pergamon
Lorenz, E. N. 1963, Journal of Atmospheric Sciences 20, 130
Maeder, Andr´ e 1999, Physics, Formation and Evolution of Rotating Stars , Springer
Magic, Z., Collet, R., Asplund, M., et al. 2013, A&A 557, A26
Magic, Z., Weiss, A., & Asplund, M. 2014, ArXiv e-prints
Meakin, C. 2015, in Proceedings of this conference
Meakin, C. A. & Arnett, D. 2007, ApJ 667, 448
Meakin, C. A. & Arnett, W. D. 2010, Ap&SS 328, 221
Moc´ ak, M., Meakin, C., Viallet, M., & Arnett, D. 2014, ArXiv e-prints
Prandtl, L. & Tietjens, O. G. 1934, Applied Hydro-& Aeromechanics , Dover Publications, Inc.
Smith, N. & Arnett, W. D. 2014, ApJ 785, 82
Stein, R. F. & Nordlund, ˚ . 1998, A ApJ 499, 914
Viallet, M., Meakin, C., Arnett, D., & Moc´k, M. 2013, a ApJ 769, 1
Vitense, E. 1953, ZfA 32, 135
Wood, T. S., Garaud, P., & Stellmach, S. 2013, ApJ 768, 157
## Discussion
Andr´ e Maeder: Analytical studies suggest that transport processes by meridional circulation and by shear diffusion are strongly affected by the horizontal turbulence in differentially rotating stars. What can the 3D simulations tell us about the possible value of the horizontal turbulence?
David Arnett: Great question! We do not seem to see Jean-Paul Zahn's strongly asymmetric diffusive mixing velocities. However the overturn is effective at mixing the whole convective region, except for Si burning, which has 'cellular' burning regions. The shellular approximation may be fine for earlier stages, up to and including oxygen burning, if we can get the boundary motion-the entrainment rates-right. The shellular approximation will break down during explosive or eruptive events due to 3D instabilities, of course.
Now suppose we spin up a non-rotating star. Turbulence is turbulence, whether driven by buoyancy or by differential rotation. At first we get slight distortions of spherical equipotentials, leading to meridional circulation, and the solenoidal flow also reacts to the inertial accelerations. Eq. 2.8 is a 3D vector equation, and contains these effects; the velocity includes the velocity of meridional circulation and the horizontal turbulence, at least in principle. An interesting issue is the relative importance of dissipation in the boundary layers versus in the bulk of the turbulent flow.
As the spin increases (Rossby number decreases) it will have an increasingly important influence on flow, convection, and angular momentum transport, and the importance of MHD becomes an issue (our code does not have MHD active at present). One can think of a continuous sequence from non-rotating star to accretion disk, parameterized by angular momentum, and the literature contains examples of 3D simulations all along this sequence. That said, the results are complex, and it seems that we still have much to learn.
Victor Khalak: How may stellar rotation be introduced into the equation for turbulence?
David Arnett: It is already in Eq. 2.8 if we do the usual transformation to a rotating frame. This is a deceptively simple vector equation (the Navier-Stokes equation in the turbulent limit), which has deep connections to a lot of theoretical work on rotating stars. Andr´ Maeder has written a beautiful development of the physics of rotating stars e (Maeder, Andr´ e 1999) which starts from the Navier-Stokes equation (his Eq. 1.2).
Ehsan Moravveji: In the very vicinity of the fully mixed core and the radiative layers (braking and/or entrainment layer?), what is the behavior of ∆ ∇ term in your energy balance equation?
David Arnett: Thanks, this is an important point that I went through too quickly. The answer may be implicit in § 2.3 and § 2.4. The ∆ ∇ is a factor in the buoyant acceleration, and can change sign. It depends upon both the temperature gradient and the composition gradient. However mixing makes the composition gradient tend to zero, which changes ∆ ∇ itself; this problem must be solved implicitly. In the simulations there is a transition layer in which the composition does change, but it may not yet be resolved numerically (the 'boundary layer' in Fig. 1). This is the layer where the velocity field changes from solenoidal to potential flow, i.e., from convection to waves. See also the contribution by Arlette Grolsch (Grolsch 2015).
The simulations show additional complexity: the boundary is dynamic and has vigorous and fluctuating wave motion, features not in MLT. If we assume that the boundary dynamics has only a slow secular variation on average (an 'entrainment' model, Meakin & Arnett (2007)), we might use Eq. 2.9 and asteroseismology to try to make progress. | 10.1017/S1743921314007406 | [
"W. David Arnett"
] | 2014-08-02T02:04:58+00:00 | 2014-08-02T02:04:58+00:00 | [
"astro-ph.SR",
"astro-ph.HE"
] | 3D and Some Other Things Missing from the Theory of Massive Star Evolution | This is a sketch of a 321D approximation for stellar convection which is
nonlocal, and thus has nonzero fluxes of KE (to be published in more detail
elsewhere). Boundary conditions are discussed in a fluid dynamics context
(i.e., predictions for overshoot, semiconvection and entrainment are analyzed).
We plan to add this as an option to MESA. Inclusion of KE fluxes seems to help
resolve the solar abundance problem (Asplund 2009). Smaller cores may ease the
explosion problems with core collapse supernova simulations. |
1408.0327v3 | ## Crossover from Room-temperature Double-channel Ballistic
## Transport to Single-channel Ballistic Transport in Zigzag Graphene
## Nanoribbons
Zhao-Dong Chu and Lin He* Republic of China
Department of Physics, Beijing Normal University, Beijing, 100875, People's
## Abstract
Very recently, it was demonstrated explicitly that a zigzag graphene nanoribbon (GNR) exhibits a crossover of conductance from G 0 to G 0 /2 with increasing the length ( G 0 = 2 e 2 / h is the quantum of conductance) even at room-temperature [Baringhaus, et al. Nature 506, 349 (2014)]. Such a result is puzzling as none of previous theories seem to match the experimental observations. Here, we propose a model to explain the crossover from double-channel to single-channel ballistic transport in zigzag GNR. The sp 3 distortion of carbon atoms at the GNR edges induces a large spin-orbit coupling on the edges atoms, which enhances spin-flip scattering of edge states of the zigzag GNR. With sufficient spin-flip scattering, the wave-function of edge states becomes a superposition of the spin-up and spin-down components. Then the coupling of the two edges becomes important. This removes the edge degree of freedom in the zigzag GNR and results in the evolution of the conductance from G 0 to G 0 /2 with increasing the length.
## Significance statement
A very recent experiment reported the observation of the crossover from double-channel ballistic transport to single-channel ballistic transport in zigzag graphene nanoribbon (GNR) with increasing the length. This result has attracted much attention, however, none of previous theories seem to match the experimental observations. Here, we propose a simple model that could explain quite well the puzzling experimental result. According to our analysis, the decrease of conductance from G 0 to G 0 /2 with increasing the probe spacing direct arises from the joint effect of the reduction of the spin degree of freedom and the edge-edge coupling in the GNRs. Our result indicates that the spin degree of freedom plays an important role in the transport properties of graphene systems.
Ballistic transport of electrons has fascinated researchers across various disciplines for several decades 1-7 . Looking for systems where ballistic transport can be observed at room temperature boosts rapid developments of several leading topics in condensed matter physics in the past few years 8-10 . Graphene nanoribbon (GNR) with perfect crystallography is believed to be one of the most promising candidates as a perfect conductor where electrons can travel long distances without dissipation. However, all experiments in lithographically patterned exfoliated GNRs demonstrated that transport is dominated by diffusive mode rather than the ballistic mode 11,12 . This stimulates many groups to develop different methods in the fabrication of high-quality graphene nanoribbons 13,14 .
Until very recently, a great breakthrough in the growth of GNRs with crystallographically perfect edges is achieved and room-temperature ballistic transport with a conductance of G 0 in the GNRs is realized by Baringhaus and colleagues for the probe spacing L ≤ 160 nm (Here G 0 = 2 e 2 / h is the quantum of conductance and the factor 2 comes from the spin degree of freedom) 15 . These zigzag GNRs are grown on the slope of a terraced silicon carbide surface and the transport properties of the GNRs were measured in situ with variable probe spacing. Figure 1 summarizes the room-temperature conductance G versus the probe spacing L of the zigzag GNRs in a range of energies around charge neutrality 15 . For L ≤ 160 nm, a conductance of G 0 is observed. With increasing the probe spacing to L ~ 1 µ m, the conductance decreases
gradually to G 0 /2, then it keeps a robust value of G 0 /2 for 1 µ m ≤ L ≤ 16 µ m. The most surprising result of their experiments is the observation of the single quantum mechanical channel for transport in these high-quality zigzag GNRs with 1 µ m ≤ L ≤ 16 µ m. This result is amazing since that at least an edge-degenerate channel, giving rise to a conductance of G 0 , should be involved in the transport of the GNRs in a range of energies around charge neutrality point 16,17 . The observed conductance of G 0 /2, as shown in Fig. 1, is obviously beyond the description of any previous theories 18 .
In this contribution, we propose how the single quantum mechanical channel can readily arise in the zigzag GNR. Even under ultraclean high vacuum conditions, there are many adatoms (impurities), such as hydrogen, that hybridize directly with carbon atoms at edges of GNRs. This hybridization-induced sp 3 distortion of the carbon atoms leads to a strong enhancement of spin-orbit coupling 19 , which consequently enhances spin-flip scattering of edge states of the zigzag GNRs. In a long zigzag GNR, the wave-function of the edge states becomes a superposition state of the spin-up and spin-down components because of sufficient spin-flip scattering. Then the edge-edge coupling becomes important, which removes the edge degree of freedom and leads to the crossover of the conductance from G 0 to G 0 /2.
## Results
Hubbard model of zigzag GNR. At the charge neutrality level of graphene
monolayer, there are four identical channels, two from the spin and two from the valley (K and K'), for ballistic transport 20 . In zigzag GNRs, the valley degree of freedom is removed around the charge neutrality point and only spin-polarized channels located at the two edges presenting a conductance of G0 are expected to be observed 16,17 . To explicitly illustrate this, we show electronic structures of a zigzag graphene nanoribbon in Fig. 2. The electronic structures can be described quite well by a tight-binding model Hamiltonian 21-24
$$H = - t \sum _ { \sigma = \uparrow, \downarrow } \sum _ { < i, j > } [ c ^ { \dagger } _ { i, \sigma } c _ { j, \sigma } + H. c ] + U \sum _ { i } n _ { i, \uparrow } n _ { i, \downarrow } \.$$
Here, the first term is the nearest neighbor hopping term on the honeycomb lattice. The operator † , i c σ ( , j c σ ) creates (annihilates) an electron with spin σ at site i ( ), j t = 2.7 eV represents the nearest-neighbor hopping amplitude of the π electrons, and < , > i j are nearest neighbors on a honeycomb lattice. The second (Hubbard) term represents electron-electron interaction with the strength of on-site Coulomb repulsion U = 6 eV and † , , , i i i n c c σ σ σ = the spin-resolved electron density at atom i . According to the Hamiltonian (1), the zigzag GNRs feature a gap with spin-polarized edge states at each edge and the spin polarizations are in opposite directions in the two edges of a GNR 21 , as shown in Fig. 2. Additionally, the states of opposite spin orientation in the zigzag GNRs are degenerate in all bands.
Conductance of a zigzag GNR described by Hubbard model. In a transport
measurement, an external voltage V creates a difference of chemical potential between two contacts -+ u u eV -= and generates a net current 25,26
$$I = I ^ { \cdot } - I ^ { + } = \frac { e } { L } \sum _ { k } \nu f ^ { \cdot } - \frac { e } { L } \sum _ { k } \nu f ^ { + } \.$$
Here I -and I + are the currents of left-moving and right-moving electrons respectively, and f -and f + are the corresponding Fermi-Dirac distributions describing the electron density of state. With considering
$$\sum _ { k } \rightarrow 2 \times \frac { L } { 2 \pi } \int d k \$$
the net current can be written as -+ 2 ( ) e I u u h = × -25,26 . The factor 2 in Eq. (3) comes from the spin degree of freedom. In the zigzag GNRs described by Hamiltonian (1), the factor 2 also refers to the two spin-polarized edges. Consequently we obtain the conductance around the charge neutrality as 2 2 I e G V h = = , which could explain quite well the result of the zigzag GNRs with L ≤ 160 nm, as shown in Fig. 1. An in-plane electric field applied across a zigzag GNR lifts the spin degeneracy of edge states and the GNR could be forced into a half-metallic state 21 . Then only one spin state (either spin-up or spin-down state) at the two edges contributes to the transport around the charge neutrality point. In such a case, the factor 2, referring to the two edge channels in the zigzag GNR, still plays its role in Eq. (3). Consequently, the conductance is still G 0 . Obviously, the above analysis cannot explain the result of the zigzag GNRs with
L > 160 nm, not to mention the crossover of the conductance between G 0 and G 0 /2 in the GNRs as a function of the probe spacing, as shown in Fig. 1. The experimental result in Ref. [15] indicates that the probe spacing plays a vital role in the emergence of the single-channel ballistic transport in the GNRs.
Spin-orbit coupling of edge atoms in zigzag GNR. To fully understand the puzzling experimental result in Ref. [15], we propose that effect of hybridization-induced sp 3 distortion of edge atoms should be taken into account in the transport properties of the zigzag GNRs. The existence of substantial amounts of adatoms (impurities) is inevitable even under ultraclean high vacuum conditions. These impurities hybridize directly with carbon atoms at edges of GNRs and induce a distortion of the graphene lattice from sp 2 to sp 3 19,27,28 . Such a distortion of edge atoms results in a large enhancement of the spin-orbit coupling on the edges 3,4,19,29
$$\mathcal { I }. c _ { \text{ \ } },$$
where N is the number of zigzag chains in the GNR, z s ↑ ↓ , is a Pauli matrix representing the electron's spin, λ = 10 meV is the strength of spin-orbit coupling, and n is the number of atoms with sp 3 orbital in the 1 st and N th zigzag chains, i.e. , in the two edges of the GNR. Then the total Hamiltonian of the system can be written as
$$H ^ { \prime } = H + H _ { e d g e } \.$$
The second term of Hamiltonian (5) enhances spin-flip scattering in zigzag GNRs and is expected to affect the transport properties of the edge states dramatically. As an example, if there is an atom at site i' with the sp 3 orbital in the 1 st zigzag chain, the spin-flip scattering induced by this atom will force the edge states to be a superposition of the spin-up and spin-down components because 1 , , , , , 1, edge i i i i i i i i i i H a c c a c c ϕ λ′ ϕ ′ ′ ′ ′+ ↑ ↑ ↑ ↑ ↑ ↓ Ψ = ∑ ⋅ + ∑ ⋅ (here
1 , i i i a ϕ ↑ ↑ Ψ = Ψ = ∑ is the initial ground state of the 1 st zigzag chain).
Edge state in a zigzag GNR with spin-orbit coupling. For generally, the wave functions of the edge states localized in the 1 st and N th zigzag chains can be written as a superposition of the spin-up and spin-down components:
$$\left | \Psi _ { _ { 1 } } \right \rangle = a \left | \Psi _ { _ { \uparrow } } \right \rangle + b \right | \Psi _ { _ { \downarrow } } \right \rangle,$$
$$\left | \Psi _ { _ { N } } \right \rangle = a \right | \Psi _ { _ { \downarrow } } \right \rangle + b \left | \Psi _ { _ { \uparrow } } \right \rangle \.$$
$$\Psi _ { \downarrow } \rangle, \\ \left | \Psi _ { \uparrow } \right \rangle \.$$
Here a and b are positive normalized parameters. The wave functions with a and b in opposite signs are not the eigenstates of Hamiltonian (5) with a positive spin-orbit coupling. To describe the effect of sp 3 distortion of edge atoms, we define p as the ratio of edge atoms with sp 3 orbital with respect to all of the edge atoms in GNR. For p = 0, we have a 2 = 1 and b 2 = 0 ( 1 ↑ Ψ = Ψ and N ↓ Ψ = Ψ ). Then Eq. (1) describes quite well the ground state of zigzag GNR, as shown in Fig. 2. When p ≠ 0, the wave functions of edge states depend on the probe spacing and we can obtain the
coefficients of the wave functions by numerically solving the Hamiltonian (5) in real space.
Figure 3(a) shows the coefficients, a 2 and b 2 , as a function of the probe spacing, i.e. , a 2 ( L ) and b 2 ( L ), for different p . Obviously, the values of a 2 and b 2 depend on both the probe spacing and the value of p . Here we take the result of p = 0.8 as an example to illustrate the effect of Hamiltonian (4) on the transport properties of zigzag GNRs. When L is of the order 100 nm or shorter, we still have a 2 ≈ 1 and b 2 ≈ 0, i.e. , 1 ↑ Ψ ≈ Ψ and N ↓ Ψ ≈ Ψ . It indicates that the effect of spin-flip scattering is negligible in such a length region even with p = 0.8. With increasing L from about 100 nm to about 1 μm, a 2 ( b 2 ) decreases (increases) gradually towards 0.5. For the case that L > 1 μm, we have a 2 ≈ b 2 ≈ 0.5, which means that the wave functions of edge states at the two edges become identical and can be written as 1 2 2 2 2 N ↑ ↓ Ψ ≈ Ψ ≈ Ψ + Ψ , as shown in Fig. 3(b). Then there is no spin degree of freedom in the zigzag GNR.
Edge-edge coupling in a zigzag GNR. Along with the variation of wave functions of the edge states, the coupling strength between the two edges also changes as a function of the probe spacing. The metallic probes, with their size of the tip much larger than the width of the GNRs, are expected to connect the two edges of the GNRs and therefore act as a medium for the edge-edge coupling during the transport
measurements. An edge-state-only model 22 with the form
$$( 7 )$$
can capture the essential features of this effect. Here describes the magnitude of edge-edge coupling and I is the identity matrix. Obvious, the edge-edge coupling is zero for the zigzag GNRs without sp 3 orbital in their edge atoms because that the two edge states are orthogonal to each other. However, for the superposition states, and , the magnitude of edge-edge coupling depends on the coefficients of the wave functions and will reach its maximum when . Figure 4(a) shows the relationship between t' and the probe spacing of a zigzag GNR with p = 0.8. For small probe spacing, i.e. , L < 100 nm, the edge-edge coupling is rather weak and it is expected that we could obtain the conductance G 0 around the charge neutrality. Whereas, in the strong coupling region, i.e. , L > 1 µ m, the factor 2 is removed from Eq. (3) and only a conductance of G 0 /2 is expected to be observed in the zigzag GNRs around the charge neutrality. Obviously, our model not only explains the single-channel ballistic transport reported in Ref. [15], but also provides a natural and reasonable explanation in understanding the crossover of the conductance between G 0 and G 0 /2 in the GNRs. The joint effect of the reduction of the spin degree of freedom and the edge-edge coupling in the GNRs leads to the decrease of conductance from G 0 1 a b ↑ ↓ Ψ = Ψ + Ψ N a b ↓ ↑ Ψ = Ψ + Ψ 1 2 2 2 2 N ↑ ↓ Ψ ≈ Ψ ≈ Ψ + Ψ
to G 0 /2 with increasing the probe spacing, as shown in Fig. 1. According to our calculation, the curve of a 2 ( L ) with p = 0.7, as shown in Fig. 1, matches the experimental result quite well.
Spin-spin correlation in a zigzag GNR. To further confirm our analysis, we calculate the spin-spin correlation χ of the edge states for the zigzag GNRs described by Hamiltonian (5). The spin-spin correlation is defined as 30-33
$$\chi = \left \langle \Psi \right | F \left | \Psi \right \rangle,$$
where is the wave function of the ground edge state, n' is the total number of atoms in one edge (the 1 st or the N th zigzag chain), and F is the spin-spin correlation function between two lattice sites i and j following the form (here we calculate that of the N th zigzag chain as an example) 32,33
$$\begin{smallmatrix} & & ( 9 ) \\ \vdots \end{smallmatrix} \cdot$$
Here denotes the local magnetic moments in the N th zigzag chain, ( ) is the number of spin-up (spin-down) electrons at site i and i z i i c s c S ↓ ↑ + = , 2 . In Fig. 4(b), the solid curve shows the spin-spin correlation as a function of the probe spacing in the N th chain of a zigzag GNR with p = 0.8. The value of χ decreases dramatically with increasing L because of the spin-flip scattering, and it approaches zero when L > 1 μm. The spin-spin correlation between the two
edges is also shown in Fig. 4(b) (the dashed curve). Its value increases substantially with increasing L and approaches 1 when L > 1 μm. Both the spin-spin correlation of each edge and the spin-spin correlation between the two edges show that the spin-flip scattering completely removes the spin degree of freedom of the GNR when the probe spacing is larger than 1 μm. This agrees quite well with the calculated result in Fig. 3(a).
## Discussion
Here, we should point out that all of our analyses are based on the GNRs with crystallographically perfect edges. A few imperfections of the edges could lead to dissipations of the edge states, which may account for the further reduction of the conductance in the zigzag GNRs with probe spacing larger than 16 μm (Fig. 1).
In summary, we propose a simple model that could explain quite well the puzzling result observed very recently in the zigzag GNR. According to our analysis, the decrease of conductance from G 0 to G 0 /2 with increasing the probe spacing, as observed experimentally in Ref. [15], direct arises from the joint effect of the reduction of the spin degree of freedom and the edge-edge coupling in the GNRs. The hybridization-induced sp 3 distortion of the carbon atoms enhances spin-flip scattering of edge states of the zigzag GNRs, which consequently removes the spin degree of freedom in the GNRs with sufficient long probe spacing. Then the wave function of
the edge states in these GNRs becomes a superposition of the spin-up and spin-down components. In the near future, more theoretical studies should be carried out to calculate transport properties of the superposition wave-function in zigzag GNRs. On the one hand, this could help us to quantitative understand the transport properties of the zigzag GNRs with probe spacing larger than 16 μm (as shown in Fig. 1); on the other hand, it may provide unprecedented opportunities to uncover novel electronic behaviors in zigzag GNRs.
*Email: [email protected]
## References:
- 1. Klitzing K, Dorda G, Pepper M (1980) New Method for High-Accuracy Determination of the Fine-Structure Constant Based on Quantized Hall Resistance. Phys Rev Lett 45(6):494-497.
- 2. Haldane FDM (1988) Model for a Quantum Hall Effect without Landau Levels: Condensed-Matter Realization of the 'Parity Anomaly.' Phys Rev Lett 61(18):20152018.
- 3. Kane CL, Mele EJ (2005) Quantum Spin Hall Effect in Graphene. Phys Rev Lett 95(22):226801.
- 4. Kane CL, Mele EJ (2005) Z2 Topological Order and the Quantum Spin Hall Effect. Phys Rev Lett 95(14):146802.
- 5. Bernevig BA, Hughes TL, Zhang S-C (2006) Quantum spin Hall effect and topological phase transition in HgTe quantum wells. Science 314(5806):1757-61.
- 6. König M, et al. (2007) Quantum spin hall insulator state in HgTe quantum wells. Science 318(5851):766-70.
- 7. Chang C-Z, et al. (2013) Experimental observation of the quantum anomalous Hall effect in a magnetic topological insulator. Science 340(6129):167-70.
- 8. Liang W, et al. (2001) Fabry - Perot interference in a nanotube electron waveguide. Nature 411(6838):665-9.
- 9. Li X, Wang X, Zhang L, Lee S, Dai H (2008) Chemically derived, ultrasmooth graphene nanoribbon semiconductors. Science 319(5867):1229-32.
- 10. Qi X-L, Zhang S-C (2011) Topological insulators and superconductors. Rev Mod Phys 83(4):1057-1110.
- 11. Han M, Özyilmaz B, Zhang Y, Kim P (2007) Energy Band-Gap Engineering of Graphene Nanoribbons. Phys Rev Lett 98(20):206805.
- 12. Wang X, et al. (2011) Graphene nanoribbons with smooth edges behave as quantum wires. Nat Nanotechnol 6(9):563-7.
- 13. Kosynkin D V, et al. (2009) Longitudinal unzipping of carbon nanotubes to form graphene nanoribbons. Nature 458(7240):872-6.
- 14. Cai J, et al. (2010) Atomically precise bottom-up fabrication of graphene nanoribbons. Nature 466(7305):470-3.
- 15. Baringhaus J, et al. (2014) Exceptional ballistic transport in epitaxial graphene nanoribbons. Nature 506(7488):349-54.
- 16. Nakada K, Fujita M, Dresselhaus G, Dresselhaus M (1996) Edge state in graphene ribbons: Nanometer size effect and edge shape dependence. Phys Rev B 54(24):1795417961.
- 17. Wakabayashi K, Fujita M, Ajiki H, Sigrist M (1999) Electronic and magnetic properties of nanographite ribbons. Phys Rev B 59(12):8271-8282.
- 18. Palacios JJ (2014) Graphene nanoribbons: Electrons go ballistic. Nat Phys 10(3):182-183.
- 19. Castro Neto A, Guinea F (2009) Impurity-Induced Spin-Orbit Coupling in Graphene. Phys Rev Lett 103(2):026804.
- 20. Gusynin V, Sharapov S (2005) Unconventional Integer Quantum Hall Effect in Graphene. Phys Rev Lett 95(14):146801.
- 21. Son Y-W, Cohen ML, Louie SG (2006) Half-metallic graphene nanoribbons. Nature 444(7117):347-9.
- 22. Jung J, MacDonald AH (2010) Magnetoelectric coupling in zigzag graphene nanoribbons. Phys Rev B 81(19):195408.
- 23. Wimmer M, Adagideli İ, Berber S, Tománek D, Richter K (2008) Spin Currents in Rough Graphene Nanoribbons: Universal Fluctuations and Spin Injection. Phys Rev Lett 100(17):177207.
- 24. Wang ZF, Jin S, Liu F (2013) Spatially Separated Spin Carriers in Spin-Semiconducting Graphene Nanoribbons. Phys Rev Lett 111(9):096803.
- 25. Ruan M (2012) Structured Epitaxial Graphene for Electronics. PhD thesis, Georgia Inst.Technol. available at http://hdl.handle.net/1853/45596.
- 26. Datta S (1995) Electronic Transport in Mesoscopic Systems (Cambridge University Press).
- 27. Duplock E, Scheffler M, Lindan P (2004) Hallmark of Perfect Graphene. Phys Rev Lett 92(22):225502.
- 28. Hong X, Zou K, Wang B, Cheng S-H, Zhu J (2012) Evidence for Spin-Flip Scattering and Local Moments in Dilute Fluorinated Graphene. Phys Rev Lett 108(22):226602.
- 29. Dugaev VK, Katsnelson MI (2014) Spin relaxation related to the edge scattering in graphene. arXiv:1404.6740v1.
- 30. Meng ZY, Lang TC, Wessel S, Assaad FF, Muramatsu A (2010) Quantum spin liquid emerging in two-dimensional correlated Dirac fermions. Nature 464(7290):84751.
- 31. Hikihara T, Hu X, Lin H-H, Mou C-Y (2003) Ground-state properties of nanographite systems with zigzag edges. Phys Rev B 68(3):035432.
- 32. Hirsch J (1985) Two-dimensional Hubbard model: Numerical simulation study. Phys Rev B 31(7):4403-4419.
- 33. Liang S, Doucot B, Anderson PW (1988) Some New Variational Resonating-Valence-Bond-Type Wave Functions for the Spin-½ Antiferromagnetic Heisenberg Model on a Square Lattice. Phys Rev Lett 61(3):365-368.
## Acknowledgments
We thank Prof. Yugui Yao and Prof. Hua Jiang for helpful discussions. We are grateful to National Key Basic Research Program of China (Grant No. 2014CB920903, No. 2013CBA01603), National Science Foundation (Grant No. 11422430, No. 11374035, No. 11004010), the program for New Century Excellent Talents in University of the Ministry of Education of China (Grant No. NCET-13-0054), and Beijing Higher Education Young Elite Teacher Project (Grant No. YETP0238).
## Author contributions
Z.D.C. performed the theoretical calculations. L.H. conceived and provided advice on the analysis and theoretical calculation. L.H. and Z.D.C. wrote the paper.
Competing financial interests: The authors declare no competing financial interests.
## Figure Legends:
FIG. 1. Room-temperature conductance G versus the probe spacing L in zigzag GNRs (the solid circles). The experimental data are reproduced from Ref. [15]. The zigzag GNRs are 40-nanometer-wide and have crystallographically perfect edges. There is a spin-degenerate channel exhibiting a conductance G = G0 for L ≤ 160 nm. With increasing the probe spacing from
160 nm to 1 µ m, the conductance decreases gradually from G0 to G0 /2, then, the conductance keeps a robust value of G0 /2 for 1 µ m ≤ L ≤ 16 µ m. The solid curve is the valve of a 2 (right-Y axis) calculated according to our model (Eq. (5)), which is explained in main text.
FIG. 2. Band structure of a zigzag GNR. (a) The spatial distribution of the charge difference between spin-up and spin-down for the ground states of an ideal zigzag GNR. The red and green balls represent the charge density of spin-down and spin-up electrons, respectively, and the radius of the circle reflects the magnitude of the charge density. (b) The band structure of a zigzag GNR obtained from Eq. (1). It is reproduced from Refs. [21,23].
FIG. 3. Spatial distribution of wave functions of edge states in zigzag GNR . (a) The relationship between coefficients, a 2 and b 2 , of edge wave functions and the probe spacing of zigzag GNR for different p . (b) The schematic spatial distribution of wave functions of edge states in zigzag GNR with p ≠ 0. The red and green balls represent the charge density of spin-down and spin-up electrons, respectively, the purple balls represent the superposition state of the spin-up and spin-down components. The radius of the circle reflects the magnitude of the charge density.
FIG. 4. The edge-edge coupling and spin-spin correlation in zigzag GNR. (a) The relationship between the magnitude of the edge-edge coupling (normalized) and the probe spacing in a zigzag GNR with p = 0.8. (b) The solid curve refers to the relationship between the spin-spin correlation and the probe spacing in a zigzag GNR with p = 0.8. The dashed curve represents the spin-spin correlation between the two edges of the same zigzag GNR.
Figure 1
Figure 2
Figure 3
Figure 4
 | 10.1007/s40843-015-0081-y | [
"Zhao-Dong Chu",
"Lin He"
] | 2014-08-02T03:01:18+00:00 | 2015-01-08T15:18:56+00:00 | [
"cond-mat.mes-hall",
"cond-mat.mtrl-sci"
] | Crossover from Room-temperature Double-channel Ballistic Transport to Single-channel Ballistic Transport in Zigzag Graphene Nanoribbons | Very recently, it was demonstrated explicitly that a zigzag graphene
nanoribbon (GNR) exhibits a crossover of conductance from G0 to G0/2 with
increasing the length (G0 = 2e2/h is the quantum of conductance) even at
room-temperature [Baringhaus, et al. Nature 506, 349 (2014)]. Such a result is
puzzling as none of previous theories seem to match the experimental
observations. Here, we propose a model to explain the crossover from
double-channel to single-channel ballistic transport in zigzag GNR. The sp3
distortion of carbon atoms at the GNR edges induces a large spin-orbit coupling
on the edges atoms, which enhances spin-flip scattering of edge states of the
zigzag GNR. With sufficient spin-flip scattering, the wave-function of edge
states becomes a superposition of the spin-up and spin-down components. Then
the coupling of the two edges becomes important. This removes the edge degree
of freedom in the zigzag GNR and results in the evolution of the conductance
from G0 to G0/2 with increasing the length. |
1408.0328v1 | ## Weakly monotone averaging functions
Tim Wilkin, Gleb Beliakov School of Information Technology, Deakin University, 221 Burwood Hwy, Burwood 3125, Australia [email protected], [email protected]
## Abstract
Monotonicity with respect to all arguments is fundamental to the definition of aggregation functions. It is also a limiting property that results in many important non-monotonic averaging functions being excluded from the theoretical framework. This work proposes a definition for weakly monotonic averaging functions, studies some properties of this class of functions and proves that several families of important non-monotonic means are actually weakly monotonic averaging functions. Specifically we provide sufficient conditions for weak monotonicity of the Lehmer mean and generalised mixture operators. We establish weak monotonicity of several robust estimators of location and conditions for weak monotonicity of a large class of penalty-based aggregation functions. These results permit a proof of the weak monotonicity of the class of spatial-tonal filters that include important members such as the bilateral filter and anisotropic diffusion. Our concept of weak monotonicity provides a sound theoretical and practical basis by which (monotone) aggregation functions and non-monotone averaging functions can be related within the same framework, allowing us to bridge the gap between these previously disparate areas of research.
Keywords: aggregation functions, monotonicity, means, penalty-based functions, non-monotonic functions
## 1. Introduction
The aggregation of several input variables into a single representative output arises naturally as a problem in many practical applications and domains. The research effort has been disseminated throughout various fields including economics, computer science, mathematics and engineering, with the subsequent mathematical formulation of aggregation problems having coalesced into a significant body of knowledge concerning aggregation functions. A wide range of aggregation functions are presented in the literature, including the weighted quasi-arithmetic means, ordered weighted averages, triangular norms and co-norms, Choquet and Sugeno integrals and many more. Several recent books provide a comprehensive overview of this field of study (Beliakov et al. [5], Grabisch et al. [17], Torra [32]).
Aggregation functions are commonly used within fuzzy logic, where logical connectives are typically modeled using triangular norms and triangular co-norms. Beyond this field the averaging functions - more commonly known as means -that are frequently applied in decision problems, statistical analysis and in image and signal processing. Means have been an important tool and topic of study for over two millennia, with examples such as the arithmetic, geometric and harmonic means known to the Greeks (Rubin [28]). Each of these means shares a fundamental property with the broader class of aggregation functions; that of monotonicity with respect to all arguments (Beliakov et al. [5], Grabisch et al. [17], Torra [32]). There are though many means of significant practical and theoretical importance that are non-monotonic and hence not classified as aggregation functions. For example, a non-monotonic average of pixel intensities within an image subset is used to perform tasks such as image reduction (Wilkin [35]), filtering (van den Boomgaard and van de Weijer [34], Sylvain et al. [30]) or smoothing (Barash and Comaniciu [3]). Within statistics, robust estimators of location are used to estimate the central tendency of a data set and the mode, an average possibly known to the Greeks (Rubin [29]), is a classic example of a non-monotonic average.
Monotonicity with respect to all arguments has an important interpretation in decision making problems: an increase in one criterion should not lead to the decrease of the overall score or utility. However, in image processing an increase in only one pixel value above its neighbours may be due to noise or corruption and should not necessarily increase the intensity value that represents that region. Accordingly, the averaging functions used in
such applications do not fit within the established theories regarding aggregation functions and are typically dealt with only from the signal processing perspective.
There are also many non-monotonic means appearing in the literature, with the mode, Gini means, Lehmer means, Bajraktarevic means (Beliakov et al. [5], Bullen [8]) and mixture functions (Ribeiro and Marques Pereira [24], Marques Pereira and Ribeiro [20]) being particularly well known cases. Ideally we would like a formal framework for averaging functions that encompasses non-monotonic means and places them in context with existing monotonic aggregation functions, enabling us to better understand the relationships within this broad class of functions. In so doing we are then able to broaden our understanding of non-monotonic averaging as an aggregation problem.
We achieve this aim herein by relaxing the monotonicity requirement for averaging aggregation functions and propose a new definition that encompasses many non-monotonic averaging functions. We justify this approach by the following interpretation of averaging: while we accept that an increase in one input, or coordinate, may lead to a decrease of the aggregate value, we argue that the same increase coincident in all inputs should only lead to an increase of the aggregate value. This is akin to the property of shiftinvariance, which along with homogeneity is one of the basic requirements of the non-monotonic location estimators (Rousseeuw and Leroy [27]). We do not impose shift-invariance though, as that would severely limit the range of averaging functions that fall under our definition averaging functions (for instance, the only shift invariant quasi-arithmetic means are weighted arithmetic means). Rather we consider the property of directional monotonicity in the direction of the vector (1 , 1 , . . . , 1), which is obviously implied by shiftinvariance as well as by the standard definition of monotonicity. We call this property weak monotonicity within the context of aggregation functions and we investigate it herein.
The remainder of this article is structured as follows. In Section 2 we provide the necessary mathematical foundations that underpin the subsequent material. Section 3 provides the main definitions and presents various properties of weakly monotone aggregation functions. Within Section 4 we examine several non-monotonic means and prove that they are, in fact, weakly monotonic. In Section 5 we draw our conclusions and discuss future research directions arising as a result of this investigation.
## 2. Preliminaries
## 2.1. Aggregation functions
In this article we make use of the following notations and assumptions. Without loss of generality we assume that the domain of interest is any closed, non-empty interval I ⊆ ¯ R = [ -∞ ∞ , ] and that tuples in I n are defined as x = ( x i,n | n ∈ N , i ∈ { 1 , ..., n } ). We write x i as the shorthand for x i,n such that it is implicit that i ∈ { 1 , ..., n } . Furthermore, I n is ordered such that for x y , ∈ I n , x ≤ y implies that each component of x is no greater than the corresponding component of y . Unless otherwise stated, a constant vector given as a is taken to mean a = a (1 , 1 , . . . , 1) = ︸ ︷︷ ︸ n times a 1 , where a ∈ R is a constant and n is implicit within the context of use.
The vector x ↗ denotes the result of permuting the vector x such that its components are in non-decreasing order, that is, x ↗ = x σ , where σ is the permutation such that x σ (1) ≤ x σ (2) ≤ . . . ≤ x σ n ( ) . Similarly, the vector x ↘ denotes the result of permuting x such that x σ (1) ≥ x σ (2) ≥ . . . ≥ x σ n ( ) . We will make use of the common shorthand notation for a sorted vector, being x () = ( x (1) , x (2) , . . . , x ( n ) ). In such cases the ordering will be stated explicitly and then x ( k ) represents the k th largest or smallest element of x accordingly. Consider now the following definitions:
Definition 1. A function F : I n → ¯ is R monotonic (non-decreasing) if and only if, ∀ x y , ∈ I n , x ≤ y then F ( x ) ≤ F ( y ).
Definition 2. A function F : I n → I is an aggregation function in I n if and only if F is monotonic non-decreasing in I and F ( a ) = a , F ( b ) = b , with I n = [ a, b ] n .
Definition 3. A function F is called idempotent if for every input x = ( t, t, ... , t ) , t ∈ I the output is F ( x ) = t .
The functions of most interest in this article are those that have averaging behaviour.
Definition 4. A function F has averaging behaviour (or is averaging) if for every x it is bounded by min( x ) ≤ F ( x ) ≤ max( x ) .
Aggregation functions that have averaging behaviour are idempotent, whereas idempotency and monotoicity imply averaging behaviour.
Definition 5. A function is called internal if its value coincides with one of the arguments.
Of particular relevance is the notion of shift-invariance Calvo et al. [10], L´ azaro et al. [19] (which is also called difference scale invariance Grabisch et al. [17]). A constant change in every input should result in a corresponding change of the output.
Definition 6. A function F : I n → I is shift-invariant (stable for translations) if F ( x + a 1 ) = F ( x ) + a whenever x x , + a 1 ∈ I n .
Definition 7. A function F is homogeneous (with degree one) if F a ( x ) = aF ( x ) for all a x ∈ I n .
Aggregation functions that are shift-invariant and homogeneous are known as linear aggregation functions . The canonical example of a linear aggregation function is the arithmetic mean.
## 2.2. Means
The term mean is used synonymously with averaging aggregation functions. Chisini's definition of a mean as an average states that the mean of n independent variables ( x , ..., x 1 n ), with respect to a function F , is a value M for which replacement of each value x i in the input by M , results in the output M (Chisini [12], stated in Grabisch et al. [17]). I.e.,
$$F ( x _ { 1 }, \dots, x _ { n } ) = F ( M, \dots, M ) = M$$
As was noted by de Finetti (de Finetti [13], stated in Grabisch et al. [17]), Chisini's definition does not necessarily satisfy Cauchy's requirement that a mean be an internal value (Cauchy [11]). However, by assuming that F is a non-decreasing, idempotent function, then existence, uniqueness and internality of M are restored to Chisini's definition. Gini ( Gini [16], p.64), writes that an average of several quantities is a value obtained as a result of a certain procedure, which equals to either one of the input quantities, or a new value that lies in between the smallest and the largest input. The requirement that F be non-decreasing is too strict given the aims of this article and as such, following many authors (e.g., Gini [16], Bullen [8]), we take the definition of a mean to be any averaging (and hence idempotent) function.
Definition 8. A function M : I n → I is called a mean if and only if it is averaging.
The basic examples of (monotonic) means found within the literature include weighted arithmetic mean, weighted quasi-arithmetic mean, ordered weighted average (OWA), order statistic S k ( x ) = x ( k ) , and the median. Less known examples include Choquet and Sugeno integrals and their special cases; the logarithmic means, Heronian means, Bonferroni means and others Bullen [8], Grabisch et al. [17].
In continuing, we wish to consider a broader class of means to include those that are not necessarily monotonic. A classic example is the mode, being the most frequent input, which is routinely used in statistics. 1 The mode is not monotonic as the following example shows. Taking the vectors x = (1 1 2 2 3 3 3) , , , , , , , y = (1 1 0 0 0 0 0), and , , , , , , z = (1 1 1 1 1 1 1), then , , , , , , Mode ( x ) = 3 , Mode ( x + y ) = 2 and Mode ( x + ) = 4. z
An important class of means that are not always monotonic are those expressed by the Mean of Bajraktarevic, which is a generalisation of the weighted quasi-arithmetic means.
Definition 9. Mean of Bajraktarevic . Let w ( ) = ( t w 1 ( ) t , ..., w n ( )) be a t vector of weight functions w i : I → [0 , ∞ ), and let g : I → ¯ R be a strictly monotonic function. The mean of Bajraktarevic is the function
$$\text{union. The mean of Bajraktarevic is the function} \\ M _ { w } ^ { g } ( x ) = g ^ { - 1 } \left ( \frac { \sum _ { i = 1 } ^ { n } w _ { i } ( x _ { i } ) g ( x _ { i } ) } { \sum _ { i = 1 } ^ { n } w _ { i } ( x _ { i } ) } \right ) & ( 2. 1 ) \\ x _ { i } ) = x _ { i }, \, \text{and all weight functions are the same, the Bajraktarevic} \\ \text{ed a mixture function (or mixture another} ) \text{ and is given}$$
When g x ( i ) = x i , and all weight functions are the same, the Bajraktarevic mean is called a mixture function (or mixture operator ) and is given by
$$\dots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \\ M _ { w } ( \mathbf x ) = \frac { n } { \sum _ { i = 1 } ^ { n } w ( x _ { i } ) x _ { i } } \\ \sum _ { i = 1 } ^ { n } w ( x _ { i } )$$
1 In general the mode is multivalued, so in order to make it a single-variate function, a convention is needed to select one of the multiple outputs, e.g. the smallest.
For the case where the weight functions are are distinct w x i ( i ), the operator M w ( x ) is a generalised mixture function . A particularly interesting sub-class of Bajraktarevic means are Gini means, obtained by setting w t i ( ) = w t i q and g t ( ) = t p when p = 0, or g t ( ) = log( ) if t p = 0.
/negationslash
$$\ddot { \ } r ^ { \prime } \ \times, \ \triangleleft ( \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \right ) \ \triangleleft ( \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \xi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ } \xi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \chi _ { 0 } \\ \text{$s$ generalise the (weighted)} \ \text{power means} \ ( \text{for} \, q = 0 ) \ \text{and hence} \ \\ \text{inimum} \. \ \maximum \ \text{and} \ the arithmetic mean as special cases.}$$
Gini means generalise the (weighted) power means (for q = 0) and hence include the minimum, maximum and the arithmetic mean as special cases. Another special case of the Gini mean is the Lehmer, or counter-harmonic mean, obtained when p = 1. The contra-harmonic mean is the Lehmer mean with q = 1. We will investigate the Lehmer mean and its properties further in Section 4.
## 2.3. Penalty based functions
In Calvo and Beliakov [9] it was demonstrated that averaging aggregation functions can be expressed as the solution of a minimisation problem of the form
$$F ( { \mathbf x } ) = \arg \min _ { y } \mathcal { P } ( { \mathbf x }, y )$$
where P ( x , y ) is a penalty function satisfying the following definition:
Definition 10. Penalty function . The function P : I n +1 → R is a penalty function if and only if it satisfies:
- 1. P ( x , y ) ≥ c ∀ x ∈ I n , y ∈ I ;
- 3. P ( x , y ) is quasi-convex in y for any x ,
- 2. P ( x , y ) = c if and only if all x i = y ; and,
for some constant c ∈ R and any closed, non-empty interval I .
A function P is quasi-convex if all its sublevel sets are convex, that is S α ( P ) = { x P | ( x ) ≤ α } are convex sets for all α , see Rockafellar [25]. The first two conditions ensure that P has a strict minimum and that a consensus of inputs ensures minimum penalty, providing idempotence of F ( x ). The third condition implies a unique minimum (but possibly many minimisers
that form a convex set). Since multiplication by, or addition of a constant to P will not change the minimisation, P may be shifted (if desired) so that c = 0. One can think of P as describing the dissimilarity or disagreement between the inputs x and the value y . It follows that F is a function that minimises the chosen dissimilarity. It is not necessary to explicitly state F , provided a suitable penalty function is given and the optimisation problem solvable. Subsequently it is sufficient to solve (2.4) to obtain the aggregate µ = F ( x ).
Non-monotonic averaging functions can also be represented by a penalty function. For penalty-based functions we have the following results due to Calvo and Beliakov [9].
Theorem 1. Any idempotent function F : I n → I can be represented as a penalty based function P : I n +1 → I such that
$$F ( { \mathbf x } ) = \arg \min _ { y } \mathcal { P } ( { \mathbf x }, y ).$$
Corollary 1. Any averaging function can be expressed as a penalty based function.
As mentioned in Mesiar et al. [22], mixture functions can be written as a penalty function with
$$\mathcal { P } ( \mathbf x, y ) = \sum _ { i = 1 } ^ { n } w ( x _ { i } ) ( x _ { i } - y ) ^ { 2 }. \\ \text{ar condition of the minimum is}$$
Clearly the necessary condition of the minimum is
$$\mathcal { P } _ { y } ( \mathbf x, y ) = - 2 \left ( \sum _ { i = 1 } ^ { n } w ( x _ { i } ) x _ { i } - y \sum _ { i = 1 } ^ { n } w ( x _ { i } ) \right ) = 0. \\ \mathcal { P } ( \mathbf x, u ) \text{ defines a mixture function. } A \text{ representation of a funct}$$
Hence P ( x , y ) defines a mixture function. A representation of a function as a penalty based function sometimes can simplify technical proofs, as we shall see later in the paper.
It is apparent given the examples presented that many means are nonmonotonic and thus not aggregation functions according to Definition 2. In the next section we introduce weak monotonicity and consider some properties of weakly monotonic averaging functions. We subsequently investigate several important examples and show that they are indeed weakly monotonic functions, allowing us to place them in a new framework with existing averaging aggregation functions.
## 3. Weak monotonicity
## 3.1. Main definition
As mentioned in Section 1 we are motivated by two important issues. The first one is that there exist many means that are not generally monotonic and hence not aggregation functions, while the second one is that there are many practical applications in which non-monotonic means have shown to provide good aggregate values commensurate with the objectives of the aggregation. To encapsulate these non-monotonic means within the framework of aggregation functions we aim to relax the monotonicity condition and present the class of weakly monotonic averaging functions . The definition of weak monotonicity provided herein is prompted by applications and intuition, which suggests that it is reasonable to expect that a representative value of the inputs does not decrease if all the inputs are increased by the same amount (or shifted uniformly) as the relative positions of the inputs are not changed. A formal definition that conveys this property is as follows.
Definition 11. A function f is called weakly monotonic non-decreasing (or directionally monotonic) if F ( x + a 1 ) ≥ F ( x ) for any a > 0 , (1 , 1 , . . . , 1) ︸ ︷︷ ︸ n times , such that x x , + a 1 ∈ I n .
Remark 1 . If F is directionally differentiable in its domain then weak monotonicity is equivalent to non-negativity of the directional derivative D 1 ( F )( x ) ≥ 0.
Remark 2 . Evidently monotonicity implies weak monotonicity, hence all aggregation functions are weakly monotonic. By Definition 6 all shift-invariant functions are also weakly monotonic. It is self evident that weakly monotonic non-decreasing functions form a cone in the linear vector space of weakly monotonic (increasing or decreasing) functions.
## 3.2. Properties
Let us establish some useful properties of weakly monotonic averages. Consider the function F : I n → I formed by the composition F ( x ) = A B ( 1 ( x ) , B 2 ( x )), where A, B 1 and B 2 are means.
Proposition 1. If A is monotonic and B , B 1 2 are weakly monotonic, then F is weakly monotonic.
Proof. By weak monotonicity B i ( x + a 1 ) ≥ B i ( x ) implies that ∃ δ i ≥ 0 such that B i ( x + a 1 ) = B i ( x ) + δ i , with x x , + a 1 ∈ I n . Thus F ( x + a 1 ) = A b ( 1 + δ , b 1 2 + δ 2 ), where b i = B i ( x ). The monotonicity of A ensures that A b ( 1 + δ , b 1 2 + δ 2 ) ≮ A b , b ( 1 2 ) and hence F ( x + a 1 ) ≥ F ( x ) and F is weakly monotonic.
By trivial extension, since all monotonic functions are also weakly monotonic, then if either of B 1 or B 2 is monotonic, then F is again weakly monotonic.
Proposition 2. If A is weakly monotonic and B , B 1 2 are shift invariant, then F is weakly monotonic.
Proof. Shift invariance implies that ∀ a : B i ( x + a 1 ) = B i ( x ) + a , with x x , + a 1 ∈ I n . Thus F ( x + a 1 ) = A b ( 1 + a, b 2 + a ), where b i = B i ( x ). The weak monotonicity of A ensures that A b ( 1 + a, b 2 + ) a ≮ A b , b ( 1 2 ) and hence F ( x + a 1 ) ≥ F ( x ) and F is weakly monotonic.
Consider functions of the form ϕ ( x ) = ( ϕ x ( 1 ) , ϕ ( x 2 ) , ..., ϕ ( x n )).
Proposition 3. If A is weakly monotonic and ϕ x ( ) is a linear function then the ϕ -transform A ϕ ( x ) = F ( x ) = ϕ -1 ( A ϕ ( ( x ))) is weakly monotonic.
Proof. ϕ x ( ) = αx + β and hence ϕ x ( + ) = a α x ( + ) + a β = αx + β + αa = ϕ x ( ) + c . Hence
$$c. \text{ENE} \\ F ( x + a \mathbf 1 ) & = \varphi ^ { - 1 } \left ( A ( \varphi ( x _ { 1 } + a ), \dots, \varphi ( x _ { n } + a ) ) \right ) \\ & = \varphi ^ { - 1 } \left ( A \left ( \varphi ( \mathbf x ) + c \mathbf 1 \right ) \right ) \\ & = \frac { A \left ( \varphi ( \mathbf x ) + c \mathbf 1 \right ) - \beta } { \alpha } \\ & \geq \frac { A \left ( \varphi ( \mathbf x ) \right ) - \beta } { \alpha } \\ & = \varphi ^ { - 1 } \left ( A ( \varphi ( \mathbf x ) ) \right ) \\. \text{$1,\quad \, \stackrel { \sim } { \longrightarrow } \, \stackrel { \sim } { \longrightarrow } \, \stackrel { \sim } { \longrightarrow } \, \stackrel { \sim } { \longrightarrow } \, \stackrel { \sim } { \longrightarrow } \, \stackrel { \sim } { \longrightarrow } \, \stackrel { \sim } { \longrightarrow } \, \stackrel { \sim } { \longrightarrow } \, \stackrel { \sim } { \longrightarrow } \, \stackrel { \sim } { \longrightarrow } \, \stackrel { \sim } { \longrightarrow } \, \stackrel { \sim } { \longrightarrow } \, \stackrel { \sim } { \longrightarrow } \, \stackrel { \sim } { \longrightarrow } \, \stack | \, \stackrel { \sim } { \longrightarrow } \, \stackrel { \sim } { \longrightarrow } \, \stackrel { \sim } { \longrightarrow } \, \stackrel { \sim } { \longrightarrow } \, \stackrel { \sim } { \longrightarrow } \, \stackrel { \sim } { \longrightarrow } \, \stackrel { \sim } { \longrightarrow } \, \stackrel { \sim } { \longrightarrow } \, \stackrel { \sim } { \longrightarrow } \, \stackrel { \sim } { \longrightarrow } \, \stackrel { \sim } { \longrightarrow } \, \stackrel { \sim } { \longrightarrow } \, \stackrel { \sim } { \longrightarrow } \, \stack.$$
by weak monotonicity of A . Hence F ( x + a 1 ) ≥ F ( x ) and F is weakly monotonic.
Note that unlike in the case of standard monotonicity, a nonlinear ϕ -transform does not always preserve weak monotonicity.
Corollary 2. The dual A d of a weakly monotonic function A is weakly monotonic under standard negation.
The following result is relevant to an application of weakly monotonic averages in image processing, discussed in Section 4.3.
Theorem 2. Let f : I n → I be a shift invariant function, and g be a function. Let F be a penalty based averaging function with the penalty P depending on the terms g ( x i -f ( x )) ( x i -y ) 2 . Then F is shift-invariant and hence weakly monotonic.
## Proof. Let
$$\mu & = \arg \min _ { y } \mathcal { P } \left ( g \left ( x _ { 1 } - f ( \mathbf x ) \right ) \left ( x _ { 1 } - y \right ) ^ { 2 }, \dots, g \left ( x _ { n } - f ( \mathbf x ) \right ) \left ( x _ { n } - y \right ) ^ { 2 } \right ). \\ \text{Then}$$
Then
$$\arg \min _ { y } \mathcal { P } ( \mathbf x + a \mathbf 1, y ) = \arg \min _ { y } \mathcal { P } \left ( g \left ( x _ { 1 } + a - f ( \mathbf x + a \mathbf 1 ) \right ) ( x _ { 1 } + a - y ) ^ { 2 }, \dots \\ \dots, g \left ( x _ { n } + a - f ( \mathbf x + a \mathbf 1 ) \right ) ( x _ { n } + a - y ) ^ { 2 } \right ) = \\ \hat { \lambda } \quad \dots \. \quad \hat { \lambda }$$
(by shift invariance)
$$& = \arg \min _ { y } \mathcal { P } \left ( g \left ( x _ { 1 } - f ( \mathbf x ) \right ) \left ( x _ { 1 } + a - y \right ) ^ { 2 }, \dots, g \left ( x _ { n } - f ( \mathbf x ) \right ) \left ( x _ { n } + a - y \right ) ^ { 2 } \right ) \\ & = \mu + a.$$
Remark 3 . Indeed we need not restrict ourselves to penalty functions with terms depending on ( x i -y ) 2 . Functions D that depend on the differences x i -y with the minimum D (0) will satisfy the above proof and satisfy the conditions on P with regards to the existence of solutions to (2.4). In particular, Huber type functions used in robust regression can replace the squares of the differences.
## 3.3. Counter-cases
For the ϕ -transform, if ϕ is nonlinear then F may or may not be weakly monotonic for all x , which can be observed by example.
Example 1. Take x = (1 8 16 35 47 9) and , , , , . ϕ t ( ) = √ t , then ϕ ( x ) = (1 , 2 √ 2 4 , , √ 35 , √ 47 9) and . ϕ ( x + 1 ) = ( √ 2 3 , , √ 17 6 , , √ 48 9). . If A is the shorth (we prove the weak monotonicity of the shorth in Section 4) then A ϕ ( ( x )) = 5 61 and . A ϕ ( ( x + )) = 2 84. As 1 . ϕ -1 ( ) = t t 2 clearly 5 61 . 2 > 2 84 . 2 and F = A ϕ is not weakly monotonic.
Internal means are not necessarily weakly monotonic, as illustrated by the following example.
$$\text{Example 2. Take } x = ( x _ { 1, } x _ { 2 } ) \in [ 0, 1 ] ^ { 2 } \text{ and }$$
$$F ( { \mathbf x } ) = \begin{cases} \min ( { \mathbf x } ) & \text{if $x_{1}+x_{2}\geq 1$} \\ \max ( { \mathbf x } ) & \text{otherwise} \end{cases}$$
which is internal with values in the set { min( x ) , max( x ) } . Consider the points x = ( 1 / , 4 0) and y = ( 3 / , 4 0), then F ( x ) = 1 / 4 and F ( y ) = 3 / 4 . It follows that F ( x + 1 / 4 1 ) = 1 / 2 > F ( x ), however F ( y + 1 / 4 1 ) = 1 / 4 < F ( y ). Hence this F is not weakly monotonic for all x ∈ I 2 .
## 4. Examples of weakly monotonic means
In this section we look at several examples of weakly monotonic, but not necessarily monotonic averaging functions. We begin by considering several of the robust estimators of location, then move on to mixture functions and some interesting cases of means from the literature. While some of the examples involve shift-invariant functions, many of their nonlinear ϕ -transforms yield proper weakly monotonic functions.
The functions presented below are defined through penalties that are not quasi-convex, therefore we need to drop the condition that P is quasi-convex from Definition 10.
Definition 12. Quasi-penalty function . The function P : I n +1 → R is a quasi-penalty function if and only if it satisfies:
- 1. P ( x , y ) ≥ c ∀ x ∈ I n , y ∈ I ;
- 3. P ( x , y ) is lower semi-continuous in y for any x ,
- 2. P ( x , y ) = c if and only if all x i = y ; and,
for some constant c ∈ R and any closed, non-empty interval I .
Note that the third condition ensures the existence of the minimum and a non-empty set of minimisers. In the case where the set of minimisers of P is not an interval, we need to adopt a reasonable rule for selecting the value of the penalty-based function F . We suggest stating in advance that in such cases we choose the infimum of the set of minimisers of P . From now one P will refer to quasi-penalty functions.
## 4.1. Estimators of Location
Perhaps the most widely used estimator of location is the mode, being the most frequent input.
Example 3. Mode : The mode is the minimiser of the (quasi)penalty function
$$\mathcal { P } ( \mathbf x, y ) = \sum _ { i = 1 } ^ { n } p ( x _ { i }, y ) \quad \text{ where } \ p ( x _ { i }, y ) = \begin{cases} 0 & x _ { i } = y \\ 1 & \text{ otherwise} \end{cases}.$$
It follows that F ( x + a 1 ) = arg min y P ( x + a 1 , y ) = arg min y n ∑ i =1 p x ( i + a, y ), which is minimised for the value y = F ( x ) + a . Hence, F ( x + a 1 ) = F ( x ) + a and thus the mode is shift invariant. By Definition 6 the mode is weakly monotonic.
Remark 4 . Note that the mode may not be uniquely defined, e.g., the mode of (1 , 1 2 2 3 4 5), in which case we use a suitable convention. , , , , , The quasipenalty P associated with the mode is not quasi-convex, and as such it may have several minimisers. A convention is needed as to which minimiser is selected, e.g., the smallest or the largest. Other examples of non-monotonic means that follow also involve quasi-penalties, and the same convention as for the mode is adopted. Then also discrete scales can be considered, compare to, e.g., the paper of Koles´rov´ et al. [18]. a a
The Least Trimmed Squares estimator (Rousseeuw and Leroy [27]) rejects up to 50% of the data values as outliers and minimises the squared residual using the remaining data.
Example 4. Least Trimmed Squares (LTS): The LTS uses the (quasi)penalty function
$$\mathcal { P } ( \mathbf x, y ) & = \sum _ { i = 1 } ^ { h } r _ { ( i ) } ^ { 2 } \\ \dots &$$
where r ( ) i = S i ( r ) is the i th order statistic of r , r k = x k -y and h = ⌊ n 2 ⌋ +1. If σ is the order permutation of { 1 , ..., n } such that r σ = r ↗ , then the minima value is µ = 1 h h ∑ i =1 x σ i ( ) . Since S k ( x ) is shift invariant then S i ( r + a 1 ) = r σ i ( ) + a and thus
$$\Pi \circ \infty \, \text{and} \, \text{p} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \, \text{in} \,
\text{of} \, \mathcal { P } \, \text{occur when} \, P _ { y } = - 2 \sum _ { i = 1 } ^ { h } ( x _ { \sigma ( i ) } - y ) = 0, \, \text{which implies that the minimum} \\ \xi \quad \cdot \quad \frac { h } { 1 } \sum _ { i = 1 } ^ { h } ( x _ { \sigma ( i ) } - y ) \cdot \quad \alpha / 2 \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \,.$$
$$\mathcal { P } ( \mathbf x + a \mathbf 1, y ) & = \sum _ { i = 1 } ^ { h } v _ { \sigma ( i ) } ^ { 2 } \\ ) \, - \, y ). \quad \text{It follows that the v}$$
where v k = (( x k + a ) -y ). It follows that the value y that minimises P ( x + a 1 , y ) is y = µ + , hence the LTS is shift invariant and thus weakly a monotonic.
The remaining estimators of location presented compute their value using the shortest contiguous sub-sample of x containing at least half of the values. The candidate sub-samples are the sets X k = { x j : j ∈ { k, k + 1 , ..., k + ⌊ n 2 ⌋ } , k = 1 , ..., ⌊ n +1 2 ⌋ . The length of each set is taken as ‖ X k ‖ = ∣ ∣ ∣ x k + /floorleft n 2 /floorright -x k ∣ ∣ ∣ and thus the index of the shortest sub-sample is
$$k ^ { * } & = \arg \min _ { i } \left \| X _ { i } \right \|, \ i = 1, \dots, \left \lfloor \frac { n + 1 } { 2 } \right \rfloor. \\ \text{ren-action} \bar { \nu } - \mathbf v \perp \alpha \mathbf 1 \ t h a \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{} \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{$$
Under the translation ¯ = x x + a 1 the length of each sub-sample is unaltered since ∥ ∥ ¯ X k ∥ ∥ = ¯ ∣ ∣ ∣ x k + /floorleft n 2 /floorright -¯ x k ∣ ∣ ∣ = ( ∣ ∣ ∣ x k + /floorleft n 2 /floorright + ) a -( x k + ) a ∣ ∣ ∣ = ∣ ∣ ∣ x k + /floorleft n 2 /floorright -x k ∣ ∣ ∣ = ‖ X k ‖ and thus k ∗ remains the same.
Consider now the Least Median of Squares estimator (Rousseeuw [26]), which is the midpoint of X k ∗ .
Example 5. Least Median of Squares (LMS): The LMS can be computed by minimisation of the (quasi)penalty function
$$\Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lamda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda\quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad\Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \ Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambdelta \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \ \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \ LABda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda \quad \Lambda
\text{rule $y$ minimises the penalty $P(x+a$1,y)$}, given by$$
$$\begin{array} { c } \lambda & \cdot & \cdot & \cdot & \cdot & \cdot & \cdot \\ \min _ { y } \mathcal { P } ( \mathbf x + a \mathbf 1, y ) = \min _ { y } \ m e d i a n \left \{ ( x _ { j } + a - y ) ^ { 2 } \, | y \in \mathbb { I }, x _ { j } \in X _ { k ^ { * } } \, \right \} = \mathcal { P } ( \mathbf x, \mu ), \\ \text{is clearly } y \, = \, \mu + a. \, \text{ Hence, } F ( \mathbf x + a \mathbf 1 ) \, = \, F ( \mathbf x ) + a \text{ and the LMS is shift} \end{array}$$
The value y minimises the penalty P ( x + a 1 , y ), given by is clearly y = µ + a . Hence, F ( x + a 1 ) = F ( x ) + a and the LMS is shift invariant and weakly monotonic.
The Shorth (Andrews et al. [1]) is the arithmetic mean of X k ∗
Example 6. Shorth: The shorth is given by
$$F ( \mathbf x ) = \frac { 1 } { h } \sum _ { i = 1 } ^ { h } x _ { i }, \ x _ { i } \in X _ { k ^ { * } }, \ h = \left \lfloor \frac { n } { 2 } \right \rfloor + 1 \\ \det X.. \text{is maltered under transient and the arithmetic}$$
Since the set X k ∗ is unaltered under translation and the arithmetic mean is shift invariant, then the shorth is shift invariant and hence weakly monotonic.
## Example 7. OWA Penalty Functions : Penalty functions having the form
$$\mathcal { P } ( \mathbf x, y ) = \sum _ { i = 1 } ^ { n } w _ { i } S _ { i } \left ( \left ( \mathbf x - y \mathbf 1 \right ) ^ { 2 } \right ) \\ \text{operators, } F ( \mathbf x ) \text{ (Yager and Beliakov [3} ; }$$
define regression operators, F ( x ) (Yager and Beliakov [37]). Consider the following results dependent on the weight vector ∆ = ( w , ..., w 1 n ).
- 1. ∆ = 1 generates Least Squares regression and F is monotonic and hence weakly monotonic;
- 2. ∆ = (0 , ..., 0 1) generates Chebyshev regression and , F is monotonic and hence weakly monotonic;
$$4. \text{ For } \Delta = \begin{cases} ( 0,.., 0 _ { k - 1 }, 1 / 2, 1 / 2, 0,.. 0 ) & n = 2 k \text{ is even} \\ ( 0,.., 0 _ { k - 1 }, 1, 0,.., 0 ) & n = 2 k - 1 \text{ is odd} \end{cases} \text{ then } F \text{ is the} \\ \text{Least Median of Squares operator and hence shift invariant and weakly}$$
- 3. Since all the terms S i ( ( x -y 1 ) 2 ) are constant under transformation ( x , y ) → ( x + a 1 , y + ) (cf Theorem 2), the OWA regression operators a are shift-invariant for any choice of the weight vector ∆.
Least Median of Squares operator and hence shift invariant and weakly monotonic; and
- 5. For ∆ = (1 , ..., 1 h , 0 , ..., 0) , h = ⌊ n 2 ⌋ + 1 then F is the Least Trimmed Squares operator and hence is shift invariant and weakly monotonic.
In the cases 3-5 the OWA regression operators are not monotonic.
Example 8. Density based means: The density based means were introduced in Angelov and Yager [2]. Let d ij denote the distance between inputs x i and x j . The density based mean is defined as
$$y = \sum _ { i = 1 } ^ { n } w _ { i } ( \mathbf x ) x _ { i }, \\ \quad \ \. \. \.$$
where and where K C is Cauchy kernel given by
$$w _ { i } ( { \mathbf x } ) = \frac { u _ { i } ( { \mathbf x } ) } { \sum _ { j = 1 } ^ { n } u _ { j } ( { \mathbf x } ) } = \frac { K _ { C } ( \frac { 1 } { n } \sum _ { j = 1 } ^ { n } d _ { i j } ^ { 2 } ) } { \sum _ { k = 1 } ^ { n } K _ { C } ( \frac { 1 } { n } \sum _ { j = 1 } ^ { n } d _ { k j } ^ { 2 } ) }, \\ \text{where } K _ { C } \text{ is Cauchy kernel given by} \\.$$
$$K _ { C } ( t ) = ( 1 + t ) ^ { - 1 }.$$
As shown in Beliakov and Wilkin [6] density based means are shiftinvariant and hence weakly monotonic. Extensions of the formulas (4.2), (4.3) are also presented.
It may appear that the class of weakly monotonic averages consists mostly of shift-invariant functions, as the above examples illustrate. This impression is due to the fact that such examples came from robust regression, where the very definition of robust estimators of location involve shift-invariance Rousseeuw and Leroy [27]. However, the class of weakly monotonic functions is reacher, as various (but not all) ϕ -transforms of shift-invariant functions (with non-linear ϕ ) are weakly monotonic but not shift-invariant. Some results on the conditions on ϕ which preserve weak monotonicity are presented in Wilkin et al. [36]. A few more examples are presented in the sequel.
## 4.2. Mixture Functions
The mixture functions were given by Eqn. (2.2), which we recall here for clarity
Mesiar et al. [22] have shown that under the constraint that w is nondecreasing and differentiable, if w x ( ) ≥ w x ′ ( ) · ( b -x , x ) ∈ [ a, b ] = I , then M w is an aggregation function and hence monotonic (and by extension, also weakly monotonic). Additionally, M w is invariant to scaling of the weight functions (i.e., M αw = M w ∀ α ∈ R \{ } 0 ). In Mesiar and Spirkova [21], it was shown that the dual, M d w , of M w is generated by w (1 -x ).
$$\lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad
\intertext { a v e s h o w n t h a t u d e n t h e c o } \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \ \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \quad \lambda \$$
As mentioned in Section 2, a special case of the Gini means (with p = 1) are the Lehmer means, which are generally not monotonic. Lehmer means are mixture functions with weight function w t ( ) = t q , which is neither increasing
for all q ∈ R nor shift invariant. Note that for q < 0 the value of Lehmer means at x with at least one component x i = 0 is defined as the limit when x i → 0 + , so that L q is continuous on [0 , ∞ ) n
We begin by establishing some general properties of Lehmer means.
Lemma 1. The Lehmer mean L q : [0 , ∞ ) n → [0 , ∞ ) , given by is
$$L _ { q } ( \mathbf x ) = \frac { \sum _ { i = 1 } ^ { n } x _ { i } ^ { q + 1 } } { \sum _ { i = 1 } ^ { n } x _ { i } ^ { q } }, \quad \ q \in \mathbb { R } \quad \quad ( 4. 4 )$$
- 1. homogeneous;
- 2. monotonic (and linear) along the rays emanating from the origin;
- 3. averaging;
- 4. idempotent;
- 5. not generally monotonic in x ;
- 6. has neutral element 0 for q > 0 ; and,
- 7. has absorbing element 0 for q < 0 .
The proof is presented in the Appendix.
We now establish a sufficient condition for weak monotonicity of Lehmer means, which depends on both q and the number of arguments n . We provide a relation between these two quantities.
Theorem 3. The Lehmer mean of n arguments, is weakly monotonic on [0 , ∞ ) n if n ≤ 1 + ( q +1 q -1 ) q -1 , q ∈ R \ (0 , 1) .
Proof. The Lehmer mean for q ∈ -[ 1 0] is known to be monotonic (Farnsworth , and Orr [15]) and hence weakly monotonic in that parameter range. In the range q ∈ (0 , 1) the Lehmer mean is not weakly monotonic, because it's partial derivative at x = ( a, b ) when a → 0 + tends to -∞ . Hence we focus on the cases q ≥ 1 and q < -1. The proof is easier to present in penalty-based representation, as the partial derivatives have more compact form. As stated
in Section 2.3, L q ( x ) can be written as a penalty-based function (2.4) with penalty P ( x , y ) = n ∑ i =1 x q i ( x i -y ) 2 . Differentiation w.r.t y yields
$$\mathcal { P } _ { y } ( \mathbf x, y ) = - 2 \sum _ { i = 1 } ^ { n } \left ( x _ { i } ^ { q + 1 } - x _ { i } ^ { q } y \right ). \\ \intertext { w o h o r o n i b i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t om i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o t m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c om i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o co m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m i t o c o m ierrow.$$
At the minimum we have the implicit equation P y = F ( x , y ) = 0, with the necessary condition that yields y = L q ( x ). We remind that for any x i = 0 the Lehmer mean is defined in the limit as x i → 0 + . The partial derivatives ∂L q ( x ) ∂x i are given by the implicit derivative ∂y ∂x i = -F x i F y , with
$$F ( \mathbf x, y ) & = \sum _ { i = 1 } ^ { n } x _ { i } ^ { q + 1 } - y \sum _ { i = 1 } ^ { n } x _ { i } ^ { q } = 0.$$
By differentiation F y ( x , y ) = -n ∑ i =1 x q i ≤ 0 , ∀ x i ∈ [0 , ∞ ) and thus the sign of the partial derivatives depends on the sign of F x i , which is given by
$$F _ { x _ { i } } ( { \mathbf x }, y ) = ( q + 1 ) x _ { i } ^ { q } - q x _ { i } ^ { q - 1 } y.$$
These derivatives can be either positive or negative. To establish weak monotonicity we require that the directional derivative of L q ( x ) in the direction (1 , 1 , ..., 1) be non-negative. We have that (D 1 L q ) ( x ) = 1 √ n ∇ L q ( x ) · 1 =
mined only by the sign of n ∑ i =1 F x i ( x , y ). We will henceforth work with the sorted inputs, x () = x ↘ such that x (1) is thus the largest input and x ( n ) the smallest.
n √ nF y ( x ,y ) n ∑ i =1 F x i ( x , y ) and thus the sign of the directional derivative is deter-
Consider first the case: q ≥ 1.
We examine the term F x (1) and note that y ≤ x (1) for any input x since L q ( x ) is averaging (condition 3 of Lemma 1). Then it follows that
$$F _ { x _ { ( 1 ) } } = ( q + 1 ) x _ { ( 1 ) } ^ { q } - q x _ { ( 1 ) } ^ { q - 1 } y \geq ( q + 1 ) x _ { ( 1 ) } ^ { q } - q x _ { ( 1 ) } ^ { q - 1 } x _ { ( 1 ) } = x _ { ( 1 ) } ^ { q } \geq 0.$$
For the remaining x i we compute the smallest possible value of F x i by selecting the point of minimum value, which is attained for
$$\frac { \partial F _ { x _ { i } } } { \partial x _ { i } } = q \left ( q + 1 \right ) x _ { i } ^ { q - 1 } - q \left ( q - 1 \right ) x _ { i } ^ { q - 2 } y = 0.$$
At the optimum either x ∗ i = 0 or
$$q ( q + 1 ) \, ( x _ { i } ^ { * } ) ^ { q - 1 } - q ( q - 1 ) \, ( x _ { i } ^ { * } ) ^ { q - 2 } \, y = 0 \\ \Rightarrow & x _ { i } ^ { * } = \left ( \frac { q - 1 } { q + 1 } \right ) y \geq 0. \\ \left ( \lambda _ { i } \right ) * \quad \cap \dots \, 1 \dots \, \Gamma \right ) \, \Gamma \right ) \quad \alpha \, / \dots \, - \, 1 \dots \, \Gamma \right ) \, \dots \, - \, 1 \dots \, \dots$$
$$\cdot$$
At x ∗ i = 0 we have that F x i = 0 (for q > 1) and F x i = -y (for q = 1), and at x ∗ i = ( q -1 q +1 ) y we have that
$$x _ { i } ^ { * } = \left ( \frac { q - 1 } { q + 1 } \right ) y \text{ we have that} \\ F _ { x _ { i } } ( x _ { i } ^ { * } ) & = ( q + 1 ) \left ( \left ( \frac { q - 1 } { q + 1 } \right ) y \right ) ^ { q } - q \left ( \left ( \frac { q - 1 } { q + 1 } \right ) y \right ) ^ { q - 1 } y \\ & = ( q - 1 ) \left ( \frac { q - 1 } { q + 1 } \right ) ^ { q - 1 } y ^ { q } - q \left ( \frac { q - 1 } { q + 1 } \right ) ^ { q - 1 } y ^ { q } \\ & = y ^ { q } \left ( \frac { q - 1 } { q + 1 } \right ) ^ { q - 1 } ( q - 1 - q ) \\ & = - y ^ { q } \left ( \frac { q - 1 } { q + 1 } \right ) ^ { q - 1 } \\ & \geq - x _ { ( 1 ) } ^ { q } \left ( \frac { q - 1 } { q + 1 } \right ) ^ { q - 1 }. \\ \text{Since (D_1 L_{q}) (x) } \in \sum ^ { n } F _ { x _ { i } } \text{ then}$$
$$\text{Since $(D_{1}L_{q})(\mathbf x) \in \sum _ { i = 1 } ^ { n } F _ { x _ { i } } $ then }.$$
$$( \mathbf D _ { \mathbf 1 } L _ { q } ) ( \mathbf x ) & = c \left ( F _ { x _ { ( 1 ) } } + \sum _ { i = 2 } ^ { n } F _ { x _ { ( i ) } } \right ), \\ \nu \quad & \ \smallsetminus a - 1$$
and since each F x ( ) i ≥ -x q (1) ( q -1 q +1 ) q -1 then
$$( D _ { 1 } L _ { q } ) ( \mathbf x ) & \geq c \left ( F _ { x _ { ( 1 ) } } + ( n - 1 ) \left ( - x _ { ( 1 ) } ^ { q } \left ( \frac { q - 1 } { q + 1 } \right ) ^ { q - 1 } \right ) \right ) \\ & = c \left ( x _ { ( 1 ) } ^ { q } - ( n - 1 ) \left ( \frac { q - 1 } { q + 1 } \right ) ^ { q - 1 } x _ { ( 1 ) } ^ { q } \right ) \\ & = c x _ { ( 1 ) } ^ { q } \left ( 1 - ( n - 1 ) \left ( \frac { q - 1 } { q + 1 } \right ) ^ { q - 1 } \right ). \\ \text{This expression is non-negative and hence } L _ { q } ( \mathbf x ) \text{ is weakly monotonic } \text{pr}$$
This expression is non-negative and hence L q ( x ) is weakly monotonic provided that
$$& ( n - 1 ) \left ( \frac { q - 1 } { q + 1 } \right ) ^ { q - 1 } \leq 1 \quad \text{ or } \quad n \leq 1 + \left ( \frac { q + 1 } { q - 1 } \right ) ^ { q - 1 }, q > 1. \\ \text{$row$} \quad \cdot \quad \cdot \quad \cdot \end{matrix} \right )$$
For q = 1 we get n ≤ 2.
Now consider the case: q < -1. We have that
$$F _ { x _ { i } } = \frac { ( 1 - p ) x _ { i } + p y } { x _ { i } ^ { p + 1 } }, \quad p = | q | > 1$$
and note that these derivatives are defined in the limit for the case where x i = 0. I.e., F + x i ∣ ∣ x i =0 = lim x i → 0 + F x i . We now examine the term F x ( n ) and note that y ≥ x ( n ) since L q ( x ) is averaging. Thus
$$F _ { x _ { ( n ) } } & = \frac { ( 1 - p ) x _ { ( n ) } + p y } { x _ { ( n ) } ^ { p + 1 } } \\ & \geq \frac { ( 1 - p ) x _ { ( n ) } + p x _ { ( n ) } } { x _ { ( n ) } ^ { p + 1 } } = \frac { 1 } { x _ { ( n ) } ^ { p } }.$$
Again we consider the remaining x i by seeking the minimum of F x i , given by
$$\frac { \partial F _ { x _ { i } } } { \partial x _ { i } } = - \frac { p ( 1 - p ) } { x _ { i } ^ { p + 1 } } - \frac { p ( p + 1 ) } { x _ { i } ^ { p + 2 } } y = 0.$$
This attains a minimum at x i = ( p +1 p -1 ) y and substitution into F x i gives
$$\text{a minimum at } x _ { i } = \left ( \frac { p + 1 } { p - 1 } \right ) y \text{ and substitution into} \\ F _ { x _ { i } } \left ( \frac { p + 1 } { p - 1 } y \right ) & = \frac { ( 1 - p ) \left ( \frac { p + 1 } { p - 1 } y \right ) + p y } { \left ( \frac { p + 1 } { p - 1 } y \right ) ^ { p + 1 } } \\ & = \frac { - 1 } { y ^ { p } } \left ( \frac { p + 1 } { p - 1 } \right ) ^ { - ( p + 1 ) } \\ & \geq \frac { - 1 } { x _ { ( n ) } ^ { p } } \left ( \frac { p + 1 } { p - 1 } \right ) ^ { - ( p + 1 ) }. \\ \text{al derivative of } L _ { q } ( \text{x} ) \text{ can be written as}$$
The directional derivative of L q ( x ) can be written as
$$\text{directional derivative of } L _ { q } ( { \mathbf x } ) \, \text{can be written as} \\ ( D _ { 1 } L _ { q } ) ( { \mathbf x } ) & = c \left ( F _ { x _ { ( n ) } } + \sum _ { i = 1 } ^ { n - 1 } F _ { x _ { ( i ) } } \right ) \\ & \geq c \left ( \frac { 1 } { x _ { ( n ) } ^ { p } } - \frac { n - 1 } { x _ { ( n ) } ^ { p } } \left ( \frac { p + 1 } { p - 1 } \right ) ^ { - ( p + 1 ) } \right ) \\ & = \frac { c } { x _ { ( n ) } ^ { p } } \left ( 1 - ( n - 1 ) \left ( \frac { p + 1 } { p - 1 } \right ) ^ { - ( p + 1 ) } \right ). \\ \text{ote that the sign of this derivative does not change in the lim.} \, \stackrel { \sim } { \sim } + \, \cdots \, \cdots \, \cdots \,.$$
We note that the sign of this derivative does not change in the limit as x ( n ) → 0 + and is non-negative for
$$\omega \colon \colon & \colon \colon \colon \colon \colon \\ n & \leq 1 + \left ( \frac { p + 1 } { p - 1 } \right ) ^ { p + 1 } \\ & = 1 + \left ( \frac { q - 1 } { q + 1 } \right ) ^ { - q + 1 }, \quad q = - p. \\ \text{oth cases } ( q < - 1, q \geq 1 ) \text{ we obtain the re}$$
Hence, in both cases ( q < -1 , q ≥ 1) we obtain the requirement for a non-negative directional derivative - and hence weak monotonicity of L q ( x ) -as being n ≤ 1+ ( q +1 q -1 ) q -1 . For the case - ≤ 1 q ≤ 0 this remains a sufficient condition for weak monotonicity, although clearly overly restrictive.
Remark 5 . As suggested to us, as
$$\left ( \frac { q + 1 } { q - 1 } \right ) ^ { q - 1 } = \left ( \left ( 1 + \frac { 2 } { q - 1 } \right ) ^ { \frac { q - 1 } { 2 } } \right ) ^ { 2 },$$
and the right hand side is increasing (with q ) and approaches e 2 as q →∞ , we have a restriction that for all q > 1 weak monotonicity holds for at most n < 9 arguments.
A further suggestion was to show that Lehmer means are weakly monotonic for any number of arguments for negative powers q . This can be achieved by examining the directional derivative (D 1 L q )( x ) directly, and in the near future we shall formalise this result.
So while Lehmer means are an interesting example of mixture functions (with familiar power functions as the weights) their usefulness in many applications would be limited, as they are weakly monotonic only for a restricted range of (positive) powers and the number of arguments.
Corollary 3. The contra-harmonic mean ( q = 1 ) is weakly monotonic only for two arguments.
## 4.3. Spatial-Tonal Filters
The well known class of spatial-tonal filters includes the mode filter (van de Weijer and van den Boomgaard [33]), bilateral filter (Tomasi and Manduchi [31]) and anisotropic diffusion (Perona and Malik [23]) among others. This is an important class of filters developed to preserve edges within images when performing tasks such as filtering or smoothing. While these filters are commonly expressed in integral notation over a continuous space, they are implemented in discrete form over a finite set of pixels that take on finite values in a closed interval. It can be shown that the class of functions is given (in discrete form) by the averaging function
$$\dots \dots \dots \dots \dots \dots \dots \dots \\ F _ { \Delta } ^ { g } ( \mathbf x ; x _ { 1 } ) = \frac { \sum _ { i = 1 } ^ { n } w _ { i } g ( | x _ { i } - x _ { 1 } | ) x _ { i } } { \sum _ { i = 1 } ^ { n } w _ { i } g ( | x _ { i } - x _ { 1 } | ) }, \\ \text{eights} \, w _ { i } \, \text{are nonlinear and non-convex functions of the locations} \dots \dots \dots \dots \dots \dots \dots \dots \dots$$
where the weights w i are nonlinear and non-convex functions of the locations of the pixels, which have intensity x i . In all practical problems the locations are constant and hence can be pre-computed to produce the constant weight vector ∆ = ( w , w , ..., w 1 2 n ). The pixel x 1 is the pixel to be filtered/smoothed such that its new value is ¯ x 1 = F g ∆ ( x ; x 1 ).
The function F g ∆ is nonlinear and not monotonic. It is trivially shown to be expressed as a penalty-based function with penalty
$$\mathcal { P } ( \mathbf x, y ) = \sum _ { i = 1 } ^ { n } w _ { i } g ( | x _ { i } - x _ { 1 } | ) ( x _ { i } - y ) ^ { 2 }. \\ \intertext { m a n d i n d i n d i n d } \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \codeot = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon = \, \colon =$$
In image filtering applications it is known that this penalty minimises the mean squared error between the filtered image and the noisy source image (Elad [14]). By Proposition 2 it follows directly that the filter F g ∆ is shift invariant and hence weakly monotonic. Furthermore, Theorem 2 permits us to generalise this class of filters to be those penalty based averaging functions having penalty function
$$\mathcal { P } ( \mathbf x, y ) = \sum _ { i = 1 } ^ { n } w _ { i } g ( | x _ { i } - f ( \mathbf x ) | ) ( x _ { i } - y ) ^ { 2 } & & ( 4. 6 ) \\ \text{further using other bivariate function } D \, \colon \mathbb { T } ^ { 2 } \to \mathbb { R } \, ( \text{as discussed in} }$$
or even further using other bivariate function D : I 2 → R (as discussed in Remark 3)
$$\mathcal { P } ( \mathbf x, y ) = \sum _ { i = 1 } ^ { n } w _ { i } g ( | x _ { i } - f ( \mathbf x ) | ) D ( x _ { i }, y ) & & ( 4. 7 ) \\ \text{implication of replacing } x _ { 1 } \text{ with } f ( \mathbf x ) \text{ in the scaling function } q \text{ is}$$
The implication of replacing x 1 with f ( x ) in the scaling function g is that we may use any shift-invariant aggregation of x , which allows us to account for the possibility that x 1 is itself an outlier within the local region of the image. For example, we could use the median, the mode or the shorth for f ( x ). This provides an interesting result and invites further research in the application of weakly monotonic means to spatial-tonal filtering and smoothing problems.
## 5. Conclusion
In this article we have introduced the concept of weakly monotonic averaging functions and examined some of the properties of these functions. We have studied several families of means previously considered to be simply non-monotonic, and shown them to be weakly monotonic. Specifically we have established a sufficient condition for the weak monotonicity of the Lehmer mean - which is an important subclass of the Mean of Bajraktarevic -and shown that several important non-monotonic regression operators are
actually weakly monotonic. Additionally we have proven that a large class of penalty-based functions are also weakly monotonic, which admits a very large class of aggregation functions. This has permitted a simple proof that the class of image processing filters known as spatial-tonal filters are weakly monotonic averaging functions. This class subsumes the class of spatial averaging filters, such as the Gaussian blur filter. Most importantly, given the definition of weak monotonicity, all aggregation functions are weakly monotonic and thus we have not needed to redefine monotonic aggregation in order to relate it to weakly monotonic averaging.
This study was prompted by two issues. First, that there exist several important classes of means that fall outside of the current definition of aggregation functions, which requires monotonicity in all arguments. These include the examples presented in Section 4: the robust estimators of location (such as the mode and the shorth), mixture functions and the spatial-tonal filters used extensively in image processing. It appears reasonable to treat these functions within the same framework that includes the monotonic means.
The second issue is that applications such as image processing and robust statistics require non-monotone averaging, where the main concern is that of noise (outliers) within the data. While the average must be a representative value of the inputs, we wish to avoid the possibility that one or more erroneous inputs drives the value of the output. A small increase above the average may be reasonable, however a large increase should permit that input to be discounted or ignored and the average to possibly decrease.
The concept of weakly monotone aggregation addresses both of these issues, bringing the existing (monotone) aggregation functions and many of the non-monotone means into the same framework. Our proposal then is to redefine the class of averaging aggregation functions to be not those functions that are monotonic, but rather the class of weakly monotonic functions that are averaging.
Proposal: A function F : I n → I is an averaging aggregation function on I n if and only if it is weakly monotonic non-decreasing on I and averaging.
It remains to be seen whether or not weak monotonicity is the minimal requirement for defining averaging aggregation functions and whether this weaker definition is justified for other types of aggregation, such as conjunctive and disjunctive functions. Furthermore, are there other possibilities for the relaxation of monotonicity that provide for a unified framework of aggregation theory and practice? We leave these questions for future work.
## References
## References
| [1] | Andrews, D., Bickel, P., Hampel, F., Huber, P., Rogers, W., Tukey, J., 1972. Robust Estimates of Location: Surveys and Advances. Princeton University Press, Princeton, N.J. |
|-------|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [2] | Angelov, P., Yager, R., 2013. Density-based averaging - a new operator for data fusion. Information Sciences 222, 163-174. |
| [3] | Barash, D., Comaniciu, D., 2004. A common framework for nonlinear diffusion, adaptive smoothing, bilateral filtering and mean shift. Image and Vision Computing 22, 73-81. |
| [4] | Beliakov, G., James, S., Li, G., 2011. Learning Choquet-integral-based metrics for semisupervised clustering. IEEE Trans. on Fuzzy Syst. 19 (3), 562-574. |
| [5] | Beliakov, G., Pradera, A., Calvo, T., 2007. Aggregation Functions: A Guide for Practitioners. Vol. 221 of Studies in Fuzziness and Soft Com- puting. Springer-Verlag, Berlin. |
| [6] | Beliakov, G., Wilkin, T., 2014. On some properties of weighted averaging with variable weights. Information Sciences 281, 1-7. |
| [7] | Blumenson, L. E., 1960. A derivation of n-dimensional spherical coordi- nates. The American Mathematical Monthly 67 (1), 63-66. |
| [8] | Bullen, P., 2003. Handbook of Means and Their Inequalities. Kluwer, Dordrecht. |
| [9] | Calvo, T., Beliakov, G., 2010. Aggregation functions based on penalties. Fuzzy Sets and Systems 161 (10), 1420-1436. |
| [10] | Calvo, T., Koles´rov´, a a A., Komorn´kov´, ı a M., Mesiar, R., 2002. Aggrega- tion operators: properties, classes and construction methods. In: Calvo, T., Mayor, G., Mesiar, R. (Eds.), Aggregation Operators. New Trends and Applications. Physica-Verlag, Heidelberg, New York, pp. 3-104. |
| [11] | Cauchy, A. L., 1821. Cours d'analyse de l'ecole royale polytechnique. Analyse alg´brique e 1. |
| [12] | Chisini, O., 1929. Sul concetto di media. Periodico di Matematiche 4, 106-116. |
|--------|--------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [13] | de Finetti, B., 1931. Sul concetto di media. Giorn. Ist. Ital. Attuari 2 (3), 369-396. |
| [14] | Elad, M., 2002. On the origin of the bilateral filter and ways to improve it. IEEE Trans. on Image Processing 11 (10), 1141-1151. |
| [15] | Farnsworth, D., Orr, R., 1986. Gini means. American Mathematical Monthly 93 (8), 603-607. |
| [16] | Gini, C., 1958. Le Medie. Unione Tipografico-Editorial Torinese, Milan (Russian translation, Srednie Velichiny, Statistica, Moscow, 1970). |
| [17] | Grabisch, M., Marichal, J.-L., Mesiar, R., Pap, E., 2009. Aggregation Functions. Encyclopedia of Mathematics and Its Foundations. Cam- bridge University Press. |
| [18] | Koles´rov´, a a A., Mayor, G., Mesiar, R., 2007. Weighted ordinal means. Inf. Sciences 177, 3822-3830. |
| [19] | L´zaro, a J., R¨ckschlossov´, u a T., Calvo, T., 2004. Shift invariant binary aggregation operators. Fuzzy Sets and Systems 142, 51-62. |
| [20] | Marques Pereira, R., Ribeiro, R., 2003. Aggregation with generalized mixture operators using weighting functions. Fuzzy Sets and Systems 137, 43-58. |
| [21] | Mesiar, R., Spirkova, J., 2006. Weighted means and weighting functions. Kybernetika 42 (2), 151-160. |
| [22] | Mesiar, R., Spirkova, J., Vavrikova, L., 2008. Weighted aggregation op- erators based on minimization. Inform. Sci. 178, 113-1140. |
| [23] | Perona, P., Malik, J., 1990. Scale space and edge detection using anisotropic diffusion. IEEE Trans. on Pattern Analysis and Machine Intelligence 12 (7), 629-639. |
| [24] | Ribeiro, R., Marques Pereira, R., 2003. Generalized mixture operators using weighting functions. Europ. J. Oper. Research 145, 329-342. |
| [25] | Rockafellar, R., 1970. Convex Analysis. Princeton University Press, Princeton. |
|--------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [26] | Rousseeuw, P., 1984. Least median of squares regression. Journal of the American Statistical Association 79 (388), 871-880. |
| [27] | Rousseeuw, P., Leroy, A., 1987. Robust Regression and Outlier Detec- tion. John Wiley and Sons. |
| [28] | Rubin, E., 1968. The statistical world of herodotus. The American Statistician 22 (1), 31-33. |
| [29] | Rubin, E., 1971. Quantitative commentary on thucydides. The American Statistician 25 (4), 52-54. |
| [30] | Sylvain, P., Kornprobst, P., Tumblin, J., Durand, F., 2008. Bilateral filtering: Theory and applications. Foundations and Trends in Computer Graphics and Vision 4 (1), 1-73. |
| [31] | Tomasi, C., Manduchi, R., January 1998. Bilateral filtering for gray and color images. In: Proc. of the 6th Int. Conf. on Computer Vision. Bombay, India, pp. 839-846. |
| [32] | Torra, V. Narukawa, Y., 2007. Modeling Decisions. Information Fusion and Aggregation Operators. Springer, Berlin, Heidelberg. |
| [33] | van de Weijer, J., van den Boomgaard, R., 2001. Local mode filtering. In: Computer Vision and Pattern Recognition. Vol. 2. IEEE Computer Society, pp. 428-433. |
| [34] | van den Boomgaard, R., van de Weijer, J., 2002. On the equivalence of local-mode finding, robust estimation and mean-shift analysis as used in early vision tasks. In: Proc. of the 16th Int. Conf. on Pattern Recog- nition. Vol. 3. pp. 927-930. |
| [35] | Wilkin, T., 2013. Image reduction operators based on non-monotonic averaging functions. In: Proceedings of the 10th IEEE Int. Conf. on Fuzzy Systems (to appear). Hyderabad, India. |
| [36] | Wilkin, T., Beliakov, G., Calvo, T., 2014. Weakly monotone averag- ing functions. In: Proceedings of IPMU'2014 Conference. Montpellier, |
[37] Yager, R., Beliakov, G., 2010. OWA operators in regression problems. IEEE Trans. on Fuzzy Syst. 18, 106-113.
## Appendix
Proof of Lemma 1.
Proof. Consider each of the following:
- 1. Homogeneous : Set x = λ u then
$$L _ { q } ( \mathbf x ) = L _ { q } ( \lambda \mathbf u ) & = \frac { \sum _ { i = 1 } ^ { n } ( \lambda u _ { i } ) ^ { q + 1 } } { \sum _ { i = 1 } ^ { n } ( \lambda u _ { i } ) ^ { q } } = \lambda \frac { \sum _ { i = 1 } ^ { n } u _ { i } ^ { q + 1 } } { \sum _ { i = 1 } ^ { n } u _ { i } ^ { q } } = \lambda L _ { q } ( \mathbf u ). \\ \text{ Hence } L _ { q } & \text{is homogeneous with degree } 1.$$
Hence L q is homogeneous with degree 1.
- 2. Monotonic (and linear) along the rays : Consider the generalised spherical coordinates (Blumenson [7]) ( r, θ, φ 1 , ..., φ n -2 ), r ≥ 0, 0 ≤ θ ≤ 2 π , 0 ≤ φ i ≤ π for the hypersphere S n -1 = { x ∈ R n : ‖ x ‖ = r } . We will restrict the angle variables so that x i ∈ [0 , ∞ ). The transformation to an orthonormal Euclidean basis E n produces the vector x of length r having components
$$\dashrightarrow \underset { r } { \times } \underset { r } { \times } \dashrightarrow \underset { r } { \times } \dashrightarrow \underset { k = 1 } { \times } } \\ x _ { 1 } & = r \cos ( \phi _ { 1 } ) \\ x _ { j } & = r \cos \left ( \phi _ { j } \right ) \prod _ { k = 1 } ^ { j - 1 } \sin \left ( \phi _ { k } \right ), \quad j = 2, \dots, n - 1 \\ x _ { n } & = r \prod _ { k = 1 } ^ { n - 1 } \sin \left ( \phi _ { k } \right ) \\ ; \phi _ { n - 1 } & = \theta \text{ and } 0 \leq \phi _ { i } \leq \pi / 2, \ i = 1, \dots, n - 1, \ r \geq 0. \text{ The } L \text{}$$
where φ n -1 = θ and 0 ≤ φ i ≤ π / , 2 i = 1 , ..., n -1 , r ≥ 0. The Lehmer
mean of the Euclidean vector x is therefore
$$\begin{array} { l } \text{mean of the Euclidean vector } x \text{ is therefore} \\ \text{mean of the Euclidean vector } x \text{ is therefore} \\ \quad & [ r \cos ( \phi _ { 1 } ) ] ^ { q + 1 } + \sum _ { j = 2 } ^ { n - 1 } \left [ r \cos ( \phi _ { j } ) \prod _ { k = 1 } ^ { j - 1 } \sin ( \phi _ { k } ) \right ] ^ { q + 1 } + \left [ r \prod _ { k = 1 } ^ { n - 1 } \sin ( \phi _ { k } ) \right ] ^ { q + 1 } \\ [ r \cos ( \phi _ { 1 } ) ] ^ { q } + \sum _ { j = 2 } ^ { n - 1 } \left [ r \cos ( \phi _ { j } ) \prod _ { k = 1 } ^ { j - 1 } \sin ( \phi _ { k } ) \right ] ^ { q } + \left [ r \prod _ { k = 1 } ^ { n - 1 } \sin ( \phi _ { k } ) \right ] ^ { q } \\ = r \left [ \frac { \left [ \cos ( \phi _ { 1 } ) ] ^ { q + 1 } + \sum _ { j = 2 } ^ { n - 1 } \left [ \cos ( \phi _ { j } ) \prod _ { k = 1 } ^ { j - 1 } \sin ( \phi _ { k } ) \right ] ^ { q + 1 } + \left [ \prod _ { k = 1 } ^ { n - 1 } \sin ( \phi _ { k } ) \right ] ^ { q + 1 } \\ [ \cos ( \phi _ { 1 } ) ] ^ { q } + \sum _ { j = 2 } ^ { n - 1 } \left [ \cos ( \phi _ { j } ) \prod _ { k = 1 } ^ { j - 1 } \sin ( \phi _ { k } ) \right ] ^ { q } + \left [ \prod _ { k = 1 } ^ { n - 1 } \sin ( \phi _ { k } ) \right ] ^ { q } \\ = r f ( \phi _ { 1 }, \dots, \phi _ { n - 1 } ) \\ \end{array}$$
Along rays emanating from the origin each f ( ) is · constant and hence L q ( x ) = α r φ is linear.
- 3. Averaging : Let x σ = x ↘ and take a = x (1) and b = x ( m ) denote the value of the largest and the smallest non-zero elements of x respectively. By homogeneity
$$L _ { q } ( { \mathbf x } ) = a \frac { 1 + \sum _ { i = 2 } ^ { n } \left ( \frac { x _ { ( i ) } } { a } \right ) ^ { q + 1 } } { 1 + \sum _ { i = 2 } ^ { n } \left ( \frac { x _ { ( i ) } } { a } \right ) ^ { q } } & = a \frac { 1 + \alpha } { 1 + \beta } \\ \leq \ a \text{ for all } i \, = \, 1, \dots, n \text{ then } \alpha \, \leq \, \beta. \text{ Hence } L$$
Since x ( ) i ≤ a for all i = 1 , ..., n then α ≤ β . Hence L q ( x ) ≤ a . Similarly,
$$L _ { q } ( { \mathbf x } ) = b \frac { 1 + \sum _ { i = 1 } ^ { m - 1 } \left ( \frac { x _ { ( i ) } } { b } \right ) ^ { q + 1 } } { 1 + \sum _ { i = 1 } ^ { m - 1 } \left ( \frac { x _ { ( i ) } } { b } \right ) ^ { q } } = b \frac { 1 + \gamma } { 1 + \delta } \\ \geq b \text{ for all } i = 1, \dots, m - 1 \text{ and } x _ { ( i ) } = 0 \text{ for all } i = \\.. \text{ } \mathbf u \underset { \mathbf u \sim \mathbf u } { \mathbf r } \, \wr \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \, \mathbf t \} }$$
Since x ( ) i ≥ b for all i = 1 , ..., m -1 and x ( ) i = 0 for all i = m +1 , ..., n then δ ≤ γ . Hence L q ( x ) ≥ b . Thus, min( x ) ≤ L q ( x ) ≤ max( x ) and L q ( x ) is averaging.
- 4. Idempotent : For any vector x = ( t, t, ..., t ) we have that
$$L _ { q } ( \mathbf x ) = \frac { t \sum _ { i = 1 } ^ { n } t ^ { q } } { \sum _ { i = 1 } ^ { n } t ^ { q } } = t \\ \text{otent}.$$
and hence L q is idempotent.
- 5. Not generally monotonic in x : Take x = (1 0) and , y = (1 , 1 / 2 ), then for q > 0, L q ( x ) = 1 and L q ( y ) = 1+( 1 2 / ) q +1 1+( 1 2 / ) q = ( 2 q +1 +1 ) (2 q +1 +2) < 1. Thus x < y and L q ( x ) > L q ( y ), hence L q ( x ) is not generally monotonic in x for all q ∈ R .
- 7. Has absorbing element of 0 for q < 0: Consider x = ( a, 0) then L q ( x ) = lim x 2 → 0 + L q (1 , x 2 ) = lim x 2 → 0 + a + x q +1 2 a + x q 2 = a x q 2 + x 2 a x q 2 +1 = 0.
- 6. Has neutral element of 0 for q > 0: Consider x = ( a, 0) then L q ( x ) = lim x 2 → 0 + a q +1 + x q +1 2 a q + x q 2 = a for q > 0.
 | null | [
"Tim Wilkin",
"Gleb Beliakov"
] | 2014-08-02T03:03:33+00:00 | 2014-08-02T03:03:33+00:00 | [
"cs.AI",
"math.CA",
"03E72, 03B52, 94D05, 26E60",
"I.5"
] | Weakly monotone averaging functions | Monotonicity with respect to all arguments is fundamental to the definition
of aggregation functions. It is also a limiting property that results in many
important non-monotonic averaging functions being excluded from the theoretical
framework. This work proposes a definition for weakly monotonic averaging
functions, studies some properties of this class of functions and proves that
several families of important non-monotonic means are actually weakly monotonic
averaging functions. Specifically we provide sufficient conditions for weak
monotonicity of the Lehmer mean and generalised mixture operators. We establish
weak monotonicity of several robust estimators of location and conditions for
weak monotonicity of a large class of penalty-based aggregation functions.
These results permit a proof of the weak monotonicity of the class of
spatial-tonal filters that include important members such as the bilateral
filter and anisotropic diffusion. Our concept of weak monotonicity provides a
sound theoretical and practical basis by which (monotone) aggregation functions
and non-monotone averaging functions can be related within the same framework,
allowing us to bridge the gap between these previously disparate areas of
research. |
1408.0330v1 | ## Role of Self-Assembled Monolayers on Improved Electrical Stability of Amorphous In-Ga-Zn-O Thin-Film Transistors
Xiaosong Du, Brendan T. Flynn, Joshua R. Motley, William F. Stickle, Hendrik Bluhm, 1 1 1 2 3 Gregory S. Herman 1,a)
1 Oregon State University, School of Chemical, Biological, and Environmental Engineering, Corvallis, Oregon, 97331, USA
2 Hewlett Packard Company, 1000 Northeast Circle Boulevard,
Corvallis, Oregon 97330, USA
3 Lawrence Berkeley National Laboratory, Chemical Sciences Division, One Cyclotron Road,
Berkeley, CA 94720, USA
a) Author to whom correspondence should be addressed. Electronic mail: [email protected]
KEYWORDS: thin film transistor, back channel passivation, bias stress test.
## ABSTRACT
Self-assembled monolayers (SAMs) have been used to improve both the positive and negative bias-stress stability of amorphous indium gallium zinc oxide (IGZO) bottom gate thin film transistors (TFTs). N-hexylphosphonic acid (HPA) and fluorinated hexylphosphonic acid (FPA) SAMs adsorbed on IGZO back channel surfaces were shown to significantly reduce biasstress turn-on voltage shifts compared to IGZO back channel surfaces with no SAMs. FPA was found to have a lower surface energy and lower packing density than HPA, as well as lower biasstress turn-on voltage shifts. The improved stability of IGZO TFTs with SAMs can be primarily attributed to a reduction in molecular adsorption of contaminants on the IGZO back channel surface and minimal trapping states present with phosphonic acid binding to the IGZO surface.
Indium gallium zinc oxide (IGZO) is a very promising active material for thin film transistors (TFTs), especially when incorporated into flat panel and flexible displays. 1-4 IGZOTFTs have relatively high average electron mobilities ( avg >10 cm 2 /Vs) despite being amorphous, and have been demonstrated to have excellent performance even after low temperature processing that is compatible with polymeric substrates. 1,5 IGZO-TFTs may eventually replace amorphous silicon TFT technologies for many display applications. 2,3 High electrical stability is an important requirement for display applications, since bias, current, and/or light induced instabilities may cause non-uniformities in relative brightness of individual pixels. High electrical stability is also critical for IGZO field effect sensors used for detecting gases, 6 temperature, light, or for deoxyribonucleic acid. For these reasons the bias stress stability of 7 8 9 IGZO-TFTs has been intensely investigated where effects of composition, process temperature, process conditions, gate dielectric, back channel passivation, and many other factors have been evaluated. 10-15
Proposed mechanisms for bias stress threshold voltage (Vth) shifts include electron trapping at the semiconductor/dielectric interface, electron injection into the dielectric, or formation of sub-band gap states in the bulk of films. It has also been suggested that adsorbed species (e.g., O2 and H2O) on the back channel of the IGZO-TFTs can be the dominant mechanism for device instability, by providing acceptor or donor states or through field induced adsorption or desorption of these species. 17-20 For example, molecular oxygen can form a depletion layer at the back channel surface through electron acceptor states, while absorbed water can form an accumulation layer at the back channel surface through electron donor states. A depletion layer results in a positive Vth shift, whereas an accumulation layer results in a
negative Vth shift. Recently it has been demonstrated that dense passivation layers 21 or selfassembled monolayers (SAMs) 22 on IGZO-TFT back channels can significantly minimize these threshold voltage shifts. A potential benefit of using SAMs for surface passivation is that one can obtain well controlled chemistries at the back channel interface, even when using polymeric dielectrics for flexible electronic applications or integrating polymeric sensing layers.
Herein we investigate the use of n-hexylphosphonic acid (n-HPA) and (3,3,4,4,5,5,6,6,6nonaflurohexyl)phosphonic acid (FPA) self-assembled monolayers (SAMs) as molecular passivation layers for IGZO-TFT back channel surfaces. The goal of these studies is to evaluate the effects of surface energy and SAM molecular dipole on device performance. IGZO surface chemistries were characterized using contact angle measurements, positive and negative bias stress turn-on voltage shifts, and X-ray photoelectron spectroscopy (XPS). Significant improvements in IGZO-TFT turn-on voltage (VON) stabilities were observed for both the HPA and FPA passivated surfaces compared to the unpassivated surfaces. These results provide insight into the role of defect passivation at back channel surfaces of IGZO and other transparent amorphous oxide semiconductors (TAOS).
IGZO-TFT test structures were fabricated using a heavily p-doped Si substrate as the gate and thermally grown SiO2 (100 nm) as the gate dielectric. Substrates were cleaned prior to deposition of IGZO with acetone, isopropyl alcohol, and DI water. Amorphous IGZO films (~50 nm thick) were deposited using RF magnetron sputter deposition with a 3 in. IGZO target (molar composition: In2O3:Ga2O3:ZnO), 100W RF power, ~4 mTorr chamber pressure, and 20 sccm flow rate with a 1:19 (O :Ar) ratio. IGZO active layers were patterned using a shadow mask 2
during deposition, and the films were subsequently annealed in air to 300 ° C. Source and drain electrodes were patterned using a shadow mask during thermal evaporation of Al (~500 nm thick) giving a W/L ratio of 1000 m/100 m. Back channel surface passivation using SAMs was performed by UV-Ozone treatment for 5 minutes followed by immediate immersion of the IGZO-TFTs in 2mM n-HPA or FPA solution in 95% ethanol. Samples were removed from the solution, rinsed with 95% ethanol, and then dried with flowing nitrogen. The water contact angle was found to increase for increasing soak times due to adsorption of n-HPA or FPA molecules and a corresponding reduction in surface energy as a result of the alkane or fluorinated alkane groups on the SAMs. The highest contact angles for n-HPA and FPA treated IGZO surfaces, after a 24 h soak, was 84°and 94° , respectively, compared to 23° which was obtained for an IGZO surface after only the UV/Ozone treatment. A 24 h soak was used for all the IGZO-TFTs in this study. In order to remove physisorbed SAMs the samples were further sonicated in a 5% triethylamine/ethanol solution, rinsed with ethanol, and then dried with nitrogen. Duplicate IGZO-TFTs were fabricated where they have gone through identical processing steps as the passivated samples, other than n-HPA or FPA soak. A schematic of the device structure and possible bonding of n-HPA or FPA to the IGZO surface is shown in Figure 1.
All IGZO-TFT bias-stress measurements were performed in the dark at room temperature in air using an Agilent 4155C precision semiconductor parameter analyzer. Single sweep drainto-source current versus gate voltage (IDS-VGS) transfer curves were measured with the drain voltage (VDS) set to 1 V. Stress measurements were performed up to 10 s using an applied VGS 4 of ± 10 V (1 MV/cm) with VDS set to 0 V. Synchrotron-based XPS measurements were performed using a specially designed endstation at the undulator beamline 11.0.2 at the
Advanced Light Source. 23 A photon energy of h = 735 eV was used to obtain surface sensitive spectra. The data were acquired with a 90º emission angle and an electron analyzer pass energy of 10 eV. Laboratory-based XPS measurements were performed with a PHI Quantera Scanning ESCA system using monochromatized Al K radiation with a 200 µ m spot size. The data were acquired with a 45º emission angle and an electron analyzer pass energy of 69 eV. The XPS data were fit using CasaXPS, where the most intense peak in the spectrum was used to define the core-level full-width-half-maximum (FWHM) and Gaussian-Lorentzian mixing. A linear background was used to fit all spectra. The binding energies were charge corrected to the C 1s aliphatic carbon peak at 284.6 eV.
In Table 1 we provide avg , V ON, drain current on-to-off ratio (ION/IOFF), and hysteresis (difference in the up and down sweeps in the IDS-VGS data) for unstressed devices with different surface treatments. Both avg and VON were extracted using methods described previously. 24 Only very small changes were observed in avg , VON, and ION/IOFF, irrespective of surface treatment, where the major difference is a decrease in VON and a reduction in hysteresis with adsorption of SAMs on the back channel surface. In Figure 2, we show positive and negative bias stress I DS-VGS data for IGZO-TFTs without and with n-HPA or FPA passivation for bias stress times of 0, 10, 100, 1,000, and 10,000 s. The arrows in the figure indicate the direction of turn on voltage shift ( VON) for increasing bias stress time. It was observed that as the duration of bias stress increased the VON continued to shift. However, we found that both sub-threshold slope and avg remained unchanged after either positive or negative bias stress.
In Figure 3, we plot VON versus bias stress time for the IGZO-TFTs. The experimental data in the figure is given as open and closed symbols, which represent positive and negative bias stress measurements, respectively. Devices without passivation had the largest bias stress VON, where the devices had shifts of 7.2 and -8.0 V after 10,000 s of positive and negative bias stress, respectively. For bias stress induced trapping and meta-stability, a stretched exponential model has been previously employed for amorphous silicon25 and IGZO-TFTs. 15,18,20,21
$$\Delta V _ { t h } = \Delta V _ { o } \left \{ 1 - \exp \left [ - \left ( \frac { t } { \tau } \right ) ^ { \beta } \right ] \right \}$$
The threshold voltage shift ( Vth) is related to the effective voltage drop across the gate oxide ( VO), which is approximately given by the difference between gate stress voltage and initial threshold voltage (VG,stress -Vth,i ), the characteristic trapping time of carriers ( ), and stretched exponential exponent ( ). Both and are dependent on stress bias, materials, and device properties. For more accuracy we used VON in place of Vth, where VON was obtained by extrapolation from log IDS versus VGS plots where log (IDS/(W/L)) (A) is equal to -11. The stretched exponential model for IGZO devices provided excellent fits to bias stress VON shifts for devices without passivation (dashed line through the circle data points in Figure 3). Values of and for different surface treatments are summarized in Table 2, and were consistent with literature values. For example, characteristic trapping times for IGZO TFTs in the literature are between ~10 to 10 s, 3 6 15,18,20,21 and can vary substantially depending on whether measuring negative or positive bias stress, the gate dielectric used, and the use of passivation layers to prevent chemistry induced trapping at the back channel interface. The stretched exponential model is primarily used to describe VON shifts due to trapping at the dielectric interface or due to metastability in the channel material, although this model has also been used to describe VON
shifts due to back channel chemical effects for IGZO 20 and zinc tin oxide (ZTO) 26 TFTs. In Figure 3 we show the bias stress results for an IGZO-TFT with n-HPA adsorbed on the surface (triangle data points). There is a significant reduction in bias stress shift, especially for negative biases, when compared to devices without SAMs on the back channel surface. Similar improvements in positive bias stress stability have previously been shown for ZTO-TFTs when n-HPA is adsorbed on the backchannel surface. 26 In Figure 3, we show that IGZO TFTs with FPA SAM had the lowest positive and negative bias stress shifts (square data points). The observed shifts for FPA passivated surfaces are significantly reduced compared to the other surface preparations, and we find that the stretched exponential model no longer provides an adequate fit to the data. However, we now find that a linear relationship between Vth and logarithmic time provides a good fit to our data for FPA passivated IGZO TFTs. This relationship can be described as following the total charge that becomes trapped at the channel/dielectric interface or in the dielectric for any given time . t 12 The positive and negative stress test showed a slope of 0.35 and -0.23, respectively.
Since devices with and without SAM passivation had identical preparation of the dielectric/IGZO interface and the IGZO post-deposition annealing conditions we are able to make several observations. In the case of the unpassivated surface the bias stress VON shifts were most likely due to chemistry at the back channel interface as opposed to trapping at the dielectric interface or due to metastability in the channel material. Since the trends for the nHPA passivated IGZO TFTs are very similar to the unpassivated IGZO TFTs (both fit the stretched exponential model), it is likely that chemistry at the back channel interface still allow trap or donor states that lead to VON shifts during bias stress. Finally, it appears that the FPA
passivated IGZO TFT stabilities switch to a different mechanism (no longer fits the stretched exponential model), and in this case we believe that the device stability is no longer dictated by the back channel interface, but is now dominated by charge injection into the dielectric.
In the lower portion of Figure 4 we show surface-sensitive O 1s XPS data for an as deposited blanket IGZO film using a photon energy of h = 735 eV. The sample was stored in ambient air for several days prior to analysis and had substantial carbon contamination (spectra not shown). To fit the O 1s core level spectrum it was necessary to use three components, similar to what has been done previously for IGZO. 27-29 From our analysis we have determined that the binding energies for the three components were at 530.1, 531.2, and 532.2 eV. For several prior IGZO XPS studies the three O 1s peak components were assigned as follows. 27,28 The low-energy peak corresponds to lattice oxygen ions, which have neighboring cations. The mid-energy O 1s peak corresponds to oxygen vacancies or more accurately, lattice oxygen ions in oxygen deficient regions of IGZO. The high-energy O 1s components were due to adsorbed species.
To assess these assignments we have performed in-situ annealing experiments with the goal to desorb impurities from the surface and/or increase the number of oxygen vacancies within the film. In the upper portion of Figure 4 we show surface-sensitive O 1s XPS data for the IGZO sample while annealing to 300 º in ultrahigh vacuum. During the anneal there was a C significant reduction in the intensities of the midand high-energy O 1s components (quantitatively summarized in Table 3). Since the annealing was performed in an oxygen deficient environment the number of oxygen vacancies should increase, resulting in an increase in relative intensity for the mid-energy O 1s component based on the assignments given above.
However, we have found a significant decrease in intensity for both the mid- and high-energy O 1s components suggesting that both components are likely due to adsorbed species that desorb during the anneal. Our interpretation of the O 1s components follows the convention that the mid-energy O 1s peak corresponds to surface hydroxyls, and not oxygen deficient regions in the material. This interpretation of the O 1s spectra has been suggested by several other groups for IGZO as well. 29-31 Thus, the mid- and high-energy peak intensity is correlated with the number of adsorbed species on the IGZO back surface and the corresponding number of adsorbateinduced traps (i.e., not the number of oxygen vacancies in the IGZO film and the corresponding number of traps), and it is these adsorbed species that influence the bias stress stability of the devices fabricated using TAOS materials. Based on the bias stress data this would suggest that the FPA passivated surface has significantly fewer absorbed species on the IGZO back channel surface, which is consistent with the significant reduction in surface energy compared to the unpassivated and n-HPA passivated surfaces.
In the lower portion of Figure 5 we show O 1s XPS data for a blanket IGZO film after being annealed to 300 ° C for 60 min in air, while in the upper portion of Figure 5 we show O 1s XPS data of samples with n-HPA and FPA to investigate changes in IGZO-TFT back channel chemistry. As mentioned above, the low-energy peak corresponds to lattice oxygen ions in IGZO. The n-HPA and FPA SAMs adsorbed on IGZO the mid-energy O 1s components can be attributed to P-O-In/P-O-Ga/P-O-Zn and P=O, which was consistent with prior studies. 32-34 The third peak at a higher binding energy of 532.3 eV can be assigned to surface contamination, as will be discussed below. The mid-energy peak intensity/low-energy peak intensity ratio (Ime/Ile) gives a qualitative estimate of the relative amount of SAMs adsorbed on the surface. As can be
seen in Figure 5, and also given in Table 4, the Ime/Ile intensity ratios are 70% and 49% for nHPA and FPA, respectively. This suggests that there is more n-HPA adsorbed to the IGZO surface, which would be expected based on the relative sizes of the alkane and fluorinated alkane chains and the associated packing densities. 35
In the lower portion of Fig. 6, we show C 1s spectra for a blanket IGZO film after being annealed to 300 ° C . The C 1s spectra had three fairly well-separated peaks with the lower energy peak having an energy of 284.6 eV, which is assigned to aliphatic carbon (C-C or C-H bonds), and the two higher energy peaks at 286.2 eV and 288.8 eV can be assigned to C-O and O-C=O groups on the surface, respectively. 36,37 As shown in the middle portion of Fig. 6, IGZO sample with n-HPA has much less C-O and O-C=O groups on the surface, where most of the intensity comes from aliphatic carbon in HPA molecules, and two higher energy peaks at 286.4 and 288.5 eV due to -CH2- bonded to the phosphonic acid group and surface contamination, respectively. In the upper portion of Fig. 6, we can see that the C 1s spectrum for FPA has five components, where the -CF2and -CF3 groups in FPA molecules give peaks at 290.8 eV and 293.3 eV, respectively. The CH2- groups bound to CF2- and PO3 give peaks at 286.1 eV and 288.4 eV, ---respectively.
These results strongly suggest that the back channel surface chemistry was the dominant reason for bias-stress VON shifts for the unpassivated and n-HPA passivated IGZO-TFTs. There are several mechanisms that have been proposed to describe device instabilities due to the back channel surface chemistry, including field-induced adsorption/desorption 16 and dipole interactions. 5,19 A significant difference between n-HPA and FPA is the orientation of the
molecular dipole at the surface. 34 For example, n-HPA has a polar head group attached to the surface and a non-polar alkyl chain pointing away from the surface. FPA has a polar head group attached to the surface, but also has highly electronegative fluorine atoms on the alkyl chains pointing away from the surface. This results in a molecular dipole that points towards and away from the surface, respectively.
Recent studies suggested that water adsorption was the dominant species for ZTO TFT instabilities leading to significant shifts in threshold voltage, 38 and increased hysteresis in the TFT transfer characteristics. 39 The saturation coverage of SAMs used in these studies significantly limits the interaction of impurity species, especially H 2 O, with the IGZO surface (i.e. it does not provide electron trap and/or donor states that would result in device instabilities). The FPA surface has a significantly lower surface energy compared to n-HPA, and as a result may further reduce the interaction of H2O with the IGZO back channel surface. This may help explain why the IGZO-TFT with FPA has lower VON bias stress shifts, and lower device hysteresis than the n-HPA passivated devices, or the unpassivated devices. Finally, we have not observed significant differences in the IGZO TFT hysteresis for the different surface treatments, which suggests that SAMs do not degrade device performance.
In summary, highly stable IGZO TFTs were fabricated through molecular passivation of the oxide back channel surface. Both the positive and negative bias stress dependence of the turn-on voltage shifts for unpassivated IGZO-TFTs and n-HPA passivated IGZO-TFTs was best described by a stretched exponential model, whereas FPA passivated IGZO TFTs had a linear relationship between turn-on voltage shifts and logarithmic time for the same bias stress
conditions. Bias stress measurements and XPS analysis suggest that reductions in the turn-on voltage shifts were due to well defined molecular passivation at the backchannel surface, which limits electrical modification of the channel due to the adsorption of impurity species (i.e., O2, H2O, etc.).
## Acknowledgements
This work was funded in part by the Oregon Nanoscience and Microtechnologies Institute (ONAMI). X. Du acknowledges funding support from the Juvenile Diabetes Research Foundation (3-PDF-2014-113-A-N). The Advanced Light Source is supported by the Director, Office of Science, Office of Basic Energy Sciences, of the U.S. Department of Energy under Contract No. DE-AC02-05CH11231.
## References:
- 1 K. Nomura, H. Ohta, A. Takagi, T. Kamiya, M. Hirano, and H. Hosonon, Nature 432 , 488 (2004).
- 2 J. F. Wager, D. A. Keszler, and R. E. Presley, Transparent Electronics (Springer, New York, 2008).
- 3 T. Kamiya, K. Nomura, and H. Hosono, Sci. Technol. Adv. Mater. 11 , 044305 (2010).
- 4 E. Fortunato, P. Barquinha, and R. Martins, Adv. Mater. 24 , 2945 (2012).
- 5 J.-S. Park, T.-W. Kim, D. Stryakhilev, J.-S. Lee, S.-G. An, Y.-S. Pyo, D.-B. Lee, Y. G. Mo, D.-U. Jin, and H. K. Chung, Appl. Phys. Lett. , 95 , 013503 (2009).
- 6 H.-W. Zan, C.-H. Li, C.-C. Yeh, M.-Z. Dai, H.-F. Meng, and C.-C. Tsai, Appl. Phys. Lett. 98 , 253503 (2011).
- 7 H. Jeong, C. S. Kong, S. W. Chang, K. S. Park, S. G. Lee, Y. M. Ha, J. Jang, IEEE Electr. Dev. Lett. 34 , 1569 (2013).
- 8 H.-W. Zan, W.-T. Chen, H.-W. Hsueh, S.-C. Kao, M.-C. Ku, C.-C. Tsai, and H.-F. Meng, Appl. Phys. Lett. 97 , 203506 (2010).
- 9 J. Jung, S. J. Kim, D. H. Yoon, B. Kim, S. H. Park, H. J. Kim, ACS Appl. Mater. Interfaces 5 , 98 (2013).
- 10 K. Nomura, T. Kamiya, H. Ohta, M. Hirano, and H. Hosono, Appl. Phys. Lett. 93 , 192107 (2008).
- 11 J. Lee, J.-S. Park, Y. S. Pyo, D. B. Lee, E. H. Kim, D. Stryakhilev, T. W. Kim, D. U. Jin, and Y.-G. Mo, Appl. Phys. Lett. , 95 , 123502 (2009).
- 12 A. Suresh and J. F. Muth, Appl. Phys. Lett. 92 , 033502 (2008).
- 13 M. Fakhri, N. Babin, A. Behrendt, T. Jakob, P. Goerrn, and T. Riedl, Adv. Mat. 25 , 2821 (2013).
- 14 J. C. Park, S.-E. Ahn, and H.-N. Lee, ACS Appl. Mater. Interfaces 5 , 12262 (2013).
- 15 X. Xiao, W. Deng, S. Chi, Y. Shao, X. He, L. Wang, S. Zhang, IEEE Trans. Electron Dev. 60 , 4159 (2013).
- 16 J. K. Jeong, H. W. Yang, J. H. Jeong, Y.-G. Mo, and H. D. Kim, Appl. Phys. Lett. 93 , 123508 (2008).
- 17 J.-S. Park, J. K. Jeong, H.-J. Chung, Y.-G. Mo, and H. D. Kim, Appl. Phys. Lett. 92 , 072104 (2008).
- 18 S.-Y. Sung, J. H. Choi, U. B. Han, K. C. Lee, J.-H. Lee, J.-J. Kim, W. Lim, S. J. Pearton, D. P. Norton, and Y.-W. Heo, Appl. Phys. Lett. 96 , 102107 (2010).
- 19 W.-F. Chung, T.-C. Chang, H.-W. Li, C.-W. Chen, Y.-C. Chen, S.-C. Chen, T.-Y. Tseng, and Y.-H. Tai, Electrochem. Solid-State Lett. 14 , H114 (2011).
- 20 M. E. Lopes, H. L. Gomes, M. C. R. Medeiros, P. Barquinha, L. Pereira, E. Fortunato, R. Martins, and I. Ferreira, Appl. Phys. Lett. 95 , 063502 (2009).
- 21 W.-T. Chen, S.-Y. Lo, S.-C. Kao, H.-W. Zan, C.-C. Tsai, J.-H. Lin, C.-H. Fang, and C.C. Lee, IEEE Electron Dev. Lett. 32 , 1552 (2011).
- 22 P. Xiao, L. Lan, T. Dong, Z. Lin, W. Shi, R. Yao, X. Zhu, and J. Peng, Appl. Phys. Lett. 104 , 051607 (2014).
- 23 D. F. Ogletree, H. Bluhm, G. Lebedev, C. S. Fadley, Z. Hussain, and M. Salmeron, Rev. Sci. Instrum. 73 , 3872 (2002).
- 24 R. L. Hoffman, J. Appl. Phys. 95 , 5813 (2004).
- 25 F. R. Libsch and J. Kanicki, Appl. Phys. Lett. 62 , 1286 (1993).
- 26 M. S. Rajachidambaram, A. Pandey, S. Vilayurganapathy, P. Nachimuthu, S. Thevuthasan, and G. S. Herman, Appl. Phys. Lett. 103 , 171602 (2013).
- 27 Y.-C. Huang, P.-Y. Yang, H.-Y. Huang, S.-J. Wang, and H.-C. Cheng, J. Nanosci. Nanotechnol. 12 , 5625 (2012).
- 28 T. T. T. Nguyen, O. Renault, B. Aventurier, G. Rodriguez, J. P. Barnes, and F. Templier, J. Displ. Tech. 9 , 770 (2013).
- 29 M. J. Lee, S. J. Kang, J. Y. Baik, K.-J. Kim, H.-D. Kim, H.-J. Shin, J. Chung, J. Lee, and J. Lee, J. Appl. Phys. 108 , 024507 (2010).
- 30 K. Nomura, T. Kamiya, E. Ikenaga, H. Yanagi, K. Kobayashi, and H. Hosono, J. Appl. Phys. 109 , 073726 (2011).
- 31 N. Oka, T. Aoi, R. Hayashi, H. Kumomi, and Y. Shigesto, Appl. Phys. Exp. 5 , 075802 (2012).
- 32 P. J. Hotchkiss, M. Malicki, A. J. Giordano, N. R. Armstrong, and S. R. Mader, J. Mater. Chem. 21 , 3107 (2011).
- 33 C. L. Perkins, J. Phys. Chem. C 113 , 18276 (2009).
- 34 P. J. Hotchkiss, S. C. Jones, S. A. Paniagua, A. Sharma, B. Kippelen, N. R. Armstrong, and S. R. Marder, Acct. Chem Res. 45 , 337 (2012).
- 35 D. Barriet and T. R. Lee, Curr. Opin. Colloid Interface Sci. 8 , 236 (2003).
- 36 X.Y. Deng, A. Verdaguer, T. Herranz, C. Weis, H. Bluhm, and M. Salmeron, Langmuir 24 , 9474 (2008).
- 37 Y. Xie, X. Zhao, Y. Chen, Q. Zhao, and Q.Yuan, J. Solid State Chem. 180 , 3546 (2007).
- 38 D. Kim, S. Yoon, Y. Jeong, Y. Kim, and M. Hong, Appl. Phys. Express 5 , 021101 (2012).
- 39 M. Fakhri, H. Johann, P. Gorrn, and T. Riedl, ACS Appl. Mater. Interfaces 4 , 4453 (2012).
Table 1. Average mobility, turn on voltage, drain current on-to-off ratio and hysteresis for unstressed IGZO TFTs with/without n-HPA or FPA.
| Surface treatment | avg (cm 2 /Vs) | V on (V) | I on /I off | Hysteresis (V) |
|---------------------|--------------------|------------|---------------|------------------|
| Annealed | 20.2 | -0.7 | 5.0× 10 6 | 0.6 |
| n-HPA | 20.6 | -1.4 | 3.0× 10 6 | 0.4 |
| FPA | 21.4 | -2.1 | 4.0× 10 6 | 0.3 |
Table 2. Characteristic trapping time of carriers, and stretched exponential exponent, for IGZO TFTs.
| Stress test | (s) | |
|---------------------------|-----------|------|
| Annealed, positive stress | 1.0× 10 4 | 0.45 |
| Annealed, negative stress | 5.5× 10 3 | 0.74 |
| HPA, positive stress | 2.1× 10 5 | 0.33 |
| HPA, negative stress | 7.7× 10 6 | 0.22 |
Table 3. Surface-sensitive XPS binding energies and peak areas obtained from blanket IGZO films for as deposited and while annealing to 300 º in ultrahigh vacuum. C
| | O 1s B.E. (eV) | Area (%) |
|--------------|------------------|------------|
| As deposited | 530.1 | 53.4 |
| As deposited | 531.3 | 28.2 |
| As deposited | 532.2 | 18.4 |
| | 530.1 | 88.7 |
| | 531.3 | 11.3 |
Table 4. XPS binding energies and peak areas obtained from blanket IGZO films that have been annealed and treated with n-HPA or FPA.
| | O 1s B.E. (eV) | Area (%) | C 1s B.E. (eV) | Area (%) |
|----------|------------------|------------|------------------|------------|
| | | | 284.6 | 46.1 |
| | 530.1 | 61.9 | 286.1 | 11.8 |
| FPA | 531.2 | 30.5 | 288.4 | 9.5 |
| | 532.3 | 7.6 | 290.8 | 22.7 |
| | | | 293.3 | 9.9 |
| | 530.1 | 53.3 | 284.6 | 90.6 |
| HPA | 531.2 | 37.4 | 286.4 | 6.9 |
| | 532.3 | 9.3 | 288.5 | 2.5 |
| | 530.1 | 65.8 | 284.6 | 72.4 |
| Annealed | 531.2 | 23.8 | 286.2 | 10.1 |
| | 532.3 | 10.4 | 288.8 | 17.5 |
FIG. 1. Schematic illustration of SAM functionalized IGZO TFT structure showing the mono-, bi-, and tri-dentate binding of SAM to IGZO.
FIG. 2. Plot of ID vs VGS for various treatments of the IGZO TFTs as a function of positive and negative biased stress time of 0, 10, 100, 1,000 and 10,000 s. (a) Annealed sample, positive bias, (b) annealed sample, negative bias, (c) Annealed sample with FPA, positive bias, (d) Annealed sample with FPA, negative bias, (e) Annealed sample with n-HPA, positive bias, (f) Annealed sample with n-HPA, negative bias.
FIG. 3. VON shift vs positive and negative bias stress time (s) for IGZO TFTs with and without n-HPA and FPA. For annealed devices without SAMs and for a n-HPA passivated surface, a fit of the stretched exponential model (dashed lines) is shown. For devices with a FPA passivated surface, a linear fit (dashed lines) is shown.
FIG. 4. Surface sensitive O 1s XPS data obtained from blanket as deposited IGZO films and the same film during an in-situ anneal to 300 ° C.
FIG. 5. O 1s XPS data obtained from blanket IGZO films that have been annealed and treated with n-HPA or FPA.
FIG. 6. C 1s XPS data obtained from blanket IGZO films that have been annealed and treated with n-HPA or FPA.
Figure 1:
Figure 2:
Figure 3:
Figure 4:
Figure 5:
Figure 6:
2
298 296 294 292 290 288 286 284 282 Binding Energy (eV)
 | 10.1149/2.010409jss | [
"Xiaosong Du",
"Brendan T. Flynn",
"Joshua R. Motley",
"William F. Stickle",
"Hendrik Bluhm",
"Gregory S. Herman"
] | 2014-08-02T03:18:26+00:00 | 2014-08-02T03:18:26+00:00 | [
"physics.chem-ph",
"cond-mat.mtrl-sci"
] | Role of Self-Assembled Monolayers on Improved Electrical Stability of Amorphous In-Ga-Zn-O Thin-Film Transistors | Self-assembled monolayers (SAMs) have been used to improve both the positive
and negative bias-stress stability of amorphous indium gallium zinc oxide
(IGZO) bottom gate thin film transistors (TFTs). N-hexylphosphonic acid (HPA)
and fluorinated hexylphosphonic acid (FPA) SAMs adsorbed on IGZO back channel
surfaces were shown to significantly reduce bias stress turn-on voltage shifts
compared to IGZO back channel surfaces with no SAMs. FPA was found to have a
lower surface energy and lower packing density than HPA, as well as lower bias
stress turn-on voltage shifts. The improved stability of IGZO TFTs with SAMs
can be primarily attributed to a reduction in molecular adsorption of
contaminants on the IGZO back channel surface and minimal trapping states
present with phosphonic acid binding to the IGZO surface. |
1408.0332v1 | ## Social Media Impact on Website Ranking
## Dushyant Vaghela 1
## http://explorequotes.com
Department of Computer Engineering RK University, Rajkot, India.
Abstract -Internet is fast becoming critically important to commerce, industry and individuals. Search Engine (SE) is the most vital component for communication network and also used for discover information for users or people. Search engine optimization (SEO) is the process that is mostly used to increasing traffic from 'free', 'organic' or 'natural' listings on search engines and also helps to increase website ranking. It includes techniques like link building, directory submission, classified submission etc. but SMO, on the other hand, is the process of promoting your website on social media platforms. It includes techniques like RSS feeds, social news and bookmarking sites, video and blogging sites, as well as social networking sites, such as Facebook, Twitter, Google+, Tumblr, Pinterest, Instagram etc. Social media optimization is becoming increasingly important for search engine optimization, as search engines are increasingly utilizing the recommendations of users of social networks to rank pages in the search engine result pages. Since it is more difficult to tip the influence the search engines in this way. Social Media Optimization (SMO) may also use to generate traffic on a website, promote your business at the center of social marketing place and increase ranking.
Key words : Social media, SMO, Page Ranking, SERP, Social media Optimization, SEO.
## I. INTRODUCTION
Social activities and thinking outline have usually been spread via radio, newspapers and television. However, a new trend has evolved: social media. Social media is the approach to socialize. In early, the web was plays a vital role to retrieval of discrete information but now web has become a social place that connect people worldwide. Now use of the social media becomes common most of users and business person as different ways. [1]
Social media is two-way messaging between company and follower/ People. The term 'social media' can be dividing into two parts. Media usually refers to publicity and the communication of information through publications and channels. Social involve the interaction of user within a group or community. [2]
Any website which permit user to like, dislike, tweets and share their content, and allow making community can be classified as a social media. Some social media platforms are: (1) Social Bookmarking site such as delicious, stumble upon. (2) Social Networking sites such as Facebook, Google+. (3) Photos and video sharing such as Flickr and YouTube (4) Microblogging site such as Twitter and (5) Blogs such as WordPress and Blogger.
This is a good Wiki explanation of that direction...
Social media optimization is becoming increasingly important for search engine optimization, as search engines are increasingly utilizing the recommendations of users of social networks such as Facebook, Twitter, and Google+ to rank pages in the search engine result pages... since it is more difficult to tilt the scales search engines in this way, search engines are putting more stock into social search. This, coupled with increasingly personalized search based on interests and location, has significantly increased the importance of a social media presence in search engine optimization. [3]
Rohit Bhargava was first used the term 'Social Media Optimization' in 2006 and also discovers five rules. (1) Increase your link ability (2) Make tagging and bookmarking easy (3) Reward inbound links (5) help your content travel (6) Encourage the mash up [4]
## II. SOCIAL MEDIA PLATFORM
## A. Type of Social Media and its Factors
It's one thing to know that social factors are increasingly affecting search engine rankings, but it's another thing entirely to see what integrations are already in place and influencing results pages. There are some social media Platform describes below.
## 1) Facebook
Facebook is the largest social platform of the social networks, with worldwide usage approaching 700 million users as of June 2011. This represents nearly 10% of the total world population. Facebook has the largest audience of any social network and is certainly worth including as a key part of your social media strategy. People can implement links on their Facebook pages. If the user's news page is public, these links will be visible to the search engines and can and will be counted as links. Not only that, when people share links they will also show up on the news pages of some of their friends. Publishers can encourage the sharing of their content by placing a Facebook Share button on their pages. [5]
In the current Google Social Search implementation, content shared by Facebook contacts receives preferential positions in the search engine results, as well as with a picture of the person who shared it. By increasing the social proof of these results, Google increases the chances that this content will receive clicks.
Facebook is the best social media site to increase traffic to your website by using the quality. Facebook groups, building up your Facebook page likes, and creating engagement that you create on your wall and the Facebook page wall. Images post work really well. So add images to your profile and the Fan Page wall every day. Add targeted people as friends and always message them with a message that positively encourages them to like your Facebook page. [7]
## 2) Twitter
One of the clearest integrations between social media and the search engines is the Twitter.
In April 2011, Twitter was in use by approximately 7% of the US population and this number was expected to grow to 11% by the end of 2011. Twitter market share may be small compared to the penetration of Facebook, but it certainly includes a large number of influencers that can help create visibility for your organization.
One of the unique ways to clutch traffic from Twitter, and it was by following new people and getting them to visit your blog via your profile links. Engage with as many other tweeters and you may have them become a visitor to your blog! [7]
## 3) Google +
Google+ was first announced on June 28, 2011. In Google's earnings announcement on January 12, 2012, it indicated that the social network had 90 million users, and that 60% of those people were active on the site every day. This rapid growth and broad adoption suggests that Google itself may be a strong player in the social field of the future and it now has a major impact on Google search rankings.
Google+ allows sharing in a manner similar to what you can do on Facebook. Once you insert a link into a post on the site, Google+ recognizes the URL and extracts content.
If you want to getting traffic from Google +, you should provide the name and contribute to Google +'s Communities. Google +'s communities superb way to more traffic to your site. Make sure you don't post into them without some sort of contribution or engagement you do for them, no matter how small. Make a practice to use the right Hashtags (#) when you are posting in the communities. [7]
## 4) LinkedIn
As of January 2011, LinkedIn had more than 101 million members worldwide. This makes it a powerful social network. LinkedIn can be used to build a strong network of connections, and paid versions of the service allow its. In Mail functionality to be used to send unsolicited emails to members; if done judiciously, this can be used to initiate new relationships with influencers of interest. It can also use to make a strong profile of users. [5]
## 5) YouTube
YouTube can be thought of as a search engine, but it is also a social network. People love to share videos, and this is happening in volume. Video hosting service from YouTube is a social networking website, which almost every person or organization with Internet access can upload videos that can be seen by the general public almost immediately. As the world's largest video platform, YouTube has had an impact in many areas, with some individual YouTube videos directly happens worldwide have formed. [5]
Below there are some factors used by several web services and tools that affect the authority, the trust of social media links and also assess Social Media reputation [11]
Table. 1: Social media Factors Moreover
| Sr. No | Factors |
|----------|--------------------------------------------------------------------|
| 1 | Followers on Twitter |
| 2 | Fans on Facebook Fan Pages |
| 3 | Shares & Likes on Facebook |
| 4 | The number of tweets & RTs on Twitter (Reply Tweets of your Tweet) |
| 5 | ratio between Followers-Friends |
| 6 | The authority of the people that follow you |
| 7 | The authority of the people that share your content |
|-----|------------------------------------------------------------------|
| 8 | The average quality of your previously shared messages |
| 9 | The number of unique mentions/shares (similar to Link Diversity) |
| 10 | The rate and the source of links that you share |
search engines are expected to use Social Media buzz to determine:
- · How a particular post is fresh
- · Is Particular page still useful/valid/up-to-date
- · What are the current trends
- · For the particular post which terms are related
- · They are likely to use the keywords and the hash tags that can be found in the content of the shared messages.[11]
## B. Why Rely On Social Media?
If you are still wondering why Google is pushing so hard with its new product Buzz, it is because it wants in on social traffic. For many sites on the Web, social traffic coming through Facebook, Twitter, and Myspace is beginning to rival, and in some cases overtake, search traffic as the single biggest source of traffic. This traffic comes from shared links, photos, and videos. By its own numbers, 5 billion pieces of content are shared on Facebook every week.
Social media used in business, ranking and generate traffic.
The top two benefits of social media marketing are increasing exposure and increasing traffic. A significant 89% of all marketers indicated that their social media efforts have generated more exposure for their businesses. Increasing traffic was the second major benefit, with 75% reporting positive results.
You can use social media to drive traffic to your website. Many companies from small to large are using social media to drive traffic to their websites and products. As a part of study made 71% percent of social media users are more likely to purchase from a brand they follow online. Social media is increasing every day and companies not using social media will be left behind. Finding different ways to promote your website through social media is very crucial to the success of your site. [6]
III. IMPACT OF SOCIAL MEDIA There are many Survey by different SEO Company / expert about social media such as Facebook, Twitter and Google +.
Fig. 1 represents Summary of Search metrics SEO Ranking Factors - Rank Correlation 2013. In the Figure 1, the correlation coefficient is plan on the x-axis, which measures its size. The longer/shorter the bar in the positive part of the x-axis that represent the higher/lower the correlation between that particular factor and a good Google ranking. Factors with a correlation factor of zero do not have any considerable correlation with Google's results. [8]
In this chart, there are some factor of social networking such as Facebook, Twitter and Google + with its ranking possibility which platform is more important.
It is noticeable that social signals correlate well with better rankings and content quality as well as the number and diversity of backlinks seem to have a huge impact on search result rankings. [10]
Fig. 1: Study of Google Search Result
Below you will also find the hottest questions that have been answered from interview along with the replies of each Search Engine: Danny Sullivan asks to both Google and Bing during interview whether the data of Twitter and Facebook affect the regular search Result or not. [9]
Table. 2: Google and Bing Interview
| Sr.No | Question | Google | Bing |
|---------|---------------------------------------------------------------------|----------------------------------------------------------------------------------------------------------------|---------------------------------------------------------------------------------------------------------------------------------------------------------------|
| 1 | Are Regular Search Results affected by social media buzz? | Yes. It is used as a signal especially for news. | Yes it is a signal. Some weight is passed and regular results are affected. |
| 2 | Are Social/Realtime Search Results affected by buzz? | Heavily affected | Heavily affected, Authority metrics are used to determine the |
| 3 | Are Twitter links taken into account (aka do they pass link juice)? | In some limited situations the data are used. | hot posts. The data are used. The weight depends on how often a link is posted, the number of tweets & retweets and the authority of the people that post it. |
| 4 | Are Facebook links taken into account? | The shared links from Fan pages are treated similarly to Twitter links. No links from personal walls are used. | Only the publicly shared links from Fan pages and Profiles are taken into account. |
| 5 | Is there an Authority Rank for Twitter | Yes, the author quality is measured. Google | Yes. Several metrics are taken into account. |
Based on the interview by Danny Sullivan, Chief Editor of Search engine land which confirms the doubts of many SEOs: Social Media does affect the organic rankings of Google and Bing. Some social media play a vital role for search engine but does it directly influence the ranking. IV. CONCLUSION
In this review paper, I concluded that Social media and user engagement have become significant factors for those
| | Profiles? | calls this Author Authority. | Bing calls this Social Authority. |
|----|----------------------------------------------------------------------------------------|-----------------------------------------------------------------------------------------------------------------------------------------------|------------------------------------------------------------------------------------------------------------------------------------|
| 6 | Is there an Authority Rank for Facebook Profiles? | They are treated similarly to Twitter Profiles. Note: Normally if this is true, they measure only public profiles like Fan pages. | No. They take into account only the Facebook data that are publicly available. |
| 7 | Does the Twitter Authority have an impact on the weight of the links that are shared? | Yes. The weight depends on the authority of the person. The Author Authority is used in limited situations and it is independent of PageRank. | The Author Authority affects the link weight. |
| 8 | Does the Facebook Authority have an impact on the weight of the links that are shared? | Similarly to Twitter. The weight of each link depends on the authority of the profile. | They don't use Facebook authority. Instead in order to find if something is really popular they compare Twitter & Facebook shares. |
practicing SEO. Links may still be the strongest source of signals to search engines. A balanced SEO strategy needs to take social media, user data, and document analysis techniques into account. Publishers must consider these four areas as a part of their web promotional strategies: (1) Participating in social media communities (2) Providing an engaging user experience (3) Offering unique and differentiated content (4) Building a brand.
I focused on social media platform such as Social Networking site, Social Bookmarking, microblogging, all have different functionality and also the use of social media goal to improve business, increase / generate traffic to your website and ranking in SERP. Some social networking site factor most likely used such as Facebook Share, Twitter tweets but some are less likely used such as Facebook Like, Google+ for increase traffic and ranking.
Based on the survey, Impact of Google + is very strong compare to other to ranking but Facebook like, share and twitter tweet are helps to generate traffic but continuously engagement is required. If we focus on Facebook's like, comment and Google+, tweets and social bookmarking factor, we can easily promote business, generate traffic and improve ranking in SERP.
## REFERENCES
- [1] "Social Media - Is It Worth the Trouble?", Iris Uitz, IBIMA Publishing,Journal of Internet Social Networking & Virtual Communities,Vol. 2012 (2012), Article ID 313585,
- [2] "SOCIAL MEDIA AND ITS ROLE IN MARKETING", Ms. Sisira Neti, International Journal of Enterprise Computing and Business Systems ISSN (Online) : 2230-8849,Vol. 1 Issue 2 July 2011
- [3] "Search Engine Optimization in an Increasingly Social World", Enge, Eric (30 November 2010), Search Engine Watch. Retrieved 17 July 2012. http://www.searchmetrics.com/en/whitepaper/googleranking-factors-us-2012/
- [4] "5 Rules of Social Media Optimization (SMO)",Rohit
Bhargava, August 10, 2006, http://www.rohitbhargava.com/2006/08/5\_rules\_of\_soci .html
- [5] "Art Of SEO",Eric Enge, Stephan Spencer, Jessie Stricchiola, and Rand Fishkin, March 2012
- [6] Use Social Media To Drive Traffic To Your Website" Posted on December 2, 2013 by Nancy Badillo
- [7] "How Social Media Affects SEO" www.quicksprout.com/2011/08/24/how-socialmediaaffects-seo, 12/4/2013.
- [8] 'SEO Ranking Factor - Rank Correlation 2013 for Google USA' http://www.searchmetrics.com/en/ services/ ranking-factors-2013/
- [9] 'Social network sites: Definition, history, and scholarship', Danah Boyd and Nicole B. Ellison, Journal of Computer-Mediated Communication, 13(1), 2007.
- [10] WHITE PAPER SEO Ranking Factors -Rank Correlation 2013 - Google USA -,Marcus Tober, Dr. Leonhard Hennig, Daniel Furch
- [11] Twitter and Facebook links affect SEO on Google and Bing Posted by Vassilis Vryniotis on Tuesday, December 7 at 11:46 to Marketing, SEO, and Social Media. | null | [
"Dushyant Vaghela"
] | 2014-08-02T03:45:53+00:00 | 2014-08-02T03:45:53+00:00 | [
"cs.SI"
] | Social Media Impact on Website Ranking | Internet is fast becoming critically important to commerce, industry and
individuals. Search Engine (SE) is the most vital component for communication
network and also used for discover information for users or people. Search
engine optimization (SEO) is the process that is mostly used to increasing
traffic from free, organic or natural listings on search engines and also helps
to increase website ranking. It includes techniques like link building,
directory submission, classified submission etc. but SMO, on the other hand, is
the process of promoting your website on social media platforms. It includes
techniques like RSS feeds, social news and bookmarking sites, video and
blogging sites, as well as social networking sites, such as Facebook, Twitter,
Google+, Tumblr, Pinterest, Instagram etc.Social media optimization is becoming
increasingly important for search engine optimization, as search engines are
increasingly utilizing the recommendations of users of social networks to rank
pages in the search engine result pages. Since it is more difficult to tip the
influence the search engines in this way. Social Media Optimization (SMO) may
also use to generate traffic on a website, promote your business at the center
of social marketing place and increase ranking. |
1408.0333v1 | An introduction to spectral data for Higgs bundles
## Outline
These notes have been prepared as reading material for the mini-course ' An introduction to spectral data for Higgs bundles ' that the author gave at the National University of Singapore as part of the Summer school on the moduli space of Higgs bundles .
## Lecture 1
The first lecture shall introduce classical Higgs bundles and the Hitchin fibration, and describe the associated spectral data in the case of principal Higgs bundles for classical complex Lie groups. Whilst bibliography is provided in the text, the main references followed are Hitchin's papers [Hit87, Hit87a, Hit92, Hit07].
## Lecture 2
During the second lecture we shall construct Higgs bundles for real forms of classical complex Lie groups as fixed points of involutions, and describe the corresponding spectral data when known, as appearing in [Sch11, Sch13, Sch13b] and [HitSch]. Along the way, we shall mention different applications and open problems related to the methods introduced in both lectures.
## Exercises
Each lecture contains exercises of varying difficulty, whose solutions can be found in [Sch13], and commented in the .tex file. Open problems which might be tackled with methods similar to the ones introduced in the lectures shall also be mentioned, and appear indicated with ((*)) . For each of these problems we suggest references which feature approaches that may be useful.
## Bibliography
We shall highlight the main references considered, as well as the precise places where the methods used were developed. Since it proves to be very difficult to give a comprehensive and exhaustive account of research in tangential areas, we shall restrain ourselves to mentioning related work only when it directly involves methods using spectral data . The reader should refer to references in the bibliography for further research in related topics (e.g., see references in [Ap09, P13, Sch13]).
## Acknowledgments
The author would like to thank the Institute of Mathematical Sciences at the National University of Singapore and the organisers of the program The Geometry, Topology and Physics of Moduli Spaces of Higgs Bundles for providing an ideal environment for research and collaborations. The author also acknowledges support from U.S. National Science Foundation grants DMS 1107452, 1107263, 1107367 'sRNMS: GEometric structures And Representation varieties' (the GEAR Network).
## CHAPTER 1
## Spectral data for G c -Higgs bundles
The art of doing mathematics consists in finding that special case which contains all the germs of generality.
David Hilbert
Following [Hit87, Hit87a, Hit07] we dedicate this lecture to overview classical Higgs bundles as well as G c -Higgs bundles for the groups G c = SL n, ( C ), Sp n, ( C ), SO (2 n +1 , C ) and SO (2 n, C ). In each case we introduce the Hitchin fibration and describe the generic fibres through spectral data , i.e., an associated spectral curve and a line bundle on it.
## 1.1 G c -Higgs bundles
Consider Σ a compact Riemann surface of genus g ≥ 2 with canonical bundle K = T ∗ Σ. Classically, a Higgs bundle on Σ is defined as follows:
Definition 1.1. A Higgs bundle is a pair ( E, Φ) for E a holomorphic vector bundle on Σ , and Φ , the Higgs field , a holomorphic section in H 0 (Σ , End ( E ) ⊗ K ) .
In order to understand better what Higgs bundles are and how to generalise the definition, we shall first look at the moduli space of vector bundles and then study the moduli space of classical Higgs bundles and its associated spectral data. For more details the reader should refer to [Hit87, Hit87a, D87, Cor88, S88, N91, S92].
## 1.1.1 Moduli space of vector bundles
Holomorphic vector bundles E on a compact Riemann surface Σ of genus g ≥ 2 are topologically classified by their rank rk E ( ) and their degree deg E ( ).
Definition 1.2. The slope of a holomorphic vector bundle E is µ E ( ) := deg E /rk E ( ) ( ) and is used to define stability conditions: A vector bundle E is said to be stable ( semi-stable ) if for any proper, non-zero sub-bundle F ⊂ E one has µ F ( ) < µ E ( ) ( µ F ( ) ≤ µ E ( ) ). It is polystable if it is a direct sum of stable bundles whose slope is the same as E .
It is known that the space of holomorphic bundles of fixed rank and fixed degree, up to isomorphism, is not a Hausdorff space. However, through Mumford's Geometric Invariant Theory one can construct the moduli space N ( n, d ) of stable bundles of fixed rank n and degree d , which has the natural structure of an algebraic variety.
Remark 1.3. For coprime n and d , the moduli space N ( n, d ) is a smooth projective algebraic variety of dimension n 2 ( g -1) + 1 .
Remark 1.4. All line bundles are stable, and thus N (1 , d ) contains all line bundles of degree d , and is isomorphic to Jac d (Σ) of Σ , an abelian variety of dimension g .
Let G c be a complex semisimple Lie group. Following [Ra75] one can define stability for principal G c -bundles as follows (see [Ap09, Section 1.1] for a comprehensive study):
Definition 1.5. A holomorphic principal G c -bundle P is said to be stable ( semi-stable ) if for every reduction σ : Σ → P/Q to maximal parabolic subgroups Q of G c one has deg σ T ∗ rel > 0 ( ≥ 0) , where T rel is the relative tangent bundle for the projection P/Q → Σ .
The notion of polystability may be carried over to principal G c -bundles, allowing one to construct the moduli space of isomorphism classes of polystable principal G c -bundles of fixed topological type over the compact Riemann surface Σ.
## 1.1.2 Moduli space of classical Higgs bundles
In order to define the moduli space of Higgs bundles, the following stability condition is considered:
Definition 1.6. A vector subbundle F of E for which Φ( F ) ⊂ F ⊗ K is said to be a Φ invariant -subbundle of E . A Higgs bundle ( E, Φ) is said to be
- · stable ( semi-stable ) if for each proper Φ -invariant F ⊂ E one has µ F ( ) < µ E ( ) ( equiv. ≤ ) ;
- · polystable if ( E, Φ) = ( E , 1 Φ ) 1 ⊕ ( E , 2 Φ ) 2 ⊕ . . . ⊕ ( E , r Φ ) r , where ( E , i Φ ) i is stable with µ E ( i ) = µ E ( ) for all i .
Remark 1.7. The characteristic polynomial of Φ restricted to an invariant subbundle divides the characteristic polynomial of Φ .
Example 1.8. Choose a square root K 1 / 2 of the canonical bundle K , and a section ω of K 2 . A family of classical Higgs bundles ( E, Φ ) ω may be obtained by considering the vector bundle E = K 1 2 ⊕ K -1 2 and the Higgs bundle Φ ω given by
$$\Phi _ { \omega } = \left ( \begin{array} { c c } 0 & \omega \\ 1 & 0 \end{array} \right ) \in H ^ { 0 } ( \Sigma, E n d ( E ) \otimes K )$$
Exercise 1. Show that the pairs ( E, Φ ) ω from Example 1.8 are stable.
Exercise 2. Prove that if a Higgs bundle ( E, Φ) is stable, then for λ ∈ C ∗ and α a holomorphic automorphism of E , the induced Higgs bundles ( E,λ Φ) and ( E,α ∗ Φ) are stable.
In order to define the moduli space of classical Higgs bundles, we shall first define an appropriate equivalence relation. For this, consider a strictly semi-stable Higgs bundle ( E, Φ). As it is not stable, E admits a subbundle F ⊂ E of the same slope which is preserved by Φ. If F is a subbundle of E of least rank and same slope which is preserved by Φ, it follows that F is stable and hence the induced pair ( F, Φ) is stable. Then, by induction one obtains a flag of subbundles F 0 = 0 ⊂ F 1 ⊂ . . . ⊂ F r = E where µ F /F ( i i -1 ) = µ E ( ) for 1 ≤ i ≤ r , and where the induced Higgs bundles ( F /F i i -1 , Φ ) are stable. i This is the Jordan-H¨ older filtration of E , and it is not unique. However, the graded object Gr( E, Φ) := ⊕ r i =1 ( F /F i i =1 , Φ ) is unique up to isomorphism. i
Definition 1.9. Two semi-stable Higgs bundles ( E, Φ) and ( E , ′ Φ ) ′ are said to be S -equivalent if Gr ( E, Φ) = ∼ Gr ( E , ′ Φ ) ′ .
Exercise 3. If a pair ( E, Φ) is strictly stable, what is the induced Jordan-H¨lder filtration? o
Following [N91, Theorem 5.10] we let M ( n, d ) be the moduli space of S -equivalence classes of semi-stable Higgs bundles of fixed degree d and fixed rank n . The moduli space M ( n, d ) is a quasi-projective scheme, and has an open subscheme M ′ ( n, d ) which is the moduli scheme of stable pairs. Thus, every point is represented by either a stable or a polystable Higgs bundle. When d and n are coprime, the moduli space M ( n, d ) is smooth.
The cotangent space of N ( n, d ) over the stable locus is contained in M ( n, d ) as a Zariski open subset. The moduli space M ( n, d ) is a non-compact variety which has complex dimension 2 n 2 ( g -1) + 2. Moreover, it is a hyperk¨hler manifold with natural symplectic form a ω defined on the infinitesimal deformations ( ˙ A, ˙ Φ) of a Higgs bundle ( E, Φ) by
$$\omega ( ( \dot { A } _ { 1 }, \dot { \Phi } _ { 1 } ), ( \dot { A } _ { 2 }, \dot { \Phi } _ { 2 } ) ) = \int _ { \Sigma } \text{tr} ( \dot { A } _ { 1 } \dot { \Phi } _ { 2 } - \dot { A } _ { 2 } \dot { \Phi } _ { 1 } ), & & ( 1. 1 )$$
where ˙ A ∈ Ω 0 1 , (End 0 E ) and Φ ˙ ∈ Ω 1 0 , (End 0 E ) (see [Hit87, Hit87a] for details). For simplicity, we shall fix n and d and write M for M ( n, d ).
## 1.1.3 Moduli space of G c -Higgs bundles
The notion of Higgs bundle can be generalized to encompass principal G c -bundles, for G c a complex semi-simple Lie group. For more details, the reader should refer to [Hit87a].
Definition 1.10. A G c -Higgs bundle is a pair ( P, Φ) where P is a principal G c -bundle over Σ , and the Higgs field Φ is a holomorphic section of the vector bundle ad P ⊗ C K , for ad P the vector bundle associated to the adjoint representation.
When G c ⊂ GL n, ( C ), a G c -Higgs bundle gives rise to a Higgs bundle in the classical sense, with some extra structure reflecting the definition of G c . In particular, classical Higgs bundles are given by GL n, ( C )-Higgs bundles.
Example 1.11. The Higgs bundles in Example 1.8 have traceless Higgs field, and the determinant Λ 2 E is trivial. Hence, for each quadratic differential ω one has an SL (2 , C ) -Higgs bundle ( E, Φ ) ω .
By extending the stability definitions for principal G c -bundles, one can define stable , semistable and polystable G c -Higgs bundles. Since in these notes we shall be working with Higgs pairs which do not preserve any subbundle, they will be automatically stable and thus we shall not dedicate time to recall the main study of stability for principal Higgs bundles. For details about the corresponding constructions, the reader should refer for example to [BiGo08, Section 3], [Ap09, Section 1]. We denote by M G c the moduli space of S -equivalence classes of polystable G c -Higgs bundles.
In the remainder of this Section, following [Hit87a] and [Hit07] we introduce the Hitchin fibration and describe the generic fibres for G c -Higgs bundles where G c = GL n, ( C ), SL n, ( C ), Sp (2 n, C ), SO (2 n + 1 , C ) and SO (2 n, C ). We shall cover with more detail the initial cases, and leave as an exercise to the reader some of the results for the latter groups.
## 1.2 The Hitchin fibration
A natural way of studying M G c is through the Hitchin fibration, as introduced in [Hit87a]. We shall denote by p i , for i = 1 , . . . , k , a homogeneous basis for the algebra of invariant polynomials of the Lie algebra g c of G c , and let d i be their degrees. Then, the Hitchin fibration is given by
$$h \, \colon \, \mathcal { M } _ { G _ { c } } \, \longrightarrow \, \mathcal { A } _ { G _ { c } } \coloneqq \bigoplus _ { i = 1 } ^ { k } H ^ { 0 } ( \Sigma, K ^ { d _ { i } } ), \quad & & ( 1. 2 ) \\ \nu \, \nu \, \times \, \quad. \, \quad \nu \, \nu \times \, \quad & \nu \, \infty$$
↦
$$( E, \Phi ) \ \mapsto \ \ ( p _ { 1 } ( \Phi ), \dots, p _ { k } ( \Phi ) ).$$
The map h is referred to as the Hitchin map , and is a proper map for any choice of basis (see [Hit87a, Section 4] for details). Furthermore, A G c always satisfies dim A G c = dim M G c / 2 , making the Higgs bundle moduli space into an integrable system.
Remark 1.12. A homogenous basis of invariant polynomials for classical Higgs bundles ( E, Φ) of rank n can be taken as tr(Φ ) i for 1 ≤ i ≤ n .
Remark 1.13. Whilst a formal definition of the smooth locus of the Hitchin base can be given (e.g., see [Do95]) in these lectures we shall note that the generic fibres of the Hitchin fibration are smooth, and thus generic points in the Hitchin base are in the smooth locus.
## 1.2.1 GL n, ( C ) -Higgs bundles
As before, let K be the canonical bundle of Σ, and X its total space with projection π : X → Σ. We shall denote by η the tautological section of the pull back π K ∗ on X . Abusing notation we denote with the same symbols the sections of powers K i on Σ and their pull backs to X . The characteristic polynomial of a Higgs bundle ( E, Φ) in a generic fibre h -1 ( a ) defines a smooth curve π : S a → Σ in X , the spectral curve of Φ, whose equation is
$$\det ( \eta I d - \pi ^ { * } \Phi ) = \eta ^ { n } + a _ { 1 } \eta ^ { n - 1 } + a _ { 2 } \eta ^ { n - 2 } + \dots + a _ { n - 1 } \eta + a _ { n } = 0, \quad \quad ( 1. 4 )$$
for a i ∈ H 0 (Σ , K i ) (for simplicity, we shall write det( η -Φ) for the characteristic polynomial of the Higgs field Φ, and drop the subscript a of S a ). By the adjunction formula on X (see e.g. [GH11]), since the canonical bundle K has trivial cotangent bundle one has K S ∼ = π K ∗ n , and hence the genus of S is
$$g _ { S } & = 1 + n ^ { 2 } ( g - 1 ). & ( 1. 5 )$$
The spectral data classical Higgs bundles in a smooth fibre of the Hitchin fibration is given by a spectral curve S defined as in (1.4) and a line bundle L ∈ Jac( S ).
In order to see that the smooth fibres of the Hitchin fibration are Jacobians, starting with a line bundle L on the smooth curve π : S → Σ with equation as in (1.4), we shall obtain a classical Higgs bundle by considering the direct image π L ∗ of L . Recall that by definition of direct image, given an open set U ⊂ Σ, one has H 0 ( π -1 ( U ) , L ) = H 0 ( U , π ∗ L . ) Multiplication by the the tautological section η induces the map
$$H ^ { 0 } ( \pi ^ { - 1 } ( \mathcal { U } ), L ) \stackrel { \eta } { \rightarrow } H ^ { 0 } ( \pi ^ { - 1 } ( \mathcal { U } ), L \otimes \pi ^ { * } K ),$$
which by definition of direct image can be pushed down to give
$$\Phi \colon \pi _ { * } L \to \pi _ { * } L \otimes K.$$
Then, one obtains a Higgs field Φ ∈ H 0 (Σ End , E ⊗ K ) for E := π L ∗ .
Exercise 4. Use Grothendieck-Riemann-Roch to see that deg ( E ) = deg ( L ) + ( n 2 -n )(1 -g . )
Moreover, the Higgs field satisfies its characteristic equation, which by construction is given by η n + a η 1 n -1 + a η 2 n -2 + . . . + a n -1 η + a n = 0 . Furthermore, since S is irreducible, from Remark 1.7 there are no invariant subbundles of the Higgs field, making the induced Higgs bundle ( E, Φ) stable.
Conversely, let ( E, Φ) be a classical Higgs bundle. The characteristic polynomial is given by det( x -Φ) = x n + a x 1 n -1 + a x 2 n -2 + . . . + a n -1 x + a n , and its coefficients define the spectral curve S in the total space X whose equation is (1.4).
From [BNR, Proposition 3.6], there is a bijective correspondence between Higgs bundles ( E, Φ) and the line bundles L on the spectral curve S described previously. This correspondence identifies the fibre of the Hitchin map with the Picard variety of line bundles of the appropriate degree. By tensoring the line bundles L with a chosen line bundle of degree -deg( L ), one obtains a point in the Jacobian Jac( S ), the abelian variety of line bundles of degree zero on S , which has dimension g S as in (1.5). In particular, the Jacobian variety is the connected component of the identity in the Picard group H 1 ( S, O ∗ S ). Thus, the fibre of the classical Hitchin fibration h : M→A is isomorphic to the Jacobian of the spectral curve S . For more details, the reader should refer for example to [Hit07, Section 2].
↦
Example 1.14. In the case of a classical rank 2 Higgs bundle ( E, Φ) , the characteristic polynomial of Φ defines a spectral curve π : S → Σ . This is a 2-fold cover of Σ in the total space of K , and has equation η 2 + a 2 = 0 , for a 2 a quadratic differential and η the tautological section of π K ∗ . By [BNR, Remark 3.5] the curve is smooth when a 2 has simple zeros, and in this case the ramification points are given by the divisor of a 2 . For z a local coordinate near a ramification point, the covering is given by z → z 2 := w. In a neighbourhood of z = 0 , a section of the line bundle M looks like f ( w ) = f 0 ( w ) + zf 1 ( w . ) Since the Higgs field is obtained via multiplication by η , one has
$$\Phi ( f _ { 0 } ( w ) + z f _ { 1 } ( w ) ) = w f _ { 1 } ( w ) + z f _ { 0 } ( w ),$$
and thus a local form of the Higgs field Φ is given by
$$\Phi = \left ( \begin{array} { c c } 0 & w \\ 1 & 0 \end{array} \right ).$$
Remark 1.15. When G c ⊂ GL n, ( C ) , the spectral data of a G c -Higgs bundle is given by the spectral data of the pair as a classical Higgs bundle, satisfying extra conditions.
Remark 1.16. For generic G c , a description of the fibres can be obtained by means of Cameral covers [DoMa96] (see also [Do95]) which is equivalent to the one given in the next sections for classical Lie groups.
## 1.2.2 SL n, ( C ) -Higgs bundles
When G c = SL n, ( C ) we apply Definition 1.10 to obtain the following:
Definition 1.17. An SL n, ( C ) -Higgs bundle is a classical Higgs bundle ( E, Φ) where the rank n vector bundle E has trivial determinant and the Higgs field has zero trace.
A basis for the invariant polynomials on the Lie algebra sl ( n, C ) is given by the coefficients of the characteristic polynomial of a trace-free matrix A ∈ sl ( n, C ). In this case, the spectral curve π : S → Σ associated to the Higgs bundle has equation
$$\eta ^ { n } + a _ { 2 } \eta ^ { n - 2 } + \dots + a _ { n - 1 } \eta + a _ { n } = 0,$$
where a i ∈ H 0 (Σ , K i ) are the coefficients of the characteristic polynomial of the Higgs field Φ. Generically S is a smooth curve of genus g S = 1 + n 2 ( g -1) , and the coefficients define the corresponding Hitchin fibration
$$h \, \colon \, \mathcal { M } _ { S L ( n, \mathbb { C } ) } \longrightarrow \mathcal { A } _ { S L ( n, \mathbb { C } ) } \coloneqq \bigoplus _ { i = 2 } ^ { n } H ^ { 0 } ( \Sigma, K ^ { i } ).$$
In this case the generic fibres of the Hitchin fibration are given by the subset of Jac( S ) of line bundles L on S for which π L ∗ = E and Λ n π L ∗ is trivial. By understanding these conditions in terms of L one has the following:
The generic fibre h -1 ( a ) of the SL n, ( C ) Hitchin fibration is biholomorphically equivalent to the Prym( S, Σ) variety of the spectral curve S a defined as in (1.4).
In order to see why one has to take the Prym variety, recall that the Norm map
$$N m \, \colon \text{Pic} ( S ) \to \text{Pic} ( \Sigma ),$$
associated to π is defined on divisor classes by Nm( ∑ n p i i ) = ∑ n π p i ( i ). In particular,
$$N m ( \pi ^ { - 1 } ( x ) ) = \pi ( \pi ^ { - 1 } ( x ) ) = n x.$$
The kernel of the Norm map is the Prym variety , and is denoted by Prym( S, Σ). From [BNR, Section 4], the determinant bundle of L satisfies
$$\Lambda ^ { n } \pi _ { * } L \cong N { \mathrm m } ( L ) \otimes K ^ { - n ( n - 1 ) / 2 }.$$
Thus, Λ n π L ∗ is trivial if and only if
$$N m ( L ) & \cong K ^ { n ( n - 1 ) / 2 }.$$
Equivalently, since Nm( ∑ n π i -1 ( p i )) = n ∑ n p , i i the determinant bundle Λ n π L ∗ is trivial if the line bundle M := L ⊗ π K ∗ -( n -1) / 2 is in the Prym variety.
Remark 1.18. In the case of even rank, equation (1.9) implies a choice of a square root of K (see [Hit87] and [Hit07, Section 2.2] for more details).
## 1.2.3 Sp (2 n, C ) -Higgs bundles
Let G c = Sp (2 n, C ), and let V be a 2 n dimensional vector space with a non-degenerate skewsymmetric form <,> . For v , v i j eigenvectors of A ∈ sp (2 n, C ) with eigenvalues λ i and λ j ,
$$\lambda _ { i } < v _ { i }, v _ { j } > \ = \ < \lambda _ { i } v _ { i }, v _ { j } >$$
$$= \ < A v _ { i }, v _ { j } >$$
$$= \ - < v _ { i }, A v _ { j } >$$
$$= \ - < v _ { i }, \lambda _ { j } v _ { j } > = - \lambda _ { j } < v _ { i }, v _ { j } >.$$
From the above one has that < v , v i j > = 0 unless λ i = -λ j . Since < v , v j j > = 0, from the non-degeneracy of the symplectic inner product it follows that if λ i is an eigenvalue so is -λ i . Thus, distinct eigenvalues of A must occur in ± λ i pairs, and the corresponding eigenspaces are paired by the symplectic form. The characteristic polynomial of A must therefore be of the form
$$\det ( x - A ) = x ^ { 2 n } + a _ { 1 } x ^ { 2 n - 2 } + \dots + a _ { n - 1 } x ^ { 2 } + a _ { n },$$
and a basis for the invariant polynomials on the Lie algebra sp (2 n, C ) is given by a , . . . , a 1 n .
Definition 1.19. An Sp (2 n, C ) -Higgs bundle is a pair ( E, Φ) for E a rank 2 n vector bundle with a symplectic form ω ( , ) , and the Higgs field Φ ∈ H 0 (Σ , End ( E ) ⊗ K ) satisfying
$$\omega ( \Phi v, w ) = - \omega ( v, \Phi w ).$$
The volume form ω n trivialises the determinant bundle Λ 2 n E ∗ . The characteristic polynomial det( η -Φ) defines a spectral curve π : S → Σ in X with equation
$$\eta ^ { 2 n } + a _ { 1 } \eta ^ { 2 n - 1 } + \dots + a _ { n - 1 } \eta ^ { 2 } + a _ { n } = 0,$$
whose genus is g S := 1 + 4 n 2 ( g -1) . The curve S has a natural involution σ η ( ) = -η and thus one can define the quotient curve ¯ : π S = S/σ → Σ, of which S is a 2-fold cover
$$p \colon S \rightarrow \overline { S }.$$
Note that the Norm map associated to p satisfies p ∗ Nm( ) = x x + σx , and thus the Prym variety Prym( S, S ) is given by the line bundles M ∈ Jac( S ) for which σ M ∗ ∼ = M ∗ .
As in the case of classical Higgs bundles, the characteristic polynomial of a Higgs field Φ gives the Hitchin fibration
$$h \, \colon \, \mathcal { M } _ { S p ( 2 n, C ) } \longrightarrow \mathcal { A } _ { S p ( 2 n, C ) } \coloneqq \bigoplus _ { i = 1 } ^ { n } H ^ { 0 } ( \Sigma, K ^ { 2 i } ),$$
and one has the following:
The generic fibres h -1 ( a ) of the Hitchin fibration for Sp (2 n, C )-Higgs bundles is given by Prym varieties Prym( S, S ) where S and its quotient ¯ are the curves defined by S a as above.
The spectral data described above for an Sp (2 n, C )-Higgs bundle ( E, Φ) can be obtained by looking at the extra conditions needed on L ∈ Jac( S ) associated to the corresponding classical Higgs pair for which π L ∗ = E . In order to understand this, note that for V ⊂ S an open set, we have V ⊂ π -1 ( π ( V )) and hence a natural restriction map H 0 ( π -1 ( π ( V )) , L ) → H 0 ( V , L ) , which gives the evaluation map ev : π π L ∗ ∗ → L . Multiplication by η commutes with this linear map and so the action of π ∗ Φ on the dual of the vector bundle π π L ∗ ∗ preserves a one-dimensional subspace. Hence L ∗ is an eigenspace of π ∗ Φ , with eigenvalue t η . Equivalently, L is the cokernel of π ∗ Φ -η acting on π E ∗ ⊗ π K ∗ ∗ . By means of the Norm map for π , this correspondence can be seen on the curve S via the exact sequence
$$0 \to L \otimes \pi ^ { * } K ^ { 1 - 2 n } \to \pi ^ { * } E \xrightarrow { \pi ^ { * } \Phi - \eta } \pi ^ { * } ( E \otimes K ^ { * } ) \xrightarrow { e v } L \otimes \pi ^ { * } K \to 0, \text{ \quad \ } ( 1. 1 5 )$$
and its dualised sequence
$$0 \to L ^ { * } \otimes \pi ^ { * } K ^ { * } \to \pi ^ { * } ( E ^ { * } \otimes K ^ { * } ) \to \pi ^ { * } E ^ { * } \to L ^ { * } \otimes \pi ^ { * } K ^ { 2 n - 1 } \to 0. \quad \quad ( 1. 1 6 )$$
In particular, from the relative duality theorem one has that
$$\pi _ { * } ( L ) ^ { * } & \cong \pi _ { * } ( K _ { S } \otimes \pi ^ { * } K ^ { - 1 } \otimes L ^ { * } ), & ( 1. 1 7 )$$
and thus E ∗ is the direct image sheaf π ∗ ( L ∗ ⊗ π K ∗ 2 n -1 ).
Given an Sp (2 n, C )-Higgs bundle ( E, Φ), one has Φ t = -Φ and an eigenspace L of Φ with eigenvalue η is transformed to σ L ∗ for the eigenvalue -η . Moreover, since the line bundle L is the cokernel of π ∗ Φ -η acting on π ∗ ( E ⊗ K ∗ ), one can consider the corresponding exact sequences (1.15) and its dualised sequence, which identify L ∗ with L ⊗ π K ∗ 1 -2 n , or equivalently, L 2 = π K ∗ 2 n -1 . By choosing a square root K 1 / 2 one has a line bundle M := L ⊗ π K ∗ -n +1 2 / for which σ M ∗ ∼ = M ∗ , i.e., which is in the Prym variety Prym( S, S ).
Conversely, an Sp (2 p, C )-Higgs bundle can be recovered from a line bundle M ∈ Prym( S, S ), for S a smooth curve with equation (1.13) and ¯ its quotient curve. Indeed, by Bertini's theorem, such S a smooth curve S with equation (1.13) always exists. Letting E := π L ∗ for L = M ⊗ π K ∗ n -1 / 2 , one has the exact sequences (1.15) and its dualised on the curve S . Moreover, since L 2 ∼ = π K ∗ 2 n -1 , there is an isomorphism E ∼ = E ∗ which induces the symplectic structure on E . Hence, the generic fibres of the corresponding Hitchin fibration can be identified with the Prym variety Prym( S, S ).
## 1.2.4 SO (2 n +1 , C ) -Higgs bundles
We shall now consider the special orthogonal group G c = SO (2 n + 1 , C ) and the corresponding Higgs bundles. Following a similar analysis as in the previous case, one can see that for a generic matrix A ∈ so (2 n + 1 , C ), its distinct eigenvalues occur in ± λ i pairs. Thus, the characteristic polynomial of A must be of the form
$$\det ( x - A ) = x ( x ^ { 2 n } + a _ { 1 } x ^ { 2 n - 2 } + \dots + a _ { n - 1 } x ^ { 2 } + a _ { n } ), & & ( 1. 1 8 )$$
where the coefficients a , . . . , a 1 n give a basis for the invariant polynomials on so (2 n +1 , C ).
Definition 1.20. An SO (2 n + 1 , C ) -Higgs bundle is a pair ( E, Φ) for E a holomorphic vector bundle of rank 2 n +1 with a non-degenerate symmetric bilinear form ( v, w ) , and Φ a Higgs field in H 0 (Σ , End 0 ( E ) ⊗ K ) which satisfies
$$( \Phi v, w ) = - ( v, \Phi w ).$$
The moduli space M SO (2 n +1 , C ) has two connected components, characterised by a class w 2 ∈ H 2 (Σ , Z 2 ) = ∼ Z 2 , depending on whether E has a lift to a spin bundle or not. The spectral curve induced by the characteristic polynomial in (1.18) is a reducible curve: an SO (2 n + 1 , C )-Higgs field Φ always has a zero eigenvalue, and from [Hit07, Section 4.1] the zero eigenspace E 0 is given by E 0 ∼ = K -n .
From (1.18), the characteristic polynomial det( η -Φ) defines a component of the spectral curve, which for convenience we shall denote by π : S → Σ, and whose equation is given by η 2 n + a η 1 2 n -2 + . . . + a n -1 η 2 + a n = 0 , where a i ∈ H 0 (Σ , K 2 i ). This is a 2 n -fold cover of Σ, with genus g S = 1 + 4 n 2 ( g -1) . The Hitchin fibration in this case is given by the map
$$h \ \colon \ \mathcal { M } _ { S O ( 2 n + 1, C ) } \longrightarrow \mathcal { A } _ { S O ( 2 n + 1, C ) } \coloneqq \bigoplus _ { i = 1 } ^ { n } H ^ { 0 } ( \Sigma, K ^ { 2 i } ), & & ( 1. 1 9 )$$
which sends each pair ( E, Φ) to the coefficients of det( η -Φ). As in the case of Sp (2 n, C ), the curve S has an involution σ which acts as σ η ( ) = -η . Thus, we may consider the quotient curve S = S/σ in the total space of K 2 , for which S is a double cover p : S → S. In this case the regular fibres can be seen as follows:
The regular fibres h -1 ( a ) of the SO (2 n +1 , C ) Hitchin fibration are given by Prym varieties Prym( S, S ¯ ) together with a trivialization of each M ∈ Prym( S, S ¯ ) over the zeros of a n defining S as in (1.18).
Following [Hit07], the symmetric bilinear form ( v, w ) canonically defines a skew form (Φ v, w ) on E/E 0 with values in K . Moreover, choosing a square root K 1 / 2 one can define
$$V = E / E _ { 0 } \otimes K ^ { - 1 / 2 },$$
on which the corresponding skew form is non-degenerate. The Higgs field Φ induces a transformation Φ ′ on V which has characteristic polynomial
$$\det \left ( x - \Phi ^ { \prime } \right ) = x ^ { 2 n } + a _ { 1 } x ^ { 2 n - 2 } + \dots + a _ { n - 1 } x ^ { 2 } + a _ { n }.$$
Note that this is exactly the case of Sp (2 n, C ) described in Section 1.2.3, and thus we may describe the above with a choice of a line bundle M 0 in the Prym variety Prym( S, S ). In particular, S corresponds to the smooth spectral curve of an Sp (2 n, C )-Higgs bundle.
When reconstructing the vector bundle E with an SO (2 n +1 , C ) structure from an Sp (2 n, C )-Higgs bundle ( V, Φ ) as in [Hit07, Section 4.3], there is a mod 2 invariant associated to each zero ′ of the coefficient a n of the characteristic polynomial det( η -Φ ). ′ This data comes from choosing a trivialisation of M ∈ Prym( S, S ) over the zeros of a n , and defines a covering P ′ of the Prym variety Prym( S, S ). The covering has two components corresponding to the spin and non-spin lifts of the vector bundle. The identity component of P ′ , which corresponds to the spin case, is isomorphic to the dual of the symplectic Prym variety, and this is the generic fibre of the SO (2 n +1 , C ) Hitchin map.
## 1.2.5 SO (2 n, C ) -Higgs bundles
Lastly, we consider G c = SO (2 n, C ). As in previous cases, the distinct eigenvalues of a matrix A ∈ so (2 n, C ) occur in pairs ± λ i , and thus the characteristic polynomial of A is of the form det( x -A ) = x 2 n + a x 1 2 n -2 + . . . + a n -1 x 2 + a n . In this case the coefficient a n is the square of a polynomial p n , the Pfaffian, of degree n . A basis for the invariant polynomials on the Lie algebra so (2 n, C ) is a , a 1 2 , . . . , a n -1 , p n , (the reader should refer, for example, to [A89] and references therein for further details).
Definition 1.21. An SO (2 n, C ) -Higgs bundle is a pair ( E, Φ) , for E a holomorphic vector bundle of rank 2 n with a non-degenerate symmetric bilinear form ( , ) , and Φ ∈ H 0 (Σ , End 0 ( E ) ⊗ K ) satisfying (Φ v, w ) = -( v, Φ w . )
Considering the characteristic polynomial det( η -Φ) of a Higgs bundle ( E, Φ) one obtains a 2 n -fold cover π : S → Σ whose equation is given by
$$\det ( \eta - \Phi ) = \eta ^ { 2 n } + a _ { 1 } \eta ^ { 2 n - 2 } + \dots + a _ { n - 1 } \eta ^ { 2 } + p _ { n } ^ { 2 },$$
for a i ∈ H 0 (Σ , K 2 i ) and p n ∈ H 0 (Σ , K n ). Note that this curve has always singularities, which are given by η = 0. The curve S has a natural involution σ η ( ) = -η , whose fixed points in this case are the singularities of S . The virtual genus of S can be obtained via the adjunction formula, giving g S = 1 + 4 n 2 ( g -1) .
In order to define the spectral data, one may consider its non-singular model ˆ : π ˆ S → Σ, whose genus is given by
$$g _ { \hat { S } } & \ = \ \ g _ { S } - \# \text{singularities} \\ & = \ \ 1 + 4 n ^ { 2 } ( g - 1 ) - 2 n ( g - 1 ) \\ & = \ \ 1 + 2 n ( 2 n - 1 ) ( g - 1 ).$$
As the fixed points of σ are double points, the involution extends to an involution ˆ on σ ˆ which S does not have fixed points. Considering the associated basis of invariant polynomials for each Higgs field Φ, one may define the Hitchin fibration
$$h \, \colon \, \mathcal { M } _ { S O ( 2 n, C ) } \longrightarrow \mathcal { A } _ { S O ( 2 n, C ) } \coloneqq H ^ { 0 } ( \Sigma, K ^ { n } ) \oplus \bigoplus _ { i = 1 } ^ { n - 1 } H ^ { 0 } ( \Sigma, K ^ { 2 i } ).$$
In this case the line bundle associated to an SO (2 n, C )-Higgs bundle is defined on the desingularisation ˆ of S S :
The smooth fibres h -1 ( a ) of the SO (2 n, C ) Hitchin fibration are given by Prym( ˆ S, S/σ ˆ ˆ), for ˆ the desingularisation of the curve S S associated to the regular base point a .
Starting with an SO (2 n, C )-Higgs bundle, since ˆ is smooth we obtain an eigenspace bundle S L ⊂ ker( η -Φ) inside the vector bundle E pulled back to ˆ . S In particular, this line bundle satisfies ˆ σ L ∗ ∼ = L ∗ ⊗ ( K ˆ S ⊗ π K ∗ ∗ ) -1 , thus defining a point in Prym( ˆ S, S/σ ˆ ˆ) given by
$$M \coloneqq L \otimes ( K _ { \hat { S } } \otimes \pi ^ { * } K ^ { * } ) ^ { 1 / 2 }.$$
Conversely, a Higgs bundle ( E, Φ) may be recovered from a curve S with has equation η 2 n + a η 1 2 n -2 + . . . + a n -1 η 2 + p 2 n = 0 , and a line bundle L on its desingularisation ˆ . S Note that given the sections
$$s = \eta ^ { 2 n } + a _ { 1 } \eta ^ { 2 n - 2 } + \dots + a _ { n - 1 } \eta ^ { 2 } + p _ { n } ^ { 2 }$$
for fixed p n with simple zeros, one has a linear system whose only base points are when η = 0 and p n = 0. Hence, by Bertini's theorem the generic divisor of the linear system defined by the sections s has those base points as its only singularities. Moreover, as p n is a section of K n , in general there are 2 n g ( -1) singularities which are generically ordinary double points. A generic divisor of the above linear system defines a curve S which has an involution σ η ( ) = -η whose only fixed points are the base points.
The involution σ induces an involution ˆ σ on the desingularisation ˆ S of S which has no fixed points, and thus we may consider the quotient ˆ S/σ ˆ and the corresponding Prym variety Prym( ˆ S, S/σ ˆ ˆ). Following a similar procedure as before, a line bundle M ∈ Prym( ˆ S, S/σ ˆ ˆ) induces a Higgs bundle ( E, Φ) where E is the direct image sheaf of L = M ⊗ ( K ˆ S ⊗ π K ∗ ∗ ) -1 / 2 . It is thus the Prym variety of ˆ which is a generic fibre of the corresponding Hitchin fibration. S
Exercise 5. Show that the genus g ˆ S/σ ˆ of ˆ S/σ ˆ is n (2 n -1)( g -1) .
## CHAPTER 2
## Spectral data for G -Higgs bundles
But most of all a good example is a thing of beauty. It shines and convinces. It gives insight and understanding. It provides the bedrock of belief.
Sir Michael Atiyah
Higgs bundles for real forms were first studied by N. Hitchin in [Hit87], and the results for SL (2 , R ) were generalised in [Hit92], where Hitchin studied the case of G = SL n, ( R ). Using Higgs bundles he counted the number of connected components and, in the case of split real forms, he identified a component homeomorphic to R dim G (2 g -2) and which naturally contains a copy of a Teichm¨ uller space. The aim of this Lecture is to introduce principal Higgs bundles for real forms and their corresponding spectral data as studied in [Sch13] and further developed in [HitSch, Hit13]. We begin by reviewing definitions and properties related to real forms of Lie algebras and Lie groups (see e.g., [FS03, He01, OnVi, Kn02, SW]), and then define G -Higgs bundles for real forms G of classical complex Lie groups G c . Through the approach of [Hit92], we describe these Higgs bundles as the fixed points of a certain involution on the moduli space of G c -Higgs bundles. In later sections we study G -Higgs bundles for non-compact real forms G and in each case give an overview of the corresponding spectral data when available.
## 2.1 G -Higgs bundles
A theorem by Hitchin [Hit87a] and Simpson [S88] gives the most important property of stable Higgs bundles on a compact Riemann surface Σ of genus g ≥ 2:
Theorem 2.1. If a Higgs bundle ( E, Φ) is stable and deg E = 0 , then there is a unique unitary connection A on E , compatible with the holomorphic structure, such that
$$F _ { A } + [ \Phi, \Phi ^ { * } ] = 0 \ \in \Omega ^ { 1, 1 } ( \Sigma, E n d \ E ),$$
where F A is the curvature of the connection.
Equation (2.1) together with the holomorphicity condition d ′′ A Φ = 0 give the Hitchin equations , where d ′′ A Φ is the anti-holomorphic part of the covariant derivative of Φ. Following [Hit92], the
above equations can also be considered when A is a connection on a principal G -bundle P , where G is the compact real form of a complex Lie group G c , and a →-a ∗ is the compact real structure on the Lie algebra. This motivates the study of Higgs bundles for real forms.
## 2.1.1 Real forms
Let g c be a complex Lie algebra with complex structure i , whose Lie group is G c .
Definition 2.2. A real form of g c is a real Lie algebra which satisfies g c = g ⊕ i g .
Given a real form g of g c , an element Z ∈ g c may be written as Z = X + iY for X,Y ∈ g . The mapping X + iY → X -iY is called the conjugation with respect to g .
↦
Remark 2.3. Any real form g of g c is given by the fixed points set of an antilinear involution τ on g c . In particular the conjugation with respect to g satisfies these properties.
Definition 2.4. A real form of a complex Lie group G c is an antiholomorphic Lie group automorphism τ : G c → G c of order two, i.e., τ 2 = 1 .
Every X ∈ g c defines an endomorphism ad X of g c given by ad X Y ( ) = [ X,Y ] for Y ∈ g c . For Tr the trace of a vector space endomorphism, B X,Y ( ) = Tr(ad X ad Y ) is a the bilinear form on g c × g c called the Killing form of g c .
Definition 2.5. A real Lie algebra g is called compact if the Killing form is negative definite on it. The corresponding Lie group G is a compact Lie group.
Definition 2.6. Let g be a real form of a complex simple Lie algebra g c , given by the fixed points of an antilinear involution τ . Then, if there is a Cartan subalgebra invariant under τ on which the Killing form is negative definite, the real form g is called a compact real form . Such a compact real form of g c corresponds to a compact real form G of G c ; if there is an invariant Cartan subalgebra on which the Killing form is positive definite, the form is called a split (or normal) real form . The corresponding Lie group G is the split real form of G c .
Any complex semisimple Lie algebra g c has a compact and a split real form which are unique up to conjugation via Aut C g c (e.g., for sl ( n , C ) these are su n ( ) and sl ( n, R ) respectively).
Remark 2.7. Recall that all Cartan subalgebras h of a finite dimensional Lie algebra g have the same dimension. The rank of g is defined to be this dimension, and a real form g of a complex Lie algebra g c is split if and only if the real rank of g equals the complex rank of g c .
An involution θ of a real semisimple Lie algebra g such that the symmetric bilinear form B θ ( X,Y ) = -B X,θY ( ) is positive definite is called a Cartan involution . Any real semisimple Lie algebra has a Cartan involution, and any two Cartan involutions θ , θ 1 2 of g are conjugate via an automorphism of g , i.e., there is a map ϕ in Aut g such that ϕθ ϕ 1 -1 = θ 2 . The decomposition of g into eigenspaces of a Cartan involution θ is called the Cartan decomposition of g .
Proposition 2.8 ([Kn02]) . Let g c be a complex semisimple Lie algebra, and ρ the conjugation with respect to a compact real form u of g c . Then, ρ is a Cartan involution.
Proposition 2.9 ([He01]) . Any non-compact real form g of a complex simple Lie algebra g c can be obtained from a pair ( u , θ ) , for u the compact real form of g c and θ an involution on u .
For completion, we shall recall here the construction of real forms from [He01]. Let h be the +1-eigenspaces of θ and i m the -1-eigenspace of θ acting on u , thus having
$$u & = \mathfrak { h } \oplus i \mathfrak { m },$$
Since g c = h ⊕ m ⊕ i ( h ⊕ m ) , there is a natural non-compact real form g of g c given by
$$\mathfrak { g } = \mathfrak { h } \oplus \mathfrak { m }.$$
Moreover, if a linear isomorphism θ 0 induces the decomposition as in (2.3), then θ 0 is a Cartan involution of g and h is the maximal compact subalgebra of g .
Following the notation of Proposition 2.9, let ρ be the antilinear involution defining the compact form u of a complex simple Lie algebra g c whose decomposition via an involution θ is given by equation (2.2). Moreover, let τ be an antilinear involution which defines the corresponding noncompact real form g = h ⊕ m of g c . Considering the action of the two antilinear involutions ρ and τ on g c , we may decompose the Lie algebra g c into eigenspaces
$$\mathfrak { g } ^ { c } = \mathfrak { h } ^ { ( +, + ) } \oplus \mathfrak { m } ^ { ( -, + ) } \oplus ( i \mathfrak { m } ) ^ { ( +, - ) } \oplus ( i \mathfrak { h } ) ^ { ( -, - ) },$$
where the upper index ( · , · ) represents the ± -eigenvalue of ρ and τ respectively. From the decomposition (2.4), the involution θ on the compact real form u giving a non-compact real form g of g c can be seen as acting on g c as σ := ρτ. Moreover, this induces an involution on the corresponding Lie group σ := G c → G . c
Remark 2.10. The fixed point set g σ of σ is given by g σ = h ⊕ i h , and thus it is the complexification of the maximal compact subalgebra h of g . Equivalently, the anti-invariant set under the involution σ is given by m C .
## 2.1.2 G -Higgs bundles through involutions
As mentioned previously, non-abelian Hodge theory on the compact Riemann surface Σ gives a correspondence between the moduli space of reductive representations of π 1 (Σ) in a complex Lie group G c and the moduli space of G c -Higgs bundles. The anti-holomorphic operation of conjugating by a real form τ of G c in the moduli space of representations can be seen via this correspondence as a holomorphic involution Θ of the moduli space of G c -Higgs bundles.
Following [Hit87a], in order to obtain a G -Higgs bundle, for A the connection which solves Hitchin equations (2.1), one requires the flat GL n, ( C ) connection
$$\nabla = \nabla _ { A } + \Phi + \Phi ^ { * }$$
to have holonomy in a non-compact real form G of GL n, ( C ), whose real structure is τ and Lie algebra is g . Equivalently, for a complex Lie group G c with non-compact real form G and real structure τ , one requires
$$\nabla = \nabla _ { A } + \Phi - \rho ( \Phi )$$
to have holonomy in G , where ρ is the compact real structure of G c . Since A has holonomy in the compact real form of G c , we have ρ ( ∇ A ) = ∇ A . Hence, requiring ∇ = τ ( ∇ ) is equivalent to ∇ A = τ ( ∇ A ) and Φ -ρ (Φ) = τ (Φ -ρ (Φ)) . In terms of σ = ρτ , these two equalities are given by σ ( ∇ A ) = ∇ A and Φ -ρ (Φ) = τ (Φ -ρ (Φ)) = τ (Φ) -σ (Φ) = σ ρ ( (Φ) -Φ) . Hence, ∇ has holonomy in the real form G if ∇ A is invariant under σ , and Φ anti-invariant. In terms of a G c -Higgs bundle ( P, Φ), one has that for U and V two trivialising open sets in the compact Riemann surface Σ, the
↦
↦
involution σ induces an action on the transition functions g uv : U ∩V → G c given by g uv → σ g ( uv ) , and on the Higgs field by sending Φ →-σ (Φ) .
Concretely, for G a real form of a complex semisimple lie group G c , we may construct G -Higgs bundles as follows. For H the maximal compact subgroup of G , we have seen that the Cartan decomposition of g is given by g = h ⊕ m , for h the Lie algebra of H , and m its orthogonal complement. This induces the following decomposition of the Lie algebra g c of G c in terms of the eigenspaces of the corresponding involution σ as defined before: g c = h C ⊕ m C . Note that the Lie algebras satisfy [ h h , ] ⊂ h , [ h m , ] ⊂ m , and [ m m , ] ⊂ h . Hence there is an induced isotropy representation given by Ad | H C : H C → GL ( m C ) . Then, Definition 1.10 generalises to the following (see e.g. [G95]):
Definition 2.11. A principal G -Higgs bundle is a pair ( P, Φ) where P is a holomorphic principal H C -bundle on Σ , and Φ is a holomorphic section of P × Ad m C ⊗ K .
Example 2.12. For a compact real form G , one has G = H and m = { 0 } , and thus σ is the identity and the Higgs field must vanish: a G -Higgs bundle becomes a principal G c - bundle.
In terms of involutions, following [Hit92] and recalling the previous analysis leading to Remark 2.10, we have the following:
Proposition 2.13. Let G be a real form of a complex semi-simple Lie group G c , whose real structure is τ . Then, G -Higgs bundles are given by the fixed points in M G c of the involution Θ G acting by
↦
$$\Theta _ { G } \colon \, ( P, \Phi ) \mapsto ( \sigma ( P ), - \sigma ( \Phi ) ),$$
where σ = ρτ , for ρ the compact real form of G c .
Similarly to the case of G c -Higgs bundles, there is a notion of stability, semi stability and polystability for G -Higgs bundles. Following [BGG03, Section 3] and [BGG06, Section 2.1], one can see that the polystability of a G -Higgs bundle for G ⊂ GL n, ( C ) is equivalent to the polystability of the corresponding GL n, ( C )-Higgs bundle. However, a G -Higgs bundle can be stable as a G -Higgs bundle but not as a GL n, ( C )-Higgs bundle. We shall denote by M G the moduli space of polystable G -Higgs bundles on Σ.
Exercise 6 ((*)) . Considering the notion of 'strong real form' from [Ad05], describe the corresponding Higgs bundles and give a definition of Θ G for which one does not have the problem described in the above paragraph.
One should note that a fixed point of Θ G in M G c gives a representation of π 1 (Σ) into the real form G up to the equivalence of conjugation by the normalizer of G in G c . This may be bigger than G itself, and thus two distinct classes in M G could be isomorphic in M G c via a complex map. Hence, although there is a map from M G to the fixed point subvarieties in M G c , this might not be an embedding. The reader may refer to [GGM09] for the Hitchin-Kobayashi type correspondence for real forms.
Remark 2.14. A description of the above phenomena in the case of rank 2 Higgs bundles is given in [Sch11, Section 10], where one can see how the SL (2 , R ) -Higgs bundles which have different topological invariants lie in the same connected component as SL (2 , C ) -Higgs bundles.
Remark 2.15. The point of view of Proposition 2.13, which is considered throughout [Sch13], fits into a more global picture where Θ G is one of three natural involutions acting on the moduli space of Higgs bundles [BaSch13a, BaSch13b], giving three families of ( B,A,A ) , ( A,B,A ) and ( A,A,B ) branes in M G c as the fixed point sets. One should note that the fixed point sets of these involutions are of great importance when studying the relation of Langlands duality with Higgs bundles, as initiated in [HT03, KW07] and [Hit07].
## 2.2 Spectral data for G -Higgs bundles
↦
As mentioned in the first Lecture, the moduli spaces M G c have a natural symplectic structure, which we denoted by ω . Moreover, following [Hit87a], the involutions Θ G send ω →-ω . Thus, at a smooth point, the fixed point set must be Lagrangian and so the expected dimension of M G is half the dimension of M G c . In order to describe the spectral data for real G -Higgs bundles, one considers the moduli space M G sitting inside M G c as fixed points of Θ G in the Hitchin base A G c and the corresponding preserved fibres.
Remark 2.16. Let g c be one of the classical Lie algebras sl ( n, C ) , so (2 n +1 , C ) , and sp (2 n, C ) . Then, for π : g c → gl ( V ) a representation of g c , the ring of invariant polynomials of g c is generated by Tr ( π X ( ) ) i , for i ∈ N and X ∈ g c .
By considering Cartan's classification of classical Lie algebras, we shall now describe G -Higgs bundles and their spectral data for non-compact real forms of a classical complex Lie algebra g c . For I n the unit matrix of order n , we denote by I p,q , J n and K p,q the matrices
$$I _ { p, q } = \left ( \begin{array} { c c c } -$$
Following Proposition 2.9, we shall study each complex Lie algebra g c and compact form u with different involutions θ which give decompositions u = h ⊕ i m . Then the corresponding natural noncompact real form is g = h ⊕ m , and to make sense of Proposition 2.13 we consider the following Lie algebras, Lie groups, real forms, and holomorphic and anti holomorphic involutions:
Table 2.1: Compact forms u of classical Lie algebras
| g c | Lie group G c | Split form | Compact form u | anti-involution ρ fixing u | dim u |
|-------|------------------|---------------|------------------|------------------------------|-------------|
| a n | SL ( n, C ) | sl ( n, R ) | su ( n ) | ρ ( X ) = - X t | n ( n +1) |
| b n | SO (2 n +1 , C ) | so ( n, n +1) | so (2 n +1) | ρ ( X ) = X | n (2 n +1) |
| c n | Sp (2 n, C ) | sp (2 n, R ) | sp ( n ) | ρ ( X ) = J n XJ - 1 n | n (2 n +1) |
| d n | SO (2 n, C ) | so ( n, n ) | so (2 n ) | ρ ( X ) = X | n (2 n - 1) |
Table 2.2: Non-compact forms G of classical Lie algebras G c
| g c | Real form G | τ fixing G | Involution θ on u | σ = ρτ |
|-------|-------------------------------------|--------------------------------------------------------------------|------------------------------------------------------------|--------------------------------------------------------------------|
| a n | SL ( n, R ) SU ∗ (2 m ) SU ( p, q ) | ρ ( X ) = - X t τ ( X ) = J m XJ - 1 m τ ( X ) = - I p,q X t I p,q | θ ( X ) = X. θ ( X ) = J m XJ - 1 m θ ( X ) = I p,q XI p,q | σ ( X ) = - X t σ ( X ) = - J m X t J - 1 m σ ( X ) = I p,q XI p,q |
| b n | SO ( p, q ) | τ ( X ) = I p,q XI p,q . | θ ( X ) = I p,q XI p,q | σ ( X ) = I p,q XI p,q |
| c n | Sp (2 n, R ) Sp (2 p, 2 q ) | τ ( X ) = X τ ( X ) = - K p,q X ∗ K p,q . | θ ( X ) = X θ ( X ) = K p,q XK p,q , | σ ( X ) = J n XJ - 1 n σ ( X ) = K p,q XK p,q |
| d n | SO ( p, q ) SO ∗ (2 m ) | τ ( X ) = I p,q XI p,q . τ ( X ) = J m XJ - 1 m . | θ ( X ) = I p,q XI p,q θ ( X ) = J m XJ - 1 m | σ ( X ) = I p,q XI p,q σ ( X ) = J m XJ - 1 m |
In the case of split real forms, following the methods of [Hit92] one obtains a description of real Higgs bundles which we shall use in subsequent sections:
Theorem 2.17 ([Sch13]) . For G the split real form of G c , the fixed points of Θ G in the smooth fibres of the Hitchin fibration for G c -Higgs bundles are given by points of order two.
## 2.2.1 G = SL n, ( R ) -Higgs bundles
Higgs bundles for SL n, ( R ) were first considered in [Hit92], where Hitchin studied a copy of Teichm¨ller space inside the moduli space of Higgs bundles for split real forms. u Following Definition 2.11, an SL n, ( R )-Higgs bundle is a pair ( E, Φ) where E is a rank n orthogonal vector bundle and Φ : E → E ⊗ K is a symmetric and traceless holomorphic map.
Proposition 2.18. SL n, ( R ) -Higgs bundles are given by the fixed points of
↦
$$\Theta _ { S L ( n, \mathbb { R } ) } \colon ($$
in M SL n, ( C ) corresponding to automorphisms f : E → E ∗ giving a symmetric form on E .
Exercise 7. Find the decomposition of u = su ( n ) induced by the corresponding θ in Table 2.2, and use this to deduce Proposition 2.18.
Recalling that the trace is invariant under transposition, one has that the ring of invariant polynomials of g c = sl ( n, C ) is acted on trivially by the involution -σ , and thus the Hitchin base is preserved by Θ SL n, ( R ) . In order to find the spectral data for SL n, ( R )-Higgs bundles, following Theorem 2.17 we look at elements of order two in the fibres of the Hitchin fibration for SL n, ( C )-Higgs bundles:
Over a smooth point in the Hitchin base A SL n, ( C ) , Higgs bundles for SL n, ( R ) correspond to line bundles L ∈ Prym( S, Σ) such that L 2 ∼ O = .
↦
In the case of n = 2, the SL (2 , C )-spectral curve S given as in (1.7) has a natural involution σ : η → -η and Prym( S, Σ) = { L ∈ Jac( S ) : σ L ∗ ∼ = L ∗ } . Hence, points in the smooth fibres corresponding to SL (2 , R )-Higgs bundles are given by line bundles L ∈ Jac( S ) such that σ L ∗ ∼ = L .
Exercise 8. Let L ∈ Prym ( S, Σ) be a line bundle of order two. Then, its direct image is a rank 2 bundle on Σ which decomposes into the sum of two line bundles V ⊕ V ∗ . How can the Lefschetz fixed point formula from [AB68] be used to relate the degree of V and the action of σ on L in the sprit of [Sch13b]?
The topological invariant associated to SL n, ( R )-Higgs bundles is the characteristic class ω 2 ∈ Z 2 which is the obstruction to lifting the orthogonal bundle to a spin bundle, and its study was carried through in [Hit13, Section 5].
Exercise 9. For n = 2 , use the approach of [Hit13, Proposition 3] to relate ω 2 to the invariants in Exercise 8.
The spectral data of SL n, ( R )-Higgs bundles gives a finite cover of the smooth locus of the Hitchin fibration. For n = 2, an explicit description of the monodromy action whose orbits are the connected components of M SL (2 , R ) is given in [Sch11].
Exercise 10 ((*)) . How can the methods in [Sch11] be extended to study monodromy for SL n, ( R ) -Higgs bundles for n ≥ 3 ?
## 2.2.2 SU ∗ (2 m ) -Higgs bundles
The group SU ∗ (2 m ) is the subgroup of SL (2 m, C ) which commutes with an antilinear automorphism J of C 2 m such that J 2 = -1. At the level of the Lie algebras we have that the involution θ decomposes u = h ⊕ i m where h = sp ( m . ) The induced non-compact real form g = h ⊕ m is
$$\begin{array} {$$
Definition 2.19. An SU ∗ (2 m ) Higgs bundle on Σ is given by
$$( E, \Phi ) \text{ for } \left \{ \begin{array} {$$
Proposition 2.20. Isomorphism classes of SU ∗ (2 m ) -Higgs bundles are given by fixed points of the involution
↦
$$\Theta _ { S U ^ { * } } \, \colon ( E, \Ph$$
on SL (2 m, C ) -Higgs bundles corresponding to pairs which have an automorphism f : E → E ∗ endowing it with a symplectic structure, and which trivialises its determinant bundle.
As the trace is invariant under conjugation and transposition, one has that the involution -σ X ( ) = J m X J t -1 m acts trivially on the ring of invariant polynomials of sl (2 m, C ), and thus preserves the Hitchin base. The spectral data associated to SU ∗ (2 m )-Higgs bundles ( E, Φ) was studied in [HitSch, Section 3], and we shall describe here its main features.
The characteristic polynomial of an SU ∗ (2 m )-Higgs bundle ( E, Φ) can be seen to be the square of a Pfaffian, det( η -Φ) = p η ( ) 2 and thus all fixed points of Θ SU ∗ lie over singular points of the SL (2 m, C ) Hitchin fibration. With a slight abuse of notation, we denote by S the spectral curve in the total space of K defined by
$$p ( \eta ) = \eta ^ { m } + a _ { 2 } \eta ^ { m - 2 } + \cdots + a _ { m } = 0$$
where the coefficients a i ∈ H 0 (Σ , K i ). It is a ramified m -fold cover of Σ whose ramification points are the zeros of a m . As in the case of complex groups, we interpret p η ( ) = 0 as the vanishing of a section of π K ∗ m over the total space of the canonical bundle π : K → Σ, where η is the tautological section of π K ∗ , and Bertini's theorem assures us that for generic a i the curve is nonsingular.
## Exercise 11. What is the genus g S of S ?
On the spectral curve S , the cokernel of ( η -Φ) is a rank two holomorphic vector bundle V on S. Then, following [BNR] (and using p (Φ) = 0 instead of the Cayley-Hamilton theorem), we can identify E with the direct image π V ∗ and Φ as the direct image of η : V → V ⊗ π K ∗ .
Exercise 12. Given a proper Φ -subbundle F ⊂ E , the characteristic polynomial of Φ | F divides the one of Φ . Use this together with Grothendieck-Riemann-Roch to show that semi-stability of V implies semi-stability of ( E, Φ) .
Following [HitSch, Proposition 1] one obtains a description of the spectral data:
The fixed point set of Θ SU ∗ (2 m ) in a smooth fibre of the SL (2 m, C )-Hitchin fibration is the moduli space of semi-stable rank 2 vector bundles on S with fixed determinant π K ∗ m -1 .
Exercise 13. Follow the approach of SL n, ( C ) -Higgs bundles to note that by fixing the determinant of V one obtains a trivialization of the determinant of π V ∗ on Σ .
## 2.2.3 SU p, q ( ) -Higgs bundles
Definition 2.21. An SU p,q ( ) -Higgs bundle over Σ is a pair ( E, Φ) where E = W 1 ⊕ W 2 for W ,W 1 2 vector bundles over Σ of rank p and q such that Λ p W 1 ∼ = Λ q W ∗ 2 , and the Higgs field Φ is given by Φ = ( 0 β γ 0 ) , for β : W 2 → W 1 ⊗ K and γ : W 1 → W 2 ⊗ K .
Exercise 14. Find the decomposition u = h ⊕ i m via the action of θ in Table 2.2 and deduce that θρ is the anti-holomorphic involution fixing the non-compact real form u ( p, q ) .
↦
Proposition 2.22. SU p,q ( ) -Higgs bundles are fixed points of Θ SU p,q ( ) : ( E, Φ) → ( E, -Φ) on SL p ( + q, C ) -Higgs bundles corresponding to bundles E which have an automorphism conjugate to I p,q sending Φ to -Φ , and whose ± 1 eigenspaces have dimensions p and q .
/negationslash
↦
The involution -σ acts trivially on the polynomials of even degree. Whilst the spectral data is not known for p = q , in the case of p = q it has been described in [Sch13, Chapter 6] and [Sch13b] by looking at U p, p ( )-Higgs bundles ( W 1 ⊕ W , 2 Φ), which when satisfying Λ p W 1 ∼ = Λ q W ∗ 2 correspond to SU p,p ( )-Higgs bundles. In this case, the characteristic polynomial defines a spectral curve π : S → Σ through the equation det( η -Φ) = η 2 p + a η 2 2 p -2 + . . . + a 2 p -2 η 2 + a 2 p = 0 , where η is the tautological section of π K ∗ and a i ∈ H 0 (Σ , K i ). This is a 2 p -fold cover of Σ, ramified over the 4 p g ( -1) zeros of a 2 p , and has a natural involution η →-η which has as fixed points the ramification points of the cover, and which by abuse of notation, we shall call σ .
The involution σ plays an important role when constructing the spectral data as described in [Sch13b, Proposition 2.1]. A line bundle L on S which defines a classical Higgs bundle induces a U p, p ( )-Higgs bundle if and only if σ L ∗ ∼ = L . In this case, at a fixed point x ∈ S of the involution, there is a linear action of σ on the fibre L x given by scalar multiplication of ± 1. This description of the spectral data can be then seen in terms of Jacobians through [Sch13b, Theorem 3.2]:
The fixed point set of Θ U p,p ( ) in a smooth fibre of the classical Hitchin fibration can be seen in terms of pull backs of Jac( S/σ ).
The topological invariants associated to U p, p ( )-Higgs bundles ( W 1 ⊕ W , 2 Φ) are the degrees deg( W 1 ) and deg( W 2 ), and can be seen in terms of the degree of the line bundle L on S and the number of ramification points of S over which the linear action of σ on the fibre of L is -1.
Exercise 15. Use the Lefschetz fixed point formula in [AB68, Theorem 4.12] to see that the parity of the degree of L and the number of points over which σ acts as -1 needs to be the same.
Exercise 16. Following [BGG03], the Toledo invariant τ ( deg ( W , 1 ) deg ( W 2 )) associated to U p, p ( ) -Higgs bundles is defined as τ ( deg ( W , 1 ) deg ( W 2 )) := deg ( W 1 ) -deg ( W 2 ) . Use Exercise 15 to express the invariant in terms of fixed points of σ and obtain natural bounds.
In the case of SU p,p ( )-Higgs bundles, for maximal Toledo invariant the fixed point set of Θ SU p,p ( ) in a smooth fibre of the SL (2 p, C )-Hitchin fibration is given by a covering of Prym( S/σ, Σ), the Prym variety of the quotient curve S/σ . For SU p,p ( + 1)-Higgs bundles, the Cameral data can be used to obtain the spectral data as seen in [P13].
Exercise 17 ((*)) . How can the methods from [Sch13b] together with the approach of [Xia03] be used to obtain the spectral data for SU p, ( 1) -Higgs bundles?
Remark 2.23. It is interesting to note that the spectral data for SU p,p ( ) -Higgs bundles from [Sch13b] is used to conjecture a dual space to M SU p,p ( ) through Langlands duality in [Hit13].
## 2.2.4 SO p,q ( ) -Higgs bundles
In this case, if p + q is even, g is a split real form if and only if p = q ; if p + q is odd, g is a split real form if and only if p = q + 1. Whilst we shall give some details on the construction of SO p,q ( )-Higgs bundles, for a more detailed description of the approach needed to understand groups with signature the reader should refer to the following section on Sp (2 p, 2 )-Higgs bundles. q
Proposition 2.24. SO p,q ( ) Higgs bundles are fixed points of
↦
$$\Theta _ { S O ( p, q ) } \colon \left ( E, \Phi \right ) \mapsto \left ( E, - \Phi \right )$$
on the moduli space of SO p ( + q, C ) corresponding to vector bundles E which have an automorphism f conjugate to I p,q sending Φ to -Φ and whose ± 1 eigenspaces have dimensions p and q .
Exercise 18. The associated involution θ from Table 2.2 decomposes u = h ⊕ i m . Give an explicit description of m and h and of the real form g = h ⊕ m .
The vector space V associated to the standard representation of h C can be decomposed into V = V p ⊕ V q , for V p and V q complex vector spaces of dimension p and q respectively, with orthogonal structures. The maximal compact subalgebra of so ( p, q ) is h = so ( p ) × so ( q ) and the Cartan decomposition of so ( p + q, C ) is given by ( so ( p, C ) ⊕ so ( q, C )) ⊕ m C , for
$$\mathfrak { m } = \left \{ \left ( \begin{array} { c c } 0 & X _ { 2 } \\ X _ { 2 } ^ { t } & 0 \end{array} \right ) \right | X _ { 2 } \text{ real } p \times q \text{ matrix} \right \}. \\ \text{. } A n \text{ } S O ( p, q ) \text{ Higgs bundle is a pair } ( E, \Phi ) \text{ where } E = V _ { p } \oplus \text{ } \text{ } \text{ }$$
∣ Definition 2.25. An SO p,q ( ) Higgs bundle is a pair ( E, Φ) where E = V p ⊕ V q for V p and V q complex vector spaces of dimension p and q respectively, with orthogonal structures, and the Higgs field is a section in H 0 (Σ ( , Hom ( V , V q p ) ⊕ Hom ( V , V p q )) ⊗ K ) given by
$$\Phi = \left ( \begin{array} { c c } 0 & \beta \\ \gamma & 0 \end{array} \right ) \quad \text{for } \gamma \equiv - \beta ^ { T }, \ a n d \, \beta ^ { T } t h e \ o r t h o g o n a l \ t r a n s p o s e \ o f \, \beta.$$
Since the ring of invariant polynomials of g c = so (2 m +1 , C ) is generated by Tr( X i ) for X ∈ g c , for p + q = 2 m +1 one has that the induced action of the involution Θ SO p,q ( ) is trivial on the ring of invariant polynomials of the Lie algebra so (2 m +1 , C ), i.e., when p and q have different parity.
Exercise 19. In the case of so (2 m, C ) , for 2 m = p + q , the ring of invariant polynomials is generated by Tr ( X i ) for X ∈ g c and i < 2 m , together with the Pfaffian p m , which is of degree m . Under which conditions on p and q is the induced action of Θ SO p,q ( ) trivial on the ring of invariant polynomials?
The spectral data for SO p,q ( )-Higgs bundles when p = q or p = q +1 can be seen through Theorem 2.17 from [Sch13] as points of order two in the smooth fibres of the SO p ( + q, C )-Hitchin fibration.
In both cases a key ingredient is the double cover p : S → S/σ given by the spectral curve (the desingularised curve in the case of SO (2 n, C )) over the quotient curve, which through K -theoretic methods allows one to express the topological invariants involved in terms of the action of σ [BaSch14].
## 2.2.5 G = SO ∗ (2 m )
The action of θ in Table 2.2 decomposes u = h ⊕ i m for h = u ( m ) = ∼ so (2 m ) ∩ sp ( m , ) and
$$i \mathfrak { m } \ = \ \left \{ \, \left ( \begin{array} { c c } X _ { 1 } & X _ { 2 } \\ X _ { 2 } & - X _ { 1 } \end{array} \, \right ) \right | X _ { 1 }, \ X _ { 2 } \in \mathfrak { s o } ( m ) \right \}, & ( 2. 8 ) \\ \lceil \, \text{ced non-compact real form } \mathfrak { g } = \mathfrak { h } \oplus \mathfrak { m } \, \text{is}$$
∣ and the induced non-compact real form g = h ⊕ m is
$$\mathfrak { g } = \mathfrak { s o } ^ { * } ( 2 m ) = \left \{ \left ( \begin{array} { c c c } Z _ { 1 } & Z _ { 2 } \\ - \overline { Z } _ { 2 } & \overline { Z } _ { 1 } \end{array} \right ) \ \left | \begin{array} { c c c } Z _ { 1 }, Z _ { 2 } \, m \times m \, \text{ complex matrices} \\ Z _ { 1 } \, \text{ skew symmetric, } Z _ { 2 } \, \text{ Hermitian} \end{array} \right \}.$$
Proposition 2.26. SO ∗ (2 m ) -Higgs bundles are fixed points of the involution
↦
$$\Theta _ { S O ^ { * } ( 2 m ) } \colon \left ( E, \Phi \right ) \mapsto \left ( E, - \Phi \right )$$
on the moduli space of SO (2 m, C ) -Higgs bundles corresponding to vector bundles E which have an orthogonal automorphism f conjugate to J m , sending Φ to -Φ and which squares to -1 , equipping E with a symplectic structure.
The vector space associated to the standard representation of h C has an orthogonal and symplectic structure J . Since J -1 = J t and J 2 = -1, the vector space may be expressed in terms of the ± i eigenspaces of J as V ⊕ V ∗ , for V a rank m vector space. Thus, we have the following definition:
Definition 2.27. An SO ∗ (2 m ) -Higgs bundle is given by a pair ( E, Φ) where E = V ⊕ V ∗ for V a rank m holomorphic vector bundle, and where the Higgs field Φ is given by
$$\Phi = \left ( \begin{array} { c c } 0 & \beta \\ \gamma & 0 \end{array} \right ) \quad \text{for} \quad \begin{cases} \gamma \, \colon \, V \to V ^ { * } \otimes K \, \ s a t i s f y i n g \, \gamma = - \gamma ^ { t } \\ \beta \, \colon \, V ^ { * } \to V \otimes K \, \ s a t i s f y i n g \, \beta = - \beta ^ { t } \end{cases}.$$
As in the previous case, the involution induced action of Θ SO ∗ (2 m ) is trivial on the ring of invariant polynomials of g c . The spectral data for these Higgs bundles is studied in [HitSch], and we shall give a short description bellow.
In order to understand the associated spectral data, one notes that SO ∗ (2 m )-Higgs bundles ( E, Φ) are in fact SU ∗ (2 m )-Higgs bundle with extra conditions. Hence, one may define a natural m cover of the Riemann surface π : S → Σ by taking
$$\sqrt { \text{char} ( \Phi ) } = \eta ^ { m } + a _ { 2 } \eta ^ { 2 m - 2 } + \dots + a _ { m }, \\ \dots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots.$$
and a rank 2 vector bundle V on S whose direct image on Σ gives E . Since in this case the equation of the spectral curve only has even coefficients, there is a natural involution σ : η →-η and one may consider the induced action of σ on V and on its determinant bundle. In particular, from [HitSch, Proposition 2] the vector bundle V gives an SO ∗ (2 m )-Higgs bundle if and only if it is preserved by the involution and the induced action on it satisfies some conditions:
The fixed point set of Θ SO ∗ (2 m ) in a smooth fibre of the SO (2 m, C )-Hitchin fibration is given by the moduli space of semi-stable rank 2 vector bundles V on S with fixed determinant π K ∗ 2 m -1 , whose induced action by σ on the determinant bundle is trivial.
Exercise 20. The relative duality theorem gives
$$( \pi _ { * } ( V ) ) ^ { * } \cong \pi _ { * } ( V ^ { * } \otimes K _ { S } ) \otimes K ^ { * }.$$
Use this to see that in order to have E ∼ = E ∗ through a skew form, the action of σ needs to be trivial on the determinant bundle of V for π V ∗ = E .
Exercise 21 ((*)) . Describe how the vector bundles of rank 2 from [HitSch] appear in the description of the connected components of M SO ∗ (2 m ) in [BGG13].
## 2.2.6 G = Sp (2 n, R ) -Higgs bundles
In this section and the one which follows we consider the non-compact real forms of the complex Lie group Sp (2 n, C ). For this, recall that the symplectic Lie algebra sp (2 n, C ) is given by the set of 2 n × 2 n complex matrices X that satisfy J X n + X J t n = 0 or equivalently, X = -J -1 n X J . t n
Let u be the compact real form u = sp ( n ) and θ X ( ) = X = J XJ n -1 n . The Lie algebra sp ( n ) is given by the quaternionic skew-Hermitian matrices; that is, the set of n × n quaternionic matrices X which satisfy X = -X t . The compact form may be decomposed as u = h ⊕ i m , for h = u ( n ) = ∼ so (2 n ) ∩ sp ( n , ) which leads to the split real form g = h ⊕ m defined by
$$\mathfrak { g } = \mathfrak { s p } ( 2 n, \mathbb { R } ) = \left \{ \left ( \begin{array} { c c } X _ { 1 } & X _ { 2 } \\ X _ { 3 } & - X _ { 1 } ^ { t } \end{array} \right ) \ \left | \ X _ { i } \ \text{real} \ n \times n \ \text{matrices} \ \right \}. \\ \text{friction} \, 2. 2 8. \ A n \ S p ( 2 n, \mathbb { R } ) - H i q q s \ b u n d l e \ i s \ q i v e n \ b y \ a \ p a i r \left ( E, \Phi \right ) \ \text{where} \ E = V \oplus V ^ { * } \ f o \cdot$$
Definition 2.28. An Sp (2 n, R ) -Higgs bundle is given by a pair ( E, Φ) where E = V ⊕ V ∗ for V a rank n holomorphic vector bundle, and for Φ the Higgs field given by
$$\Phi = \left ( \begin{array} { c c } 0 & \beta \\ \gamma & 0 \end{array} \right ) \quad \text{for} \ \left \{ \begin{array} { c c } \gamma \colon V \to V ^ { * } \otimes K \ s a t i s f y i n g \ \gamma = \gamma ^ { t } \\ \beta \colon V ^ { * } \to V \otimes K \ s a t i s f y i n g \ \beta = \beta ^ { t } \ \. \end{array}$$
Proposition 2.29. Sp (2 n, C ) Higgs bundles, and Sp (2 n, R ) -Higgs bundles are given by the fixed points of
↦
$$\Theta _ { S p ( 2 n, \mathbb { R } ) } \colon \left ( E, \Phi \right ) \mapsto \left ( E, - \Phi \right )$$
on Sp (2 n, C ) -Higgs bundles corresponding to vector bundles E which have a symplectic isomorphism sending Φ to -Φ , and whose square is the identity, endowing E with an orthogonal structure.
The invariant polynomials of g c are of even degree, and hence the involution -σ acts trivially on them, making Θ Sp (2 n, R ) preserve the whole Hitchin base A Sp (2 n, C ) . In the case of rank 4 Higgs bundles, the spectral data was first consider in P. Gothen's thesis [G95], and through Theorem 2.17 and the spectral data for complex Higgs bundles one has the following:
The fixed points of Θ Sp (2 n, R ) in the smooth fibres of the Sp (2 n, C )-Hitchin fibration are given by line bundles L ∈ Prym( S, S/σ ) such that L 2 ∼ O = .
In particular, since S is a ramified double cover of S/σ , one has that L ∈ Prym( S, S/σ ) if and only if σ L ∗ ∼ = L ∗ . Hence, by considering points of order two one has that σ L ∗ ∼ = L and thus there is a natural induced action of σ on the line bundle L . The topological invariants associated to these Higgs bundles were studied in [Hit13] through the natural action of σ .
Exercise 22. Compare the calculations in [G95, Section 4.1] which lead to Milnor-Wood type inequalities for Sp (2 n, R ) -Higgs bundles, with the inequalities one obtains by using the involution σ as in [Hit13].
Exercise 23 ((*)) . Express the invariants from [G95, Section 4.2.4-4.2.6] in terms of different choices of the natural involution σ on S as well as in terms of the action of a second natural involution appearing in some situations on S/σ .
## 2.2.7 G = Sp (2 p, 2 ) q -Higgs bundles
The induced non-compact real form g = h ⊕ m is
$$\text{sp} ( 2 p, 2 q ) = \left \{ \left ( \begin{array} { c c c c } Z _ { 1 1 } & Z _ { 1 2 } & Z _ { 1 3 } & Z _ { 1 4 } \\ \overline { Z } _ { 1 2 } ^ { t } & Z _ { 2 2 } & Z _ { 1 4 } ^ { t } & Z _ { 2 4 } \\ - Z _ { 1 3 } ^ { t } & Z _ { 1 4 } ^ { t } & - Z _ { 1 2 } ^ { t } \\ \overline { Z } _ { 1 4 } ^ { t } & - Z _ { 2 4 } ^ { t } & - Z _ { 1 2 } ^ { t } & Z _ { 2 2 } \end{array} \right ) \ \left | \begin{array} { c } Z _ { i, j } & \text{complex matrices,} \\ Z _ { 1 1 }, \, Z _ { 1 3 } \text{ order } p, \\ Z _ { 1 2 }, \, Z _ { 1 4 } \text{ } p \times q \text{ matrices,} \\ Z _ { 1 3 }, \, Z _ { 2 4 } \text{ symmetric.} \end{array} \right \}. \\ \text{series 24. Show that $\mathbb{M}^{C}$ can be expressed as subset of certain off-diagonal matrices.}$$
∣ Exercise 24. Show that m C can be expressed as subset of certain off-diagonal matrices.
Proposition 2.30. Sp (2 p, 2 ) q -Higgs bundles are given by the fixed points of
↦
$$\Theta _ { S p ( 2 p, 2 q ) } \colon \left ( E, \Phi \right ) \mapsto \left ( E, - \Phi ^ { T } \right )$$
on the moduli space of Sp (2 p +2 q, C ) -Higgs bundles corresponding to symplectic vector bundles E which have an endomorphism f : E → E conjugate to ˜ K p,q , sending Φ to the symplectic transpose -Φ T , and whose ± 1 eigenspaces are of dimension 2 p and 2 q .
Definition 2.31. An Sp (2 p, 2 ) q -Higgs bundle is given by a pair ( E, Φ) for E = V 2 p ⊕ V 2 q is a direct sum of symplectic vector spaces of rank 2 p and 2 q , and
$$\Phi = \left ( \begin{array} { c c } 0 & - \gamma ^ { T } \\ \gamma & 0 \end{array} \right ) \text{ for } \left \{ \begin{array} { c c } \gamma & \colon V _ { 2 p } \to V _ { 2 q } \otimes K \\ - \gamma ^ { T } & \colon V _ { 2 q } \to V _ { 2 p } \otimes K \end{array}, \text{ for } \gamma ^ { T } \text{the symmlectic transpose of } \gamma.$$
As the trace is invariant under conjugation and transposition, the induced action of Θ Sp (2 p, 2 ) q is trivial on the ring of invariant polynomials of g c = sp (2( p + ) q , C ). In the case of p = , one can q see that Sp (2 p, 2 p )-Higgs bundles are a particular case of SU ∗ (2 p )-Higgs bundles, and thus one needs to understand which extra conditions to the spectral data for SU ∗ (2 p )-Higgs bundles needs to be added in order to have the Higgs bundles for the symplectic real form.
From the previous section, when p = q the corresponding spectral curve is a 2 p -fold cover of the Riemann surface Σ whose equation is given by the square root of the characteristic polynomial of the Higgs field. Moreover, it has a natural involution σ whose action determines the associated spectral data. More precisely, the following is shown in [Sch13, Chapter 7], and reinterpreted in [HitSch, Section 5]:
The fixed point set of Θ Sp (2 p, 2 p ) in a smooth fibre of the Sp (4 p, C )-Hitchin fibration is given by the moduli space of semi-stable rank 2 vector bundles V on S with fixed determinant π K ∗ 2 p -1 , whose induced action by σ on Λ 2 V is -1.
Since the action on Λ 2 V is -1, the involution σ : S → S acts with different eigenvalues ± 1 on the fibres of V over the ramification points of S , and thus through [AG06], the spectral data relates to the moduli space of admissible parabolic rank 2 bundles on S/σ as seen in [Sch13, Section 8.3].
Exercise 25 ((*)) . Nonabelianization of the fixed point set of Θ Sp (2 p, 2 p ) can also be seen through Cameral covers, as shown in [P13]. Realise the action of σ in terms of Cameral covers.
.
## Bibliography
[Ad05] J. Adams (2005), Strong real forms and the Kac classification , At. of Lie Groups and Rep.
- [AG06] J.E. Andersen, J. Grove (2006), Automorphism fixed points in the moduli space of semi-stable bundles , Q J Math 57 1 1-35. arXiv:math/0205099.
[Ap09] M.A. Arroyo (2009), The Geometry of SO p,q ( ) -Higgs Bundles , Ph.D., Univ. de Salamanca.
- [A89] H. Aslaksen (1989), SO (2) invariants of a set of 2 × 2 matrices MATH. SCAND. 65 , 59-66.
- [AB68] M.F. Atiyah, R. Bott (1968), A Lefschetz Fixed Point Formula for Elliptic Complexes: II. Applications , Ann. Math. 88, 3 , 451-491.
[BaSch13a] D. Baraglia, L.P. Schaposnik (2014), Higgs bundles and ( A,B,A ) -branes , Communications in Mathematical Physics. arXiv:math/1305.4638.
[BaSch13b] D. Baraglia, L.P. Schaposnik (2014), Real structures on moduli spaces of Higgs bundles , Advances in Theoretical and Mathematical Physics. arXiv:math/1309.1195.
[BaSch14] D. Baraglia, L.P. Schaposnik (2014), Higgs bundles and split real forms , preprint.
[BNR] A. Beauville, M.S. Narasimhan and S. Ramanan (1989), Spectral curves and the generalised theta divisor , J. reine angew. Math. 398, 169-179.
- [BiGo08] I. Biswas, T.L. G´mez (2008), o Connections and Higgs fields on a principal bundle , Ann. Global Anal. Geom. 33, 1 , 19-46.
- [BGG03] S. Bradlow, O. Garc´ ıa-Prada, P.G. Gothen (2003), Surface group representations and U p, q ( ) -bundles , J. Diff. Geom. 64 , 111-170.
- [BGG06] S. Bradlow, O. Garc´ ıa-Prada, P.G. Gothen (2006), Maximal surface group representations in isometry groups of classical Hermitian symmetric spaces , Geom. Ded. 122 , 185-213, arXiv:0511415.
[BGG13] S. Bradlow, O. Garc´ ıa-Prada, P.G. Gothen (2013), Higgs bundles for the non-compact dual of the special orthogonal group, arXiv:math/1303.1058.
[Cor88] K. Corlette (1988), Flat G -bundles with canonical metrics , J. Differential Geom. 28, 361-382.
- [Do95] R. Donagi (1995), Spectral Covers , MSRI Series, 28 arXiv:alg-geom/9505009.
- [DoMa96] R. Donagi, E. Markman (1993), Spectral covers, algebraically completely integrable, Hamiltonian systems, and moduli of bundles , Integrable Systems and Quantum Groups.
- [DoPa12] R. Donagi, T. Pantev (2012), Langlands duality for Hitchin systems , Invent. math. arXiv:math/0604617.
- [D87] S.K. Donaldson (1987), Twisted harmonic maps and the self-duality equations , Proc. London Math. Soc. 55, 3, 127-131.
- [FS03] J. Fuchs, C. Schweigert (2003), Symmetries, Lie Algebras and Representations: A Graduate Course for Physicists , Cambridge Monographs on Mathematical Physics, Camb. Univ. Press.
- [GGM09] O. Garc´ ıa-Prada, P.B. Gothen, I. Mundet i Riera (2009), The Hitchin-Kobayashi correspondence, Higgs pairs and surface group representations , arXiv:math/0909.4487
[G95] P.B. Gothen (1995), The topology of Higgs bundle moduli spaces , Ph.D. thesis, Mathematics Institute, University of Warwick.
- [GH11] P. Griffiths, J. Harris (2011), Principles of Algebraic Geometry ,
Wiley Class. Library 52.
- [HT03] T. Hausel, and M. Thaddeus (2003) , Mirror symmetry, Langlands duality, and the Hitchin system , Invent. Math. 153 197-229. arXiv:math/0205236.
[Hit87] N.J. Hitchin (1987), The self-duality equations on a Riemann surface , Proc. LMS 55, 3, 59-126. [Hit87a] N.J. Hitchin (1987), Stable bundles and integrable systems , Duke Math. J. 54, 1 , 91-114.
- [Hit92] N.J. Hitchin (1992), Lie Groups and Teichm¨ller Space u , Topology 31, 3 , 449-473.
- [Hit07] N.J. Hitchin (2007), Langlands duality and G 2 spectral curves , Q.J. Math., 58 , 319-344. arXiv:0611524.
- [HitSch] N.J. Hitchin, L.P. Schaposnik (2014), Nonabelianization of Higgs bundles J. Differential Geometry, 97 79-89. arXiv:1307.0960.
- [Hit13] N.J. Hitchin (2013), Higgs bundles and characteristic classes . arXiv:math/1308.4603.
- [He01] S. Helgason (2001), Differential geometry, Lie groups, and symmetric spaces , Graduate Studies in Mathematics, AMS 34.
- [KW07] A. Kapustin, E. Witten (2007), Electric-magnetic duality and the geometric Langlands program . Commun. Number Theory Phys. 1 1 1-236. arXiv:hep-th/0604151.
- [Kn02] A.W. Knapp (2002), Lie groups beyond an introduction , Progress in mathematics, 140 .
- [N91] N. Nitsure (1991), Moduli spaces of semistable pairs on a curve , Proc. Lon. Math. Sot. 62
- [OnVi] A.L. Onishchik, E.B. Vinberg (1994), Lie groups and Lie algebras III: structure of Lie groups and Lie algebras , Encyclopaedia of mathematical sciences: Lie groups and Lie algebras Volume 3, Springer.
- [P13] A. Pe´n (2013), o Higgs bundles, real forms and the Hitchin fibration , PhD Thesis UAM.
- [Ra75] A. Ramanathan (1975), Stable principal bundles on a compact Riemann surface , Math. Ann. 213 , 129-152.
- [SW] A.A. Sagle, R.E. Walde (1973), Introduction to Lie groups and Lie algebras , Pure and Applied Mathematics, Academic Press 51.
- [Sch11] L.P. Schaposnik (2013), Monodromy of the SL 2 Hitchin fibration , IJM, ArXiv:1111.2550.
- [Sch13] L.P. Schaposnik (2013), Spectral data for G-Higgs bundles , University of Oxford. arXiv:1301.1981.
- [Sch13b] L.P. Schaposnik (2014), Spectral data for U(m,m)-Higgs bundles IMRN, arXiv:1307.4419.
- [S88] C.T. Simpson (1988), Constructing variations of Hodge structure using Yang-Mills theory and applications to uniformisation , Journal of the AMS 1 867-918.
- [S92] C.T. Simpson (1992), Higgs bundles and local systems , Publ. Math. I.H.E.S. 75 , 5-95.
- [Xia03]
- E.Z. Xia (2003), The moduli of flat U p, ( 1) structures on Riemann surfaces , Geometriae Dedicata 97 , 33-43. arXiv:math/9910037. | null | [
"Laura P. Schaposnik"
] | 2014-08-02T03:50:50+00:00 | 2014-08-02T03:50:50+00:00 | [
"math.AG",
"math-ph",
"math.DG",
"math.MP",
"14D20, 14D21, 20C33"
] | An introduction to spectral data for Higgs bundles | These notes have been prepared as reading material for the mini-course that
the author gave at IMS, National University of Singapore, as part of the
"Summer school on the moduli space of Higgs bundles". |
1408.0334v1 | ## Complex Two-Graphs via Equiangular Tight Frames
Thomas R. Hoffman a , James P. Solazzo
a
a Coastal Carolina University, Conway, SC 29528
## Abstract
In 'A survey of two-graphs' [24], J.J. Seidel lays out the connections between simple graphs, two-graphs, equiangular lines and strongly regular graph. It is well known that there is a one-to-one correspondence between regular two-graphs and equiangular tight frames. This article gives a generalization of two-graphs for which these connections can be mimicked using roots of unity beyond ± 1 .
Keywords: equiangular tight frame, two-graph, roots of unity, Seidel adjacency matrix,
2010 MSC:
05C50, 05E30
## 1. Introduction
Two-graphs play a wide and varied role in several areas of mathematics. To quote J.J. Seidel from his well-known paper, A survey of two-graphs [24], 'Twographs provide a good example of combinatorial geometry and group theory.' The study of two-graphs is equivalent to the study of sets of equiangular lines in Euclidean geometry, sets of equidistant point sets in elliptic geometry, binary maps of triples with vanishing co-boundary, and double coverings of complete graphs.
Applications include but are not limited to network theory [1] and coding theory [10]. At the beginning of the 21 st century, R. Holmes and V. Paulsen in [17] and T. Strohmer and R. Heath in [28], discovered the work done by J.J. Seidel and others regarding two-graphs had found another application. In particular, the existence and construction of real equiangular tight frames (ETFs) was expedited by their discovery of the fact that there is a in one-to-one correspondence between real ETFs and regular two-graphs . This one-to-one correspondence is a wellknown fact in the frame theory community [1, 2, 3, 4, 7, 15, 8, 27, 28, 29, 4, 19, 30]. Indeed extending the already lengthy list of applications of two-graphs to now include such areas as signal processing and communication theory.
In this article, we present an alternate yet equivalent definition of a two-graph. This new definition allows us to generalize the definition of a two-graph in a natural and intuitive way to what we refer to as a complex two-graph . Associated to each two-graph is a set of Seidel adjacency matrices , that is, a set of symmetric matrices whose diagonal entries are all zero and off diagonal entries are ± 1 . Similarly, associated to each complex two-graph is a set of complex Seidel adjacency matrices ,that is, a set of self-adjoint matrices whose diagonal entries are all zero and off diagonal entries are m th roots of unity for a fixed m in N . The fact that the off diagonal entries of a 'real' Seidel adjacency matrix are square roots of unity is a trivial yet surprisingly useful observation. This observation coupled with our equivalent definition of a two-graph is the key to this extension. Furthermore, many of the results regarding complex two-graphs mirror the analogous results pertaining to two-graphs .
For example, it is well-known that for a set of equiangular lines in R k to meet the absolute or relative bounds, the associated two-graph must be regular , i.e., the associated Seidel adjacency matrix has precisely two distinct eigenvalues. These results extend naturally to C k . That is, regular complex two-graphs produce sets of equiangular lines that meet either the absolute or relative bounds. Consequently, associated with each regular complex two-graph is a complex ETF.
This article is organized as follows. In Section 2, the motivation underlying the definition of a two-graph is presented. Section 3 discusses the relationship between two-graphs, equiangular lines, and ETFs, comparing the real case to the complex case. Section 4 presents the reader with a careful introduction to complex two-graphs via the cube roots of unity. Section 5 extends the definitions and results from Section 4 to include the m th roots of unity for a fixed m in N .
For the reader familiar with two-graphs this paper is self-contained. For the reader not as familiar with two-graphs many of the definitions and results in this article are accompanied by examples intended to motivate said definitions or results.
## 2. Motivating the Definition of a Two-Graph
In this section we summarize the first four sections of J.J. Seidel's, A Survey of Two-Graphs, [24]. Lemma 2.7 lays the foundation for understanding the generalization of a two-graph presented in Section 4.
A graph is a pair (Ω , E ) where Ω is a set of vertices , and E is a set of unordered pairs of vertices, whose elements are called edges . For the purposes of this paper,
graphs do not have loops or multiple edges. A complete graph on n vertices is a graph with | Ω = | n and E contains every possible unordered pair of vertices.
Denote by A X and V X the adjacency matrix, and the set of vertices of the graph X , respectively. We also use I n for the n × n identity matrix and J n for the n × n matrix of all ones.
Definition 2.1. Given a graph X on n vertices, the Seidel adjacency matrix of X is defined to be the n × n matrix S X := ( s ij ) where s i,j is defined to be -1 when i and j are adjacent vertices, +1 when i and j are not adjacent, and 0 when i = j .
The Seidel adjacency matrix of X is related to the usual adjacency matrix A X by
$$S _ { X } = J _ { n } - I _ { n } - 2 A _ {$$
Definition 2.2. Let X be a graph and τ ⊆ V X . Now define the graph X τ to be the graph arising from X by changing all of the edges between τ and V X -τ to nonedges and all the nonedges between τ and V X -τ to edges. This operation is called switching on the subset τ , see [10].
The operation of switching is an equivalence relation on the collection of graphs on n vertices. This can be seen by observing if τ ⊆ V X , then switching on τ is equivalent to conjugating S X by the diagonal matrix D with D ii = -1 when i ∈ τ and 1 otherwise. The switching class of X , denoted [ X ] , is the collection of graphs obtained from X by switching on every subset of V X .
Example 2.3. The graph in Figure 1 will be denoted as X S . This graph will be referred to frequently throughout the paper.
Figure 1: Star graph on 6 vertices
The graph in Figure 2 can be obtained by switching X S on the set τ = { 2 3 , } .
4
Figure 2: X S switched on { 2 3 , } .
As stated above, switching on τ is equivalent to conjugating S X by the diagonal matrix D with D ii = -1 when i ∈ τ and 1 otherwise. This is demonstrated in Example 2.4.
Example 2.4. The Seidel matrix for X S is
$$S _ { X _ { S } } = \begin{pmatrix} 0 & 1 & -$$
The diagonal matrix D corresponding to switching on the set τ = { 2 3 , } is
$$D = \begin{pmatrix} 1 & 0 & 0 & 0 & 0 \\ 0 & -$$
The result of this conjugation is Seidel matrix for the graph in Figure 2.
$$D S _ { X s } D = \begin{pmatrix} 0 & - 1 & 1$$
Definition 2.5. The graphs X and Y on n vertices are called switching equivalent if Y is isomorphic to X τ for some τ ⊂ V X , see [10].
Switching equivalent defines a second yet coarser equivalence relation on the collection of graphs on n vertices. The switching equivalent class of X , denoted [[ X ]] , is the collection of graphs obtained from X by conjugating S X by a signed permutation matrix, i.e. the product of a permutation matrix and a diagonal matrix of ± 1 ′ s . Thus, the spectrum of the Seidel adjacency matrices of switching equivalent graphs are identical. Note that [ X ] is a subset of [[ X ]] for any graph. For the complete graph and empty graph on n vertices, their switching classes are equal to their switching equivalent classes.
Corollary 2.6 (Corollary 3.5 in [24]) . Switching does not change the parity of the number of adjacencies among any 3 vertices of a graph.
/negationslash
Proof. On 3 vertices there are 4 non-isomorphic graphs, 2 distinct switching classes of graphs, and 2 distinct switching equivalent classes of graphs. The 4 non-isomorphic graphs X 1 , X 2 , X 3 , and X 4 are given in Figure 3. Clearly, [ X 1 ] = [[ X 1 ]] = [ X 2 ] and [ X 3 ] = [[ X 3 ]] = [ X 4 ] but [ X 1 ] = [ X 3 ] .
Figure 3: Nonisomorphic graphs on 3 vertices.
Lemma 2.7 (Lemma 3.8 in [24]) . For any graph on 4 vertices the number of subgraphs on 3 vertices, having an odd number of edges, is even.
Proof. On 4 vertices there are 11 non-isomorphic graphs, 8 distinct switching classes of graphs, and 3 distinct switching equivalent classes of graphs. The 11 non-isomorphic graphs are X , ..., X 1 6 , shown in Figure 4, and their complements X , ..., X 6 11 . The distinct switching classes are [ X 1 ] , [ X 2 ] , [ X 4 ] , and each 1 -edge graph contributes a distinct switching class. The distinct switching equivalent classes are [[ X 1 ]] the empty graph, [[ X 2 ]] the 1 -edge graph, and [[ X 4 ]] the complete graph.
Figure 4: Nonisomorphic graphs on 4 vertices.
Lemma 2.7 is the motivation behind the definition of a two-graph . Let Ω be a finite set and ∆ a set of triples of elements from Ω .
Definition 2.8. A two-graph (Ω ∆) , is a pair of a vertex set Ω and a triple set ∆ ⊂ Ω 3 , such that each set of four element subset from Ω contains an even number of triples of ∆ .
Lemma 2.9 is necessary to prove Theorem 2.10 below which states there is one-to-one correspondence between two-graphs and the switching classes of graphs on n vertices.
Lemma 2.9 (Lemma 3.9 in [24]) . The graphs (Ω , E ) and (Ω , E ′ ) are switching equivalent if the parity of the number of edges among each triple of vertices is the same for both graphs.
Proof. Let v be any vertex in Ω and S the set of vertices in Ω which have different adjacency with v in (Ω , E ) and (Ω , E ′ ) . Switching (Ω , E ′ ) on the set S gives a new graph (Ω , E ′′ ) such that the adjacencies of v with every other vertex are the same in (Ω , E ) and (Ω , E ′′ ) . Consider a pair of vertices { u, w } from Ω for which neither is equal to v . By hypothesis, the triangles { v, u, w } in (Ω , E ) and (Ω , E ′ ) have the same parity of edges. Switching on S preserves the parity of these triangles, so the triangles { v, u, w } in (Ω , E ) and (Ω , E ′′ ) have the same parity of edges and the adjacencies between v and u , and v and w are equal. Thus, the adjacency between u and w must also be the same for (Ω , E ) and (Ω , E ′′ ) . Therefor, these two graphs are isomorphic and the original two are switching equivalent.
Theorem 2.10 (Theorem 4.2 in [24]) . Given n , there is a one-to-one correspondence between the two-graphs and the switching classes of graphs on n vertices.
The following is Seidel's proof and is included for later reference.
Proof. Let (Ω , E ) be any graph. Define ∆ as the triples of Ω which correspond to triangles containing an odd number of edges. By Corollary 2.6, ∆ is invariant under switching. Lemma 2.7 proves (Ω ∆) , is a two-graph.
Conversely, let (Ω ∆) , be a two-graph, satisfying Definition 2.8. Select any ω in Ω and partition Ω \ { ω } into any 2 disjoint sets Ω 1 and Ω 2 . Let E consist of the following pairs:
$$\begin{array} {$$
Thus, we associate to (Ω ∆) , a class of graphs (Ω , E ) . By construction, ∆ is the set of triangles in (Ω , E ) which have an odd number of edges. So, by Lemma 2.9, the class of graphs constructed from (Ω ∆) , is a switching class and distinct switching classes yield distinct two-graphs. This proves the theorem.
Table 1 provides partial data on the number of non-isomorphic graphs, switching classes (two-graphs), and switching equivalent classes (non-isomorphic twographs) on n vertices up to n = 12 [21]. Indeed for Ω = { 1 2 3 , , } there are 2 two-graphs which are non-isomorphic. For Ω = { 1 2 3 4 , , , } there are 8 twographs but only 3 non-isomorphic two-graphs. Two of the three aforementioned non-isomorphic two-graphs correspond to the empty and complete graphs on 4 vertices and the third non-isomorphic two-graph corresponds to any one of the six 1-edge graphs on 4 vertices. This is precisely Lemma 2.7.
Table 1: Class Sizes
| n | non-isomophic | switching classes | switching equivalent classes |
|-----|------------------|----------------------|--------------------------------|
| 3 | 4 | 2 | 2 |
| 4 | 11 | 8 | 3 |
| 5 | 34 | 64 | 7 |
| 6 | 156 | 1024 | 16 |
| 7 | 1044 | 32,768 | 54 |
| 8 | 12,346 | 2 21 | 243 |
| 9 | 274,668 | 2 28 | 2038 |
| 10 | 12,005,168 | 2 36 | 33,120 |
| n | no known formula | 2 ( n - 1)( n - 2) 2 | See Proposition Appendix A.2 |
## 3. Equiangular Lines in R k and C k
This section reviews the process which takes a two-graph to a set of equiangular lines and vice versa. This process provides both the insight and the underlying motivation for our generalization of the definition of a Seidel matrix to allow m th -roots of unity in the off diagonal entries as well as our generalization of the definition of a two-graph.
## 3.1. Equiangular Lines in R k to a Two-Graph
Given a set Γ = { x , ..., x 1 n } of vectors in R k , let U be the k × n matrix with the elements of Γ as its columns. Then
$$G \coloneqq U ^ { T } U$$
is the Gram matrix of the vectors in Γ .
If Γ is set of unit vectors representing a set of equiangular lines in R k with x x T i j = ± α , then the n × n Gram matrix associated with Γ has the form
$$G = I + \alpha S$$
where S is an n × n Seidel adjacency matrix. Let X be the graph associated to the matrix S . If Ω := { 1 2 3 , , , ..., n } and ∆ is the set of all triples of vertices of X whose induced subgraph on three vertices has either 1 or 3 edges, the ordered pair (Ω ∆) , is a two-graph by Lemma 2.7.
Thus, every set of n -equiangular lines in R k yields a two-graph using the previously described process. It is worth noting there are 2 n -1 distinct Seidel adjacency matrices associated with a given set of n -equiangular lines. However, this set of Seidel adjacency matrices belong to the same switching class.
## 3.2. A Two-Graph to Equiangular Lines in R k
Constructing a graph (Ω , E ) from a two-graph (Ω ∆) , is not a well-defined process. Indeed there is a one to many correspondence. Fortunately the many are in the same switching class. The proof of Theorem 2.10 includes a process of how to build a graph (Ω , E ) given a two-graph (Ω ∆) , . We review this process below as well as include an example.
Pick v in Ω and a subset Ω 1 of Ω \ { v } . Define Ω 2 as the complement of Ω 1 in Ω \ { v } .
- · Start with E = {} .
- · For each ω in Ω 1 , add { v, ω } into E .
- · For each pair { ω, ω ′ } of elements in Ω 1 , if { v, ω, ω ′ } is in ∆ , add { ω, ω ′ } into E .
- · For each pair { ω, ω ′ } with ω in Ω 1 and ω ′ in Ω 2 , if { v, ω, ω ′ } is not in ∆ , add { ω, ω ′ } into E .
- · For each pair { ω, ω ′ } of elements in Ω 2 , if { v, ω, ω ′ } is in ∆ , add { ω, ω ′ } into E .
The resulting set E is the edge set for a graph (Ω , E ) . Example 3.1 illustrates this process.
## Example 3.1. Consider the two-graph
$$( \{ 1, 2, 3, 4, 5, 6 \}, \{ \{ 1, 2, 3 \}, \{ 1, 2, 6 \}, \{ 1, 3, 4 \}, \{ 1, 4, 5 \}, \{ 1, 5, 6 \}, \\ \{ 2, 3, 5 \}, \{ 2, 4, 5 \}, \{ 2, 4, 6 \}, \{ 3, 4, 6 \}, \{ 3, 5, 6 \} \} ).$$
Let v = 2 and Ω = 1 { 5 6 , } , so Ω = 2 { 1 3 4 , , } .
- · Start with E = {} .
- · Add { 2 5 , } and { 2 6 , } into E .
- · Since { 1 2 5 , , } and { 2 3 6 , , } are not in ∆ , we include { 1 5 , } and { 3 6 , } in E .
- · Since { 1 2 3 , , } is in ∆ , we include { 1 3 , } in E .
The resulting graph is
$$( \{ 1, 2, 3, 4, 5, 6 \}, \{ \{ 2, 5 \}, \{ 2, 6 \}, \{ 1, 5 \}, \{ 3, 6 \}, \{ 1, 3 \} \} ),$$
## or as in Figure 5.
One should notice the choice of v , Ω 1 and Ω 2 will possibly result in different graphs, but they will be in the same switching class. Applying a permutation from S | Ω | to the labels in the triple sets of ∆ will result in a graph switching equivalent to X S .
Figure 5: Graph resulting from two-graph construction, X S .
/negationslash
Given a two-graph (Ω ∆) , construct a graph, say X , on n vertices using this process. Again any graph constructed using this process must be in the same switching class as any other graph constructed via the given two-graph (Ω ∆) , . Consequently, the spectrum of the associated Seidel matrix denoted, S X , of any such graph X , remains constant. Since trS X = 0 and S X = 0 , the least eigenvalue of S X is negative. It follows that
$$G \coloneqq I + \frac { 1 } { \alpha } S _ { X }$$
is a positive semi-definite matrix where -α denotes the least eigenvalue of S X . Thus if the G has rank k , then there is a k × n matrix U such that G = U U T , where the n columns of this matrix U are the vectors in Γ which generate the n -equiangular lines in R k . Once again G is the Gram matrix associated with Γ .
## 3.3. Equiangular Lines in C k and Complex Seidel Adjacency Matrices
Now consider a set of equiangular lines in C k . If Λ = { z , ..., z 1 n } is a set of vectors representing this set of equiangular lines in C k with | z z ∗ i j | = α , then the n × n Gram matrix associated with Λ has the form
$$G = I + \alpha Q$$
where Q is a Hermitian matrix with all diagonal entries zero and all off-diagonal entries have modulus 1 . In [17], Holmes and Paulsen call such a matrix Q a signature matrix . However, some authors refer to this matrix as a Seidel matrix due to the connection to two-graphs. For the remainder of this paper we define a complex Seidel adjacency matrix as follows.
/negationslash
Definition 3.2. An n × n Hermitian matrix S such that s ii = 0 and | s ij | = 1 for all i = j is called a complex Seidel adjacency matrix .
Thus, there is a one-to-one correspondence between sets of n -equiangular lines in C k and complex Seidel adjacency matrices. In Sections 4 and 5 we prove a given complex Seidel adjacency matrix which has only roots of unity for its nonzero entries gives a natural way to generalize the definition of a two-graph to what we refer to as a complex two-graph . Moreover, we show complex regular two-graphs are precisely the complex two-graphs for which the relative or absolute bounds are met for the associated set of equiangular lines.
## 3.4. Relative, Absolute, and Welch Bounds
The maximal number of equiangular lines in either R k and C k occurs precisely when the associated Seidel adjacency matrix has exactly two distinct eigenvalues, e.g., [17, 10, 24]. In addition, the vectors associated with the maximal set of equiangular lines span the ambient space. This is a particularly valuable fact in frame theory since it guarantees this set of vectors, with a slight modification to their length, will be an ETF in either R k or C k .
It is well known that the maximal number of equiangular lines is k k ( +1) 2 in R k and k 2 in C k . One way to prove this in R k is to show the projections corresponding to the equiangular lines form a linearly independent set inside the vector space of symmetric k × k matrices which has dimension k k ( +1) 2 . One difference in the complex setting is that the Hermitian k × k matrices do not form a vector space over C . However, the Hermitian k × k matrices do form a vector space over R with dimension equal to k 2 . In Proposition 3.3, we derive the known upper bound, k 2 , for the number of equiangular lines in C k using this idea.
We begin by noting if z is a unit vector in C k , then Z = zz ∗ is a Hermitian k × k matrix and Z 2 = Z . It is also worth noting replacing z by e iθ z does not change the matrix Z . To compare with the real case, for a line through the origin in R k there are two distinct unit vectors which can be used to represent the given line. However, in the complex case, for a line through the origin in C k there are infinitely many unit vectors one can choose to represent the given line.
Now if W is also a unit vector in C k and W = ww ∗ , then
$$Z W = z z ^ { * } w w ^ { * } = ( z ^ { * } w ) z w ^ { * },$$
and so
$$\text{tr} ( Z W ) = | \langle z, w \rangle | ^ { 2 }.$$
Proposition 3.3 and it's proof closely resemble Theorem 11.2.1 in [10]. For the remainder of this section Γ = { z , ..., z 1 n } will denote a set of unit vectors associated with a set of equiangular lines in C k , and Z , ..., Z 1 n will denote the projections onto this set of equiangular lines, i.e., Z i = z z i ∗ i for each i = 1 , ..., n .
Proposition 3.3. (The Absolute Bound) Let Z , ..., Z 1 n be the projections onto a set of equiangular lines in C k . Then these matrices form a linearly independent set in the vector space of Hermitian matrices over the R , and consequently n ≤ k 2 .
/negationslash
$$\ r r o o { J. \text{ Leu} } & \alpha = \left | \langle z _ { i }, z _ { j } \rangle \right | \text{ for } t \neq \\ \text{lines.} & \text{If } W = \sum _ { i = 1 } ^ { n } c _ { i } Z _ { i }, \text{then} \\ & \quad \vdots \left | \mathcal { m } { V } ^ { 2 } \right | \cdot \sum _ { i = 1 } ^ { n } \Sigma$$
Proof. Let α = |〈 z , z i j 〉| for i = j , the cosine of the smaller angle between the
/negationslash
$$c _ { i } ^ { 2 }.$$
$$\overline { i = 1 } ^ { \quad } & \\ & \text{tr} ( W ^ { 2 } ) & = \sum _ { i, j } c _ { i } c _ { j } \text{tr} ( Z _ { i } Z _ { j } ) \\ & = \sum _ { i } c _ { i } ^ { 2 } + \sum _ { i, j \colon i \neq j } c _ { i } c _ { j } \alpha ^ { 2 } \\ & = \alpha ^ { 2 } \left ( \sum _ { i } c _ { i } \right ) ^ { 2 } + ( 1 - \alpha ^ { 2 } ) \sum _ { i } c _ { i } ^ { 2 }. \\ \\ \cdot \quad \cdot \quad \stackrel { \sim } { \vec { \sim } \cdot \vec { \sim } \cdot \vec { \sim } \cdot \vec { \sim } \cdot \vec { \sim } \cdot \vec { \sim } } \cdot \cdot \cdot \cdot \cdot \cdot$$
It follows that the tr( W 2 ) = 0 if and only if c i = 0 for all i . So, the Z i are linearly independent. The space of Hermitian k × k matrices over R has dimension k 2 , and the result follows.
The following two propositions are Lemmas 11.3.1 and 11.4.1 in [10]. The proofs of these propositions are not included since they are identical to the proofs given in [10] and the idea is similar to the proof of Proposition 3.3.
Proposition 3.4. (Lemma 11.3.1 [10]) Suppose Z , ..., Z 1 n are the projections onto a set of equiangular lines in C k and |〈 z , z i j 〉| = α . If I = ∑ i c Z i i , then c i = k/n for all i and
$$n = \frac { d - d \alpha ^ { 2 } } { 1 - d \alpha ^ { 2 } }.$$
The Seidel matrix determined by any set of n unit vectors spanning these lines has eigenvalues
$$- \frac { 1 } { \alpha }, \underset { \dots } { \frac { n - k } { k \alpha } }$$
with multiplicities n -k and k , respectively.
Proposition 3.5. (Lemma 11.4.1 [10]) Suppose { z , ..., z 1 n } is a set of n equiangular lines in C k and |〈 z , z i j 〉| = α . If α -2 > k , then
$$n \leq \frac { k - k \alpha ^ { 2 } } { 1 - k \alpha ^ { 2 } }.$$
If Z , ..., Z 1 n are the projections onto these lines, then equality holds if and only if ∑ i Z i = ( k/n I ) . Corollary 3.6. (Welch bound) Given a set { z , ..., z 1 n } of n vectors in C k or R k set
$$\alpha \coloneqq \max _ { i \neq j } | \langle z _ { i }, z _ { j } \rangle |.$$
/negationslash
Then
$$\alpha & \geq \left ( \frac { n - k } { k ( n - 1 ) } \right ) ^ { \frac { 1 } { 2 } }. \\ \dots & e \dots \dots \dots \ddots \ddots.$$
To summarize, given a set of equiangular lines in either C k or R k this set of lines spans the given space if and only if equality holds in Proposition 3.5. In addition, this set of equiangular lines is maximal in the space. The Welch bound plays an equivalent role in frame theory, that is, a given set of frame vectors it is necessary for equality to hold in the Welch bound for the frame vectors to be an equiangular tight frame.
## 4. Cube Root Two-Graphs
In [3], to simplify the search for complex ETFs the authors restrict the off diagonal entries of a Seidel adjacency matrix to the cube roots of unity. The fact that the Seidel adjacency matrix must have two distinct eigenvalues coupled with this restriction to the cube roots of unity introduced new constraints that must be satisfied for the frame associated with the Seidel adjacency matrix to be an ETF. These new constraints along with the fact that these 'cube root Seidel adjacency matrices' corresponded to strongly regular graphs allowed the authors to discover new complex ETFs.
Like the authors in [3], J.A. Tropp in [30] simplifies the search for complex ETFs but this time by restricting the entries of the frame vectors to m th roots of unity. D. Kalra developed a technique in [19] which similarly restricts the entries in the frame vectors. Neither Tropp's nor Kalra's techniques lead to Seidel matrices whose nonzero entries are all roots of unity. Tropp poses several open questions at the end of [30], one of which is 'Are complex ETFs equivalent to some type of graph or combinatorial object?'.
In this section, we use the techniques from [3] to extend the definition of a two-graph and answer the above question posed in [30].
Recall that associated to each Seidel adjacency matrix there is a two-graph and to each two-graph there is an associated switching class of Seidel matrices. In this
section, all nonzero entries of the 'Seidel adjacency matrix' will be restricted to the cube roots of unity. That is, ω := e 2 π/ 3 , ω 2 := e 4 π/ 3 , and 1 . Such a matrix will be called a cube root Seidel matrix . The graph associated with an n × n cube root Seidel matrix will be a complete graph on n vertices with edges weighted by 1 , ω , and ω 2 . Such graphs will be referred to as cube root edge weighted graphs or CREWgraphs . Figure 6 gives an example of such a graph.
ω
Figure 6: CREW graph on 3 vertices
Unlike the real-case, there is a choice as to which matrix will correspond to the graph given in Figure 6. For the purposes of this article, the weight of the edge { i, j } with i < j will be the ( i, j ) th entry in the corresponding cube root Seidel matrix, which means the ( j, i ) th entry will be the complex conjugate of the ( i, j ) th entry. The cube root Seidel matrix corresponding to Figure 6 is
$$\begin{pmatrix} 0 & 1 & \omega \\ 1 & 0 & \omega ^ { 2 } \\ \omega ^ { 2 } & \omega & 0 \end{pmatrix}.$$
Recall from Section 2, switching a graph X on a subset τ ⊆ V X is equivalent to conjugating S X by the diagonal matrix D with D ii = -1 when i ∈ τ and 1 otherwise.
## Definition 4.1. Let X be a CREW graph and define
D 3 := { D is a diagonal matrix : D ii is a cube root of unity } .
Given D in D 3 , the graph associated with the cube root Seidel matrix D S ∗ X D is called a switch on D . The switching class of X , denoted [ X ] , is the collection of graphs obtained by switching X by every element of D 3 .
The following is an example of a switch on the graph in Figure 6 with the resulting graph.
$$\begin{pmatrix} 1 & 0 & 0 \\ 0 & \omega ^ { 2 } & 0 \\ 0 & 0 & \omega \end{pmatrix} \begin{pmatrix} 0 & 1 & \omega \\ 1 & 0 & \omega ^ { 2 } \\ \omega ^ { 2 } & \omega & 0 \end{pmatrix} \begin{pmatrix} 1 & 0 & 0 \\ 0 & \omega & 0 \\ 0 & 0 & \omega ^ { 2 } \end{pmatrix} = \begin{pmatrix} 0 & \omega & 1 \\ \omega ^ { 2 } & 0 & 1 \\ 1 & 1 & 0 \end{pmatrix}$$
Figure 7: A switch of Figure 6.
In the real case, we switched on a vertex or a set of vertices of a graph. Working with weighted graphs changes this approach. In this case, we say switching the i th vertex by weight ω is the result of conjugating by the diagonal matrix which has 1 ′ s on the diagonal with the exception that ω is in the i th position. Careful consideration of the example above suggests Proposition 4.2.
Proposition 4.2. Let G be a CREW graph. Switching G on vertex v i by ω results in a graph G ′ where edges not incident to v i are not effected and edges incident to v i have their weight multiplied by ω if there other vertex is v j with i < j and their weight is multiplied by ω when j < i .
Proof. Let S be the Seidel matrix for G and D the diagonal matrix corresponding to this switch. The resulting matrix G ′ has Seidel matrix DSD -1 . Since we weight our graphs using the upper half of the Seidel matrix, we see the entries in DSD -1 above the diagonal in the i th row ( i < j ) are multiplied by ω and in the i th column ( j < i ) are multiplied by ω .
The proof of Proposition 4.2 does not rely on cube roots of unity and extends to all complex numbers of modulus 1 .
A Seidel matrix whose nonzero entries in the first row and column are all 1 ′ s is said to be in standard form . Each Seidel matrix, with real or complex entries, can be switched to be in standard form. The three graphs in Figure 8 are representatives for distinct switching classes of CREW graphs on 3 vertices.
Proposition 4.3. There are three distinct switching classes for the CREW graphs with 3 vertices. In fact, the three graphs in Figure 8 are the unique representatives in standard form from each switching class.
Figure 8: Representatives for switching classes.
Proof. Given any CREW graph on 3 vertices the associated Seidel adjacency matrix in standard form must be associated with one of the three graphs in Figure 8. Now suppose there is a switch from one of the representatives in Figure 8 to another. This means there is a D in D 3 such that
$$\begin{pmatrix} a & 0 & 0 \\ 0 & b & 0 \\ 0 & 0 & c \end{pmatrix} \begin{pmatrix} 0 & 1 & 1 \\ 1 & 0 & \omega _ { 1 } \\ 1 & \overline { \omega } _ { 1 } & 0 \end{pmatrix} \begin{pmatrix} \overline { a } & 0 & 0 \\ 0 & \overline { b } & 0 \\ 0 & 0 & \overline { c } \end{pmatrix} = \begin{pmatrix} 0 & 1 & 1 \\ 1 & 0 & \omega _ { 2 } \\ 1 & \overline { \omega ^ { 2 } } & 0 \end{pmatrix} \\ \text{where} \, \omega _ { 1 }, \omega _ { 2 }, a, b, \text{ and } c \text{ are cube roots of unity.} \text{ This forces } a = b = c = 1 \text{ while}$$
where ω 1 , ω 2 , a b , , and c are cube roots of unity. This forces a = b = c = 1 which in return forces ω 1 = ω 2 .
Thus far, switching on a CREW graph, X , has been accomplished by conjugating the associated cube root Seidel adjacency matrix, S X , by a diagonal matrix D in D 3 . Proposition 4.4 extends switching on CREW graphs in a manner similar to Definition 2.2.
Proposition 4.4. Let G be a CREW graph with vertices labeled { 1 2 , , ..., n } . Switching vertex i of G by weight ω will change the weight of edges { i, j } by a factor of ω 2 if i < j and by a factor of ω if j < i . Edge weights for edges not incident to i will not change.
Proof. Let S be the Seidel matrix corresponding to G . Switching the vertex i by ω on the matrix becomes multiplying the i th row by ω and the i th column by ω 2 . Considering the graph corresponding to this new matrix gives the desired result since the edge weights come from the upper half of the matrix.
Proposition 4.5. Let S be an n × n cube root Seidel matrix. There are 3 n -1 elements in the switching class of S . Furthermore, there are 3 ( n -1)( n -2) 2 switching classes of n × n cube root Seidel matrices.
Proof. Let S be a n × n cube root Seidel matrix. Switches of S are the result of conjugating S by diagonal matrices D in D 3 . If ω is a cube root of unity, then
( ωD S ωD ) ( ) -1 = DSD -1 . Thus, there are 3 n -1 elements in the switching class of S . To count the number of classes, divide the number of n × n cube root Seidel matrices, which is 3 n n ( -1) 2 , by the number of elements in each class. This yields the stated result.
As in the real case, when classifying CREW graphs up to isomorphism and switching, some switching classes collapse together. Allowing for both switching and isomorphism, the new classes are called switching equivalent classes . Let X be a CREW graph, then conjugating the associated cube root Seidel adjacency matrix, S X , by the product of a diagonal matrix D in D 3 and a permutation matrix P results in a cube root Seidel matrix, PDS X ( PD ) -1 , which is switching equivalent to X . The switching equivalent class of X , denoted [[ X ]] , is the collection of all such conjugations.
While Proposition 4.5 gives the number of switching classes of CREW graphs, there is not a known formula for the number of switching equivalent classes. The sequence 2 , 4 , 14 , 120 , 3222 does not occur in the Online Encyclopedia of Integer Sequences, see [23], so the number of switching equivalence classes of CREW graphs does not match with any known sequence. However, [12] contains a formula for the number of non-isomorphic CREW graphs on n vertices which is repeated in Appendix A.4. To compare with the real case, recall that Mallows and Sloan, [22], provide a formula for the number of switching equivalent classes but the number of non-isomorphic graphs is the ever elusive graph isomorphism problem. Table 2 summarizes the data we have collected thus far.
Table 2: Cube Root Class Sizes
| n | non-isomorphic CREW graphs | switching classes | switching equivalent classes |
|-----|------------------------------|-----------------------------------|--------------------------------|
| 3 | 7 | 3 | 2 |
| 4 | 42 | 27 | 4 |
| 5 | 582 | 729 | 14 |
| 6 | 21,480 | 59,049 | 120 |
| 7 | 2,142,288 | 14,348,907 | 3222 |
| n | See Appendix A.4 | 3 ( n - 1)( n - 2) 2 by Prop. 4.5 | no known formula |
Recall in the real case, on four vertices there are 11 non-isomorphic graphs, 8 switching classes (or equivalently 8 two-graphs), 3 switching equivalent classes (or equivalently 3 non-isomorphic two-graphs). Using Table 2 and terminology in
Section 4.3 we have on four vertices there are 42 non-isomorphic CREW graphs, 27 cube root two-graphs, and 4 non-isomorphic cube root two-graphs.
## 4.1. Complex Two-Graphs with Cube Roots of Unity
Before defining a complex two-graph with cube roots of unity a further exploration of two-graphs (in the real setting) will be useful. The following 'new' yet equivalent definition of a two-graph plays a crucial role in adapting the definition of a two-graph to include not only CREW graphs but p th root edge weighted graphs as well.
Definition 4.6. A two-graph (Ω ∆ , 1 , ∆ ) 2 is a triple of a vertex set Ω and triple sets ∆ 1 and ∆ 2 such that ∆ 1 ∪ ∆ = Ω 2 3 and each set of four elements from Ω contains an even number of elements of ∆ 1 and ∆ 2 as subsets.
Comparing Definitions 4.6 and 2.8 leads to the following proposition.
Proposition 4.7. Definitions 2.8 and 4.6 are equivalent.
Proof. Let (Ω ∆) , be a two-graph according to Definition 2.8. Clearly (Ω ∆ ∆) , , ¯ satisfies Definition 4.6.
Let (Ω ∆ , 1 , ∆ ) 2 be a two-graph according to Definition 4.6. By Lemma 2.7, (Ω ∆ ) , 1 satisfies Definition 2.8.
Example 4.8 gives the two graph for Figure 1 following Definitions 2.8 and 4.6.
Example 4.8. Recall the star graph from Example 2.3, repeated in Figure 9. If Ω = { 1 2 3 4 5 6 , , , , , } and let ∆ be the set of triples of vertices of this graph whose induced subgraph on three vertices has either 1 or 3 edges. By Lemma 2.7, (Ω ∆) , is a two-graph where
$$\Delta = \{ \{ 1, 2, 3 \}, \{ 1, 2, 6 \}, \{ 1, 3, 4 \}, \{ 1, 4, 5 \}, \{ 1, 5, 6 \}, \{ 2, 3, 5 \}, \\ \{ 2, 4, 5 \}, \{ 2, 4, 6 \}, \{ 3, 4, 6 \}, \{ 3, 5, 6 \} \}.$$
Using Definition 4.6, (Ω ∆ , 1 , ∆ ) 2 is a two-graph where ∆ = ∆ 1 and
$$\Delta _ { 2 } = \{ \{ 1, 2, 4 \}, \{ 1$$
Figure 9: Star graph on 6 vertices
Theorem 2.10 connected two-graphs and switching classes of simple graphs. Restating Theorem 2.10 using Definition 4.6 requires replacing simple graphs with complete graphs whose edges are weighted by ± 1 , i.e., square root edge weighted graphs . This trivial replacement is the key to understanding the extension of the definition of a two-graph to include cube roots of unity in this section and the m th roots of unity in Section 5.
Theorem 4.9 (Theorem 2.10 restated using Definition 4.6) . Given n , there is a one-to-one correspondence between the two-graphs (Definition 4.6) and the switching classes of ± 1 edge weighted complete graphs on n vertices.
Proof. Let (Ω , E ) be a complete graph on n vertices with the edges in E weighted by -1 and the rest of the edges weighted by 1 . Define ∆ 1 as the set of triangles containing an odd number of edges weighted by -1 , i.e., the switching class of the triangle with three -1 weighted edges, and ∆ 2 the rest of the triangles, i.e., the switching class of the triangle with three 1 weighted edges. Since ∆ 1 and ∆ 2 are invariant under switching, Lemma 2.7 proves (Ω ∆ , 1 , ∆ ) 2 is a two-graph.
Conversely, let (Ω ∆ , 1 , ∆ ) 2 be a two-graph, satisfying Definition 4.6. Select any ω in Ω and partition Ω \ { ω } into any 2 disjoint sets Ω 1 and Ω 2 . Define E 1 and E -1 as the edges weighted by 1 and -1 respectively. For all ω 1 ∈ Ω 1 , the pair { ω, ω 1 } is in E -1 and for all ω 1 ∈ Ω 2 , the pair { ω, ω 2 } is in E 1 . Lastly, if { ω, ω , ω 1 2 } is in ∆ 1 , then { ω , ω 1 2 } is in E 1 , otherwise { ω , ω 1 2 } is in E -1 .
Thus, we associate to (Ω ∆ , 1 , ∆ ) 2 a class of ± 1 complete graphs (Ω , E 1 , E -1 ) . By construction, ∆ 1 is the set of triangles in (Ω , E 1 , E -1 ) which have an odd number of edges weighted by -1 . So, by Lemma 2.9, the class of graphs constructed from (Ω ∆ , 1 , ∆ ) 2 is a switching class and distinct switching classes yield distinct two-graphs. This proves the theorem.
The existence, and hence the definition, of two-graphs comes from Lemma 3.8 in [24]. Before defining cube root two-graphs, we extend this lemma.
Lemma 4.10 (Extension of Lemma 3.8 in [24]) . For any CREW graph on 4 vertices the number of induced CREW subgraphs on 3 vertices, having an odd number of edges weighted w ∈ { 1 , ω, ω 2 } , is even.
Proof. Let G be a CREW graph on 4 vertices and let G ′ be the graph obtained from G by removing edges with weight not w . By Lemma 2.7, there are an even number of induced subgraphs of G ′ on three vertices with an odd number of edges. These subgraphs correspond to the induced subgraphs of G which have an odd number of edges weighted with w .
Using Definition 4.6 as a model, we now define cube root two-graphs.
Definition 4.11. A cube root two-graph (Ω ∆ , 1 , ∆ 2 , ∆ ) 3 is a quadruple of a vertex set Ω and triple sets ∆ 1 , ∆ 2 and ∆ 3 such that ∆ 1 ∪ ∆ 2 ∪ ∆ = Ω 3 3 and each set of four element subset of Ω contains an even number of triples of ∆ 1 , ∆ 2 , or ∆ 3 .
Proving a version of Theorem 2.10 for cube root two-graphs requires extending Lemma 2.9. The idea behind the proof for Lemma 2.9 works equally as well in the cube root case, but some rewriting needs to be done since the parity of edges no longer makes sense.
Lemma 4.12 (Extension of Lemma 3.9 in [24]) . The CREW graphs (Ω , E ) and (Ω , E ′ ) are switching equivalent if the parity of the number of edges weighted by w among each triple of vertices is the same for both graphs with w ∈ { 1 , ω, ω 2 } .
Proof. Let v be any vertex in Ω . Define S 1 as the set of vertices u in Ω such that the edge weight of { v, u } in (Ω , E ′ ) is ω times the edge weight in (Ω , E ) . Similarly, define S 2 as the set of vertices u in Ω such that the edge weight of { v, u } in (Ω , E ′ ) is ω 2 times the edge weight in (Ω , E ) . Switching (Ω , E ′ ) by ω on the set S 1 and by ω 2 on S 2 results in a new graph (Ω , E ′′ ) such that the adjacencies of v with every other vertex are the same in (Ω , E ) and (Ω , E ′′ ) . Consider a pair of vertices { u, w } from Ω for which neither is equal to v . By hypothesis, the triangles { v, u, w } in (Ω , E ) and (Ω , E ′ ) have the same parity of each possible edge weight. Switching on S preserves the parity of these triangles, so the triangles { v, u, w } in (Ω , E ) and (Ω , E ′′ ) have the same parity of edge weights and the weights of the edges { v, u } , and { v, w } are equal. Thus, the edge weight of { u, w } must also be the same for (Ω , E ) and (Ω , E ′′ ) . Therefor, these two graphs are isomorphic and the original two are switching equivalent.
Theorem 4.13 (Extension of Theorem 2.10) . Given n , there is a one-to-one correspondence between the cube root two-graphs and the switching classes of CREW graphs on n vertices.
Proof. Let (Ω , E ) be a complete graph on n vertices with the edges weighted by cube roots of unity. Define ∆ 1 , ∆ 2 and ∆ 3 to be the sets of triples of vertices whose induced subgraphs are in the switching classes of the corresponding graphs given in Figure 8. Since ∆ 1 , ∆ 2 , and ∆ 3 are invariant under switching, Lemma 4.10 proves (Ω ∆ , 1 , ∆ 2 , ∆ ) 3 is a cube root two-graph.
Conversely, let (Ω ∆ , 1 , ∆ 2 , ∆ ) 3 be a cube root two-graph. Select any v in Ω and partition Ω \ { v } into any 3 disjoint sets Ω 1 , Ω 2 , and Ω 3 . For simplicity, we assume Ω = { 1 , ..., n } , v = 1 and for each i ∈ Ω 1 , j ∈ Ω 2 and k ∈ Ω 3 , i < j < k . Changing any choices of v and Ω 1 , Ω 2 , and Ω 3 to fit this description is a permutation of Ω which simplifies the following construction but does not restrict the generality of the proof.
From the partition, build a CREW graph G as follows:
Let E 1 , E 2 and E 3 be a partition of the edges of G such that an edge in E i has weight ω i . As an abuse of notation, we consider indices for E i to be modulo 3 , so E 1 = E 4 and E 2 = E 5 .
- · For every u in Ω i , put { v, u } in E i .
- · For every pair u and w in Ω i , if { 1 , u, w } is in ∆ j , put { u, w } in E j .
- · For every u in Ω 1 and w in Ω 2 , if { 1 , u, w } is in ∆ j , put { u, w } in E j +1 .
- · For every u in Ω 2 and w in Ω 3 , if { 1 , u, w } is in ∆ j , put { u, w } in E j +1 .
- · For every u in Ω 1 and w in Ω 3 , if { 1 , u, w } is in ∆ j , put { u, w } in E j +2 .
Thus, associated to (Ω ∆ , 1 , ∆ 2 , ∆ ) 3 is a class of CREW graphs of the form (Ω , E 1 , E 2 , E 3 ) . By construction, ∆ i is the set of triangles in (Ω , E 1 , E 2 , E 3 ) which have an odd number of edges weighted by ω i . So, by Lemma 4.12, the class of graphs constructed from (Ω ∆ , 1 , ∆ 2 , ∆ ) 3 is a switching class and distinct switching classes yield distinct two-graphs. This proves the theorem.
The use of a permutation in the proof of Theorem 4.13 is justified since to get around this complicates the decision process for putting edges into the E i . For example, if we don't assume v = 1 and the Ω i 's are ordered, we get for every u in
Ω 1 and w in Ω 2 , put { u, w } in E k where
$$k = \begin{cases} j + 1 & \text{if $v<u<w$$$
Writing out all of the cases in this format will lead to the same result and is not useful for understanding the proof or constructing CREW graphs from cube root two-graphs.
The following example describes a cube root two-graph.
Example 4.14. Following the proof of Theorem 4.13, the graph with cube root Seidel matrix
corresponds to the cube root two-graph with Ω = { 1 , . . . , 8 } ,
$$\begin{array} {$$
```
<_SQL_>
```
and
$$\Delta _ { 3 } = \{ \{ 1, 2, 3 \}, \{ 1$$
A regular two-graph is a two-graph such that any, and consequently all, of the associated Seidel adjacency matrices have two distinct eigenvalues. There is an intimate relationship between regular two-graphs and strongly regular graphs which is captured in Theorem 4.15.
Let Φ be a two-graph and X an associated graph. A vertex of X can be isolated by switching and removed, resulting in a graph with one fewer vertex, called a neighborhood of Φ .
Theorem 4.15 (Theorem 11 6 1 . . in [10]) . Let Φ be a nontrivial two-graph on n +1 vertices. Then the following are equivalent:
- 1. Φ is a regular two-graph.
- 2. All the neighborhoods of Φ are regular graphs.
- 3. All the neighborhoods of Φ are ( n, k, a, c ) strongly regular graphs with k = 2 c .
- 4. One neighborhood of Φ is an ( n, k, a, c ) strongly regular graph with k = 2 c .
Theorem 4.15 explains the motivation behind calling a two-graph with exactly two-eigenvalues a regular two-graph. Furthermore the strongly regular graphs have been actively studied and many of these results can be used to build regular two-graphs. Regular two-graphs are important because they are the only nontrivial two-graphs for which the corresponding set of equiangular lines meet the absolute bound or the relative bound. Thus, the set of vectors determined by choosing a unit vector to represent each line spans the ambient space. In frame theory this guarantees this set of vectors with adjusted lengths will form an equiangular tight frame.
The proof of Theorem 4.15 uses terminology such as equitable and quotient matrix which are specific to graphs as well as powerful tools. While this terminology does not apply directly to the cube root setting, the underlying linear algebra results do hold, and gives us Theorem 4.16.
Similar to two-graphs, we call a cube root two-graph regular if any, and hence all, associated CREW graph has two eigenvalues. In [3], they show if a cube root Seidel adjacency matrix with two eigenvalues, λ 1 and λ 2 , is in standard form, then all rows after the first have a constant sum. The row sum being constant is the key to proving Theorem 4.16.
Theorem 4.16. Let Φ be a nontrivial cube root two-graph on n +1 vertices. Then the following are equivalent:
- 1. Φ is a regular cube root two-graph.
- 2. All the neighborhoods of Φ are CREW graphs with vertex sums µ and eigenvalues µ λ , 1 , and λ 2 . The multiplicity of µ is 1 .
- 3. One neighborhood of Φ is a CREW graph with vertex sum µ and eigenvalues µ λ , 1 , and λ 2 . The multiplicity of µ is 1 .
Proof. 1 → 2: Let x be a vertex of Φ and S a Seidel matrix whose first row and column correspond to x . Without loss of generality, assume S is in standard form, which corresponds to x being isolated. By [3], S has two eigenvalues, λ 1 and λ 2 . Let A be the ( n -1) × ( n -1) matrix obtained from S by removing the first row and column. By the interlacing theorem, n -2 of the eigenvalues of A are λ 1 or λ 2 . By [3], the rows of A have a constant sum equal to µ , so this is an eigenvalue for A as well.
2 → 3:
Obvious.
3 → 1: Let A be a matrix corresponding to a cube root Seidel graph with vertex sum µ and eigenvalues µ λ , 1 , and λ 2 . By the Spectral Theorem, eigenvectors for λ 1 and λ 2 can be chosen to be orthogonal to 1 , the eigenvector corresponding to µ . Let S be the matrix
$$\begin{pmatrix} \text{to } \mu. \text{ Let } S \text{ be the matrix } & \begin{pmatrix} 0 & 1 ^ { t } \\ 1 & A \end{pmatrix} \\ \text{then the vectors } & \begin{pmatrix} 0 \\ v \end{pmatrix} \\ \text{where } v \text{ is an eigenvector for } A, \text{ are eigenvectors for } \end{pmatrix}$$
where v is an eigenvector for A , are eigenvectors for S , corresponding to eigenvalues λ 1 or λ 2 .
Let and
$$P = \begin{pmatrix} 1 & 0 \\ 0 & 1 \\ \vdots & \vdots \\ 0 & 1 \end{pmatrix} \\ \Gamma \quad \int 0 \quad n - 1 \, \zeta$$
$$B = \begin{pmatrix} 0 & n - 1 \\ 1 & \mu \end{pmatrix} \\ \text{is-never-for $B$ then $P$}$$
then SP = PB . If v is an eigenvector for B , then Pv is an eigenvector for S for the same eigenvalue. The characteristic polynomial for B is x 2 -µx -( n -1) , so its eigenvalues are λ 1 and λ 2 . Since the first component of Pv is nonzero, these eigenvectors are not any of the previously known eigenvectors of S , so S has just the two eigenvalues. Thus, S is a Seidel matrix which corresponds to a regular cube root two-graph.
## 4.2. Constructions for (9,6) ETF
The article [3] contains an example of a (9 , 6) -equiangular tight frame or equivalently 9 equiangular lines in C 6 . Their construction starts with the known directed strongly regular graph on 8 vertices in Figure 10.
Figure 10: Directed strongly regular graph on 8 vertices.
/negationslash
A Seidel matrix can be constructed from this graph by letting the ij th entry, with i = j , be
$$\begin{cases} \omega & \text{if there is an edge from $i$ to $j$} \\ \omega ^ { 2 } & \text{if there is an edge from $j$ to $i$} \\ 1 & \text{otherwise} \end{cases}$$
and setting the diagonal entries to 0 . Adding a first row and column of ones, with zero on the diagonal entry, completes the construction. This new Seidel matrix has two eigenvalues, and will be switching equivalent to the matrix
```
envalues, and will be switching equivalent to the matrix
```
where ω is a primitive cube root of unity. This construction is an application of proof of Theorem 4.16.
While this construction seems to connect cube root two-graphs to directed strongly regular graphs with no undirected edges, the development of cube root two-graphs describe above can be extended to m th roots of unity. Along with this development, the authors have constructed nontrivial complex two-graphs and regular complex two-graphs for many roots of unity which do not obviously connect with generalizations of strongly regular graphs, see [7, 15].
## 5. Complex Two-Graphs with m th Roots of Unity
Using the obvious definitions for m th root Seidel matrices and m th root edge weighted graphs, we reconsider the results of Section 4. Let D m be the collection of diagonal matrices whose nonzero entries are m th roots of unity, then the definitions of switching classes and switching equivalent classes make sense. With these definitions, most of the results from Section 4 are true for m th roots of unity without modification because their proofs do not depend on cube roots of unity.
Recall the representatives of switching classes for 3 vertices in Figure 8. The extension is representatives have a single edge weighted by 1 , ω ,..., ω m -1 , with ω a primitive m th root of unity, and the other two edges both weighted by 1 . Proposition 4.4 changes as follows, but the proof from Section 4 holds with the modification of the matrix entries being m th roots of unity.
Proposition 5.1 (Extension of Proposition 4.4) . There are m distinct switching classes of m th root edge weighted graphs on three vertices.
With the goal of defining m th root two-graphs, Lemmas 4.10 and 4.12 require special attention. Fortunately, the proof of Lemma 4.10 is an application of Lemma 2.7 and can be extended to any number of weights. The proof of Lemma 4.12 is extended to m th roots of unity by replacing ω 2 by ω . With these lemmas in place, we define m th root two-graphs as follows.
Definition 5.2. An m th root two-graph (Ω ∆ , 1 , . . . , ∆ ) m is a m + 1 -tuple of a vertex set Ω and triple sets ∆ 1 , ∆ 2 ,..., ∆ m such that ⋃ ∆ = Ω i 3 and each set of four element subset of Ω contains an even number of triples of ∆ i , for 1 ≤ i ≤ m .
As expected, the m th root two-graph are in one-to-one correspondence with the switching classes of m th root edge weighted graphs.
Theorem 5.3 (Extension of Theorem 4.13) . Given n , there is a one-to-one correspondence between the m th root two-graphs and the switching classes of m th root edge weighted graphs on n vertices.
The proof of this theorem follows from modifying the proof of Theorem 4.13 to use edge sets E 1 ,..., E m and extending the cases for all possible pairs of vertices. Fortunately, the edge weights are determined by the ∆ i 's, and the proof follows.
For regular m th root two-graphs, we need constant row sums in our m th root Seidel matrices. While, in [3], the authors focus only on cube roots of unity, their result that the standard form of a cube root Seidel matrix with two eigenvalues has constant row sum for every row after the first does not depend on cube roots of unity, only on the standard form and two eigenvalues, and hence, is true for m th root Seidel matrices. This gives us the Theorem 4.16.
Theorem 5.4 (Extension of Theorem 4.16) . Let Φ be a nontrivial m th root twograph on n +1 vertices. Then the following are equivalent:
- 1. Φ is a regular m th root two-graph.
- 2. All the neighborhoods of Φ are m th root edge weighted graphs with vertex sums µ and eigenvalues µ λ , 1 , and λ 2 , the multiplicity of µ is 1 .
- 3. One neighborhood of Φ is an m th root edge weighted graph with vertex sum µ and eigenvalues µ λ , 1 , and λ 2 , the multiplicity of µ is 1 .
The proof of Theorem 4.16 does not depend on cube roots of unity, only on the relationship between the row sums and the eigenvalues of the matrices corresponding to the cube root two-graph. So, replacing the cube roots by m th roots does not effect the proof and the result holds.
## Appendix A. Known Formulas for Counting Switching Equivalence Classes and Cube Root Edge Weighted Graphs
In Tables 1 and 2, formulas are used which were derived combinatorially. The techniques and terminology are different enough from the rest of this article that they deserve attention.
Suppose j is a positive integer and ( j ) denotes a partition of j . Define j k to be the number of times k appears in ( j ) , so ∑ j k =1 k ∗ j k = j . Several of the following formulas involve summations over all partitions of j which will be denoted as ∑ ( j ) .
In [22], the following formula for the number of Euler graphs on n vertices is attributed to R. W. Robinson.
Theorem Appendix A.1 (Eulerian Graphs) . The number of Euler graphs on n vertices is where
and
The central result of [22] is Appendix A.2. Seidel had proved Appendix A.2 for odd n by finding Euler graphs as representatives of switching equivalent classes. Mallows and Sloan proved the even case without making an obvious connection between Euler graphs and switching equivalent classes or two-graphs.
$$\text{eorem Appendix } A. 1 \text{ (Eulerian Graphs). The number of Euler graphs on} \\ \text{rtices is} \\ \sum _ { ( j ) } \frac { 2 ^ { v ( j ) - \lambda ( j ) } } { \prod _ { i } i ^ { j i } j _ { i }! }, \\ \text{ere} \\ v ( j ) = \sum _ { i < k } j _ { i } j _ { k } \gcd ( i, j ) + \sum _ { i } i \left ( j _ { 2 i } + j _ { 2 i + 1 } + \binom { j } { 2 } \right ), \\ d \\ \lambda ( j ) = \sum _ { i } j _ { i } - \text{sgn} \left ( \sum _ { i } j _ { 2 i + 1 } \right ). \\ \text{ The central result of [22] is Appendix } A. 2. \text{ Seidel had proved Appendix } A \\ \text{ odd } n \text{ by finding Euler graphs as representatives of switching equivalent class}$$
Proposition Appendix A.2 (Theorem 1 in [22]) . The number of two-graphs on n vertices is equal to the number of Euler graphs on n vertices.
We do not include the proof of Appendix A.2 as it is most of [22]. However, it is a nice arguement and we recommend any interested person should read it.
Harary and Palmer define a complete directed graph as a directed graph such that for any pair of vertices there is either a directed edge or two directed edges connecting them. With this definition comes the question of how many such graphs are there on n vertices. In [12], this question is answered and [13] contains a refinement of the formula. We include the refined formula.
Theorem Appendix A.3 (Page 133 of [13]) . The number of complete directed graphs on n vertices is where
$$c _ { n } = \frac { 1 } { n! } \sum _ { ( n ) } \frac { n! } { \prod k ^ { n _ { k } } n _ { k }! } 3 ^ { a ( n ) }$$
$$a ( n ) & = \sum _ { k = 1 } ^ { n } \left ( \left \lfloor \frac { k - 1 } { 2 } \right \rfloor n _ { k } + k \binom { n _ { k } } { 2 } \right ) + \sum _ { 1 \leq r < s \leq n } \gcd ( r, s ) n _ { r } n _ { s }. \\ \intertext { a n d v } \underbrace { \lambda \ 2 \, \text{bold} d \, \text{int} \, \text{int} } \underbrace { \lambda \ 2 \, \text{bold} d \, \text{int} \, \text{int} \, \text{int} \, \text{int} \, \text{int} \, \text{int} \, \text{int} \, \text{int} \, \text{int} \, \text{int} \, \text{int} \, \text{int} \, \text{int} \, \text{int} \, \text{int} \, \text{int} \, \text{int} \, \text{int} \, \text{int} \, \text{int} \, \text{int} \, \text{int} \, \text{int} \, \text{int} \, \text{int} \, \text{int} \, \text{int} \, \text{int} \, \text{int} \, \colon }$$
Appendix A.3 holds interest for this article because of Appendix A.4.
Proposition Appendix A.4. There is a one to one correspondence between cube root edge weighted graphs on n vertices and complete directed graphs on n vertices.
Proof. Given a complete directed graph G , label the vertices from 1 to n and define CRG to be a complete graph on n vertices. For any pair of vertices v and u , if two edges connect them in G , then weight the edge { v, u } of CRG with a 1 . For any pair { v, u } with a single directed edge going from v to u , weight it by ω if v < u and ω 2 if u < v . The graph with CRG with these weights is a cube root edge weighted graph. The choice of labeling does not effect the outcome of the weights, so this assignment is one to one and clearly invertible.
## References
- [1] B.G. Bodmann and H.J. Elwood, Complex equiangular Parseval frames and Seidel matrices containing pth roots of unity. Proc. Amer. Math. Soc. 138 (2010), no. 12, 43874404.
- [2] B.G. Bodmann and V.I. Paulsen, Frames, graphs and erasures, Linear Algebra Appl. 404 (2005), 118-146.
- [3] B.G. Bodmann, V.I. Paulsen, and M. Tomforde, Equiangular Tight Frames from Complex Seidel Matrices Containing Cube Roots of Unity, Linear Algebra Appl. 430 (2009), 396-417.
- [4] L. Bos, and S. Waldron, Some Remarks on Heisenberg Frames and Sets of Equiangular Lines, New Zealand Journal of Mathematics 36 (2007), 113137.
| [5] | P.G. Casazza and J. Kova ˘ c evi´, c Uniform Tight Frames with Erasures, Ad- vances in Computational Mathematics 18 (2003), 387-430. |
|-------|-----------------------------------------------------------------------------------------------------------------------------------------------------|
| [6] | J. Clark and D.A. Holton, A first look at graph theory, World Scientific Pub- lishing Co. Inc., Teaneck NJ, 1991. |
| [7] | D.M. Duncan, T.R. Hoffman, and J.P. Solazzo, Equiangular tight frames and fourth root Seidel matrices, Linear Algebra Appl. 4 32 (2010), 2816-2823. |
| [8] | M. Fickus, D.G. Mixon, and J.C. Tremain, Steiner equiangular tight frames. Linear Algebra Appl. 436 (2012), no. 5, 10141027. |
| [9] | The GAP Group, GAP - Groups, Algo- rithms, and Programming, Version 4.7.5 ; 2014 (\protect\vrule width0pt\protect\href{http://www.gap-system.org}{h |
| [10] | C.D. Godsil and G. Royle, Algebraic Graph Theory, Springer New York, 2001. |
| [11] | R.L. Graham, B.L. Rothschild, and J.H. Spencer, Ramsey Theory, Second Ed. John Wiley &Sons Inc. New York, 1990. |
| [12] | F. Harary and E. Palmer, Enumeration of Mixed Graphs, Proceedings of the American Mathematical Society 1 7 (1966), 682-687. |
| [13] | F. Harary and E. Palmer, Graphical Enumeration, Academic Press, New York-London, 1973. |
| [14] | K. Hoffman and R. Kunze, Linear Algebra, Prentice-Hall Inc. Englewood Cliffs, N.J., Second edition, 1971. |
| [15] | T.R. Hoffman and J.P. Solazzo, Complex equiangular tight frames and era- sures. Linear Algebra Appl. 437 (2012), no. 2, 549558. |
| [16] | R.B. Holmes, Optimal Frames, PhD Thesis, University of Houston, 2003. |
| [17] | R.B. Holmes and V.I. Paulsen, Optimal frames for erasures, Linear Algebra and its Applications 377 (2004), 31-51. |
| [18] | J. Jasper, D.G. Mixon, and M. Fickus, Kirkman equiangular tight frames and codes. IEEE Trans. Inform. Theory 60 (2014), no. 1, 170181. |
| [19] | D. Kalra, Complex equiangular cyclic frames and erasures, Linear Algebra and its Applications 419 (2006), 373-399. |
|--------|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [20] | J. Kova ˘ c evi´ c and A. Chebira, Life beyond bases: The advent of frames (Part I), IEEE SP Mag., vol. 24, no. 4 , Jul. 2007, pp. 86-104. Feature article. |
| [21] | B. McKay, Isomorph-Free Exhaustive Generation, Journal of Algorithms, 26 (1998), 306-324. |
| [22] | C. Mallows and N. Sloane, Two-graphs, switching classes and Euler graphs are equal in number, SIAM Journal on Applied Mathematics, 28 (1975), 876-880. |
| [23] | OEIS Foundation Inc. (2011), The On-Line Encyclopedia of Integer Se- quences, http://oeis.org. |
| [24] | J.J. Seidel, Asurvey of two-graphs, in: colloquio Internazionale sulle Teorie Combinatorie (Rome, 1973), Tomo I, in: Atti dei Convegni Lincei, Vol 17, Accad. Naz. Lincei, Rome, 1976, pp. 481-511. |
| [25] | J.J. Seidel, Geometry and combinatorics, in: Selected works of J. J. Seidel, Edited and with a preface by D. G. Corneil and R. Mathon, Academic Press, Boston, 1991. |
| [26] | J.J. Seidel and D.E. Taylor, Two-graphs, a second survey. Algebraic methods in graph theory, Vol. I, II (Szeged, 1978), pp. 689711, Colloq. Math. Soc. Jnos Bolyai, 25, North-Holland, Amsterdam-New York, 1981. |
| [27] | F. Sz¨ll˝si, o o Complex Hadamard matrices and equianguar l tight frames. Lin- ear Algebra Appl. 438 (2013), no. 4, 1962-1967. |
| [28] | T. Strohmer and R. Heath, Grassmannian frames with applications to coding and communications, Appl. Comp. Harm. Anal., 14 (2003), 257-275. |
| [29] | M.A. Sustik, J.A. Tropp, I.S. Dhillon, and R.W. Heath, Jr., On the existence of equiangular tight frames, Linear Algebra Appl., 426 , (2007), 619-635. |
| [30] | J.A. Tropp, Complex equiangular tight frames, In Proc. SPIE Wavelets XI, pp. 590412.01-11, San Diego, Aug. 2005. |
| [31] | J.H. van Lint and J.J. Seidel, Equilateral point sets in elliptic geometry, Ned- erl. Akad. Wetensch. Proc. Ser. A 69=Indag. Math., 28 , (1966), 335-348. | | null | [
"Thomas Hoffman",
"James Solazzo"
] | 2014-08-02T04:04:30+00:00 | 2017-03-13T19:47:50+00:00 | [
"math.CO"
] | Complex Two-Graphs via Equiangular Tight Frames | In `A survey of two-graphs' \cite{Sei}, J.J. Seidel lays out the connections
between simple graphs, two-graphs, equiangular lines and strongly regular
graph. It is well known that there is a one-to-one correspondence between
regular two-graphs and equiangular tight frames. This article gives a
generalization of two-graphs for which these connections can be mimicked using
roots of unity beyond $\pm 1$. |
1408.0335v1 | ## Discovery of an isostructural phase transition within orthorhombic phase field of CaTiO3
Saurabh Tripathi1, Anil Kumar2, Sharmila Shirodkar3, Masatomo Yashima4, U. V. Waghmare2, Dhananjai Pandey1
1 School of Materials Science and Technology, Indian Institute of Technology (Banaras Hindu University), Varanasi-221005, India
2 Theoretical Sciences Unit, Jawaharlal Nehru Centre for Advanced Scientific Research, Bangalore 560064, India
3 Department of Chemistry and Materials Science, Graduate School of Science and Engineering, Tokyo Institute of Technology, 2-12-1-W4-17, O-okayama, Meguro-ku, Tokyo, 152-8551, Japan
Earth's lower mantle extending from 670 to 2,990 km deep is predominantly composed of a perovskite-type (Mg,Fe)SiO 3 phase 1,2 . The perovskite phase undergoes a structural phase transition to a post-perovskite phase responsible for D'' layer seismic discontinuity 2,3 at about 2690 km depth in the lowermost region of the lower mantle. However, structural basis of other seismic discontinuities occurring in the upper region of the lower mantle (700 km to 1,200 km deep) remains unexplained 4-7 , as no apparent change in the crystal symmetry of the orthorhombic perovskite phase has been reported . 5 We present here unambiguous evidence for a non-apparent isostructural phase transition 8 in the stable orthorhombic perovskite phase of CaTiO which may have relevance to 3 phase transitions in the perovskite phase of (Mg,Fe)SiO also, as both the compounds 3 have similar structure, tolerance factor and thermochemical properties 9-11 . Our results are based on the analysis of neutron powder diffraction patterns using Rietveld and mode crystallography techniques and are supported by density functional and Landau theory calculations. The present results on CaTiO would encourage search for isostructural 3 phase transition in the perovskite phase of (Mg,Fe)SiO that may provide clue to the 3 unexplained geophysical phenomena in the upper part of the earth's lower mantle.
The family of oxide perovskites with the general formula AB O3 is named after the mineral name of CaTiO3 9-10 where A and B are larger and smaller cations, respectively. CaTiO3, whose structure and the tolerance factor are similar to the perovskite phase of the earth's lower mantle material (Mg,Fe)SiO 3 10-11 , undergoes structural phase transitions at easily achievable temperatures 12 , as compared to (Mg,Fe)SiO3 that requires extremely high temperatures and pressures . CaTiO 2 3 has therefore been used as a model system to mimic the phase transition behaviour of (Mg,Fe)SiO . The stable high temperature cubic 3 phase (space group Pm3m) of CaTiO3 transforms to a tetragonal phase (space group I 4/ mcm ) at T ~1636 K which can be visualized to occur through condensation of one of the components of the R 4+ mode at q = ‰ ‰ ‰ of the cubic Brillouin zone culminating in anti-phase rotated oxygen octahedra in the a a c 0 0 -tilt system 12-13 . This tetragonal phase on further cooling below T = 1512 K transforms to an orthorhombic phase in the Pnma space group with a b a tilt system -+ -14-15 due to the condensation of M 3+ mode 12-13 at q= ‰ ‰ 0 involving in-phase rotation and another component of the R 4+ mode causing additional anti-phase rotation of the octahedra. This orthorhombic phase is widely accepted as the stable phase of CaTiO 3 12-13 under ambient conditions and of MgSiO 3 under the lower mantle conditions . From an analysis of the temperature dependent 4 neutron diffraction patterns of CaTiO within the stability field of the orthorhombic Pnma 3 phase, we present evidence for discontinuous changes in the unit cell volume, pseudomonoclinic cell distortion angle and the in-phase as well as anti-phase tilt angles in between 1000-1069K indicative of a first order isostructural phase transition that does not involve any change of space group and/or Wyckoff site symmetries. We show that
this transition is phonon driven as the amplitudes of the R 4+ and M 3+ phonon modes, obtained from amplimode analysis of neutron powder diffraction data and first principles calculations, exhibit anomalous non-analytic behavior at the transition temperature. We also present results of Landau theory to show how a synergy between the two order parameters emerging at higher temperature phase transitions involving in-phase and antiphase tilts of oxygen octahedra can lead to such an isostructural phase transition.
In CaTiO3 12 , as also in MgSiO 3 16 , the hhh reflections are doublet and hence the equivalent elementary perovskite cell can be taken to be pseudomonoclinic (for more details, see the structure refinement section of the supplementary file). The pseudomonoclinic cell parameters of CaTiO3 at room temperature show the relationship 17 : a pm =cpm<bpm, α = =90 γ 0 ≠β . The temperature variation of the elementary perovskite cell parameters a pm = c pm , b pm and the monoclinic angle β pm of CaTiO , as obtained from 3 Rietveld refinements, shown in Fig 1 reveals that the relationship a pm =cpm<bpm changes to a pm =cpm>bpm somewhere between 1000 to 1069K. Further, the monoclinic angle γ pm changes discontinuously in between 1000-1069K (see inset 'a' in Fig. 1). The unit cell volume also shows a discontinuous change in the same temperature range (see inset 'b' in Fig. 1). These discontinuous changes in the unit cell parameters occur within the stability field of the orthorhombic phase in the Pnma space group, below the cubic to tetragonal and tetragonal to orthorhombic phase transitions at 1636 and 1512K, respectively, and therefore represent an isostructural phase transition (ISPT), which is quite rare in nature 9,18,19 .
In the a b -+ a tilt system of the Pnma phase of CaTiO , the tilt angles for the anti-phase -3 rotations (a ) of the oxygen octahedra about the pseudocubic [100] and [001] axes are -p p equal and are equivalent to a single tilt through an angle φ about one of the two diad axes parallel to the [101] and [101] directions, where the subscript p refers to the ideal high p p symmetry elementary perovskite cell. The in-phase octahedral rotation (b ) through an + angle ζ is about the [010] axis. Both the tilt angles, p φ and ζ , can be determined from the coordinates of the apical oxygen O , which changes from 0.25, 0, 0.75 in the cubic phase I to 0.25+u, v, 0.75-w in the orthorhombic structure, using the relationships given in Ref 2021 : It is evident from Fig 2 (a, b), which depicts the temperature variation of the antiphase ( φ ) and in-phase ( ζ ) tilt angles, that both the angles undergo a discontinuous change in between 1000-1069K similar to the unit cell parameters shown in Fig 1. Since the tilt angles are the macroscopic order parameters for the AFD phase transition, their discontinuous change in between 1000-1069K in Fig 2 (a, b) confirms the existence of a first order AFD transition of isostructural type. To understand the likely role of phonons in the AFD isostructural transition, we also determined the amplitudes of the R 4+ and M3+ modes of the cubic perovskite structure 22-23 from the refined positional coordinates using the Amplimodes suite 24-25 . We find that the amplitudes of R 4+ and M 3+ modes change discontinuously at the isostructural phase transition (see Fig 2 (a, b). This suggests the role of phonons in this transition, as confirmed further using first principles calculations.
For a system with more than one unstable mode, there could be competition as well as coupling between the modes to drive the system to the most stable structure. The amplitude of the unstable mode, which is a measure of deformation of the high
symmetric high temperature structure that yields the ground state structure, provides important information about the structural phase transitions. For a given structure, the amplitude of unstable mode shows a continuous behaviour as a function of temperature or external field, but shows a non-analytic behaviour close to a phase transition. The unstable modes at M and R points in the Brillouin zone of cubic CaTiO3 'freeze-in' or distort the structure leading to the orthorhombic phase at low temperatures. Our calculations of the cubic structure show that the M and R modes are unstable in CaTiO , 3 in agreement with the findings of previous workers 26-27 , and their frequency decreases monotonically with temperature as shown in Fig.3 (a). To understand the behaviour of M and R mode amplitudes and associated phase transitions with temperature, as observed in our experiments, we allow the system to relax to minimum energy by freezing the unstable zone boundary phonon modes using a 2 x 2 x 2 supercell. The energy profile as a function of atomic displacements along the soft eigen modes provides information about the mode amplitude (of atomic displacements) in the minimum energy structure. We determined energy profile as a function of atomic displacements for the experimental structural parameters at different T (see Fig. 2 (a, b) of Extended data figure file). We find that energy gain by freezing of the M 3+ and R4+ modes decreases with the increase of temperature (a similar trend can be seen in the increase of phonon frequencies for these modes with temperature in Fig 3(a)). The amplitudes of M 3+ and R 4+ modes decrease continuously with increasing temperature but show a non-analytical behaviour with a discontinuous change in between 1000 to 1069K (see Fig. 3(a)), consistent with our experimental observations shown in Fig. 2(a, b). Above the isostructural phase transition
temperature, the temperature dependence of the two amplitudes are continuous within the stability limit of the Pnma phase.
In complement to first-principles demonstration of the discontinuity in R 4+ and M 3+ phonon modes across the isostructural transition occurring somewhere in between 1000 to 1069K (as a result of observed discontinuity in volume), we now use phenomenological Landau theory and show that this isostructural phase transition originates from the synergistic coupling between order parameters emerging at higher temperature transitions. From the Landau free energy of higher temperature phase transitions (see Supplementary file for complete derivation), we obtain free energy as a function of x and y, the two order parameters related to the R 4+ mode (along (0,a,a)) and M3+ mode (along (a,0,0)), respectively 28 .
$$& F = e _ { 1 } x + \frac { a _ { 1 } } { 2 } ( \frac { T } { T _ { \text{C} } } + p ) x ^ { 2 } + \frac { b _ { 1 } } { 3 } x ^ { 3 } + \frac { c _ { 1 } } { 4 } x ^ { 4 } \\ & + e _ { 2 } y + \frac { a _ { 2 } } { 2 } ( \frac { T } { T _ { \text{H} } } + p ) y ^ { 2 } + \frac { b _ { 2 } } { 3 } y ^ { 3 } + \frac { c _ { 2 } } { 4 } y ^ { 4 } \\ & + d _ { 1 } x ^ { 2 } y + d _ { 2 } x y ^ { 2 } + d _ { 3 } x y + d _ { 4 } x ^ { 2 } y ^ { 2 },$$
where {e1, a , b , c } are the Landau coefficients for the order parameter (x) associated 1 1 1 with the R4+ mode (along (0,a,a)), and {e , a , b , c } are the Landau coefficients of the 2 2 2 2 order parameter (y) associated with M 3+ mode [along (a,0,0)], and d 's (i=1 to 4) are the i coupling constants. Renormalization of the coefficients of the quadratic and linear terms (refer to Supplementary file) gives the effective transition temperatures of T C and T H. We fit them to 1000 K to reproduce the experimentally observed isostructural phase transition (refer to Supplementary file). The values of parameters were fixed at e = 0.3, 1
a1 =1, b = 2, c = 1, T 1 1 C= 1000 K and p= 0, and e2= 0.1, a2 = 1, b = 2, c = 1, T 2 2 H= 1000 K and q= 0, such that the value of free energy functions of the decoupled x and y order parameters is 0 at T C and TH, respectively. The coupling coefficients d , d , d , d are 0.3, 1 2 3 4 0.1, 0.4 and 0.1 respectively. Equilibrium values of x and y are obtained through minimization of free energy at each temperature 29 and are depicted in Fig. 3 (b), which shows a clear discontinuity in both the x and y order parameters near 1000K, in agreement with experimental observations. Thus, the isostructural phase transition is a result of nonlinear coupling between two distinct three-component order parameters that have already broken the symmetry, and the richness in the symmetry equivalent structures possible.
We believe that the discovery of the isostructural phase transition in CaTiO will 3 encourage search for a similar transition within the stability field of the orthorhombic perovskite phase of MgSiO3/(Mg,Fe)SiO3 that may hold clue to the hitherto unexplained geophysical phenomena in the upper part of the earth's lower mantle consisting of 5 perovskite phase of (Mg,Fe)SiO predominantly. 3
Methods: The details of sample preparation, neutron data collection, Rietveld refinements and Amplimode analysis, first principles calculations and Landau theory are given in the supplementary files.
## References
- 1. Peyronneau, J. & Poirer, J. P. Electrical conductivity of the Earth's lower mantle. Nature 342 , 537-539, 1989.
- 2. Oganov, A. R. & Ono, S. Theoretical and experimental evidence for a post-perovskite phase of MgSiO3 in Earth's D" layer. Nature 430 , 445-448 (2004).
- 3. Murakami, M., Hirose, K., Kawamura, K., Sata, N. & Ohishi, Y. Post-Perovskite Phase Transition in MgSiO3. Science 304 , 855-858 (2004).
- 4. Stixrude, L. & Cohen, R. E. Stability of orthorhombic MgSiO perovskite in the 3 Earth's lower mantle. Nature 364 , 613-616 (1996).
- 5. Revenaugh, J. & Jordan, T. H. Mantle layering from ScS reverberations 2. The transition zone. J. Geophysical Research. 96 , 763-779 (1991).
- 6. Lay, T. Structure of the core-mantle transition zone: a chemical and thermal boundary layer. EOS 70 , 49, 54-55, 58-59 (1989).
- 7. Kawakatsu, H. & Niu, F. Seismic evidence for a 920-km discontinuity in the mantle. Nature 371 , 301-305 (1994).
- 8. Singh, A., Pandey, V., Kotnala, R. K. & Pandey, D. Direct Evidence for Multiferroic Magnetoelectric Coupling in 0.9BiFeO -0.1BaTiO3 3 . Phys. Rev. Lett. 101 , 247602-1247601-4 (2008).
- 9. Galasso, F. S. Structure, properties, and preparation of perovskite-type compounds. Pergamon press 1969.
- 10. Salje, E. Characteristics of perovskite-related materials. Phil. Trans. R. Soc. Lond. A 328 , 409-416 (1989).
- 11. Gautason, B. & Muehlenbachs, K. Oxygen diffusion in perovskite: implications for electrical conductivity in the lower mantle. Science 260 , 518-521 (1993).
- 12. Ali, R. & Yashima, M. J. Space group and crystal structure of the Perovskite CaTiO3 from 296K to 1720K. Solid State Chem. 178 , 2867-2872 (2005).
- 13. Yashima, M. & Ali, R. Structural phase transition and octahedral tilting in the calcium titanate perovskite CaTiO . Solid. State. Ionics. 3 180 , 120-126 (2009).
- 14. Kennedy , B. J., Howard, C. J. and Chakoumakos, B. C. Phase transitions in perovskite at elevated temperatures - a powder neutron diffraction study. J. Phys.: Condens. Matter 11 , 1479-1488 (1999).
- 15. Glazer, A.M. Simple Ways of Determining Perovskite Structures. Acta Crystallogr. Sec. A 31, 756-762 (1975).
- 16. Meade, C., Kwang Mao, H. & Hu, J. High-Temperature Phase Transition and Dissociation of (Mg,Fe)SiO3 Perovskite at Lower Mantle Pressures. Science 268, 17431745 (1995).
- 17. Ball , C. J., Begg, B. D., Cookson, D.J., Thorogood, G. J. & Vance, E. R. Structures in the System CaTiO3/SrTiO3. J. Solid. State. Chem. 139 , 238-247, (1998).
- 18. Lee, S., Pirogov, A., Kang, M., Jang, K.-H., Yonemura, M., Kamiyama, T., Cheong, S.-W., Gozzo F., Shin N., Kimura H., Noda Y., & Park, J,-G. Giant magneto-elastic coupling in multiferroic hexagonal manganites, Nature 451 , 805-808 (2008).
- 19. Bhattacharjee, S., Taji, K., Moriyoshi, C., Kuroiwa, Y., & Pandey, D. Temperature induced isostructural phase transition, associated large negative volume expansion, and the existence of a critical point in the phase diagram of the multiferroic (1-x)BiFeO 3 xPbTiO3 solid solution system, Phys. Rev. B 84 , 104116 (1-11) 2011
- 20 Ranjan, R., Pandey, D., Siruguri, V., Krishna, P. S. R. & Paranjape, S. K. Novel structural features and phase transition behaviour of (Sr 1-x Cax)TiO3: I. Neutron diffraction study. J. Phys.: Condens. Matter 11 , 2233-2246 (1999).
- 21. Ranjan, R. Ph.D. Thesis of B.H.U., Varanasi, India.
- 22. Darlington, C. N. W. Normal-mode analysis of the structures of perovskites with tilted octahedra. Acta. Cryst. A 58 , 66-71 (2002).
- 23. Darlington, C. N. W. Normal-mode analysis of the structures of perovskites with tilted octahedra. Erratum. Acta. Cryst. A 58 , 299-300 (2002).
- 24. Aroyo, M. I., Kirov, A., Capillas, C., Perez-Mato, J. M. and Wondratschek. H. Bilbao Crystallographic Server II: Representations of crystallographic point groups and space groups. Acta Cryst. A 62 , 115-128 (2006).
- 25. Aroyo, M. I., Perez-Mato, J. M., Capillas, C. Kroumova, E., Ivantchev, S., Madariaga, G., Kirov, A. & Wondratschek, H. Bilbao Crystallographic Server I: Databases and crystallographic computing programs. Zeitschrift fuer Kristallographie, 221 , 1, 15-27 (2006).
- 26. Cockayne, E. & Burton, B. P. Phonons and static dielectric constant in CaTiO from 3 first principles. Phys. Rev. B 62 , 3735-3743 (2000).
- 27. Tripathi, S., Kumar, A., Waghmare, U. V. & Pandey, D. Effect of NaNbO 3 substitution on the quantum paraelectric behavior of CaTiO . Phys. Rev. B 3 81 , 212101222108 (2010).
28. iso.byu.edu. Stokes, Harold. T., Hatch, Dorian. M., & Campbell, Branton. J. ISOTROPY Software Suite.
- 29. Wolfram Research, Inc., Mathematica, Version 7.0, Champaign, IL (2008). Registered Name: Waghmare, Umesh. V, Organization: Jawarharlal Nehru Centre for Advanced Scientific Research, Licence Number: L3283-6940.
Supplementary Information is linked to the online version of the paper at www.nature.com/nature.
Author Contributions: D. P. conceived the problem and applied the concept of isostructural phase transition, supervised the analysis of the neutron diffraction patterns
as also the theoretical calculations and prepared the final version of the manuscript. ST analysed the neutron diffraction patterns and prepared the first draft of the paper. MY provided the neutron diffraction data collected on a sample prepared in his laboratory. The details of sample preparation are given in Ref. No. 12. The data used in Ref. No. 12 was reanalyzed by ST leading to the discovery of the isostructural phase transition in the orthorhombic phase of CaTiO3. AK and UW carried out first principle calculations while SS and UW carried out the Landau theory calculations. AK, SS and UW prepared the first draft of the theoretical portion of the manuscript.
## Figure Legends
Fig. 1 Variation of equivalent elementary perovskite cell parameters a pm= c pm , b pm with temperature. Inset shows variation of (a) monoclinic angle β pm and (b) volume of CaTiO3 with temperature. Error bars are displayed in the figures.
Fig. 2 Variation of (a) inphase tilt angle ( ζ ) and amplitude of M 3+ (A) primary mode, (b) antiphase tilt ( φ ) and amplitude of R 4+ (A) primary mode of Pnma phase of CaTiO3 with temperature. Error bars for tilt angles are displayed in the figures.
Fig. 3 (a) Variation of frequency ( ω ) and amplitude (A) of soft R and M primary modes for cubic structure of CaTiO with temperature, (b) 'x' and 'y' order parameters with 3 temperature 29 . Note that since the values of 'x' and 'y' are negative, here we have plotted the absolute values.
Fig. 1
Fig. 2
-1
c
Fig. 3
## Extented data
Extended data Fig. 1 (a, b, c, and d) depicts the observed, calculated and difference profiles obtained after final cycle of refinement for T= 296K, 868K, 1069K and 1423K using Pnma space group.
Extended data Fig. 2 (a, b) Energy profile as a function of atomic displacements for the experimental structural parameters at different Temperature.
## Supplementary Information
## Structure refinement
The Rietveld refinements using the neutron powder diffraction patterns of CaTiO 3 were performed by FULLPROF suite . The complete Fullprof suite and documentation 1 is available at the web . The peak shape was assumed to be pseudo-Voigt function with 2 asymmetry. The background of each profile was approximated by a 6 coefficient polynomial. Except for the occupancy parameters of the atoms, which were kept at their nominal composition, all other parameters, i.e., scale factor, zero correction, background, half width parameters along with mixing parameters, lattice parameters, positional coordinates, and thermal parameters, were refined.
Amplitudes of phonon modes were calculated using AMPLIMODES package from Bilbao Crystallographic Server 3,4 .
The asymmetric unit of the orthorhombic phase of CaTiO3 in the Pnma space group consists of 4 atoms, viz Ca, Ti, O1 (apical oxygen) and O2 (planar oxygen) as
| Ca | 1 | 4c | 0.0+/-u | 0.25 | 0+/-w |
|------|-----|------|-----------|---------|----------|
| Ti | 1 | 4b | 0.0 | 0.0 | 0.5 |
| O | 1 | 8d | 0.25+/-u | 0.0+/-v | 0.75+/-w |
| O | 2 | 4c | 0.50+/-u | 0.75 | 0+/-w |
There are 7 refineable atomic xyz coordinates. The coordinates of all the atoms in the asymmetric unit at various temperatures from 296K to 1423K were obtained by Rietveld refinements. Fig. 1 (a, b, c, and d) of the Extended data figure file, depicts the observed, calculated and difference profiles obtained after final cycle of refinement for T= 296K,
868K, 1069K and 1423K using Pnma space group. Evidently, all the fits are very good as can be seen from the near flatness of the difference profile.
In order to explain the isostructural phase transition we have expressed orthorhombic structure (supercell) of CaTiO with Pnma space group having a b a tilt system into 3 -+ -pseudomonoclinic elementary perovskite cell (subcell). The relationships between the orthorhombic cell parameters A , B 0 0 and C0 and the elementary pseudocubic perovskite cell parameters a pm, b pm , and c pm are : (a pm) = (c 2 pm) = (A /2) + (C /2) , A 2 0 2 0 2 0 2 = a pm 2 + cpm 2 -2a pmcpmcos β , and C 0 2 = a pm 2 + c pm 2 +2apmcpmcos β , and B0=2bpm, where β pm is the angle between the a pm and c pm axes. The other two angles are α pm= γ pm=90 0 ≠ β pm making the elementary cell pseudomonoclinic. For A =C , the elementary perovskite cell 0 0 axes are orthogonal to each other. The hhh type reflections are doublet and singlet for the monoclinic and orthogonal cells, respectively. In CaTiO , as also in MgSiO , the hhh 3 3 reflections are doublet and hence the equivalent elementary perovskite cell can be taken to be pseudomonoclinic.
## Mode Crystallography:
Mode crystallographic analysis was done using AMPLIMODE suite at Bilbao Crystallographic Server 3,4 , using Rietveld refined coordinates. AMPLIMODES is a computer program which calculates the amplitude of symmetry adapted modes for the low symmetry distorted structure with respect to a high symmetry structure. These modes are given in relative units using atomic displacements of the atoms in the asymmetric unit of the distorted phase.
In order to work with the AMPLIMODE analysis, we need to know the parent [or high symmetry (HS)] structure, the [low symmetry structure (LS)] and the transformation matrix for the two phases, and the symmetry adapted primary and secondary modes that transform the HS phase into the LS phase. In the case of CaTiO , the high symmetry 3 structure is cubic with space group Pm 3 m , where A site cation, B site cation and O go to 1b, 1a and 3d Wyckoff sites, with fractional coordinates 0.5 0.5 0.5, 0.0 0.0 0.0 and 0.5 0.0 0.0, respectively.
The transformation matrices relating the unit cell of the low symmetry room temperature phase in the Pnma space group with the unit cell of the high symmetry cubic phase of CaTiO3 is given below:
$$[ \begin{matrix} 1 & 0 & 1 ] \, [ & 1 / 2 ] \\ [ & 0 & 2 & 0 ] \, [ & 0 ] \\ [ & - 1 & 0 & 1 ] \, [ & 1 / 2 ] \end{matrix} ]$$
The Wyckoff sites of the cubic phase get split via condensation of various symmetry adapted modes to give the Pnma space group for the room temperature structures of CaTiO . It has been shown that seven modes are required to arrive at the 3 Pnma structure from the Pm 3 m structure 5,6 . The irreps corresponding to these seven modes are: octahedral tilting modes R4+ (anti-phase tilt) and M3+ (in-phase tilt), octahedral distortion modes X5+, R5+ and M2+, and A-site cation displacement modes R5+ and X5+ . Using the package 'Symmodes' at the Bilbao Crystallographic server , we 7 8 find that the displacive phase transition from cubic space group Pm 3 m to the orthorhombic space group Pnma requires the splitting of the cubic Wyckoff sites through these vibrational modes in the following manner:
- (i) Ca at 1b site of Pm3m goes to 4c site via R5+(1) and X5+(1) symmodes with coordinates as (0.00+u, 0.25, 0.00-w),
- (ii) O at 3d site goes to O1 at 8d (0.25+u, 0.00+v, 0.75-w) and O2 at 4c (0.50+u, 0.75, 0.00-w) sites via R4+(1), R5+(1), X5+(1), M2+(1), and M3+(1) symmodes and
- (iii) Ti at 1a site remains undisplaced from its position and occupies 4b site with coordinates (0.00, 0.00, 0.50),
The above representations of irreps are obtained from Bilbao crystallographic server. Thus this process gives us amplitudes of phonon modes using Rietveld refined coordinates of low symmetry structure.
## First principle calculations
Density functional theory calculations were performed using ABINIT code and 9 local density approximation (LDA) for the exchange-correlation functional. We used norm conserving pseudopotentials with an energy cutoff 40 Ha and 6 × 6 × 6 MonkhorstPack mesh for the sampling of the cubic Brillouin zone. We approximate the experimental lattice constants of cubic structure for the different values of temperature (T) from the volume of orthorhombic unit cells and performed calculations using these approximated cubic lattice constants. The phonon frequencies are the eigen values of the dynamical matrix D:
$$\mathbf D _ { i \alpha j \beta } = ( 1 / \sqrt { m _ { i } } m _ { j } ) \mathbf x \left ( \partial ^ { 2 } \mathbf E / \partial \mathbf x _ { i \alpha } \partial \mathbf x _ { j \beta } \right )$$
where m is the mass of atom i, x i i α is the displacement of atom i along cartesian direction α and E is total energy of the system. The phonon dispersion relations and soft mode eigen vectors are obtained from the linear response calculations in a cubic unit cell. The
phonon frequencies are proportional to the curvature of energy surface as a function of small displacement of atoms from their equilibrium positions (see Eq. 1). The unstable modes, which correspond to the negative curvature of the energy surface, are a signature of the structural phase transitions. The presence of unstable phonon modes shows that system can lower its energy from such displacement patterns, which also lowers the symmetry of structure leading to the structural phase transitions. The more is the curvature of energy surface for a unstable mode, larger is the probability of system following that displacement pattern.
## Phenomenological theory
R4+ and M3+ are the two modes that lead to the transition of cubic Pm-3m phase to the Pnma phase. R 4+ and M 3+ modes correspond to tilting of octahedra, and are triply degenerate. Let {x 1, x 2, x 3 } and {y1, y 2, y 3 } be the order parameters associated with the R4+ and M3+ modes, respectively. The free energy functional, F(Pm-3m) with Pm-3m as the parent structure and all the six order paramters is given by,
$$& F ( P m - 3 m ) = g _ { 1 2 } ( T ) ( x _ { 1 } ^ { 2 } + x _ { 2 } ^ { 2 } + x _ { 3 } ^ { 2 } ) + g _ { 1 4 } ( x _ { 1 } ^ { 2 } + x _ { 2 } ^ { 2 } + x _ { 3 } ^ { 2 } ) ^ { 2 } + g _ { 2 4 } ( x _ { 1 } ^ { 4 } + x _ { 2 } ^ { 4 } + x _ { 3 } ^ { 4 } ) \\ & + f _ { 1 2 } ( T ) ( y _ { 1 } ^ { 2 } + y _ { 2 } ^ { 2 } + y _ { 3 } ^ { 2 } ) + f _ { 1 4 } ( y _ { 1 } ^ { 2 } + y _ { 2 } ^ { 2 } + y _ { 3 } ^ { 2 } ) ^ { 2 } + f _ { 2 4 } ( y _ { 1 } ^ { 4 } + y _ { 2 } ^ { 4 } + y _ { 3 } ^ { 4 } ) \\ & + h _ { 1 4 } ( x _ { 1 } ^ { 2 } + x _ { 2 } ^ { 2 } + x _ { 3 } ^ { 2 } ) ( y _ { 1 } ^ { 2 } + y _ { 2 } ^ { 2 } + y _ { 3 } ^ { 2 } ) + h _ { 2 4 } ( x _ { 1 } ^ { 2 } y _ { 1 } ^ { 2 } + x _ { 2 } ^ { 2 } y _ { 2 } ^ { 2 } + x _ { 3 } ^ { 2 } y _ { 3 } ^ { 2 } ).$$
Freezing in of the R 4+ mode along (a,0,0) gives the I4/mcm phase. Hence, substituting x1 = xc (constant) and neglecting the constant terms in equation S1 we get,
$$& F ( I 4 / m c m ) = ( g _ { 1 2 } ( T ) + 2 g _ { 1 4 } x _ { c } ^ { \ 2 } ) ( x _ { 2 } ^ { \ 2 } + x _ { 3 } ^ { \ 2 } ) + g _ { 1 4 } ( x _ { 2 } ^ { \ 2 } + x _ { 3 } ^ { \ 2 } ) ^ { 2 } + g _ { 2 4 } ( x _ { 2 } ^ { \ 4 } + x _ { 3 } ^ { \ 4 } ) \\ & + ( f _ { 1 2 } ( T ) + 2 h _ { 1 4 } x _ { c } ^ { \ 2 } ) ( y _ { 1 } ^ { \ 2 } + y _ { 2 } ^ { \ 2 } + y _ { 3 } ^ { \ 2 } ) + f _ { 1 4 } ( y _ { 1 } ^ { \ 2 } + y _ { 2 } ^ { \ 2 } + y _ { 3 } ^ { \ 2 } ) ^ { 2 } + f _ { 2 4 } ( y _ { 1 } ^ { \ 4 } + y _ { 2 } ^ { \ 4 } + y _ { 3 } ^ { \ 4 } ) \\ & + h _ { 1 4 } ( x _ { 2 } ^ { \ 2 } + x _ { 3 } ^ { \ 2 } ) ( y _ { 1 } ^ { \ 2 } + y _ { 2 } ^ { \ 2 } + y _ { 3 } ^ { \ 2 } ) + h _ { 2 4 } ( x _ { c } ^ { \ 2 } y _ { 1 } ^ { \ 2 } + x _ { 2 } ^ { \ 2 } y _ { 2 } ^ { \ 2 } + x _ { 3 } ^ { \ 2 } y _ { 3 } ^ { \ 2 } ).$$
Below 1512 K, the R4+ and M3+ modes freeze into the tetragonal I4/mcm structure along (0,a,a) and (a,0,0) respectively, and give rise to the orthorhombic Pnma phase. Thus, the
temperature dependence enters F through g 12 (T)= (T/1512)-1 and f 12 (T)= (T/1512)-1. Since the isostructural phase transition involves freezing in of both R 4+ and M 3+ modes along (0,a,a) and (a,0,0) directions respectively, the order parameters x and y are variable for T <1512 K. Hence substituting x =x = x + x , and y1= y + y and y =y =0 (x and y 2 3 o o 2 3 o o are constants), equation S2 reduces to,
$$& a v \text{comma} \mathfrak { y }, \text{equation} \mathfrak { z } \text{in} \colon w, \\ & F ( P n m a ) = [ 4 x _ { o } ( g _ { 1 2 } ( T ) + 2 g _ { 1 4 } x _ { c } ^ { 2 } ) + 4 x _ { o } ^ { 3 } ( 4 g _ { 1 4 } + 2 g _ { 2 4 } ) + 4 h _ { 1 4 } x _ { o } y _ { o } ] x \\ & + [ 2 g _ { 1 2 } ( T ) + 2 g _ { 1 4 } x _ { c } ^ { 2 } + 6 x _ { o } ^ { 2 } ( 2 g _ { 1 4 } + 2 g _ { 2 4 } + h 1 4 y _ { o } ^ { 2 } ) ] x ^ { 2 } \\ & + [ 4 x _ { o } ( g _ { 1 4 } + 2 g _ { 2 4 } ) ] x ^ { 3 } + ( 4 g _ { 1 4 } + 2 g _ { 2 4 } ) x ^ { 4 } \\ & + [ ( f _ { 1 4 } + f _ { 2 4 } ) 4 y _ { o } ^ { 3 } + ( f _ { 1 2 } ( T ) + h _ { 1 4 } x _ { c } ^ { 2 } ) ^ { 2 } y _ { o } + 2 h _ { 2 4 } x _ { c } ^ { 2 } y _ { o } + 4 h _ { 1 4 } x _ { o } ^ { 3 } ] y \\ & + [ f _ { 1 2 } ( T ) + h _ { 1 4 } x _ { c } ^ { 2 } + 6 ( f _ { 1 4 } + f _ { 2 4 } ) y _ { o } ^ { 2 } + 2 h _ { 1 4 } x _ { o } ^ { 2 } ] y ^ { 2 } + [ 4 y _ { o } ( f _ { 1 4 } + f _ { 2 4 } ) ] y ^ { 3 } \\ & + ( f _ { 1 4 } + f _ { 2 4 } ) y ^ { 4 } \\ & + ( 8 h _ { 1 4 } x _ { o } y _ { o } ) x y + ( 4 h _ { 1 4 } y _ { o } ) x ^ { 2 } y + ( 4 h _ { 1 4 } x _ { o } ) x y ^ { 2 } + 2 h _ { 1 2 } x ^ { 2 } y ^ { 2 }$$
+
2
2
2
2
14
12
14
14
2
)
4
(
)
4
(
)
8
(
y
x
h
xy
x
h
y
x
y
h
xy
y
x
h
o
o
o
o
+
+
+
## Redefining the Landau coefficients we get,
$$& F = e _ { 1 } x + \frac { a _ { 1 } } { 2 } ( \frac { T } { T _ { c } } + p ) x ^ { 2 } + \frac { b _ { 1 } } { 3 } x ^ { 3 } + \frac { c _ { 1 } } { 4 } x ^ { 4 } \\ & + e _ { 2 } y + \frac { a _ { 2 } } { 2 } ( \frac { T } { T _ { H } } + p ) y ^ { 2 } + \frac { b _ { 2 } } { 3 } y ^ { 3 } + \frac { c _ { 2 } } { 4 } y ^ { 4 } \\ & + d _ { 1 } x ^ { 2 } y + d _ { 2 } x y ^ { 2 } + d _ { 3 } x y + d _ { 4 } x ^ { 2 } y ^ { 2 },$$
Since the I4/mcm to Pnma phase transition is relevant to equation S4, the critical temperature should be TC = 1512 K. However all the Landau coefficients are renormalized, and hence the temperature at which x and y order parameters will freeze into the structure is no longer 1512 K. The critical temperature is now replaced by effective temperature, which is T C= TH= 1000 K. Note that, we fit the T C= TH values to 1000 K to mimic the isostructural phase transition in between 1000-1100K.
Reference :
- 1. J. Rodríguez-Carvajal, Physica B 192 , 55 (1993).
- 2. http://www.ill.eu/sites/fullprof/.
- 3. Aroyo, M. I., Kirov, A., Capillas, C., Perez-Mato, J. M. and Wondratschek. H. Bilbao Crystallographic Server II: Representations of crystallographic point groups and space groups. Acta Cryst. A 62 , 115-128 (2006).
- 4. Aroyo, M. I., Perez-Mato, J. M., Capillas, C. Kroumova, E., Ivantchev, S., Madariaga, G., Kirov, A. & Wondratschek, H. Bilbao Crystallographic Server I: Databases and crystallographic computing programs. Zeitschrift fuer Kristallographie, 221 , 1, 15-27 (2006).
- 5. C. N. W. Darlington, Acta. Cryst. A 58 , 66(2002(a)).
- 6. C. N. W. Darlington, Acta. Cryst. A 58 , 299(2002(b)).
- 7. K. S. Knight and C. N. W. Darlington, Powder Diffr 20 , 7(2005).
- 8. C. Capillas, E. Kroumova, M. I. Aroyo, J. M. Perez-Mato, H. T. Stokes, and D. M. Hatch, J. Appl. Cryst. 36 , 953(2003).
- 9. http://www.abinit.org/. | null | [
"Saurabh Tripathi",
"Anil Kumar",
"Sharmila Shirodkar",
"Masatomo Yashima",
"U. V. Waghmare",
"Dhananjai Pandey"
] | 2014-08-02T04:09:38+00:00 | 2014-08-02T04:09:38+00:00 | [
"cond-mat.mtrl-sci"
] | Discovery of an isostructural phase transition within orthorhombic phase field of CaTiO3 | Earths lower mantle extending from 670 to 2,990 km deep is predominantly
composed of a perovskite-type (Mg,Fe)SiO3 phase1,2. The perovskite phase
undergoes a structural phase transition to a post-perovskite phase responsible
for D" layer seismic discontinuity2,3 at about 2690 km depth in the lowermost
region of the lower mantle. However, structural basis of other seismic
discontinuities occurring in the upper region of the lower mantle (700 km to
1,200 km deep) remains unexplained4-7, as no apparent change in the crystal
symmetry of the orthorhombic perovskite phase has been reported5. We present
here unambiguous evidence for a non-apparent isostructural phase transition8 in
the stable orthorhombic perovskite phase of CaTiO3 which may have relevance to
phase transitions in the perovskite phase of (Mg,Fe)SiO3 also, as both the
compounds have similar structure, tolerance factor and thermochemical
properties9-11. Our results are based on the analysis of neutron powder
diffraction patterns using Rietveld and mode crystallography techniques and are
supported by density functional and Landau theory calculations. The present
results on CaTiO3 would encourage search for isostructural phase transition in
the perovskite phase of (Mg,Fe)SiO3 that may provide clue to the unexplained
geophysical phenomena in the upper part of the earths lower mantle. |
1408.0336v1 | ## Silver environment and covalent network rearrangement in GeS -Ag glasses 3
L. Rátkai , I. Kaban 1 2,3 , T. Wágner , J. Kolář , S. Valková , Iva V oleská , B.Beuneu , P. Jóvári 4 4 4 4 5 1
- 1 Wigner Research Centre for Physics, Institute for Solid State Physics, H-1525 Budapest, POB 49, Hungary
- 2 TU Dresden, Institute of Materials Science, D-01062 Dresden, Germany
- 3 IFW Dresden, Institute for Complex Materials, P.O. Box 270116, D-01171 Dresden, Germany
- 4 University of Pardubice, Faculty of Chemical Technology, Centre of Materials and Nanotechnology,
- Čs. legií 565 square, 53210 Pardubice, Czech Republic
- 5 Laboratoire Léon Brillouin, CEA-Saclay, 91191 Gif sur Yvette Cedex, France
## Abstract
The structure of Ag-doped GeS glasses (0, 15, 20, 25 at.% Ag) was investigated by diffraction 3 techniques and extended X-ray absorption fine structure measurements. Structural models were obtained by fitting the experimental datasets simultaneously by the reverse Monte Carlo simulation technique. It is observed that Ge has mostly S neighbours in GeS , but Ge-Ge bonds appear already at 3 15% Ag content. Sulphur has ~2 S/Ge neighbours over the whole concentration range, while the S-Ag coordination number increases with increasing Ag content. Ag-Ag pairs can already be found at 15% Ag. The Ag-S mean coordination number changes from 2.17±0.2 to 2.86±0.2 between 15% and 25% Ag content. Unlike As-S network in AsS -25Ag glass, the Ge-S network is not fragmented upon Ag2 doping of GeS glass. 3
## 1. Introduction
Memristive switching [1] has attracted high attention in view of development of advanced nonvolatile memories. There exist several concepts of resistance-change memory such as for example electric switching in chalcogenide glasses [2], resistive switching in transition-metal oxides [3,4], and resistance change in solid electrolytes [3-8]. The first concept utilizes amorphous-to-crystalline phasechange, while the second one uses valency change of transition-metal oxide upon applied electrical pulse. The third concept is based on nanoscale ionic transport and electrochemical formation or removal of nanoscale conductive pathways in solid-electrolyte matrix depending on polarity of electrodes. Devices utilizing this phenomenon are called electrochemical metallization cells (EMC) or conductive bridge memories (CBM).
Different candidate materials for application in the EMC cells are under investigation at present [3-13]. Among them are Ge-Se or Ge-S chalcogenide glasses as solid electrolyte containing Ag or Cu as active metal [5-7]. In comparios to the Ge-Se-Ag glasses, Ag-doped Ge-S glasses are able
to withstand much higher temperatures, which is essential for providing long term resistanceswitching functionality. Therefore, it appears to be a better candidate for EMC memory applications. To optimize their functionality it is important to understand the structure of Ge-S-Ag glasses. In this paper we present the results of a study on GeS -Ag glasses with 0, 15, 20, 25 at.% Ag. Atomic level 3 structural models are generated by fitting diffraction and extended X-ray absorption fine structure (EXAFS) datasets with the reverse Monte Carlo (RMC) simulation technique [14-16]. This method offers a frame for combining experimental structural information with physical/chemical knowledge (e.g. density, preferred coordination numbers) available a priori. The validity of various structural models can also be tested. This approach has already been applied to several closely related systems such as AsS -Ag [17] and As S -Ag [18], GeSe -Ag [19] and Ge-Se-In glasses [20]. 2 2 3 3
## 2. Experimental
## 2.1. Sample preparation
Four glasses of the composition (GeS ) 3 100-x Agx with x = 0, 15, 20, 25 at.% were prepared from 5N purity elements. The components of total mass of 10 g were inserted into the quartz ampoules, which were then evacuated to a pressure of 10 -3 Pa, sealed, and placed in a rocking furnace. The glasses were synthetized at the well defined heating profile with a maximum temperature of 1000°C kept for 12 hours. The ampoules were quenched in iced water and then these ampoules were annealed for 3 hours at 50 K below the respective glass transition temperature.
The mass density of the glasses was determined with accuracy of 0.15% by dual weighing by the standard Archimedean method at room temperature. The measured values are given in Table 1.
## 2.2. Experiments
Neutron diffraction measurements were carried out at the 7C2 diffractometer of Laboratoire Léon Brillouin (Saclay, France). Powder samples were filled into vanadium sample holders with 0.1 mm wall thickness and 6 mm diameter. The wavelength of incident radiation was 0.72 Å. Raw data were corrected for background scattering and detector cell efficiency and normalized following standard procedures.
Ge and Ag K-edge EXAFS spectra were recorded at beamline X1 of Hasylab, Hamburg. Measurements were carried out in transmission mode. Powder samples were mixed with cellulose and pressed into tablets. The transmission of the tablets was around 1/e at the measured absorption edge. Intensities before and after the samples were recorded with ionisation chambers filled with Ar and Kr with pressures depending on the energy of the edge. The X-ray absorption cross sections μ(E) were converted to χ k ( ) by the program Viper [21].
X-ray diffraction data were taken at the BW5 high energy X-ray diffractometer [22] (Hasylab). The energy of incident radiation was 100.0 keV (λ = 0.124 Å). Raw data were corrected for background scattering, detector deadtime, and Compton scattering [23].
## 3. Reverse Monte Carlo simulation
Large scale structural models (12000 atoms) were obtained by fitting simultaneously diffraction and EXAFS datasets. Simulations were carried out with the rmcppm code [16]. Number densities used in the models were calculated from the experimental mass densities (table 1). Minimum interatomic distances (cut offs) are listed in table 2. Backscattering amplitudes and phases needed to transform partial pair correlation functions to model χ(k) curves [24] were obtained by the feff programme [25]. Besides, minimum intertomic distances coordination constraints were also used: each Ge atom was forced to have four neighbours (either S or S/Ge, see below), each S atom had at most two Ge neighbours and no 'floating' Ag or S atoms (with 0 or 1 neighbours) were allowed.
## 4. Results and discussion
## 4.1. GeS glass 3
Structural model of glassy GeS was obtained by fitting simultaneously ND, XRD and Ge K-edge 3 EXAFS datasets. As this composition is sulfur-rich, Ge-Ge bonding was forbidden while S-S bonding was allowed during the simulation. For this composition only, the coordination number of sulfur was constrained to be 2.
Experimental curves and fits are compared in figure 1 while partial pair correlation functions are shown in figure 2. The Ge-S distance is 2.23 Å which agrees well with Ge-S bond lengths (2.202.23 Å) found in GeS [26] and in Ge-In-S, Ge-In-S-AgI and Na S-GeS glasses [27-29]. The S-S 2 2 2 distance is 2.06 Å, which is close to the value found in amorphous sulfur [30].
## 4.2. GeS -Ag glasses 3
The structure of ternary GeS -Ag alloys was investigated by fitting the four measurements (XRD, ND, 3 and EXAFS at Ge and Ag absorption edges) simultaneously. The quality of the fits is demonstrated in figure 3, while coordination numbers and nearest neighbour distances are summarized in tables 3 and 4. Selected partial pair correlation functions are shown in figure 4.
Previous studies on AsS -Ag [17], GeSe -Ag [19] and GeS -Ag S [31] glassy systems revealed 2 3 2 2 that Ag prefers S/Se and tries to avoid the network forming cation (As/Ge). Our results show that from this point of view GeS -Ag alloys behave in a similar way. Ag has on the average 2.17±0.2 sulphur 3 neighbours in GeS -15Ag, while it is coordinated by 2.86±0.2 sulfur atoms in GeS -25Ag. The Ag-S 3 3 distance is 2.53-2.57 Å, which agrees with the Ag-S bond lengths found in As S -Ag S [32], but it is 2 3 2 slightly shorter than the values found in GeS -Ag S (2.58 Å [31]) and in GeS -In S -AgI glasses (2.60 2 2 2 2 3 Å [33]). The deviations may be caused partly by experimental uncertainties, partly by the composition
dependence of Ag-S interaction. The Ag-Ag distance is around 3.0 Å. A similar value was found in As2S3 -Ag2 S [32]. On the other hand, a neutron diffraction study on isotopically substituted GeS -Ag S 2 2 gave no clear conclusion: while a peak of g AgAg ( r ) was reported around 3 Å in the RMC study of Lee et al [34], direct transformation of diffraction datasets gave no evidence of nearest neighbour silver atoms [31]. It is to be mentioned that the Ag-Ag coordination number (0.73-0.93) does not change significantly in the composition range investigated.
With increasing Ag content, the initially sulfur-rich GeS -Ag system can turn to a sulfur3 deficient state where Ge-Ge bonding may be required to satisfy the four valences of germanium. Test calculations suggest that Ge-Ge bonds can be found already in GeS -15Ag (figure 5). The Ge-Ge 3 coordination number is around 0.7±0.3, the concentration dependence is not significant.
It is to be noted that the S-S coordination number is around 1 and on the average each sulfur atom takes part in ~2 covalent bonds. The existence of S-S bonds was also tested by dedicated runs. It was found that raising the S-S cut off from 1.9 Å to 2.8 Å only slightly deteriorates the neutron diffraction and Ge K-edge EXAFS fits but drastically influences the fit of Ag K-edge EXAFS data as demomstrated in figure 6. (The reason for this change is that the elimination of S-S bonds brings about a forced increase of the Ag-S coordination number.) The total coordination number of sulphur increases from about 2 in GeS glass to 3.30±0.4 in GeS -25Ag glass. 3 3
## 4.3. Comparison of AsS -25Ag and GeS -25Ag glasses 2 3
Since Ge-S-Ag and As-S-Ag glasses have similar potential applications it may be interesting to compare the environment of silver atoms and the changes of the host covalent matrices induced by alloying. The structure of glassy AsS -25Ag was investigated by diffraction techniques and EXAFS in 2 a similar way [17]. It was found that the mean Ag-S coordination number is 3.34±0.4, while the AgAg coordination number is 0.78±0.4. On the average, the total coordination number of Ag is close to 4. The Ag-S distance is practically the same in the two glasses, but the first peak is more pronounced in AsS2-25Ag (figure 7).The Ag-Ag distance is 2.92±0.03 Å, which is close to the value found in GeS 3 25Ag (~2.96-3.01 Å). The coordination number of As is close to 3 in AsS -25Ag, and no As-As bonds 2 had to be allowed to get reasonable fits. The S-As coordination number is around 1.5, while the S-S coordination number is not higher than the sensitivity of our approach (~0.3).
Thus, while Ge-Ge bonds are formed in GeS -25Ag, AsS 3 3/2 units remain intact in AsS -25Ag. 2 The other important difference between the two glasses is that due to the nonvanishing S-S bonding each sulfur atom participates in ~2 covalent bonds in GeS -25Ag. The same number is around 1.5 in 3 AsS2-25Ag. It can be concluded that even if the chemical short range order is changed (due to the formation of Ge-Ge bonds) the connectivity of the covalent network of Ge and S atoms is not altered by the addition of Ag.
## 5. Conclusions
Short range order in GeS and GeS -Ag glasses (15, 20 and 25 at.% Ag) was investigated by neutron 3 3 and X-ray diffraction as well EXAFS measurements at the Ge and Ag K-edges. Structural models were obtained by fitting experimental datasets simultaneously with the reverse Monte Carlo simulation technique. It was found that S-S bonding is preserved even in GeS -25Ag and each S atom takes part 3 in ~2 covalent bonds. The average coordination number of sulphur is 2.64±0.3 in GeS -15Ag while it 3 is 3.30±0.3 in GeS -25Ag glass. Ge-Ge bonds appear already in GeS -15Ag composition; thus 3 3 addition of silver changes chemical short range order of the host GeS matrix. On the other hand, as 3 each sulfur atom has ~2 Ge/S neighbours, the connectivity of the covalent network of Ge and S atoms is not altered by the addition of Ag. The average coordination number of Ag is 2.90±0.3 in GeS -15Ag 3 glass and 3.79±0.3 in GeS -25Ag glass. 3
## Acknowledgments
P.J. was supported by OTKA (Hungarian Basic Research Fund) Grant No. 083529. The authors thanks to project CZ.1.07/2.3.00/20.00254 'Research Team for Advanced Non-crystalline Materials" realized by European Social Fund and Ministry of Education, Youth and Sports of The Czech Republic within The Education for Competitiveness Operational Programme for financial support.
## References
- [1] Chua L O 1971 IEEE Transactions on Circuit Theory 18 507
- [2] Ovshinsky S 1968 Phys. Rev. Lett. 21 1450
- [3] Waser R, Dittmann R, Staikov G and Szot K 2009 Adv. Mater. 21 2632
- [4] Waser R, Dittmann R, Salinga M and Wuttig M 2010 Solid-State Electronics 54 830
- [5] Kozicki M N, Gopalan C, Balakrishnan M, Park M and Mitkova M 2004 Proc. Non-Volatile Memory Tech. Symp . 10
- [6] Kozicki M N, Balakrishnan M, Gopalan C, Ratnakumar C and Mitkova M 2005 Proc. NonVolatile Memory Tech. Symp . 83
- [7] Kozicki M N and Mitkova M 2006 J. Non-Cryst. Solids 352 567
- [8] Lu W, D. Jeong D S, Kozicki M and Waser R 2012 MRS Bulletin 37 124
- [9] Frumar M and Wagner T 2003 Curr. Opinion Solid St. Mat. Sci. 7 117
- [10] Frumar M, Frumarova B and Wágner T 2011 Amorphous and Glassy Semiconducting Chalcogenides. In: Bhattacharya P, Fornari R and Kamimura H, (eds.), Comprehensive Semiconductor Science and Technology 4 206 Amsterdam: Elsevier
- [11] Stehlík Š, Kolář J, Bartoš M, Vlček M., Frumar M, Zima V and Wágner T 2010 Sol. State Ionics 181 1625.
- [12] Kolář J, Wágner T, Zima V, Stehlík Š, Frumarová B, Beneš L, Vlček M and Frumar M 2011 J. Non-Cryst. Solids , 357 2223
- [13] Stehlik Š, Shimakawa K, Wágner T and M. Frumar M 2012 J. Phys. D: Appl. Phys . 45 205304
- [14] McGreevy R L, Pusztai L 1988 Mol. Simul. 1 359
- [15] McGreevy R L 2001 J. Phys.: Condens. Matter 13 R877
- [16] Gereben O, Jóvári P, Temleitner L and Pusztai L 2007 J. Optoelectron. Adv. Mater. 9 3021
- [17] Kaban I, Jóvári P, Wágner T, Bartoš M, Frumar M, Beuneu B, Hoyer W, Mattern N and Eckert J 2011 J. Non-Cryst. Solids 357 3430
- [18] Kaban I, Jóvári P, Wágner T, Frumar S, Stehlik Š, Bartos M, Hoyer W, Beuneu B and Webb M A 2009 J. Phys.: Condens. Matter 21 395801
- [19] Kumara L S R, Ohara K, Kawakita Y, Jóvári P, Hidaka M, Nark Eon Sung N E, Beuneu B and Takeda S 2011 EPJ Web of Conferences 15 02007, DOI: 10.1051/epjconf/20111502007
- [20] Kaban I, Jóvári P, Petkova T, Petkov P, Stoilova A, Hoyer W and Beuneu B 2010 J. Phys. Condensed Matter , 22 404205
- [21] Klementev K V 2001 . J Phys . D : Appl. Phys. 34 209
- [22] Bouchard R, Hupfeld D, Lippmann T, Neuefeind J, Neumann H B, Poulsen H F, Rütt U, Schmidt T, Schneider J R, Süssenbach J and von Zimmermann M 1998 J. Synchrotron Rad. 5 90
- [23] Poulsen H, Neuefeind J, Neumann H B, Schneider J R, Zeidler M D 1995 J. Non-Cryst. Solids
- 188 63
- [24] Winterer M 2000 J. Appl. Phys . 88 5635
- [25] Ankudinov A L, Ravel B, Rehr J J and Conradson S D 1998 Phys. Rev. B 58 7565
- [26] Zeidler A, Drewitt J W E, Salmon P S, Barnes A C, Crichton W A, Klotz S, Fischer H E, Benmore C J, Ramos S and Hannon A C 2009 J. Phys.: Condens. Matter 21 474217
- [27] Sen S, Aitken B G and Khalid S 2005 J. Non-Cryst. Solids 351 1710
- [28] Chrissanthopoulos A, Jóvári P, Kaban I, Gruner S, Kavetskyy T, Borc J, Wang W, Ren J, Chen G and Yannopoulos S N 2012 J. Solid State Chem . 192 7
- [29] Itoh K, Fukunaga T 2009 Solid State Ionics 180 351
- [30] Winter R, Pilgrim W C, Egelstaff P A, Chieux P, Anlauf S and Hensel F 1990 Europhys. Lett. 11 225
- [31] Lee J H, Owens A P, Pradel A, Hannon A C, Ribes M and Elliott S R 1995 J. Non-Cryst. Solids 192-193 57
- [32] Salmon P S, Liu J 1996 J. Non-Cryst. Solids 205-207 172
- [33] Chrissanthopoulos A, Jóvári P, Kaban I, Gruner S, Kavetskyy T, Borc J, Wang W, Ren J, Chen G, Yannopoulos S N 2012 J. Sol. State Chem. 192 7
- [34] Lee J H, Owens A P, Elliott S R 1993 J. Non-Cryst. Solids 164-166 139
Table 1. Mass density and number density of the GeS -Ag glasses investigated. Accuracy of the mass 3 density is about 0.15%.
| Composition | GeS 3 | GeS 3 -15Ag | GeS 3 -20Ag | GeS 3 -25Ag |
|----------------------------|---------|---------------|---------------|---------------|
| Density (g/cm 3 ) | 2.588 | 3.628 | 4.031 | 4.347 |
| Number density (atom/Å 3 ) | 0.03692 | 0.041968 | 0.043864 | 0.044654 |
Table 2. Minimum interatomic distances (in Å) used in the reverse Monte Carlo simulation runs.
Table 3. Coordination numbers of GeS -Ag glasses obtained by the simultaneous fitting of diffraction 3 and EXAFS datasets.
| Composition | Atomic pair | Atomic pair | Atomic pair | Atomic pair | Atomic pair | Atomic pair |
|---------------|---------------|---------------|---------------|---------------|---------------|---------------|
| | Ge-Ge | Ge-S | Ge-Ag | S-S | S-Ag | Ag-Ag |
| GeS 3 | 3.1 | 2.0 | - | 1.95 | - | - |
| GeS 3 -Ag | 2.35 | 2.0 | 3.15 | 1.95 | 2.2 | 2.85 |
Table 4. Nearest neighbour distances (in Å ) calculated from the models obtained by reverse Monte Carlo simulation
| Composition | Coordination number | Coordination number | Coordination number | Coordination number | Coordination number | Coordination number | Coordination number | Coordination number | Coordination number |
|---------------|-----------------------|-----------------------|-----------------------|-----------------------|-----------------------|-----------------------|-----------------------|-----------------------|-----------------------|
| | N GeGe | N GeS | N SGe | N SS | N SAg | N AgS | N AgAg | N S | N Ag |
| GeS 3 | - | 3.90 | 1.30 | 0.86 | - | - | - | 2.16 | - |
| GeS 3 -15Ag | 0.64 | 3.20 | 1.07 | 1.06 | 0.51 | 2.17 | 0.73 | 2.64 | 2.90 |
| GeS 3 -20Ag | 0.69 | 3.30 | 1.10 | 0.96 | 0.95 | 2.86 | 0.82 | 3.01 | 3.68 |
| GeS 3 -25Ag | 0.75 | 3.25 | 1.08 | 0.95 | 1.27 | 2.86 | 0.93 | 3.30 | 3.79 |
| Composition | Mean interatomic distance | Mean interatomic distance | Mean interatomic distance | Mean interatomic distance | Mean interatomic distance |
|---------------|-----------------------------|-----------------------------|-----------------------------|-----------------------------|-----------------------------|
| | r GeGe | r GeS | r SS | r SAg | r AgAg |
| GeS 3 | - | 2.23±0.02 | 2.06±0.02 | - | - |
| GeS 3 -15Ag | 2.43±0.02 | 2.22±0.02 | 2.10±0.04 | 2.53±0.03 | 2.96±0.05 |
| GeS 3 -20Ag | 2.46±0.02 | 2.22±0.02 | 2.10±0.04 | 2.57±0.03 | 3.01±0.05 |
| GeS 3 -25Ag | 2.44±0.02 | 2.22±0.02 | 2.10±0.04 | 2.56±0.03 | 2.97±0.05 |
## Figure captions
Figure 1. Comparison of experimental and RMC-model curves of GeS glass. 3
Figure 2. Partial pair correlation functions of GeS glass. 3
Figure 3. XRD, ND, Ge and Ag K-edge measurements and RMC-model curves of GeS -25% Ag glass 3 obtained by fitting the four datasets simultaneously.
Figure 4. Partial pair correlation functions of GeS -25Ag glass. 3
Figure 5. Comparison of the fits of Ge K-edge EXAFS data for the GeS -15Ag glass with and without 3 Ge-Ge bonds ( symbols : experimental data, solid line : fit).
Figure 6. Comparison of the fits of Ag K-edge EXAFS data for the GeS -25Ag glass with and without 3 S-S bonds ( symbols : experimental data, solid line : fit).
Figure 7. Comparison of the Ag-S partial pair correlation functions in GeS -25Ag ( 3 dashes ) and AsS 2 25Ag ( solid line ) glasses.
Figure 1.
Figure 2.
Figure 3.
Figure 4.
Figure 5.
Figure 6.
Figure 7.
 | 10.1088/0953-8984/25/45/454210 | [
"László Rátkai",
"Ivan Kaban",
"Tomáš Wágner",
"Jakub Kolář",
"Silviya Valková",
"Iva Voleská",
"Brigitte Beuneu",
"Pál Jóvári"
] | 2014-08-02T04:17:14+00:00 | 2014-08-02T04:17:14+00:00 | [
"cond-mat.mtrl-sci"
] | Silver environment and covalent network rearrangement in GeS3-Ag glasses | The structure of Ag-doped GeS3 glasses (0, 15, 20, 25 at.% Ag) was
investigated by diffraction techniques and extended x-ray absorption fine
structure measurements. Structural models were obtained by fitting the
experimental datasets simultaneously by the reverse Monte Carlo simulation
technique. It is observed that Ge has mostly S neighbours in GeS3, but Ge-Ge
bonds appear already at 15% Ag content. Sulfur has ~2 S/Ge neighbours over the
whole concentration range, while the S-Ag coordination number increases with
increasing Ag content. Ag-Ag pairs can already be found at 15% Ag. The Ag-S
mean coordination number changes from 2.17 +/- 0.2 to 2.86 +/- 0.2 between 15%
and 25% Ag content. Unlike the As-S network in AsS2-25Ag glass, the Ge-S
network is not fragmented upon Ag-doping of GeS3 glass. |
1408.0337v1 | ## A Bayesian estimation approach to analyze non-Gaussian data-generating processes with latent classes
Naoki Tanaka ∗
Shohei Shimizu ∗
Takashi Washio ∗
## Abstract
A large amount of observational data has been accumulated in various fields in recent times, and there is a growing need to estimate the generating processes of these data. A linear non-Gaussian acyclic model (LiNGAM) based on the non-Gaussianity of external influences has been proposed to estimate the data-generating processes of variables. However, the results of the estimation can be biased if there are latent classes. In this paper, we first review LiNGAM, its extended model, as well as the estimation procedure for LiNGAM in a Bayesian framework. We then propose a new Bayesian estimation procedure that solves the problem.
## 1 Introduction
Several methods have recently been proposed to discover a complete causal structure, that is, all the causal directions, under the assumption that disturbance variables have non-Gaussian distributions. However, the estimation results can be biased if there are 'latent classes.' Latent classes are unobserved discrete variables that have more than one observed child variables. Data that has been generated from different processes are mixed in the presence of latent classes. Several methods have been proposed to estimate the causal structure in the presence of latent classes [12], but all of these are affected by local optima. Therefore, in this paper, we propose a new estimation approach that can solve this problem.
The structure of this paper is as follows. In Section 2, we briefly review the data generating process for estimating causal structure, (LiNGAM, short for Linear Non-Gaussian Acyclic Model) [11], the LiNGAM mixture model [12], existing estimation approaches to the LiNGAM mixture model [12], and the BayesLiNGAM model. In Section 3, we extend previous research dedicated to solving the problem, and test the performance of our new method through experiments with simulated data in Section 4. Section 5 concludes the paper.
∗ The Institute of Scientific and Industrial Research, Osaka University
## 2 Background
## 2.1 Linear non-Gaussian acyclic model (LiNGAM)
We begin by introducing the basic LiNGAM model [11]. LiNGAM is a causal model with the following four assumptions.
- 1. The relations between the observed variables x i ( i = 1 , . . . , n ) can be represented by a directed acyclic graph (DAG), as shown in Fig. 1.
- 2. x i is assigned a value by a linear function of the values already assigned to the variables constituting its parents in the DAG as well as a 'disturbance' (noise) term e i and an optimal constant term µ i :
$$x _ { i } = \sum _ { k ( j ) < k ( i ) } b _ { i j } ( x _ { j } - \mu _ { j } ) + e _ { i } + \mu _ { i } \quad \quad ( 1 )$$
where k i ( ) is a causal order of x i in the DAG (so that if there is a directed edge from x j to x i in the DAG, k j ( ) < k i ( )), b ij represents the strength of the connection between x j and x i in the DAG.
- 3. The e i are all continuous random variables that follow non-Gaussian distributions with zero means and non-zero variances, and the e i are mutually independent, i.e., p e , . . . , e ( 1 n ) = ∏ i p i ( e i ).
- 4. The dataset D = { x 1 , . . . , x N } (each x contains components x i ) are observed, and each data vector x is generated according to the process described above with the same DAG, coefficients b ij , constants µ i , and disturbances e i , sampled independently from the same distributions. Note that this assumption implies that there are either no unobserved (latent) confounders [10] (hidden variables) or that the x i are causally sufficient [13].
Figure 1: An example of LiNGAM described by a DAG.
## 2.2 LiNGAM mixture model
The LiNGAM mixture model [12] is an extension of the basic LiNGAM to cases where the final assumption of the basic LiNGAM model collapses because of the presence of latent classes (hidden groups). If they exist, each latent class has a different structure. 'Structure' here consists of the causal order k i ( ), the connection strengths b ij , the densities of disturbances p i , and the means of observed variables µ i . Therefore, structure in a class c can be described using k ( c ) ( ), i b ( c ) ij , p ( c ) i , and µ ( c ) i (Note that the disturbances e ( c ) i are generated from p ( c ) i ). The data within each class ( c = 1 , . . . , l ) are assumed to be generated by the basic LiNGAM. Thus, the data-generating model within class c can be described as follows:
$$x _ { i } = \sum _ { k ^ { ( c ) } ( j ) < k ^ { ( c ) } ( i ) } b _ { i j } ^ { ( c ) } ( x _ { j } - \mu _ { j } ^ { ( c ) } ) + \mu _ { i } ^ { ( c ) } + e _ { i } ^ { ( c ) }$$
Moreover, the data vectors x are assumed to be generated by the following mixture density:
$$p ( \mathbf x | \Theta ) = \sum _ { c = 1 } ^ { l } \left \{ \begin{matrix} p ( \mathbf x | \theta ^ { ( c ) } ) \\ \theta ^ { ( c ) } \\ \end{matrix} ) \right \} P ( c ) \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \ \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \ & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad &\quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad & \quad \\ \quad \quad \quad$$
where l is the number of classes, and Θ = [ θ (1) , . . . , θ ( ) l ], and θ ( c ) contain all the parameters in (2), i.e., k ( c ) ( j ) , k ( c ) ( ) i , b ( c ) ij , µ ( c ) j , µ ( c ) i , and p ( c ) i ( e ( c ) i ). Note that if only one latent class exists, the LiNGAM mixture model is equivalent to basic LiNGAM.
## 2.3 Existing estimation approach to LiNGAM mixture model
In this subsection, we briefly introduce the existing estimation approach to the LiNGAM mixture model. In this approach, the LiNGAM mixture model is transformed into an Independent Component Analysis (ICA) [6]) model, and an ICA algorithm (for example, [7] [9]) is applied to it [12].
The LiNGAM mixture model (Equation (2)) can be written in matrix form as follows:
$$x = B ^ { ( c ) } x + ( \mathbf I - B ^ { ( c ) } ) \mu ^ { ( c ) } + \mathbf e ^ { ( c ) } \quad \quad \quad ( 4 ) \\ \mathbf r _ { \lambda } \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad. \quad.$$
where B ( c ) , µ ( c ) , and e ( c ) collect b ( c ) ij , µ ( c ) i , and e ( c ) i , respectively, and I is the identity matrix. We can then obtain the following equation by solving Equation (4) for x :
$$\mathbf x = \mu ^ { ( c ) } + \mathbf A ^ { ( c ) } \mathbf e ^ { ( c ) }$$
where A ( c ) = ( I -B ( c ) ) -1 . We can estimate B ( c ) by calculating A ( c ) using the ICA algorithm and computing B ( c ) . However, there are two important indeterminacies that ICA cannot solve: the order and the scaling of the independent components. In order to solve them, we need to permute and normalize A ( c ) appropriately before using it to compute B ( c ) (See [11] for more details).
## 2.4 The BayesLiNGAM method
Existing estimation methods for the LiNGAM mixture model [7] [9] are affected by local optima, because of which we propose a Bayesian approach. In this subsection, we summarize the BayesLiNGAM method [5], which is an estimation method of basic LiNGAM.
In Bayesian inference, the inference with the highest posterior probability for some hypotheses is selected. In BayesLiNGAM, the hypotheses are different possible DAGs. BayesLiNGAM outputs the DAG with the highest posterior probability of all possible DAGs given the data. The posterior probabilities can be calculated by Bayes' theorem:
$$P ( G _ { m } | \mathbf D ) = \frac { p ( \mathbf D | G _ { m } ) P ( G _ { m } ) } { p ( \mathbf D ) }$$
Here, G m are the different possible DAGs, m = { 1 , . . . , N g } , where N g denotes the number of different DAGs on n variables, and D is the observed dataset D = { x 1 , . . . , x N } , N : sample size). Due to our assumption of independent and identically distributed (IID) data, p ( D ) = ∏ N s =1 p ( x s ). We then need to specify the likelihoods p ( D | G m ), prior probabilities P G ( m ), and normalization constant p ( D ) to compute the posterior probabilities.
The prior probabilities P G ( m ) first incorporate any domain of knowledge and prior information. If we have no knowledge or information about the G m , all values of P G ( m ) are equal, i.e., P G ( m ) = 1 N g .
Furthermore, p ( D ) is a constant that simply normalizes posterior probabilities. Hence, we can obtain p ( D ) as follows:
$$P ( \mathbf D ) = \sum _ { m = 1 } ^ { N _ { g } } p ( \mathbf D | G _ { m } ) P ( G _ { m } )$$
The likelihoods p ( D | G m ) can be computed by marginalizing over p ( DΘ | , G m ) of Θ :
$$P ( \mathbf D | G _ { m } ) = \int p ( \mathbf D | \Theta, G _ { m } ) p ( \Theta | G _ { m } ) d \Theta$$
Here, Θ collects all the parameters in Equation (1) (i.e., k j , k ( ) ( ) i , b ij , µ j , µ i , and p i ), p ( DΘ | , G m ) denotes the likelihood of the model (here, basic LiNGAM), and p ( Θ | G m ) denotes the prior distributions of Θ . Note that if the G m are given, we can find the parents of x i in the G m and determine the causal order k j ( ) and k i ( ). That is, the first term of the right side in Equation (1) can be specified by the G m . We introduce the likelihood of the model and prior distribution for Θ in Section 3.
## 3 Proposed method
## 3.1 Model
We assume that the data within each class ( c = 1 , . . . , l ) are generated by the basic LiNGAM. Thus, the data-generating model in class c can be described by Equation (2), and the probability densities of the data x are the same as in Equation (3). We then use the Bayesian approach to estimate the LiNGAM mixture model in the same manner as BayesLiNGAM in Section 2.4.
Here, we need to parameterize the densities p i to specify p ( DΘ | , G m ) in Equation (8). Due to the assumption in LiNGAM that the e i are all continuous random variables with non-Gaussian distributions, we use a generalized Gaussian distribution [8] that incorporates a shape parameter. The generalized Gaussian distribution is symmetric and includes Gaussian, Laplacian, continuous uniform, and several non-Gaussian distributions. The probability density function is as follows:
$$p _ { i } ( e _ { i } ) = \frac { \lambda _ { i } \sqrt { \frac { \Gamma ( 3 / \lambda _ { i } ) } { \Gamma ( 1 / \lambda _ { i } ) } } \exp ( - ( \sqrt { \frac { \Gamma ( 3 / \lambda _ { i } ) } { \Gamma ( 1 / \lambda _ { i } ) } } \frac { | e _ { i } | } { \sigma _ { i } } ) ^ { \lambda _ { i } } ) } { 2 \sigma _ { i } \Gamma ( 1 / \lambda _ { i } ) }$$
Here, σ i represent the standard deviations, λ i are the shape parameters, and Γ() denotes the Gamma function.
Moreover, we need to transform the density of the e i into that of the x i . The LiNGAM mixture model (Equation (2)) can be written in matrix form as follows:
$$\mathbf x \ & = \ \mathbf f ( \mathbf e ^ { ( c ) } ) \\ & = \ \mu ^ { ( c ) } + ( \mathbf I - \mathbf B ^ { ( c ) } ) ^ { - 1 } \mathbf e ^ { ( c ) } \\ & \quad \dots \ \dots \ \dots \quad \mathbf w \ \nu$$
where f () is a mapping vector, and B ( c ) , e ( c ) , and µ ( c ) collect b ( c ) ij , e ( c ) i , and µ ( c ) i , respectively. I is the identity matrix.
$$p ( x | \theta ^ { ( c ) }, G _ { m } ) & = \frac { 1 } { | \det J f ( \mathbf f ^ { - 1 } ( x ) ) | } p _ { \mathbf e } ^ { ( c ) } ( \mathbf f ^ { - 1 } ( x ) ) \quad \quad ( 1 1 ) \\ \tau \cdot \quad \tau \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdots$$
The density of x within class c ( p ( x | θ ( c ) , G m ) ) is obtained from the density of e ( c ) ( p ( c ) e ( e ( c ) ) ) as follows [6]:
where J f is a Jacobian matrix. From equations (10) and (11) with the assumption of acyclicity, det J f ( f -1 ( x )) equals to one, and we can obtain the following equation:
$$p ( x | \theta ^ { ( c ) }, G _ { m } ) = p _ { \mathbf e } ^ { ( c ) } ( \mathbf e ^ { ( c ) } )$$
According to the assumption in LiNGAM that the e i are mutually independent,
$$p _ { \text{e} } ^ { ( c ) } ( \text{e} ) \ = \ \prod _ { i = 1 } ^ { n } p _ { i } ^ { ( c ) } ( e _ { i } ^ { ( c ) } ).$$
Then, the probability density of x within each class c , using equations (2), (12), and (13), is given by
$$. 3 ), & \text{ is given by} \\ & p ( x | \theta ^ { ( c ) }, G _ { m } ) \\ & = \prod _ { i = 1 } ^ { n } p _ { i } ^ { ( c ) } ( e _ { i } ^ { ( c ) } ) \\ & = \prod _ { i = 1 } ^ { n } p _ { i } ^ { ( c ) } ( x _ { i } - \mu _ { i } ^ { ( c ) } - \sum _ { k ^ { ( c ) } ( j ) < k ^ { ( c ) } ( i ) } b _ { i j } ^ { ( c ) } ( x _ { j } - \mu _ { j } ^ { ( c ) } ) ). \\ \ n \, \text{ now specify} \, p ( \text{D} | \Theta. G _ { m } ) \, \text{ with the assumption of IID data and equations}$$
We can now specify p ( DΘ | , G m ) with the assumption of IID data and equations (3) and (14):
$$\cos \liminf _ { s = 1 } ^ { N } p ( x ^ { s } | \Theta, G _ { m } ) & & ( 1 5 ) \\ p ( x | \Theta, G _ { m } ) & & ( 1 5 ) \\ = \sum _ { c = 1 } ^ { l } \left \{ p ( x | \theta ^ { ( c ) }, G _ { m } ) \right \} P ( c ) \\ = \sum _ { c = 1 } ^ { l } \left \{ \prod _ { i = 1 } ^ { n } p _ { i } ^ { ( c ) } ( x _ { i } - \mu _ { i } ^ { ( c ) } - \sum _ { k ^ { ( c ) } ( j ) < k ^ { ( c ) } ( i ) } b _ { i j } ^ { ( c ) } ( x _ { j } - \mu _ { j } ^ { ( c ) } ) ) \right \} \\ \times P ( c ) & & ( 1 6 ) \\ 2 \ \Pr i o r \ \text{ parameters and hyperparameters}$$
## 3.2 Priors for parameters and hyperparameters
We use multinomial distribution for P c ( ) in Equation (3). The distribution represents, for N independent trials, each of which leads to one of the possible events c = 1 , . . . , l , the number of times that each event occurs ( z = [ z (1) , . . . , z ( ) l ]). w ( c ) is the probability of the occurrence of each event. Generally, the probability mass function of a multinomial distribution is as follows:
$$P ( \mathbf z ) = \frac { N! } { \prod _ { c = 1 } ^ { l } z ^ { ( c ) }! } \prod _ { c = 1 } ^ { l } ( w ^ { ( c ) } ) ^ { z ^ { ( c ) } } & & ( 1 7 ) \\ > 0 \text{ and } \sum _ { \varepsilon = 1 } ^ { l } w ^ { ( c ) } = 1. \text{ When we use the above distribution in}$$
where w ( c ) > 0 and ∑ l c =1 w ( c ) = 1. When we use the above distribution in Equation (3), the number of trials N is one. Then, z can be regarded as the indicator vector and corresponds to variable c . For example, if x belongs to class 2 (i.e., c = 2), z (2) = 1 and the other components are 0. From the above, we can use a multinomial distribution for P c ( ) in the following form:
$$P ( c ) & = w ^ { ( c ) }$$
We use a Dirichlet distribution for the parameters of the multinomial distribution ( w (1) , . . . , w ( ) l ) because the former is a conjugate prior for the latter
and is typically used in similar contexts. The density function of the Dirichlet distribution is as follows:
$$p ( \mathbf w ) = \frac { \Gamma ( \sum _ { c = 1 } ^ { l } a ^ { ( c ) } ) } { \prod _ { c = 1 } ^ { l } \Gamma ( a ^ { ( c ) } ) } \prod _ { c = 1 } ^ { l } ( w ^ { ( c ) } ) ^ { a ^ { ( c ) } - 1 } & & ( 1 9 ) \\ \intertext { r } \, = \, w ^ { ( 1 ) } \dots \, w ^ { ( l ) } \text{ and } a ^ { ( c ) } ( > \, 0 ) \text{ are concentration parameters. } \text{ We }$$
where w = w (1) , . . . , w ( ) l and a ( c ) ( > 0) are concentration parameters. We can generate a random vector w by normalizing independent gamma random variables γ (1) , . . . , γ ( ) l with shape parameters a ( c ) (concentration parameters in the Dirichlet distribution) and scale parameter 1 [4]:
$$w ^ { ( c ) } \ & = \ \frac { \gamma ^ { ( c ) } } { \sum _ { c = 1 } ^ { l } \gamma ^ { ( c ) } } \\ p ( \gamma ^ { ( c ) } ) \ & = \ \frac { 1 } { \Gamma / \wedge ( c ) \wedge } ( \gamma ^ { ( c ) } ) ^ { a ^ { ( c ) } - 1 } \exp ( \gamma ^ { ( c ) } )$$
$$p ( \gamma ^ { ( c ) } ) \ = \ \frac { \sum _ { 1 } ^ { - \cdot - \cdot } } { \Gamma ( a ^ { ( c ) } ) } ( \gamma ^ { ( c ) } ) ^ { a ^ { ( c ) } - 1 } \exp ( \gamma ^ { ( c ) } )$$
We use Gaussian distributions for b ( c ) ij and µ ( c ) i with zero and ϕ ( c ) as their means and v 2 and τ 2 as their variances respectively.
$$p ( b _ { i j } ^ { ( c ) } ) \ = \ \frac { 1 } { \sqrt { 2 \pi v ^ { 2 } } } \exp \left \{ - \frac { ( b _ { i j } ^ { ( c ) } ) ^ { 2 } } { 2 v ^ { 2 } } \right \} \quad \ \quad \ \ ( 2 2 )$$
$$p ( \mu _ { i } ^ { ( c ) } ) \ = \ \frac { 1 } { \sqrt { 2 \pi \tau ^ { 2 } } } \exp \left \{ - \frac { ( \mu _ { i } ^ { ( c ) } - \varphi ^ { ( c ) } ) ^ { 2 } } { 2 \tau ^ { 2 } } \right \}$$
We heuristically determine the value of ϕ . We use the result of the Gaussian mixture model estimation and an expectation-maximization (EM) algorithm [2] for ϕ because the greater the number of latent classes, the more difficult it is to estimate µ ( c ) i . The EM algorithm is a method to compute maximum likelihood solutions for models with latent variables. In the algorithm, parameters and responsibilities (conditional probabilities of latent variables given the data) are updated in turn until the change in the likelihood function falls below some threshold.
We use inverse gamma distributions [3] with shape parameters α , η , and χ and scale parameters β , ζ , and /epsilon1 for ( σ ( c ) i ) 2 , λ ( c ) i , and v 2 , respectively.
$$p ( ( \sigma _ { i } ^ { ( c ) } ) ^ { 2 } ) \ = \ \frac { \beta ^ { \alpha } } { \Gamma ( \alpha ) } ( \sigma _ { i } ^ { ( c ) } ) ^ { 2 ( - \alpha - 1 ) } \exp \left \{ \frac { - \beta } { ( \sigma _ { i } ^ { ( c ) } ) ^ { 2 } } \right \} \quad \quad ( 2 4 )$$
$$p ( \lambda _ { i } ^ { ( c ) } ) \ = \ \frac { \zeta ^ { \eta } } { \Gamma ( \eta ) } ( \lambda _ { i } ^ { ( c ) } ) ^ { - \eta - 1 } \exp \left ( \frac { - \zeta } { \lambda _ { i } ^ { ( c ) } } \right ) ^ { \mu \times \imath } \, ^ { \prime } \, \cdots \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
$$p ( v ^ { 2 } ) \ = \ \frac { \epsilon ^ { \chi } } { \Gamma ( \chi ) } v ^ { 2 ( - \chi - 1 ) } \exp \left ( \frac { - \epsilon } { v ^ { 2 } } \right )$$
We can generate inverse gamma random variables using gamma random variables. If a variable X has a gamma distribution with shape parameter α and
scale parameter β , Y = 1 /X has an inverse gamma distribution with shape parameter α and scale parameter 1 /β [3].
We determine the value of a ( c ) , α β , , η , ζ , χ /epsilon1 , , and τ arbitrarily. From the above, we can compute P ( D | G m ) in Equation (8) using ordinary Monte Carlo sampling [2] to compute the integral.
## 4 Simulation
In this simulation, we assume for simplicity that we know of the existence of a causal connection between two observed variables, and that its direction is the same for all classes. We can thus estimate which direction is true ( x ( c ) 1 → x ( c ) 2 or x ( c ) 1 ← x ( c ) 2 ). We generated 1,000 datasets under every combination of sample size ( N = 50 100 500) and the number of classes ( , , l = 2 4 6). , ,
The data within each class were generated by the basic LiNGAM (Equation(1)), following which we mixed them. In all datasets, the true model was x ( c ) 1 → x ( c ) 2 and the connection strength was b ( c ) ij = [ -1 5 . , -0 5] . ∩ [0 5 . , 1 5]. . The distributions of external influences e ( c ) i were randomly selected from the following three non-Gaussian distributions with one variance: Laplace, Uniform ([ -√ 3 , √ 3]) and Student-t (with five degrees of freedom).
With regard to the priors we did not use any prior information about G m , and thus P G ( m ) = 1 2 . We randomly selected the value of a ( c ) from among 3, 5, and 7. Our choice of values for the rest of the parameters was α = 3, β = 3, η = 3, ζ = 3, χ = 3, /epsilon1 = 3, and τ = 0 5. .
We selected the number of classes l as follows. We calculated the logmarginal likelihoods of the two models for all numbers of classes and selected the class with the largest log-marginal likelihood. The maximum number of classes tested was 2log N . This was motivated by the fact that when the sample size N approaches infinity in a Dirichlet process mixture model, the number of classes l approaches log N [1].
We compared our method with the existing method [12] to determine the one that can correctly estimate causal directions more times given the above datasets. However, we use [9] in the existing method [12] instead of [7]. We show the simulation results in Table 1. The numbers in the tables denote the number of times each method correctly estimated causal directions. As shown in the table, our method has more correct answers than the existing method.
## 5 Conclusion
In this paper we proposed a new estimation approach for discovering causal structure in the presence of latent classes. In a simulation run using artificial data, our method correctly estimated more causal directions than the existing method. Our plan for future research is to evaluate our method on a wide variety of real datasets.
| | Table | 1: Simulation results Sample size | 1: Simulation results Sample size | |
|-----------------|---------|-------------------------------------|-------------------------------------|-----|
| | | 50 | 100 | 500 |
| Our method | l = 2 | 913 | 947 | 981 |
| | l = 4 | 908 | 937 | 973 |
| | l = 6 | 922 | 957 | 967 |
| Existing method | l = 2 | 649 | 657 | 684 |
| | l = 4 | 663 | 655 | 729 |
| | l = 6 | 646 | 700 | 762 |
## References
- [1] C. E. Antoniak. Mixtures of dirichlet processes with applications to bayesian nonparametric problems. The annals of statistics , pages 11521174, 1974.
- [2] C. M. Bishop. Pattern Recognition and Machine Learning . Springer, 2006.
- [3] John D Cook. Inverse gamma distribution. online: h ttp://www. johndcook. com/inverse gamma. pdf, Tech. Rep , 2008.
- [4] L Devroye. Non-uniform Random Variate Generation . Springer-Verlag, 1986.
- [5] P. O. Hoyer and A. Hyttinen. Bayesian discovery of linear acyclic causal models. In Proc. 25th Conf. on Uncertainty in Artificial Intelligence (UAI2009) , pages 240-248, 2009.
- [6] A. Hyv¨ arinen, J. Karhunen, and E. Oja. Independent component analysis . Wiley, New York, 2001.
- [7] M. N. H. Mollah, M. Minami, and S. Eguchi. Exploring latent structure of mixture ICA models by the minimum β -divergence method. Neural Computation , 18:166-190, 2006.
- [8] S. Nadarajah. A generalized normal distribution. Journal of Applied Statistics , 32(7):685-694, 2005.
- [9] J. A. Palmer, S. Makeig, K. Kreutz-Delgado, and B. D. Rao. Newton method for the ICA mixture model. In Proc. IEEE Int. Conf. on Acoustics, Speech and Signal Processing (ICASSP2008) , pages 1805-1808, 2008.
- [10] J. Pearl. Causality: Models, Reasoning, and Inference . Cambridge University Press, 2000.
- [11] S. Shimizu, P. O. Hoyer, A. Hyv¨rinen, and A. Kerminen. a A linear nongaussian acyclic model for causal discovery. J. Machine Learning Research , 7:2003-2030, 2006.
- [12] S. Shimizu and A. Hyv¨ arinen. Discovery of linear non-gaussian acyclic models in the presence of latent classes. In Proc. 14th Int. Conf. on Neural Information Processing (ICONIP2007) , pages 752-761, 2008.
- [13] P. Spirtes, C. Glymour, and R. Scheines. Causation, Prediction, and Search . Springer Verlag, 1993. (2nd ed. MIT Press 2000). | null | [
"Naoki Tanaka",
"Shohei Shimizu",
"Takashi Washio"
] | 2014-08-02T04:31:56+00:00 | 2014-08-02T04:31:56+00:00 | [
"stat.ML"
] | A Bayesian estimation approach to analyze non-Gaussian data-generating processes with latent classes | A large amount of observational data has been accumulated in various fields
in recent times, and there is a growing need to estimate the generating
processes of these data. A linear non-Gaussian acyclic model (LiNGAM) based on
the non-Gaussianity of external influences has been proposed to estimate the
data-generating processes of variables. However, the results of the estimation
can be biased if there are latent classes. In this paper, we first review
LiNGAM, its extended model, as well as the estimation procedure for LiNGAM in a
Bayesian framework. We then propose a new Bayesian estimation procedure that
solves the problem. |
1408.0339v1 | ## Beamforming for Secure Communication via Untrusted Relay Nodes Using Artificial Noise
Siddhartha Sarma
## Abstract
A two-phase beamforming solution for secure communication using untrusted relay nodes is presented. To thwart eavesdropping attempts of relay nodes, we deliberately introduce artificial noise in the source message. After pointing out the incongruity in evaluating secrecy rate in our model for certain scenarios, we provide an SNR based frame work for secure communication. We intend to bring down the SNR at each of the untrusted relay nodes below a certain predefined threshold, whereas, using beamforming we want to boost the SNR at the destination. With this motive optimal scaling vector is evaluated for beamforming phase which not only nullifies the artificial noise transmitted initially, but also maximizes the SNR at the destination. We discuss both the total and individual power constraint scenarios and provide analytical solution for both of them.
## I. INTRODUCTION
Recently omnipresence of wireless devices prompted the researchers and engineers to delve into the security issues of wireless communication. As compared to wired medium, ensuring security for wireless medium is more challenging mainly due to the broadcast nature of the transmission. But the ongoing research on physical layer security promises robust and reliable security schemes for wireless communication. Contrary to conventional cryptographic schemes physical layer security techniques are impregnable as the security is ensured by inherent randomness present in the wireless medium.
Physical layer security came to existence through the seminal work of Wyner [1], where he explored the possibility of secure communication without relying on private keys. Later Leung
S. Sarma is with the Department of Electronic Systems Engineering, Indian Institute of Science, Bangalore, Karnataka 560012, India (e-mail: { siddharth } @dese.iisc.ernet.in).
and Hellman [2] extended this idea to Gaussian channels. Following their works, in last decade several researchers have devised techniques for secure communication in single-hop single and multi-antenna systems. Recently a considerable amount of research is ongoing to extend those schemes to multi-hop network scenarios. In fact, several works have demonstrated that cooperative relaying [3, 4] can significantly improve the performance of secure communication. One such cooperative scheme is beamforming, where multiple transmitters adjust their gain and phase to improve the signal strength at the destination. Though beamforming was initially developed for multi-antenna systems, but due to energy and hardware constraints distributed beamforming solution using multiple single antenna node is proposed [3, 4]. In those papers authors have assumed that relay nodes are trustworthy and they beamform towards the destination to defeat (an) external eavesdropper(s). But due to the unprotected nature of public ad hoc network ( e.g. sensor network), an adversary can wiretap those relay nodes to obtain the information transmitted by source. This prompted us to look for a solution of one of the worst adversarial scenarios of secure communication, where one have to maintain the secrecy of data from the participating relay nodes. Secure transmission using untrusted relay nodes has appeared before in [5-8], where secure communication is obtained by simultaneously transmitting two or more signals. But multiple simultaneous transmission in the above mentioned papers were possible due to mutiple sources and/or multiple jammers. In absence of mutiple sources or jammers source has to rely on artificial noise [9]. Authors in [10] considered artificial noise based beamforming solution for AF relay network, but unlike them we have untrusted relay nodes and artificial noise is introduced by the source itself.
In our work we have considered multiple amplify-and-forward (AF) untrusted relay nodes, who help the source to deliver the message at destination, but at the same time can eavesdrop the ongoing transmission. To prevent eavesdropping the source adds artificial noise in the message and broadcast it in first phase. In second phase source and relay nodes beamform to deliver the signal at the destination. Though secrecy rate is the conventional metric to evaluate the performance of such system, but for our model in certain scenarios maximizing secrecy rate results in removal of artificial noise. Therefore, we consider a pragmatic approach based on SNR criteria, where we intend keep the SNR at the untrusted relay nodes below certain predefined threshold. Assuming perfect channel state information (CSI), we evaluate the optimal scaling factor for all the transmitting nodes which not only nullifies the artificial noise but also maximizes
Fig. 1. Simple two hop network with multiple relays
the SNR at the destination. We consider both the total and individual power constraints scenario for all transmitting nodes and provide analytical approach for optimal solutions.
## II. SYSTEM MODEL & PROBLEM FORMULATION
We consider a two-hop network shown in Figure 1 containing multiple relay nodes ( M = { 1 2 , , . . . , M } ) aiding the transmission from source node S to destination D . Adversary is using the relay nodes to passively eavesdrop the on going transmission. To prevent such eavesdropping source transmits artificial noise along with the actual signal to confuse the adversary. In next phase source along with the relay nodes performs beamforming to deliver the message to the destination. We are interested in maximizing the information rate which can be transmitted from the source to the destination securely.
In first phase source transmits a linear combination of two messages √ αP x 1 + √ (1 -α P u ) 1 using power P 1 , where α ∈ [0 , 1] , x is the Gaussian distributed message with zero mean and unit variance and u is statistically independent Gaussian distributed artificial noise with identical distribution as x . The message received at relay and the destination can be written as:
$$y _ { i } = h _ { s i } ( \sqrt { \alpha P _ { 1 } } x + \sqrt { ( 1 - \alpha ) P _ { 1 } } u ) + z _ { i }, \, i \in \mathcal { M } \bigcup \{ D \}$$
where z i are complex Gaussian noise distributed according to CN (0 , σ 2 ) . In second phase relay nodes along with the source performs beamforming at destination to nullify the artificial noise. In amplify-and-forward (AF) relaying scheme nodes transmit the received noisy message after scaling it by appropriate factor, w i ∈ C . Source simultaneously transmits √ αP w x 1 0 -
√ (1 -α P ) 1 M ∑ i =1 w h i si h id h sd u . Therefore, the received message at destination can be written as:
$$y _ { d } = \sqrt { \alpha P _ { 1 } } ( w _ { 0 } h _ { s d } + \sum _ { i = 1 } ^ { M } w _ { i } h _ { s i } h _ { i d } ) x + \sum _ { i = 1 } ^ { M } h _ { i d } w _ { i } z _ { i } + z _ { d }$$
Assuming Maximum Ratio Combining (MRC) [3], capacity of the source to destination channel can be written as:
$$C _ { d } = \frac { 1 } { 2 } \log \left ( 1 + \frac { | h _ { s d } | ^ { 2 } \alpha P _ { 1 } } { \sigma ^ { 2 } + | h _ { s d } | ^ { 2 } ( 1 - \alpha ) P _ { 1 } } + \frac { | \mathbf h ^ { \dagger } \mathbf w | ^ { 2 } \alpha P _ { 1 } } { \sigma ^ { 2 } ( 1 + \mathbf w ^ { \dagger } \mathbf D _ { \mathbf h } \mathbf w ) } \right )$$
where w = [ w , w , 0 1 · · · , w M ] † , h = [ h sd , h s 1 h 1 d , · · · , h sM
h Md ] † , and D h is ( M +1) × ( M +1) matrix with diagonal elements [0 , | h 1 d | 2 , | h 2 d | 2 , · · · , | h Md | 2 ] .
As the relays can receive only during the first phase, so the capacity of the source to relay channel is:
$$C _ { i } = \frac { 1 } { 2 } \log \left ( 1 + \frac { | h _ { s i } | ^ { 2 } \alpha P _ { 1 } } { \sigma ^ { 2 } + | h _ { s i } | ^ { 2 } ( 1 - \alpha ) P _ { 1 } } \right )$$
The total power consumed during the second phase can be calculated in following manner:
$$\alpha P _ { 1 } | w _ { 0 } | ^ { 2 } + ( 1 - \alpha ) P _ { 1 } \left | \sum _ { i = 1 } ^ { M } \frac { h _ { s i } h _ { i d } } { h _ { s d } } w _ { i } \right | ^ { 2 } + \sum _ { i = 1 } ^ { M } ( | h _ { s, i } | ^ { 2 } P _ { 1 } + \sigma ^ { 2 } ) | w _ { i } | ^ { 2 } = \mathbf w ^ { \dagger } \mathbf D \mathbf w$$
where D = αP 1 0 t 0 (1 -α P ) 1 gg † + T , g is M length vector with i th element g i = h si h id h sd and
T is a diagonal matrix with diagonal elements [ | h s, 1 | 2 P 1 + σ , 2 · · · , | h s,M | 2 P 1 + σ 2 ] . Therefore, the total power constraint can be expressed as
$$\text{w} ^ { \dagger } D \text{w} \leq P _ { t o t }$$
For several scenarios individual constraint is more relevant than total constraint. Those constraints can be written as:
$$\mathbf w ^ { \dagger } \mathbf D _ { s } \mathbf w \leq P _ { s }$$
$$| w _ { i } | ^ { 2 } \leq \frac { P _ { i } } { | h _ { s i } | ^ { 2 } P _ { 1 } + \sigma ^ { 2 } }, \, \forall i \in \{ 1, 2, \cdots, M \}$$
$$\text{where} \, D _ { s } = \begin{bmatrix} \alpha P _ { 1 } & 0 ^ { t } \\ 0 & ( 1 - \alpha ) P _ { 1 } \text{gg} ^ { \dagger } \end{bmatrix}. \text{Now for a given total or individual power budget maximum}$$
achievable secrecy rate can be formulated as following optimization problem:
$$\max \, \min _ { i } R _ { s i } ( \alpha, \mathbf w ) = C _ { d } - C _ { i }$$
Subject to: (1) or (2)
$$\alpha \in [ 0, 1 ]$$
We denote the SNR at i th relay node as Γ i = | h si | 2 αP 1 σ 2 + | h si | 2 (1 -α P ) 1 . One can easily see that Γ i is an increasing function in | h si | 2 , ∀ α ∈ [0 , 1] . Therefore, if we order the source to relay channel gains according to their absolute values, then the one with the highest value will have maximum SNR. Formally,
$$\Gamma _ { e } \geq \Gamma _ { i }, \text{ where } e = \arg \max _ { i } \left [ | h _ { s 1 } | ^ { 2 }, \dots, | h _ { s M } | ^ { 2 } \right ]$$
Therefore, the objective of the problem (3) can be rephrased as maximize C d -C e for the same constraint set. We rewrite C d -C e in following manner:
$$\text{sounds} \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \col. \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \col: \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \col. \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \col.: \, \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \, \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \col. \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \colon \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \text{s} \, \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \colon \, \text{s } \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \Text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \, \text{s} \\ \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s } \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s}" \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s}\text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \\ \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{ s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \text{s} \$$
$$\bigcup 1$$
By differentiating the objective function with respect to α , we have the following observation.
Observation. C d -C e is an increasing function of α in [0 , 1] if
$$f ( \mathbf w ) \geq \frac { ( \rho _ { d } - \rho _ { e } ) \rho _ { d } } { ( \rho _ { d } - 1 ) ^ { 2 } ( \rho _ { e } - 2 ) }$$
glyph[negationslash]
assuming ρ e = 2 .
But if we use α = 1 , then in absence of artificial noise, relay nodes can properly decode the message. So, we consider an SNR based approach, where an SNR upto γ results in significant amount of BER in eavesdropping relay nodes and therefore, for all practical purposes we can assume that the message is undecodable at relay nodes. The Optimization problem (3) can be rewritten as:
$$\max C _ { d }$$
$$\text{subject to} \colon \, \Gamma _ { e } \leq \gamma$$
,
(1) or (2)
(3b)
## III. ANALYSIS & SOLUTION
From the optimization problem (4) one can easily see that to maximize C d , Γ e should be equal to γ . Therefore, we can calculate the corresponding α and also using the constraint (3b) we can bound the γ value.
$$\alpha ( \gamma ) = \frac { 1 + \frac { \sigma ^ { 2 } } { | h _ { s e } | ^ { 2 } P _ { 1 } } } { ( 1 + \frac { 1 } { \gamma } ) }, \text{ where } \gamma \leq \frac { | h _ { s e } | ^ { 2 } P _ { 1 } } { \sigma ^ { 2 } } \\. \quad \hat { \ }. \quad \hat { \ }. \quad \hat { \ }. \quad \hat { \ }. \quad \hat { \ }. \quad \hat { \ }.$$
Total Power Constraint: Once α is fixed the first term in the destination SNR is constant, therefore, optimization problem (4) can be reduced to:
$$\max \left ( \frac { | h ^ { \dagger } w | ^ { 2 } } { ( 1 + w ^ { \dagger } D _ { h } w ) } \right )$$
$$\text{subject to} \colon w ^ { \dagger } D w \leq P _ { t o t }$$
One can check that the constraint will be satisfied with equality at optima and the objective function can be written as: w † ( hh † ) w w Dw † ˜ , where ˜ D = 1 P tot D + D h . This is indeed the well known generalized Rayleigh Quotient problem [11, p. 176] for which the maximum value of the objective function corresponds to the maximum eigenvalue of matrix ˜ D -1 hh † . The solution can be expressed as w ∗ = µ v where v = ˜ D -1 h and µ = √ P tot v Dv ′ .
Individual Power Constraints: : The individual power constraint problem for AF relaying can be written as:
$$\max \left ( \frac { | h ^ { \dagger } w | ^ { 2 } } { ( 1 + w ^ { \dagger } D _ { h } w ) } \right )$$
subject to: (2)
It can be seen that the angles of the scaling factors should be chosen in following manner to obtain the optimal value.
$$\arg ( w _ { 0 } ) = - \arg ( h _ { s d } ), \, \arg ( w _ { i } ) = - ( \arg ( h _ { s i } ) + \arg ( h _ { i d } ) ), \forall i$$
Therefore, if we denote c = [ | h sd | , | h s 1 | , · · · , | h sM | ] t and u = [ | w , w h 0 | | 1 1 d | , · · · , | w M Md h | ] t , then the optimization problem (6) can be essentially written as:
$$\max \left ( \frac { ( c ^ { t } u ) ^ { 2 } } { ( 1 + u ^ { t } I _ { 0 } u ) } \right )$$
$$\text{subject to} \colon u ^ { t } \widehat { D } _ { s } u \leq P _ { s }$$
$$0 \leq u _ { i } \leq u _ { m a x, i }, \, \forall i \in \mathcal { M }$$
$$& \text{where } I _ { 0 } = \begin{bmatrix} 0 & 0 ^ { t } \\ 0 & I \end{bmatrix}, \widehat { D } _ { s } = \begin{bmatrix} \alpha P _ { 1 } & 0 ^ { t } \\ 0 & \frac { ( 1 - \alpha ) P _ { 1 } } { | h _ { s 1 } | ^ { 2 } P _ { 1 } + \sigma ^ { 2 } } } { h _ { s } h _ { s } ^ { 2 } } \end{bmatrix}, u _ { m a x, i } = \sqrt { \frac { P _ { i } } { | h _ { s 1 } | ^ { 2 } P _ { 1 } + \sigma ^ { 2 } } }, \text{I is a } M \times M \text{ identity}$$
Proposition 1. The optimal solution for problem (7) with only source constraint i.e. (7b) is given by: u ∗ 1 = √ η 1 -η 2 || c 〈 2 〉 || 2 r ∗ 2 , u ∗ 〈 2 〉 = c 〈 2 〉 || c 〈 2 〉 || r ∗ where r ∗ = √ || c 〈 2 〉 || 2 η 1 || c 〈 2 〉 || 4 η 2 +( η 1 + || c 〈 2 〉 || 2 η 2 ) 2 c 2 1
Proof: We denote the vector [ u , u 2 3 , · · · , u M +1 ] as u 〈 2 〉 , then the objective function can be rewritten as:
$$\frac { ( c _ { 1 } u _ { 1 } + c _ { \langle 2 \rangle } ^ { t } \mathbf u _ { \langle 2 \rangle } ) ^ { 2 } } { 1 + | | \mathbf u _ { \langle 2 \rangle } | | ^ { 2 } }$$
Now it is easy to see that, if || u 〈 2 〉 || = r , then maximum value of c t 〈 2 〉 u 〈 2 〉 = c 〈 2 〉 || c 〈 2 〉 || r . As the objective function is an increasing function of u 1 , so in absence of individual constraints i.e. (7c), the source constraint (7b) will be satisfied with equality. From that we can calculate the value of u 1 in terms of r .
$$u _ { 1 } = \sqrt { \eta _ { 1 } - \eta _ { 2 } | | { \mathbf c } _ { \langle \varepsilon \rangle } | | ^ { 2 } r ^ { 2 } }, \text{ where } \eta _ { 1 } = \frac { P _ { s } } { \alpha P _ { 1 } }, \, \eta _ { 2 } = \frac { ( 1 - \alpha ) } { \alpha | c _ { 1 } | ^ { 2 } }$$
We can write the objective function (7a) in terms of r in following manner: ( c 1 √ η 1 -η r 2 2 + || c 〈 2 〉 || r ) 2 1+ r 2 By differentiating with respect to r and equating it to 0, we get the above result.
If the solution obtained in the above manner satisfy all the individual constraints, then we have indeed solved the problem (7). But, if for any u i individual constraint is violated, then we solve the problem using the following iterative approach.
- 1) Initialize parameters t 1 = 0 and t 2 = 1 .
- 2) Sort the relay nodes in descending order based on u /u ∗ i max,i , ∀ i ∈ { 2 , · · · , M } . Sort the vector c 〈 2 〉 and u max accordingly. In sorted list let us denote the first variable as u (2) . Update the parameter t 1 = t 1 + c (2) u (2) and t 2 = t 2 + u 2 (2) . Now we have to solve the following optimization problem:
$$\max \left ( \frac { ( t _ { 1 } + c _ { 1 } u _ { 1 } + \widehat { c } ^ { t } \widehat { u } ) ^ { 2 } } { t _ { 2 } + | | \widehat { u } | | ^ { 2 } } \right ) _ { v _ { 1 } \right ) v _ { 2 } } \right ) ^ { ( 8 a ) }$$
$$\text{s.t.} \ \alpha P _ { 1 } u _ { 1 } ^ { 2 } + \frac { ( 1 - \alpha ) P _ { 1 } } { | h _ { s d } | ^ { 2 } } ( t _ { 1 } + \widehat { \mathfrak { c } } ^ { t } \widehat { \mathfrak { u } } ) ^ { 2 } = P _ { s }$$
where ̂ u and ̂ c are the vectors obtained after removing u (2) and c (2) , respectively.
- 3) If the solution of problem (8) satisfies all the individual constraints, then optimal solution is obtained, otherwise repeat step 2 & 3 until the solution obtained satisfy their individual constraints.
Remark: The rationale behind picking the variable u (2) corresponding to arg max i u /u ∗ i max,i in step 2 is: for increasing P s , u (2) is the first variable to violate its individual constraint and (7a) increases with u (2) till it reaches the optimum corresponding to that P s value.
Proposition 2. The optimal solution of the problem (8) is given by u ∗ 1 = √ η 1 -η 2 ( t 1 + || ̂ u || r ∗ ) 2 , ̂ u ∗ = ̂ c || ̂ c || r ∗ where r ∗ satisfies the following quadratic polynomial:
$$q _ { 0 } r ^ { 4 } + q _ { 1 } r ^ { 3 } + q _ { 2 } r ^ { 2 } + q _ { 3 } r + q _ { 4 } = 0$$
where q 0 = η η t τ 2 3 2 1 2 , q 1 = -2 η t τ 2 1 ( η c 1 2 1 + η 3 ( t 2 τ 2 -t 2 1 )) , q 2 = η 3 ( η t 1 2 1 -η 2 (( t 2 1 -t 2 τ 2 ) 2 -2 t 2 1 t 2 τ 2 )) -η c 1 2 1 ( η 1 -η t 2 2 1 +2 η t τ 2 2 2 ) , q 3 = 2 t 1 t 2 τη 3 ( η 1 -η t 2 2 1 + η t τ 2 2 2 ) , q 4 = t 2 2 τ 2 ( η 1 -η η t 2 3 2 1 ) η 3 = (1 + η c 2 2 1 ) , τ = || ̂ c ||
Proof: Following the argument of Proposition (1) we can write ̂ ̂ c u t = || ̂ c || r , where || ̂ u || = r . Now due to equality constraint (8b), we can express u 1 in terms of r as: u 1 = √ η 1 -η 2 ( t 1 + || ̂ c || r ) 2 Therefore, optimization problem (8) can be rewritten as:
$$\max _ { r } \frac { ( t _ { 1 } + | | \widehat { c } | | r + c _ { 1 } \sqrt { \eta _ { 1 } - \eta _ { 2 } ( t _ { 1 } + | | \widehat { c } | | r ) ^ { 2 } } ) ^ { 2 } } { t _ { 2 } + r ^ { 2 } }$$
By differentiating it with respect to r and equating it to 0, we get the above polynomial equation in terms of r .
Corollary 1. If r ∗ is the real positive root of equation (9) which maximizes the objective function of problem (10) , then the solution of problem (8) is obtained using that r ∗ .
Proof: As u 1 and ̂ u calculated using r ∗ satisfy the constraint (8b) and maximizes (10), hence it is indeed the optimal solution of problem (8).
## IV. RESULTS
In Figure 2 we plot the capacity ( C d ) of source to destination channel with respect to transmit power at first stage ( P 1 ) for multiple α values. Source to relay and relay to destination channel gains were generated from complex Gaussian distribution with mean 0 and variance 1. The
distribution used for source to destination channel is CN (0 , 0 25) . . While plotting the results we averaged C d over 100 such network instances. For individual power constraints we considered P s = 5 and P i = 0 1 . , ∀ i , whereas, for total power constraint we used P tot = P s + M ∗ P i . As the SNR value at the destination is an increasing function of P 1 and α , so the channel capacity increases as we increase both the parameters. In Figure 3 we plot the capacity for both total
Fig. 2. Plot of C d with respect to P 1 for several α values in case of both total and individual constraints.
and individual constraint scenario with respect to number of the relay nodes. As the number of relay nodes increases the second term in the destination SNR also increases, which results in increase of C d .
## V. CONCLUSION
In the current article we have presented a beamforming solution for secure communication using untrusted relay nodes. After justifying SNR based approach, we formulated and solved two optimization problems - one for total power constraint and other for individual power constraints. Our current model assumes a perfect CSI for evaluation of optimal scaling vector which may not be available in several practical scenarios. Therefore, in future we would like to study the performance of our model for delayed CSI.
## REFERENCES
[1] A. Wyner, 'The Wire-tap Channel,' Bell Systems Technical Journal , vol. 54, no. 8, pp. 1355-1387, Jan 1975.
Fig. 3. Plot of C d with respect to number of relays for several P 1 values in case of both total and individual constraints.
- [2] S. Leung-Yan-Cheong and M. Hellman, 'The Gaussian Wire-tap Channel,' IEEE Transactions on Information Theory , vol. 24, no. 4, pp. 451-456, Jul 1978.
- [3] Lun Dong and Zhu Han and Petropulu, A.P. and Poor, H.V., 'Amplify-and-forward based cooperation for secure wireless communications,' in Acoustics, Speech and Signal Processing, 2009. ICASSP 2009. IEEE International Conference on , 2009, pp. 2613-2616.
- [4] J. Zhang and M. C. Gursoy, 'Collaborative relay beamforming for secrecy,' in Communications (ICC), 2010 IEEE International Conference on . IEEE, 2010, pp. 1-5.
- [5] X. He and A. Yener, 'Two-Hop Secure Communication Using an Untrusted Relay: A Case for Cooperative Jamming,' in IEEE Global Telecommunications Conference (GLOBECOM 08) , Dec 2008, pp. 1-5.
- [6] R. Zhang, L. Song, Z. Han, B. Jiao, and M. Debbah, 'Physical Layer Security for Two Way Relay Communications with Friendly Jammers,' in IEEE Global Telecommunications Conference (GLOBECOM 2010) , Dec 2010, pp. 1-6.
- [7] C. Jeong, I.-M. Kim, and D. I. Kim, 'Joint secure beamforming design at the source and the relay for an amplify-and-forward mimo untrusted relay system,' IEEE Transactions on Signal Processing , vol. 60, no. 1, pp. 310-325, Jan 2012.
- [8] X. He and A. Yener, 'Cooperation with an Untrusted Relay: A Secrecy Perspective,' IEEE Transactions on Information Theory , vol. 56, no. 8, pp. 3807-3827, 2010.
- [9] R. Negi and S. Goel, 'Secret Communication using Artificial Noise,' in Proceeding IEEE 62nd Vehicular Technology Conference (VTC-2005-Fall) , vol. 3, Sep 2005, pp. 1906-1910.
- [10] Y. Yang, Q. Li, W.-K. Ma, J. Ge, and M. Lin, 'Optimal Joint Cooperative Beamforming and Artificial Noise Design for Secrecy Rate Maximization in AF Relay Networks,' in IEEE 14th Workshop on Signal Processing Advances in Wireless Communications (SPAWC), 2013 , June 2013, pp. 360-364.
- [11] R. A. Horn and C. R. Johnson, Eds., Matrix Analysis . New York, NY, USA: Cambridge University Press, 1986. | null | [
"Siddhartha Sarma"
] | 2014-08-02T04:50:43+00:00 | 2014-08-02T04:50:43+00:00 | [
"cs.IT",
"math.IT"
] | Beamforming for Secure Communication via Untrusted Relay Nodes Using Artificial Noise | A two-phase beamforming solution for secure communication using untrusted
relay nodes is presented. To thwart eavesdropping attempts of relay nodes, we
deliberately introduce artificial noise in the source message. After pointing
out the incongruity in evaluating secrecy rate in our model for certain
scenarios, we provide an SNR based frame work for secure communication. We
intend to bring down the SNR at each of the untrusted relay nodes below a
certain predefined threshold, whereas, using beamforming we want to boost the
SNR at the destination. With this motive optimal scaling vector is evaluated
for beamforming phase which not only nullifies the artificial noise transmitted
initially, but also maximizes the SNR at the destination. We discuss both the
total and individual power constraint scenarios and provide analytical solution
for both of them. |
1408.0340v2 | ## Motion of macroscopic bodies in terms of quantum evolution
## A. Yu. Samarin
Samara Technical State University, 443100 Samara, Russia
## Abstract
The mechanical motion of the center of mass of a system in physical space is described by using Feynman's representation of quantum evolution. It is shown that alone unitary evolution of a closed system of quantum particles without decoherence and collapse phenomena generates the classical mechanical motion of the system's center of mass, when the number of these particles go to infinity. The approach, introduced here, allows considering a mechanical system with the classical properties, as the set of particles obeying the quantum laws exclusively. The rigorous mathematical image of macroscopic objects in term of quantum mechanics allows considering any non-relativistic processes involving both quantum and macroscopic objects; this is necessary, first of all, to describe quantum measurements.
Keywords: Feynmann's representation, center-of-mass, macroscopic body, Schr¨dinger o cat
## 1. Introdution
Since Schr¨dinger formulated the problem of the description of the motion of a macroo scopic body for the first time in 1935 [1], it has become one of the most widely discussed problems in quantum mechanics. Many attempts, based on different interpretations of quantum mechanics, have been made to overcome the difficulties connected with this problem [2, pp.xvii-xxxix], [3, 4]. However, there is no mathematical representation of the mechanical motion of the macroscopic object in terms of quantum mechanics [5]. Not only does this fact limits the generality of the quantum theory, but it does not also allow to describe the reduction of the wave function [6, p.351], [7, p.36], caused by quantum measurements, in the form of a causal chain of events [8]. Ability to present the macroscopic body as a collection of a large number of the quantum objects is necessary to start the measuring process study having at least some chance of success [9]. The collapse conception as a unique nonlocal physical phenomenon [10, 11, 12, 13, 14] is necessary in order to find out the domain of applicability of the notion 'observable' and, perhaps, to point out the mathematical objects that are suitable for using outside this domain [15, pp.163-195] (these quantities, of course, must be expressed in terms of observables under certain conditions). Such an approach would likely allow to study the processes, detail description of which is impossible using observables due to uncertainty principle [16, 17]. In addition, it would likely allow to describe the evolution of the quantum system in the form of mechanical motion in physical space [18] and introduce the spacetime notion for quantum objects, what is necessary for the creation of quantum gravitation [19, 20].
The macroscopic path of the particle corresponding to the least action principle has been obtained as a classical limit of the path integral [15, pp.29,30]. However this does not solve the problem of macroscopic body description: first of all, the mechanism of
localization of the wave function has not been shown and, secondly, the macroscopic body cannot be considered as a particle if we want to describe the quantum system in general case.
In order to introduce the mathematical image of a macroscopic body, laws of quantum mechanics, first of all, we have to determine the macroscopic body in a suitable manner. In general, the macroscopic body is a system of a vast number of particles. Individual motion any of these particles obeys the quantum laws. Therefore, a quantum system can posses macroscopic properties only as a whole. Moreover, in accordance with [7, p.4] the quantum equations will be in close correspondence with the equation of classical mechanics, in other words, dynamics laws of classical mechanical motion have to be a direct consequence of quantum laws. Taking this fact into account, the following statement can be made: the macroscopic body is a system of quantum particles that the center of mass moves in accordance with the laws of classical mechanics. Namely the center of mass as a formal mathematical notion characterizing a quantum system in the form of a unified object can possess classical properties.
Mechanical motion of the center of mass in physical space is the subject of consideration in this paper
## 2.The roots of the problem
The first serious problem of description of macroscopic objects by means of conventional quantum mechanics is following: the center of mass of a macroscopic body is always localized in space, whereas any quantum system in accordance with this theory is localized only immediately after the collapse. Quantum evolution within the framework of this theory does not result in localization of the center of mass of the system consisting of any large numbers of particles. In order to overcome this difficulty in [4, 21, 22] an accessory postulate has been added to the conventional quantum mechanics postulates. This postulate asserts, that there is a spontaneous localization of quantum particles taking place in the macroscopic body practically all the time. Nevertheless, even so meaningful postulate does not allow to deduce the classical law of the center of mass mechanical motion starting from the quantum mechanics dynamics.
Supposing that the collapse is a consequence of the measuring process and taking into account that the physics of classical systems can be treated without ever referring to measurement [23], the mechanical motion of the center of mass will be considered solely basing on the unitary evolution law, taken in a suitable form.
The least action principle determinimg the mechanical motion of the center of mass of a macroscopic body is formulated for physical space. To deduce this principle from quantum evolution, the latter has to be described correspondingly, i.e. in the spatio-temporal form. A major step towards the spatio-temporal description of unitary quantum evolution was taken by Dirac and Feynman [24, 25].
## 2. The Law of Dynamics
As we consider the system of quantum particles then, first of all, let us determine the notion of a quantum particle in the spatio-temporal form. The quantum particle is an indivisible quantum object, that is transformed into a mass point as the result of the collapse. Formally this means, that coordinates wave function of the quantum particle becomes δ -function as the result of the space localization in physical space. Then the particles can be considered as matter fields or collections of matter fields. This depends on whether the corresponding wave functions are nondegenerate or degenerate. In any case, it is sufficient to consider the nondegenerate case (This does not lead to the loss of generality of the consideration due to superposition principle).
The mechanical motion of these matter fields (continua) can not be described in terms of the Hilbert space: the set of basis vectors of the Hilbert space is countable, whereas
the coordinates set of physical space has cardinality of continuum. This means that the spatial variables are not observables and the coordinate form of the wave equation have to be considered as the dynamic law of mechanical motion in physical space. As we consider the path of the center of mass, the wave equation in the integral form is more suitable than the differential wave equation.
Consider the evolution of the system which consist of n quantum particles in the form of one-dimensional mechanical motion of these particles. Let x ...x 1 n be the space coordinates of these particles. Take as a postulate, the dynamic law of this motion in the form of the integral wave equation [25], [15, pp.163-195]:
$$\Psi _ { t _ { 2 } } ( x _ { 2 } ^ { 1 }, \dots, x _ { 2 } ^ { n } ) \ = \ \int \dots \int K _ { t _ { 2 }, t _ { 1 } } ( x _ { 2 } ^ { 1 }, \dots, x _ { 2 } ^ { n }, x _ { 1 } ^ { 1 }, \dots, x _ { 1 } ^ { n } ) \\ \times \Psi _ { t _ { 1 } } ( x _ { 1 } ^ { 1 }, \dots, x _ { 1 } ^ { n } ) \, d x _ { 1 } ^ { 1 } \cdots d x _ { 1 } ^ { n },$$
where Ψ t 2 ( x , . . . , x 1 2 n 2 ), Ψ ( t 1 x , . . . , x 1 1 n 1 ) are the wave functions of the system at the time t 2 and the time t 1 < t 2 correspondingly; K t 2 ,t 1 ( x , . . . , x 1 2 n 2 , x 1 1 , . . . , x n 1 ) is the kernel of the integral evolution operator. The superscript of the spatial variable denotes the number of the particle, the subscript denotes the instant of time. In accordance with the assumption that the quantum particles are matter fields, the spatial variables can be interpreted as the space coordinates of the individual particles of these continua. The wave function and the kernel of the integral evolution operator depend on time parametrically. Further, instead of the term 'kernel', the term 'transition amplitude' will be used (this amplitude formally corresponds with the transition between the states having spatial localization).
Let the configuration space has the coordinate axes corresponding with the particles of the system. Then, the transition amplitude has the form of continual integral [26] in this configuration space. By Γ denote an arbitrary virtual path in this space. Thus, we have
$$K _ { t _ { 2 }, t _ { 1 } } ( x _ { 2 } ^ { 1 }, \dots, x _ { 2 } ^ { n }, x _ { 1 } ^ { 1 }, \dots, x _ { 1 } ^ { n } ) = \int \exp \frac { i } { \hbar } S _ { 1, 2 } [ \Gamma ] \, [ d \Gamma ].$$
The subscript of the action functional denotes the positions of the system in space and time. Let γ be a virtual path of a quantum particle in space. Substituting γ for Γ in the last expression, we get
$$K _ { t _ { 2 }, t _ { 1 } } ( x _ { 2 } ^ { 1 }, \dots, x _ { 2 } ^ { n }, x _ { 1 } ^ { 1 }, \dots, x _ { 1 } ^ { n } ) \\ = \int \dots \int \exp \frac { i } { \hbar } S _ { 1, 2 } ^ { \Sigma } [ \gamma ^ { 1 }, \dots, \gamma ^ { n } ] \, [ d \gamma ^ { 1 } ] \cdots [ d \gamma ^ { n } ],$$
where glyph[negationslash]
$$S _ { 1, 2 } ^ { \Sigma } [ \gamma ^ { 1 }, \dots, \gamma ^ { n } ] & = \sum _ { j = 1 } ^ { n } \int _ { t _ { 1 } } ^ { t _ { 2 } } \left ( m ^ { j } \frac { ( v ^ { j } ) ^ { 2 } } { 2 } - U ^ { j } ( x ^ { j } ) \\ & \quad - \sum _ { k = 1, k \neq j } ^ { n } U ^ { j k } ( x ^ { j }, x ^ { k } ) \right ) d t \\ \hat { \ } \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \end{array}$$
glyph[negationslash]
is the sum of all the action functionals of the particles forming the system. In this expression m v j ( j ) 2 2 is the kinetic energy of the particle j for the path γ j ; U j ( x j ) is the potential energy of this particle in external fields for the same path; n ∑ k =1 ,k = j U jk ( x , x j k )
is the interaction energy of the particles j and k of the system for the paths γ j and γ k correspondingly.
Consider the motion of the center of mass of the system. Each virtual path of the system in the configuration space determines a unique virtual path of the center of mass in the form
$$X _ { \Gamma } ( t ) = \frac { \sum _ { j = 1 } ^ { n } m ^ { j } x _ { \Gamma } ^ { j } ( t ) } { \sum _ { j = 1 } ^ { n } m ^ { j } }.$$
Here, the path x j Γ ( ) of an individual particle corresponds with the path Γ. In this case t the total functional (3) is the sum of the functionals corresponding with the motion of the center of mass and the relative motion of the particles of the system:
$$S _ { 1 2 } ^ { \Sigma } = S _ { 1 2 } ^ { C } + S _ { 1 2 } ^ { R },$$
where
$$S _ { 1, 2 } ^ { C } = \underset { t _ { 1 } } { \overset { t _ { 2 } } { \int } } \left ( M \frac { ( V ) ^ { 2 } } { 2 } - \underset { j = 1 } { \overset { n } { \sum } } U ^ { j } ( X, \xi ^ { j } ) \right ) d t ;$$
glyph[negationslash]
$$S _ { 1, 2 } ^ { R } = \sum _ { j = 1 } ^ { n } \int _ { t _ { 1 } } ^ { t _ { 2 } } \left ( m ^ { j } \frac { ( \dot { \xi } ^ { j } ) ^ { 2 } } { 2 } - \sum _ { k = 1, k \neq j } ^ { n } U ^ { j k } ( \xi ^ { j }, \xi ^ { k } ) \right ) d t ;$$
V = ˙ - the velocity of the center of mass; X ξ j -the coordinate of the relative movement of the quantum particle j ; M -the mass of the system. Using these notation, instead of (2) we get:
$$& K _ { t _ { 2 }, t _ { 1 } } ( X _ { 2 }, \xi _ { 2 } ^ { 1 }, \dots, \xi _ { 2 } ^ { n - 1 }, X _ { 1 }, \xi _ { 1 } ^ { 1 }, \dots, \xi _ { 1 } ^ { n - 1 } ) \\ & = \int [ d X ( t ) ] \int \dots \int \exp \frac { i } { \hbar } \left ( S _ { 1, 2 } ^ { C } + S _ { 1, 2 } ^ { R } \right ) \\ & \quad \times [ d \xi ^ { 1 } ( t ) ] \cdots [ d \xi ^ { n - 1 } ( t ) ].$$
For us to be able to represent a mechanical system as a matter point, we have to suppose that the potential energy of the particles of this system in the external field does not depend on the space coordinates of the relative movement, i.e. U j ( X,ξ j ) ≈ U j ( X ). In addition, the kinetic energy of the center of mass MV 2 2 does not depend on these coordinates. Then, the motion of the center of mass does not depend on the relative movement. Thus, we have the following equation for the transition amplitude of the center of mass:
$$K _ { t _ { 2 }, t _ { 1 } } ( X _ { 2 }, X _ { 1 } ) \\ = \int \exp \left ( \frac { i } { \hbar } \int _ { t _ { 1 } } ^ { t _ { 2 } } \left ( T ( V ( t ) ) - U ( X ( t ) ) \right ) d t \right ) [ d X ( t ) ].$$
In order to review the path of a macroscopic body, it is necessary to estimate the contribution of different virtual paths to the above continual integral.
## 4. The Macroscopic body Path
There is a formal mathematical procedure which transforms the complex path integral into the real form [27]. According to this procedure, the time variable is transformed into
the complex form t = τ exp iϕ . Then the path integral is considered for the imaginary negative time ( ϕ = -π 2 ⇒ t = -iτ ):
$$K _ { \tau _ { 2 }, \tau _ { 1 } } ( X _ { 2 }, X _ { 1 } ) \\ = \int \exp \left ( - \frac { 1 } { \hbar } \underset { \tau _ { 1 } } { \overset { \tau _ { 2 } } { \int } } \left ( T ( V ( \tau ) ) + U ( X ( \tau ) ) \right ) d \tau \right ) [ d X ( \tau ) ],$$
where T V ( ( τ )) = M 2 ( dX τ ( ) dτ ) 2 is the kinetic energy expressed as the function of time module. Using the previous expression and equation (1), we obtain
$$\Psi _ { \tau _ { 2 } } ( X _ { 2 } ) \\ = C \int \left ( \int \exp \left ( - \frac { 1 } { \hbar } S _ { 1 2 } ^ { E } [ X ( \tau ) ] \right ) [ X ( \tau ) ] \right ) \Psi _ { 1 } ( X _ { 1 } ) \, d X _ { 1 }.$$
Here S E 12 [ X τ ( )] is the Euclidean action functional, C -the normalization factor. Denote by S E 12 ( X ,X 1 2 ) the action functionals that have initial and final positions of the center of mass assigned. Let S E,min 12 ( X ,X 1 2 ) be the least of these action functional. Then we have
$$\Psi _ { \tau _ { 2 } } ( X _ { 2 } ) = C \int \exp \left ( - \frac { 1 } { \hbar } S _ { 1 2 } ^ { E, m i n } ( X _ { 1 }, X _ { 2 } ) \right ) \\ \times \left ( \int \exp \left ( - \frac { 1 } { \hbar } \Delta S _ { 1 2 } ^ { E } [ X ( \tau ) ] \right ) [ X ( \tau ) ] \right ) \Psi _ { 1 } ( X _ { 1 } ) \, d X _ { 1 }.$$
Here ∆ S E 12 [ X τ ( )] = S E 12 [ X τ ( )] -S E,min 12 ( X ,X 1 2 ). Let s E,min 12 be the least of all the action functional S E,min 12 ( X ,X 1 2 ) having different coordinates. Then, we obtain
$$\mathbb { m } \sim _ { 1 2 } \mathbb { m } \times _ { 1 }, \mathbb { m } \times _ { 2 / \dots } \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \imes \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \times \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \ \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \cdot \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \quad \mathbb { m } \.$$
Here ∆ S E,min 12 ( X ,X 2 1 ) = S E,min 12 ( X ,X 2 1 ) -s E,min 12 . The exponential factor exp ( -1 ¯ h s E,min 12 ) ′ .
does not depend on space coordinates and can be included in the normalization factor C Consequently,
$$\Psi _ { \tau _ { 2 } } ( X _ { 2 } ) = C ^ { \prime } \\ \times \int \exp \left ( - \frac { 1 } { \hbar } \Delta S _ { 1 2 } ^ { E, m i n } ( X _ { 1 }, X _ { 2 } ) \right ) \\ \times \left ( \int \exp \left ( - \frac { 1 } { \hbar } \Delta S _ { 1 2 } ^ { E } [ X ( \tau ) ] \right ) [ X ( \tau ) ] \right ) \Psi _ { 1 } ( X _ { 1 } ) \, d X _ { 1 }.$$
glyph[negationslash]
A macroscopic body contains an infinite number of quantum particles. Therefore, for any finitesimal time interval, the center of mass of this system has such Euclidian action that S 12 >> h ¯ on any path. This situation can be formally expressed as ¯ h → 0. In this case the integral measure of all sets of the virtual paths for that ∆ S E,min 12 ( X ,X 1 2 ) = 0 or ∆ S E,min 12 [ X τ ( )] = 0 is equal to zero because of lim exp ¯ h → 0 ( -1 ¯ h ∆ S E,min 12 ( X ,X 1 2 ) ) = 0
glyph[negationslash]
and lim exp ¯ h → 0 ( -1 ¯ h ∆ S E 12 [ X τ ( )] ) = 0. Therefore we have a unique path that determines the last path integral. The spacial part of the wave function Ψ τ 2 ( X 2 ) is the delta-function δ X ( 2 -X min 2 ). The path X min ( τ ) corresponds with the least Euclidian action s E,min 12 . If to go back to the real time, using the analytic continuation, we obtain usual principle of least least action and the spacial wave function Ψ t 2 ( X 2 ) = δ X ( 2 -X min 2 ).
## 4. Conclusion
Thus, the classical mechanical motion of a system is the direct result solely of the quantum evolution of the system (without collapse). The macroscopic motion of the center of mass does not depend on the wave function of the relative motion of the particles forming the system. Therefore decoherence phenomenon does not affect the motion of the center of mass and, thus, cannot not define the system as a macroscopic object.
This paper approach allows us to calculate the deviation scope of the quantum particle system behavior from the classical one, namely, it enables us to find the probability of the center of mass deviation from the path corresponding to the least action principle. This would give an opportunity to establish clearly the connection between quantum and classical descriptions. This connection can be verified experimentally using mesoscopic systems.
In order to describe the collapse as a physical phenomenon it is necessary to consider interaction of the quantum object with the measurer, using the integral wave equation. The wave function collapse have to be the result of the initiation of a macroscopic process in the measuring instrument. This process can be expressed mathematically strictly by the center-of-mass motion of the measuring instrument particles.
## References
- [1] E. Schr¨dinger, o Naturwissenschaften , 23 :48 (1935), 807-812 c rs s e o rf ; 49 (1935), 823828 c rs s e o rf ; 50 (1935), 844-849 c rs s e o rf .
- [2] J. S. Bell, Speakable and Unspeakable in Quantum Mechanics , Cambridge University Press, Cambridge, 2010.
- [3] M. Schlosshauer, 'Decoherence, the measurement problem and interpretation of quantum mechanics', Rev.Mod.Phys. , 76 :1 (2004), 1267-1305, arXiv: /0312059 c rs s e o rf .
- [4] A.Bassi,G.C.Ghirardi, 'Dynamical reduction models', Phys.Rept. , 379 (2003), 257426, arXiv: /0305089 c rs s e o rf .
- [5] N. Bruno, A. Martin, P. Sekatski, N. Sangouard, R. T. Thew, N. Gisin, 'Displacement of entanglement back and forth between the micro and macro domains', Nature Phys. , 9 (2013), 545--548 c rs s e o rf .
- [6] von Johann V. Neumann, Mathematische grundlagen der quantenmechanik , Verlag von Julius Springer, Berlin, 1932.
- [7] P.A.M. Dirac,, Quantum Mechanics , Oxford University Press, Oxford, 1930.
- [8] J. S. Bell, Against 'Measurement' , http://physicsworldarchive.iop.org/ summary/pwa-xml/3/8/phwv3i8a26 , 1990.
- [9] A. Yu. Samarin, 'Space localization of the quantum particle', Vestn. Samar. Gos. Tekhn. Univ. Ser. Fiz.-Mat. Nauki , 1(30) (2013), 387-397 Math-Net.Ru .
- [10] A. Einstein, B. Podolsky, N. Rosen, 'Can quantum mechanics description be considered complete?', Phys. Rev. , 47 (1935), 777-780 c rs s e o rf .
- [11] J. S. Bell, 'On the Einstein-Podolsky-Rosen paradox', Physics , 1 (1964), 195-200.
- [12] J.F. Clauser, M.A. Horne, A. Shimony, R.A. Holt, 'Proposed Experiment to Test Local Hidden-Variable Theories', Phys. Rev. Lett. , 23 (1969), 880-883 c rs s e o rf .
- [13] S.J. Freedman and J.F. Clauser, 'Experimental Test of Local Hidden-Variable Theories', Phys. Rev. Lett. , 28 (1972), 938-941 c rs s e o rf .
- [14] A. Aspect, 'Bell's inequality test: more ideal than ever', Nature , 398 (1999), 189-190 c rs s e o rf .
- [15] R. P. Feynman, A. R. Hibbs, Quantum Mechanics and Path Integrals , International Earth & Planetary Sciences, McGraw-Hill Co., New York.
- [16] E Dzyaloshinskii, E M Lifshitz, Lev P Pitaevskii, 'General theory of Van-der-Vaals forses', Sov. Phys. Usp. , 4 (1961), 153-176 c rs s e o rf .
- [17] L. N. Cooper, 'Bound electron pairs in a degenerate Fermi gase', Physical Review , 104(4) (1956), 1189-1190 c rs s e o rf .
- [18] A. Yu. Samarin, Adv. Studies Theor. Phys. , 8 :1 (2014), 27-34, arXiv: 1407.3559 [quant-ph] c rs s e o rf .
- [19] J.B. Hartle, 'Bohmian histories and decoherent histories', Phys. Rev. A , 69 (2004), 042111-042300, arXiv: /0209104 c rs s e o rf .
- [20] F.Dowker, S. Johnston, R.D. Sorkin, 'Hilbert spaces from path integrals', J. Phys. A: Math. Theor. , 43 (2010), 275302-275323, arXiv: 1002.0589 c rs s e o rf .
- [21] G.C. Ghirardi, R. Grassi, A. Rimini, 'Continuous-spontaneous-reduction model involving gravity', Phys. Rev. A , 42 (1990), 1057-1064 c rs s e o rf http://streaming. ictp.it/preprints/P/89/105.pdf .
- [22] A. Bassi, G.C. Ghirardi, 'Dynamical reduction models with general Gaussian noises', Phys. Rev. A , 65 (2002), 042114-042300, arXiv: /0201122 c rs s e o rf .
- [23] R. F. Werner, 'Comment on 'What Bell did'', J. Phys. A: Math. Theor , 47 (2014), 424011-424012 c rs s e o rf .
- [24] P. A. M. Dirac, 'The Lagrangian in quantum mechanics', Phys. Z. Sowjetunion , 3 (1933 http://www.ifi.unicamp.br/ ~ cabrera/teaching/aula%2015%202010s1. pdf ), 64-72.
- [25] R.P.Feynmann, 'Space-time approach to non-relativistic quantum mechanics', Rev. of Mod. Phys , 20 (1948), 367-387 c rs s e o rf http://web.ihep.su/dbserv/compas/src/ feynman48c/eng.pdf .
- [26] J. Zinn-Justin, Path Integrals in Quantum Mechanics , Oxford Press, Oxford, 2004 c rs s e o rf .
- [27] M. Kac, Probability and Related Topics in Physical Sciences , Interscience Publ., New York, 1959. | null | [
"A. Yu. Samarin"
] | 2014-08-02T04:58:00+00:00 | 2015-08-13T14:11:49+00:00 | [
"quant-ph",
"81S40, 58D30"
] | Macroscopic Body Motion in Terms of Quantum Evolution | Macroscopic body is considered as a system, consisting of an infinite number
of quantum particles. Mechanical motion of the system center of mass in
physical space is described, using Feynman's representation of the quantum
evolution. It is shown that the center of mass of the system moves accordingly
with the minimum action principle if action functional is much higher then
Plank's constant. |
1408.0341v1 | Tunnel magnetoresistance and spin-transfer-torque switching in polycrystalline Co2FeAl full-Heusler alloy magnetic tunnel junctions on Si/SiO2 amorphous substrates
Zhenchao Wen, Hiroaki Sukegawa, Shinya Kasai, Koichiro Inomata, and Seiji Mitani National Institute for Materials Science (NIMS), 1-2-1 Sengen, Tsukuba 305-0047, Japan
## Abstract:
We studied polycrystalline B 2-type Co2FeAl (CFA) full-Heusler alloy based magnetic tunnel junctions (MTJs) fabricated on a Si/SiO2 amorphous substrate. Polycrystalline CFA films with a (001) orientation, a high B 2 ordering, and a flat surface were achieved using a MgO buffer layer. A tunnel magnetoresistance (TMR) ratio up to 175% was obtained for an MTJ with a CFA/MgO/CoFe structure on a 7.5-nm-thick MgO buffer. Spin-transfer torque induced magnetization switching was achieved in the MTJs with a 2-nm-thick polycrystalline CFA film as a switching layer. Using a thermal activation model, the intrinsic critical current density ( J c0 ) was determined to be 8.2 × 10 A/cm , which is lower 6 2 than 2.9 × 10 7 A/cm , the value for epitaxial CFA-MTJs [ 2 Appl. Phys. Lett. 100 , 182403 (2012)]. We found that the Gilbert damping constant ( ) evaluated using ferromagnetic resonance measurements for the polycrystalline CFA film was ~ 0.015 and was almost independent of the CFA thickness (2~18 nm). The low J c0 for the polycrystalline MTJ was mainly attributed to the low of the CFA layer compared with the value in the epitaxial one (~0.04).
## I. INTRODUCTION
Half-metallic ferromagnets (HMFs) draw great interest because of the perfect spin polarization of conduction electrons at the Fermi level, which is considered to enhance the spin-dependent transport efficiency of high-performance spintronic devices. [1-3] Cobaltbased full-Heusler alloys with the chemical formula Co2 YZ (where Y is a transition metal and Z is a main group element), are extensively studied as a type of HMFs owing to their high Curie temperature of approximately 1000 K, high spin polarization, and low damping constant. [4, 5] They exhibit great potential for applications in spintronics, including current-perpendicular-to-plane giant magnetoresistance (CPP-GMR) read heads, [6, 7] magnetoresistive random access memories (MRAMs), [8] and spin transistors such as spinfunctional metal-oxide-semiconductor field-effect transistors (spin-MOSFETs). [9, 10] In particular, magnetic tunnel junctions (MTJs) with Co-based full-Heusler alloy electrodes have been shown tremendously increasing tunnel magnetoresistance (TMR) ratios during the last decade since Inomata et al. [11] demonstrated a TMR ratio of 16% using Co2Cr0.6Fe0.4Al/AlOx/CoFe MTJs at room temperature (RT). [12-21] Recently, a remarkable TMR ratio of approximately 2000% at 4.2 K (354% at RT) was achieved using epitaxial Co2MnSi/MgO/Co2MnSi(001) MTJs, demonstrating the half-metallicity of Cobased Heusler alloys and strong 1 coherent tunneling effect in the MgO/Heusler MTJs [21].
The Co2FeAl (CFA) alloy is of particular interest because of its high spin polarization (a half-metallic electronic structure) [22] and low effective damping constant ( ) ~ 0.001 [23], which are beneficial for enhancing the TMR ratio and lowering the magnetization
switching current of spin-transfer torque (STT). CFA films prepared using sputtering deposition generally have a disordered B 2 structure (swapping between Y and Z sites) rather than an ordered L 21 structure owing to the thermodynamic stability of CFA [24]. Nevertheless, the spin polarization calculated for the L 21 structure is conserved even for the B 2 structure [25]. Importantly, a CFA film with a (001) orientation has a large in-plane lattice spacing ( d (200) / 2 = d (110) = 0.203 nm) compared with other half-metallic Heusler alloys such as Co2FeAl0.5Si0.5 ( d (110) = 0.201 nm) and Co2MnSi ( d (110) = 0.198 nm). Therefore, a nearly perfect CFA/MgO(001) heterostructure is easily achieved by the magnetron sputtering method, and this is favorable for enhancing the coherent tunneling effect [13, 18]. Recently, TMR ratios as high as 360% at RT (785% at 10 K) were demonstrated in epitaxial CFA-based MTJs with a sputter-deposited MgO barrier [17-19]. The large TMR ratio originated from the high spin polarization of the CFA layer and the strong contribution of the coherent tunneling effect through 1 Bloch states in the CFA and the MgO barrier. Moreover, (001)-textured CFA films can be grown on MgO-buffered Si/SiO2 amorphous substrates, and a relatively large TMR ratio of 166% at RT (252% at 48 K) was achieved in a (001)-textured CFA/MgO/CoFe MTJ [20]. Such polycrystalline fullHeusler MTJs on amorphous substrates are desired because of their compatibility with practical industrial applications of full-Heusler spintronic devices, while single-crystal MgO(001) substrates have a limited scope of application.
Furthermore, STT-induced magnetization switching (STT switching), a key technology for writing information in spintronics devices, was realized using MTJs with an epitaxial CFA ultrathin (~1.5 nm) layer as a switching (free) layer [26]. However, a large critical
switching current density ( J c0 ) of 2.9 × 10 A/cm was observed owing to the enhancement 7 2 of , which was ~0.04, of the epitaxial CFA film. In addition, STT switching can be disturbed by the stabilization of intermediate magnetic states possibly because of the presence of in-plane magnetocrystalline anisotropy, which is generally seen in epitaxial magnetic films (e.g., 4-fold anisotropy for cubic (001) films) [27]. Therefore, reducing the undesirable magnetic anisotropy using polycrystalline Heusler alloy films is effective for highly efficient STT switching.
In this work, we systematically studied MTJs with (001)-textured polycrystalline CFA films on Si/SiO2 amorphous substrates. A MgO buffer was introduced in the MTJs for achieving (001) texture with B 2-ordering structure of CFA layers on the amorphous substrates. The (001)-texture, B 2 order, and surface morphology of the polycrystalline CFA films and the TMR effect in the entire MTJ stacks were characterized for varying MgO buffer thickness. Also, the MgO-barrier thickness and resistance-area product ( RA ) dependence of the TMR ratios were investigated for the polycrystalline CFA-MTJs. Furthermore, STT switching was examined in lowRA MTJs with a thin polycrystalline CFA film (2.0 nm) as a free layer. J c0 of 8.2 × 10 A/cm was demonstrated by a thermal 6 2 activation model for switching current; this is far lower than the value for epitaxial CFAMTJs [26]. The values for the polycrystalline CFA films, obtained using a waveguidebased ferromagnetic resonance method, were almost constant against the CFA thickness, and a relatively low of ~0.015 was demonstrated. We attributed the reduction in the J c0 of the STT switching to the reduced of the polycrystalline CFA films.
## II. EXPERIMENT
All multilayer stacks were deposited on thermally oxidized Si/SiO2 amorphous substrates at RT using an ultra-high vacuum magnetron sputtering system with a base pressure lower than 4 10 -7 Pa. MgO layers were deposited from a sintered MgO target by RF sputtering with an RF power density of 2.19 W/cm and an Ar pressure of 10 mTorr. 2 CFA layers were deposited from a stoichiometric Co50Fe25Al25 (at.%) alloy target using DC power. The structural properties and surface morphology of CFA films on MgO buffers were characterized by out-of-plane (2 θ ω -scan) X-ray diffraction (XRD) with Cu Kα radiation (λ = 0.15418 nm) and atomic force microscopy (AFM), respectively. MTJ stacks with the structure of CFA/MgO/Co75Fe25/IrMn/Ru (unit: nm) were deposited on MgObuffered thermally oxidized Si/SiO2 amorphous substrates and patterned into junctions with active area of 5 10 m 2 by conventional lithography methods with Ar ion milling. For STT switching, spin-valve MTJs with the structure of MgO (7.5)/Cr (40)/CFA (2)/MgO (0.6-0.8)/Co75Fe25 (5)/Ru (0.8)/Co75Fe25 (5)/IrMn (15)/Ta (5)/Ru (10) (unit: nm) were prepared on the amorphous substrates and nanofabricated into 100-nm-scaled ellipses. The actual areas of the MTJ nano-pillars were obtained according to the ratio of the RA to the junction resistance; the RA was characterized by current-in-plane tunneling (CIPT) measurement [28] before patterning. The MTJ stacks were post-annealed in a vacuum furnace for 30 minutes under a magnetic field of 5 kOe. The magneto-transport properties were measured using a DC 2- or 4- probe method. The magnetic damping constant of the polycrystalline CFA film was measured by waveguide-based ferromagnetic resonance
(FMR). The films were patterned into rectangular shape elements of 600 × 20 μm using 2 UV lithography together with Ar ion milling, and then coplanar waveguides made of Au were fabricated on them. The FMR signal obtained as change of the real part of S21 signal, was determined using a network analyzer. An external magnetic field along the longitudinal axis was varied from 0 to 1.9 kOe, while the excitation power was fixed as 0 dBm. All measurements were performed at RT.
## III. RESULTS
## A. Effect of MgO buffer on polycrystalline CFA films and TMR
Before MTJ multilayer films were grown, MgO buffer was deposited on a Si/SiO2 amorphous substrate in order to establish the (001)-texture of polycrystalline CFA films by taking advantage of the unique (001)-texture property of MgO layers on an amorphous substrate. CFA films were subsequently grown on the MgO buffer layer, and the structural properties of the CFA films depending on the MgO buffer thickness were investigated by XRD. The out-of-plane XRD patterns of 30-nm-thick CFA films are shown in Fig. 1(a); the CFA films were annealed at T a = 400 ° C, and the thicknesses of the MgO buffers ( t MgO) are 2.5, 5.0, 7.5, and 10.0 nm. In addition to the peaks (denoted by 's') from Si/SiO2 substrates, MgO(002), CFA(002), and (004) peaks were observed along with the absence of other oriented peaks, demonstrating the (001)-texture established in the stacks. Figure 1(b) shows the t MgO dependence of the integrated intensity of CFA(002) peaks for as-deposited CFA films, and CFA films annealed at 400 ° C and 480 ° C. With increasing t MgO, the intensity of the peaks initially increases for all samples owing to the improved (001)-texture of the MgO buffer layer, reaching a maximum at t MgO = 7.5 nm. The reduction in intensity at t MgO
= 10 nm may be caused by the degraded surface morphology of the MgO buffer layer. In addition, the XRD intensity increases with increasing annealing temperature, which indicates that the increasing temperature improves the B 2 order and (001)-texture of the CFA films. In the XRD 2 θ ω -scan with the diffraction vector along CFA[111], (111) reflection was not detected, indicating that the CFA films have a B 2-ordering structure with swapping between Fe and Al atoms while Co atoms occupy the regular sites. The degree of B 2 ordering was estimated according to the ratio of the integrated intensity of the CFA(002) and (004) peaks. The peaks were fitted by Voigt profiles, and the ratio of their integrated intensities, i.e., the ratio of I (002) to I (004), is shown in Fig. 1(c). The maximum I (002)/ (004) value was obtained at I t MgO = 7.5 nm for all of the samples: as-deposited and annealed at 400 ° C or 480 ° C. This value is comparatively large for CFA films annealed at 480 ° C, indicating the improvement of the B 2 ordering and the mosaicity of the CFA films due to annealing at high temperature. The degree of B 2 ordering, SB 2 , can be evaluated using the ratios according to the following equation: [29]
$$S _ { B 2 } \, = \, \int _ { \real } \frac { \lfloor \iota ( 0 0 2 ) / \iota ( 0 0 4 ) \rfloor e x p. } { \iota \iota ( 0 0 0 2 \, \diagdown / \iota ( 0 0 4 \, \diagdown \iota \,. } },$$
where [ (002)/ (004)]exp. is the ratio of the integrated intensity of the (002) peak to that of I I the (004) peak as determined by experiments, and [ (002)/ (004)]cal. is the ideal ratio of the I I two peaks. For t MgO = 7.5 nm, the ordering parameter SB 2 is calculated to be 0.89, 0.95, and 0.98 for CFA films as deposited, annealed at 400 ° C, and 480 ° C, respectively. The results demonstrate a high B 2 order and an excellent (001)-texture were established in the polycrystalline CFA films on MgO-buffered Si/SiO2 amorphous substrates.
For stacking MTJ multilayers with the polycrystalline CFA films, the t MgO dependence of the surface morphology of the CFA films was investigated. Figure 2 shows the average surface roughness ( Ra ) and peak-to-valley ( P V -) value as a function of t MgO for 30-nm-thick CFA films annealed at 400 ° C and 480 ° C, respectively. For the samples annealed at 400 ° C, flat surfaces with Ra ~ 0.1 nm and P V -ranging from 1.3 to 1.5 nm were observed for all values of t MgO. The inset of Fig. 2 shows an example of AFM images of the samples (annealed at 400 ° C on a 7.5-nm-thick MgO buffer). The results indicate the feasibility of stacking MTJs with a thin MgO barrier. In addition, the samples annealed at 480 ° C with higher Ra and P V -values were observed as well as a large t MgO dependence.
The whole MTJ stacks with the structure of MgO ( t MgO)/CFA (30)/MgO ( barr)/Co75Fe25 t (5)/IrMn (15)/Ru (10) (unit: nm) were then fabricated on Si/SiO2 substrates with varying MgO buffer thickness t MgO (2.5 10.0 nm) and MgO barrier thickness t barr (1.5, 1.8, and 2.0 nm). The stacks were annealed at 370 ° C in the presence of a magnetic field of 5 kOe. Figure 3 shows TMR ratios as a function of t MgO for the polycrystalline CFA-MTJs measured at RT using CIPT. The 30-nm-thick CFA films were post-annealed at T a = 400 ° C and 480 ° C in order to improve the B 2 ordering. The TMR ratios obtained in MTJs with T a = 400 ° C were higher than those of MTJs with T a = 480 ° C, which can be attributed to a better CFA/MgO-barrier interface due to the flat CFA surface annealed at 400 ° , C although a higher degree of B 2 ordering was observed for T a = 480 ° C, as shown in Figs. 1 and 2. With increasing t MgO, the TMR ratio increases and is nearly saturated at t MgO > 5 nm, indicating that high-quality CFA films with B 2 order and (001)-texture were established with the more than 5-nm-thick MgO buffer layer, which is consistent with the XRD
analyses. A slight reduction in the TMR ratio was observed at t MgO = 10 nm, which could be caused by the reduction in the degree of (001)-orientation of the MgO buffer. The MTJs with a 1.8-nm-thick MgO barrier exhibit larger TMR ratios than those with 1.5- and 2.0nm-thick MgO barriers, which could be due to the plastic relaxation of the MgO barrier [30] and/or the oscillatory behavior of the TMR ratio as a function of MgO thickness [18].
## B. MgO barrier thickness and RA dependences of TMR
In order to realize STT switching in the polycrystalline CFA-MTJs, it is expected that introduction of a conductive underlayer, reduction in the free-layer thickness, and control of the RA of the barrier layer are required. The MgO barrier thickness and RA dependences of the TMR ratio were investigated using spin-valve MTJs with the structure of MgO buffer (7.5)/Cr (40)/CFA (30)/MgO ( t barr : 1.2-2.0)/Co75Fe25 (5)/IrMn (15)/Ta (5)/Ru (10) (unit: nm) on a Si/SiO2 substrate. A Cr underlayer was selected as the conductive electrode because Cr has a very small lattice mismatch with CFA (~0.6%) and can further facilitate the ordering structure of full-Heusler alloys [13]. The Cr layers for the samples as deposited, annealed at 400 ° C, and 600 ° C were prepared on the 7.5-nm-thick MgO buffer for the MTJ stacks. The entire stacks were annealed at 370 ° C in the presence of a magnetic field of 5000 Oe, and then their TMR ratios and RA values were characterized using CIPT measurement.
Figure 4(a) shows the dependence of the TMR ratios on the nominal thickness of the MgO barrier ( t barr ) for the MTJs with different Cr annealing conditions. For the samples that were as-deposited and annealed at 400 ° C, the TMR ratio increases with t barr , and TMR ratios greater than 100% were achieved for the whole range of t barr . This means that the
(001)-texture and B 2 ordering of CFA films can be maintained on the MgO/Cr buffer layers. We obtained the largest TMR ratio of 175% for the 400 ° C annealed sample with t barr = 1.95 nm; this TMR ratio is higher than 166%, which was observed in the MTJ without the Cr buffer, which indicates that the Cr buffer with optimal conditions promotes CFA(001) growth and improves the effective tunneling spin polarization. On the other hand, the samples annealed at a high temperature (600 ° C) exhibited smaller TMR ratios (80-120%). This was attributed to the rough surface ( Ra = 0.4 nm and P V -= 3.2 nm) of the CFA film on the Cr layer annealed at 600 ° C, which can lead to a declined crystalline orientation of the MgO(001) barrier. Furthermore, oscillation behavior of the TMR ratios as a function of t barr was observed for all of the structures. The TMR oscillation behavior is typically observed in epitaxial MTJs such as Fe/MgO/Fe [31], Co2MnSi/MgO/Co2MnSi [32], Co2Cr0.6Fe0.4Al/MgO/Co2Cr0.6Fe0.4Al [33], and CFA/MgO/CoFe [18] MTJs, while it is absent in polycrystalline MTJs such as CoFeB/MgO/CoFeB [34] MTJs. More remarkable oscillation amplitude in epitaxial full-Heusler alloy-based MTJs than that of epitaxial Fe/MgO/Fe MTJs was observed, which may be related to the electronic structures of fullHeusler alloy electrodes and the full-Heusler/MgO interface; however, the origin has not been understood yet. The unexpected oscillation behavior in the polycrystalline CFA-MTJs may be also attributed to the unique electronic structure of CFA and the interface. In addition, the flat buffer layer with a good crystallinity enabled us to achieve a well-defined layer-by-layer growth for the CFA layer and the MgO barrier, which may be advantageous for observing the oscillatory behavior. The oscillation period was approximately 0.2 nm in nominal thickness, which seems to be shorter than that for the epitaxial CFA/MgO/CoFe
(0.32 nm, short-period) [18]. Further investigation is needed to clarify the origin of the behavior.
The RA as a function of the MgO barrier thickness is plotted in Fig. 4(b). We observed a typical behavior of an exponential increase with increasing t barr . According to the WenzelKramer-Brillouin (WKB) approximation, the relationship between RA and t barr can be expressed as follows:
$$\colon e x p ( \frac { 4 \pi \sqrt { 2 m \phi } } { h } \ t _ { b a r r } ),$$
where h , m , and ϕ are Planck's constant, the effective electron mass assumed as a free electron mass (9.11 × 10 -31 kg) here, and the barrier height energy of the tunnel barrier, respectively [31]. A similar barrier height of 0.7 eV was obtained for all three samples by the fitting of RA t -barr curves. This value is greater than the reported values for Fe/MgO/Fe grown using molecular-beam-epitaxy (MBE) (0.39 eV) [31], sputtered CoFeB/MgO/CoFeB (0.29-0.39 eV) [35-37], and CoFeB/MgO (electron-beam evaporated)/CoFeB (0.48 eV) [38] MTJs. The reasons may be due to the different densities of oxygen vacancy defects in the MgO barriers and/or the deviation of the actual MgO thickness from the nominal one.
Figure 5 shows TMR ratios in a lowRA regime for the polycrystalline CFA-MTJs with 2-nm-thick CFA film as a free layer. The MTJs were annealed at 225 ° C for 30 minutes in order to reduce the influence of Cr layer to CFA. A TMR ratio of 40-60% was achieved with an RA of 7-20 m 2 (nominal MgO thickness: 0.6-0.8 nm) for the 100-nm-scaled
elliptical MTJs with the thin CFA layer, which is favorable for achieving STT switching in the MTJ stacks with a thin CFA free layer and the low RA value. In addition, the TMR ratio of the polycrystalline MTJs with 2-nm-thick CFA film is comparable to that with a thick (30 nm) CFA film at a low RA value, as shown in Fig. 5, which indicates that the (001)texture and B 2 ordering can be maintained in the thin 2-nm-thick CFA films.
## C. STT-induced magnetization switching
STT-induced magnetization switching was performed in spin-valve MTJs with the structure of Si/SiO2-substrate/MgO (7.5)/Cr (40)/CFA (2)/MgO (0.6-0.8)/Co75Fe25 (5)/Ru (0.8)/Co75Fe25 (5)/IrMn (15)/Ta (5)/Ru (10) (unit: nm). A schematic of the structure of a polycrystalline CFA-MTJ nanopillar is shown in Fig. 6(a). The synthetic antiferromagnetic coupling exchange bias of CoFe/Ru/CoFe/IrMn was employed to reduce the offset magnetic field of hysteresis loops. Figure 6(b) indicates the tunneling resistance of an MTJ nanopillar as a function of the applied magnetic field ( H ) measured with a DC bias voltage of 1 mV. A TMR ratio of 43% was observed in the MTJ with a thin CFA free layer (2.0 nm) and MgO barrier (0.75 nm). Sharp switching between parallel (P) and antiparallel (AP) magnetic configurations was observed. The RA of the MTJ was determined using CIPT measurements to be 13 m 2 , and the active area of the MTJ nanopillar was calculated to be 1.24 × 10 -2 m 2 . The hysteresis offset field ( H offset ) and the coercivity field ( H c ) were determined using the R-H loops to be -11 and 26 Oe, respectively. Figure 6(c) shows the representative resistance-current ( R-I ) loops of the CFA-MTJ nanopillar measured by a DC current with a sweep rate of 1.2 × 10 4 A/s at different magnetic fields of 0, -11 and -20
Oe, respectively. The positive current indicates that electrons flow from the bottom electrode to the top electrode. Magnetic switching between P (low-resistance) and AP (high-resistance) states was achieved owing to the current. When current was applied in the negative (positive) direction, the P (AP) state can be obtained from AP (P) state, corresponding to magnetization reversal of the CFA free layer. Also, the critical switching currents ( I c ) in both directions shift in the negative direction with the decrease of magnetic field from 0 to -20 Oe. These results indicate typical behaviors of STT-induced magnetization switching.
Since the STT switching by the DC current is a thermally activated process [39-41], we use a thermal activation model for switching currents deduced from R I - loops to evaluate the intrinsic critical switching current density ( J c0 ) and thermal stability factor 0 (= K V k T u / B ) for the MTJ, where K u is the uniaxial magnetic anisotropy, V is the volume of the free layer, k B is the Boltzmann constant, and T is the absolute temperature. In the thermal activation model, the sweep current ( ) is assumed to increase linearly with time , I t t i.e., I t ( ) = vt , and the cumulative probability distribution function P t ( ) of the switching current in H offset can be expressed as
$$P \left ( t \right ) = \ 1 - \exp \left ( - \frac { f _ { 0 } l _ { c 0 } } { v _ { \Delta _ { 0 } } } \left \{ \exp \left [ - \Delta _ { 0 } \left ( 1 - \frac { v t } { l _ { c 0 } } \right ) \right ] - \exp [ - \Delta _ { 0 } ] \right \} \right ), \quad ( 3 )$$
where f 0 is the effective attempt frequency (=10 9 Hz), I c0 is the intrinsic switching current, and v is a constant sweep rate of the sweep current in the measurement of R I - loops [41]. The distribution of the critical switching current I c was obtained by repeating the measurement of the R I -loops for 300 times. Figure 6(d) shows typical R I -loops at H offset =
-11 Oe for the polycrystalline CFA-MTJ nanopillar. The mean critical currents in the positive ( I c,P AP) and negative ( I c,AP P) directions are determined to be 530 and 400 μA, corresponding to the critical current density of 4.3 × 10 ( 6 J c,P AP) and 3.2 × 10 6 A/cm 2 ( J c,AP P), respectively. Figures 6 (e) and (f) show the switching probability for I c,P AP and I c, AP P as a function of the sweep current. Using the constant sweep rate v = 1.2 × 10 -4 A/s in the measurement of R I - loops, the switching probability was fitted using Eq. (3), as shown by the solid lines. As a result, the intrinsic current density of J c0,P AP ( J c0,AP P) = 9.1 × 10 A/cm (7.3 × 10 A/cm ) and thermal stability of 6 2 6 2 0,P AP ( 0,AP P) = 30.0 (28.4) were achieved for the polycrystalline CFA-MTJ nanopillar.
## D. Gilbert damping of the polycrystalline CFA film
The Gilbert damping parameter, , is a critical parameter for determining the critical current density of STT switching. In order to examine for the polycrystalline CFA film, waveguide-based FMR was performed in a single ferromagnetic layered sample, consisting of SiO2-substrate//MgO(7.5 nm)/Cr(40 nm)/CFA(2-18 nm) structure. Typical FMR spectra with varied external magnetic fields ( H ext ) for a sample of 18-nm-thick CFA are shown in the inset of Fig. 7(a). A clear shift in the resonant frequency can be seen as H ext increases from 500 to 1500 Oe. The peak intensity is relatively small at low magnetic field, possibly because of the anisotropy distribution inside the film.
Figures 7(a) and (b) show the H ext dependence of the resonant frequency ( f 0 ), demagnetization field ( H d ), and magnetic damping parameter ( H), respectively, estimated by fitting each spectrum using an analytical solution [42]. Here, we assume the
gyromagnetic ratio ( ) as 2 × 0.00297 GHz/Oe and neglect the in-plane magnetic anisotropic field. Both H d and H exhibit a weak dependence on H ext possibly due to the anisotropy distribution, which becomes saturated at a high magnetic field range. The f 0 was fitted using the simplified Kittel formula:
$$f _ { 0 } = \frac { \gamma } { 2 \pi } \sqrt { H _ { e x t } ( H _ { e x t } + H _ { d } ) } \quad \quad ( 4 )$$
Then, we obtained H d = 12154 ± 17.9 Oe, which agrees well with the values obtained in the individual resonant spectrum for H ext > 300 Oe. The magnetic field dependence of can be excluded by the fitting equation,
$$\alpha _ { H } = \alpha + k \exp \left ( - H _ { e x t } / H _ { 0 } \right ), \text{ \quad \ } ( 5 )$$
where k and H 0 are fitting parameters, and is estimated to be 0.0148 ± 0.0003. Figure 7(c) summarizes CFA thickness dependence of the saturation magnetization, M s , and Gilbert damping constant, , of polycrystalline CFA films. The weak thickness dependence of M s and indicates that the CFA film guarantees a good quality even in a thin thickness regime at around 2 nm.
## IV. DISCUSSION
Using polycrystalline CFA full-Heusler thin films, MTJs with the structure of CFA/MgO/Co75Fe25 were successfully fabricated on a SiO2 amorphous substrate. The effects of the MgO buffer on the structural properties of the polycrystalline CFA films and the TMR in the MTJs were investigated. Optimized (001)-texture, B 2 order, and surface
morphology of the CFA films was demonstrated on a 7.5-nm-thick MgO buffer, which contributes a large TMR ratio in the whole polycrystalline MTJ stacks. In order to achieve STT switching, Cr underlayers were utilized as a conductive electrode on the optimized MgO buffer. The Cr layer was known to have very small lattice mismatch (~0.6%) with CFA and facilitate the ordering structure of full-Heusler alloys. A TMR ratio of 175% was achieved in the polycrystalline CFA-MTJs on MgO/Cr-buffered Si/SiO2 substrate. The buffer layer dependence of the structural properties of polycrystalline CFA films and the TMR ratios in entire MTJ stacks are significant for practical spintronic applications of fullHeusler alloy materials. A proper buffer layer with minimal diffusion, and enhanced (001)texture and ordering parameter of polycrystalline full-Heusler alloys is required for further increasing the TMR ratios of the MTJs.
STT switching was performed in the CFA-MTJs with polycrystalline CFA as a free layer. The average intrinsic current density J c0 = ( J c0,P AP + J c0,AP P)/2 = 8.2 × 10 A/cm 6 2 for the polycrystalline CFA-MTJ is generally comparable to that reported for CoFeB-MTJs with in-plane magnetization [43, 44]; however, it is much lower than the value of 2.9 × 10 7 A/cm 2 for epitaxial CFA-MTJs [26]. Based on Slonczewski's model of STT switching [45, 46], the simplified J c0 , ignoring external magnetic fields, is given by,
$$\L e f f, \quad \ \ \ ( 6 )$$
where e is the electron charge, is reduced Planck's constant, M s is the saturation magnetization, is the thickness of the free layer, t is the spin-transfer efficiency, and Heff the effective field acting on the free layer, including the magnetocrystalline anisotropy field,
demagnetization field, stray field. The polycrystalline CFA free layer ( = 2.0 nm) is thicker t than the epitaxial one ( = 1.5 nm). A similar t M s (~1000 emu/cm ) at RT was observed for 3 both polycrystalline and epitaxial CFA films. The of the polycrystalline CFA-MTJs should be smaller than that of the epitaxial CFA-MTJs because the TMR ratio of the polycrystalline CFA-MTJs (43%) is lower than that of the epitaxial CFA-MTJs (60%) [26]. Accordingly, the low J c0 in the polycrystalline CFA-MTJs can be mainly attributed to the small of the polycrystalline CFA free layer (~0.015) compared with that of the epitaxial CFA films (~0.04) [26]. For the polycrystalline CFA-MTJs, annealing was performed with a low temperature of 225 ° C and a short time of 30 minutes, whereas the epitaxial CFAMTJs were annealed at 360 ° C for 1 hour; this could be a factor in the reduction in [47]. As a result, a lower J c0 was obtained in the polycrystalline CFA-MTJs than that in the epitaxial ones. However, the value is still greater than that reported for a 50-nm-thick CFA film on a MgO layer annealed at 600 C (~0.001) [23]; this may be attributed to the inter-diffusion of Cr atoms into the CFA layer and the residual magnetic moments on the Cr surface. Consequently, a proper buffer material is strongly required for CFA fullHeusler MTJs in order to reduce the of the free layer and thus the J c0 of STT switching. Another factor for reducing J c0 is the magnetic anisotropy of the free layer. In perpendicular anisotropy CoFeB/MgO/CoFeB tunnel junctions, a low current density for STT switching was demonstrated owing to the perpendicular magnetic anisotropy (PMA) [48-53]. To evaluate the contribution of interface PMA, magnetization measurements at in-plane and out-of-plane of the polycrystalline CFA film was performed. An effective anisotropy energy density ( K eff ) of -5 × 10 erg/cm was obtained for a 2-nm-thick CFA/MgO 6 3
structure where the negative K eff indicates the CFA layer is in-plane magnetized. In general, K eff can be simply expressed by the equation of K eff = K v + K t i / , where K v is the volume anisotropy energy density which can be treated as demagnetization energy density (2π M s 2 ) for simplicity, K i is the interface anisotropy energy density, and t is the thickness of the CFA layer. As a result, the value of K i can be calculated to be 0.25 erg/cm , indicating that 2 an interface PMA is induced at the CFA/MgO interface. In the CFA/MgO-MTJs, the interface PMA can cause a reduced effective anisotropy, and may also play a role for the reduction in J c0 . For further decreasing J c0 , a chosen buffer material, out-of-plane magnetization of the CFA free layer [54-56], and/or advanced fabricating techniques for high-quality CFA-MTJs are required.
## V. CONCLUSION
In conclusion, TMR ratios and STT-induced magnetization switching were studied in (001)-textured polycrystalline CFA full-Heusler based MTJs on Si/SiO2 amorphous substrates. CFA films with a good (001)-texture and high B 2 order were achieved on MgObuffered Si/SiO2 amorphous substrates. The MgO barrier thickness and RA dependences of the TMR ratio in the polycrystalline CFA-MTJs were also studied. Moreover, STT switching was achieved in the MTJs with a thin polycrystalline CFA film (2.0 nm) as a free layer. The J c0 of 8.2 × 10 A/cm was demonstrated for the polycrystalline CFA-MTJs with 6 2 in-plane magnetization using a thermal activation model for a cumulative switching probability distribution with a sweep current. The J c0 is much lower than that reported for epitaxial CFA-MTJs, which is mainly attributed to the reduced of the polycrystalline CFA free layer.
## ACKNOWLEDGMENTS
This work was partly supported by the Japan Science and Technology Agency, CREST, and JSPS KAKENHI Grant Number 23246006.
## Reference:
- [1] R. A. de Groot, F. M. Mueller, P. G. vanEngen, and K. H. J. Buschow, New class of materials: halfmetallic ferromagnets, Phys. Rev. Lett. 50 , 2024 (1983).
- [2] C. Felser, G. H. Fecher, and B. Balke, Spintronics: a challenge for materials science and solid-state chemistry, Angew. Chem., Int. Ed. 46 , 668 (2007).
- [3] T. Graf, C. Felser, and S. S. P. Parkin, Simple rules for the understanding of Heusler compounds, Prog. Solid State Chem. 39 , 1 (2011).
- [4] I. Galanakis, P. Dederichs, and N. Papanikolaou, Slater-Pauling behavior and origin of the halfmetallicity of the full-Heusler alloys, Phys. Rev. B 66 , 174429 (2002).
- [5] M. Oogane, T. Kubota, Y. Kota, S. Mizukami, H. Naganuma, A. Sakuma, and Y. Ando, Gilbert magnetic damping constant of epitaxially grown Co-based Heusler alloy thin films, Appl. Phys. Lett. 96 , 252501 (2010).
- [6] T. Furubayashi, K. Kodama, H. Sukegawa, Y. K. Takahashi, K. Inomata, and K. Hono, Currentperpendicular-to-plane giant magnetoresistance in spin-valve structures using epitaxial Co2FeAl0.5Si0.5/Ag/ Co2FeAl0.5Si0.5 trilayers, Appl. Phys. Lett. 93 , 122507 (2008).
- [7] Y. Sakuraba, K. Izumi, T. Iwase, S. Bosu, K. Saito, K. Takanashi, Y. Miura, K. Futatsukawa, K. Abe, and M. Shirai, Mechanism of large magnetoresistance in Co2MnSi/Ag/Co2MnSi devices with current perpendicular to the plane, Phys. Rev. B 82 , 094444 (2010).
- [8] Z. Bai, L. Shen, G. Han, and Y. P. Feng, Data storage: review of Heusler compounds, SPIN 02 , 1230006 (2012).
- [9] Y. Shuto, R. Nakane, W. Wang, H. Sukegawa, S. Yamamoto, M. Tanaka, K. Inomata, and S. Sugahara, A New Spin-Functional Metal-Oxide-Semiconductor Field-Effect Transistor Based on Magnetic Tunnel Junction Technology: Pseudo-Spin-MOSFET, Appl. Phys. Express 3 , 013003 (2010).
- [10] Y. Shuto, S. Yamamoto, H. Sukegawa, Z. C Wen, R Nakane, S. Mitani, M. Tanaka, K. Inomata, and S. Sugahara, Design and performance of pseudo-spin-MOSFETs using nano-CMOS devices, IEDM, 29.6 (2012).
- [11] K. Inomata, S. Okamura, R. Goto, and N. Tezuka, Large Tunneling Magnetoresistance at Room Temperature Using a Heusler Alloy with the B2 Structure, Jpn. J. Appl. Phys. 42 , L419 (2003).
- [12] S. Tsunegi, Y. Sakuraba, M. Oogane, K. Takanashi, and Yasuo Ando, Large tunnel magnetoresistance in magnetic tunnel junctions using a Co2 MnSi Heusler alloy electrode and a MgO barrier, Appl. Phys. Lett. 93 , 112506 (2008).
- [13] H. Sukegawa, W. H. Wang, R. Shan, T. Nakatani, K. Inomata, and K. Hono, Spin-polarized tunneling spectroscopy of fully epitaxial magnetic tunnel junctions using Co2FeAl0.5Si0.5 Heusler alloy electrodes, Phys. Rev. B 79 , 184418 (2009).
- [14] R. Shan, H. Sukegawa, W. H. Wang, M. Kodzuka, T. Furubayashi, T. Ohkubo, S. Mitani, K. Inomata, and K. Hono, Demonstration of Half-Metallicity in Fermi-Level-Tuned Heusler Alloy Co2FeAl0.5Si0.5 at Room Temperature, Phys. Rev. Lett. 102 , 246601 (2009).
- [15] N. Tezuka, N. Ikeda, F. Mitsuhashi, and S. Sugimoto, Improved tunnel magnetoresistance of magnetic tunnel junctions with Heusler Co2FeAl0.5Si0.5 electrodes fabricated by molecular beam epitaxy, Appl. Phys. Lett. 94 , 162504 (2009).
| [16] | M. Yamamoto, T. Ishikawa, T. Taira, G. Li, K. Matsuda and T. Uemura, Effect of defects in Heusler alloy thin films on spin-dependent tunnelling characteristics of Co 2 MnSi/MgO / Co 2 MnSi and Co 2 MnGe/MgO / Co 2 MnGe magnetic tunnel junctions, J. Phys.: Condens. Matter. 22 , 164212 (2010). |
|--------|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| [17] | W. H. Wang, H. Sukegawa, R. Shan, S. Mitani, and K. Inomata, Giant Tunneling Magnetoresistance up to 330% at Room Temperature in Sputter Deposited Co 2 FeAl/MgO/CoFe Magnetic Tunnel Junctions, Appl. Phys. Lett. 95 , 182502 (2009). |
| [18] | W. H. Wang, E. Liu, M. Kodzuka, H. Sukegawa, M. Wojcik, E. Jedryka, G. H. Wu, K. Inomata, S. Mitani, and K. Hono, Coherent tunneling and giant tunneling magnetoresistance in Co FeAl/MgO/CoFe magnetic tunneling junctions, Phys. Rev. B , 140402 (R) (2010). |
| [19] | 2 81 W. H. Wang, H. Sukegawa, and K. Inomata, Temperature dependence of tunneling magnetoresistance in epitaxial magnetic tunnel junctions using a Co 2 FeAl Heusler alloy electrode, Phys. Rev. B 82 , 092402 (2010). |
| [20] | Z. C. Wen, H. Sukegawa, S. Mitani, and K. Inomata, Tunnel magnetoresistance in textured Co 2 FeAl/MgO/CoFe magnetic tunnel junctions on a Si/SiO 2 amorphous substrate, Appl. Phys. Lett. 98 , 192505 (2011). |
| [21] | H. Liu, Y. Honda, T. Taira, K. Matsuda, M. Arita, T. Uemura, and M. Yamamotoa, Giant tunneling magnetoresistance in epitaxial Co 2 MnSi/MgO/Co 2 MnSi magnetic tunnel junctions by half- metallicity of Co 2 MnSi and coherent tunneling, Appl. Phys. Lett. 101 , 132418 (2012). |
| [22] | S. Wurmehl, G. H. Fecher, K. Kroth, F. Kronast, H. A. Dür, r Y. Takeda, Y. Saitoh, K. Kobayashi, H.-J. Lin, G. Schö nhense, and C. Felser, Electronic structure and spectroscopy of the quaternary Heusler alloy Co 2 Cr 1- x Fe x Al, J. Phys. D: Appl. Phys. 39 , 803 (2006). |
| [23] | S. Mizukami, D. Watanabe, M. Oogane, Y. Ando, Y. Miura, M. Shirai, and T. Miyazaki, Low damping constant for Co 2 FeAl Heusler alloy films and its correlation with density of states, J. Appl. Phys. 105, 07D306 (2009). |
| [24] | R. Y. Umetsu, A. Okubo, and R. Kainuma, Magnetic and chemical order-disorder transformations in Co 2 Fe(Ga 1- x Si x ) and Co 2 Fe(Al 1- y Si y ) Heusler alloys, J. Appl. Phys. 111 , 073909 (2012). |
| [25] | Y. Miura, K. Nagao, and M. Shirai, Atomic disorder effects on half-metallicity of the full-Heusler alloys Co 2 (Cr 1-x Fe x )Al: A first-principles study, Phys. Rev. B 69 , 144413 (2004). |
| [26] | H. Sukegawa, Z. C. Wen, K. Kondou, S. Kasai, S. Mitani, and K. Inomata, Spin-transfer switching in full-Heusler Co 2 FeAl-based magnetic tunnel junctions, Appl. Phys. Lett. 100 , 182403 (2012). |
| [27] | H. Sukegawa, S. Kasai, T. Furubayashi, S. Mitani, and K. Inomata, Spin-transfer switching in an epitaxial spin-valve nanopillar with a full-Heusler Co 2 FeAl 0.5 Si 0.5 alloy, Appl. Phys. Lett. 96 , 042508 (2010). |
| [28] | D. C. Worledge, and P. L. Trouilloud, Magnetoresistance measurement of unpatterned magnetic tunnel junction wafers by current-in-plane tunneling, Appl. Phys. Lett. 83 , 84 (2003). |
| [29] | S. Okamura, A. Miyazaki, N. Tezuka, S. Sugimoto, and K. Inomata, Epitaxial Growth of Ordered Co 2 (Cr 1-x Fe x )Al Full-Heusler Alloy Films on Single Crystal. Substrates, Mater. Trans. 47 , 15 (2006). |
| [30] | J. L. Vassent, M. Dynna, A. Marty, B. Gilles, and G. Patrat, A study of growth and the relaxation of elastic strain in MgO on Fe(001), J. Appl. Phys. 80 , 5727 (1996). |
| [31] | S. Yuasa, T. Nagahama, A. Fukushima, Y. Suzuki, and K. Ando, Giant room-temperature magnetoresistance in single-crystal Fe/MgO/Fe magnetic tunnel junctions, Nature Mater. 3 , 868 (2004). |
| [32] | T. Ishikawa, S. Hakamata, K. Matsuda, T. Uemura, and M. Yamamoto, Fabrication of fully epitaxial Co 2 MnSi/MgO/Co 2 MnSi magnetic tunnel junctions, J. Appl. Phys. 103 , 07A919 (2008). |
| [33] | T. Marukame, T. Ishikawa, T. Taira, K. Matsuda, T. Uemura, and M. Yamamoto, Giant oscillations in spin-dependent tunneling resistances as a function of barrier thickness in fully epitaxial magnetic tunnel junctions with a MgO barrier, Phys. Rev. B 81 , 134432 (2010). |
| [34] | D. D. Djayaprawira, K. Tsunekawa, M. Nagai, H. Maehara, S. Yamagata, N. Watanabe, S. Yuasa, |
- Y. Suzuki, and K. Ando, 230% room-temperature magnetoresistance in CoFeB/MgO/CoFeB magnetic tunnel junctions, Appl. Phys. Lett. 86 , 092502 (2005).
- [35] K. Tsunekawa, D. D. Djayaprawira, M. Nagai, H. Maehara, S. Yamagata, N. Watanabe, S. Yuasa, Y. Suzuki, and K. Ando, Giant tunneling magnetoresistance effect in low-resistance CoFeB/MgO(001)/CoFeB magnetic tunnel junctions for read-head applications, Appl. Phys. Lett. 87 , 072503 (2005).
- [36] J. Hayakawa, S. Ikeda, F. Matsukura, H. Takahashi, and H. Ohno, Dependence of Giant Tunnel Magnetoresistance of Sputtered CoFeB/MgO/CoFeB Magnetic Tunnel Junctions on MgO Barrier Thickness and Annealing Temperature, Jpn. J. Appl. Phys. 44 , L587 (2005).
- [37] H. Sukegawa, W. H. Wang, R. Shan, and K. Inomata, Tunnel Magnetoresistance in Full-Heusler Co2FeAl0.5Si0.5-Based Magnetic Tunnel Junctions, J. Magn. Soc. Jpn. 33 , 256 (2009).
- [38] H. Kurt, K. Oguz, T. Niizeki, and J. M. D. Coey, Giant tunneling magnetoresistance with electron beam evaporated MgO barrier and CoFeB electrodes, J. Appl. Phys. 107 , 083920 (2010).
- [39] R. H. Koch, J. A. Katine, and J. Z. Sun, Time-Resolved Reversal of Spin-Transfer Switching in a Nanomagnet, Phys. Rev. Lett. 92 , 088302 (2004).
- [40] Z. Li and S. Zhang, Thermally assisted magnetization reversal in the presence of a spin-transfer torque, Phys. Rev. B 69 , 134416 (2004).
- [41] T. Taniguchi, and H. Imamura, Dependence of Spin Torque Switching Probability on Electric Current, J. Nanosci. Nanotechnol. 12 , 7520 (2012); Theoretical study on dependence of thermal switching time of synthetic free layer on coupling field, J. Appl. Phys. 111 , 07C901 (2012).
- [42] N. Toda, K. Saito, K. Ohta, H. Maekawa, M. Mizuguchi, M. Shiraishi, and Y. Suzuki, Highly sensitive ferromagnetic resonance measurements using coplanar waveguides, J. Magn. Soc. Jpn . 31 , 435 (2007).
- [43] Z. Diao, D. Apalkov, M. Pakala, Y. Ding, A. Panchula, and Y. Huai, Spin transfer switching and spin polarization in magnetic tunnel junctions with MgO and AlOx barriers, Appl. Phys. Lett. 87 , 232502 (2005).
- [44] H. Kubota, A. Fukushima, Y. Ootani, S. Yuasa, K. Ando, H. Maehara, K. Tsunekawa, D. D. Djayaprawira, N. Watanabe, and Y. Suzuki, Evaluation of Spin-Transfer Switching in CoFeB/MgO/CoFeB Magnetic Tunnel Junctions, Jpn. J. Appl. Phys. 44 , L1237 (2005).
- [45] J. Z. Sun, Spin-current interaction with a monodomain magnetic body: A model study, Phys. Rev. B 62 , 570 (2000).
- [46] J. C. Slonczewski, Currents, torques, and polarization factors in magnetic tunnel junctions, Phys. Rev. B 71 , 024411 (2005).
- [47] Y. Cui, B. Khodadadi, S. Schafer, T. Mewes, J. Lu, and Stuart A. Wolf, Interfacial perpendicular magnetic anisotropy and damping parameter in ultrathin Co2FeAl films, Appl. Phys. Lett. 102 , 162403 (2013).
- [48] H. Meng, R. Sbiaa, S. Lua, C. Wang, M. Akhtar, S. K. Wong, P. Luo, C. Carlberg, and K. S. A. Ang, Low current density induced spin-transfer torque switching in CoFeB-MgO magnetic tunnel junctions with perpendicular anisotropy, J. Phys. D: Appl. Phys. 44, 405001 (2011).
- [49] W. Kim et al. , Extended scalability of perpendicular STT-MRAM towards sub-20nm MTJ node. IEDM Tech. Dig. 24.1.1 (2011).
- [50] M. T. Rahman et al. , Reduction of switching current density in perpendicular magnetic tunnel junctions by tuning the anisotropy of the CoFeB free layer. J. Appl. Phys. 111 , 07C907 (2012).
- [51] M. Gajek et al. , Spin torque switching of 20nm magnetic tunnel junctions with perpendicular anisotropy. Appl. Phys. Lett. 100 , 132408 (2012).
- [52] L. Thomas et al. , Perpendicular spin transfer torque magnetic random access memories with high spin torque efficiency and thermal stability for embedded applications (invited), J. Appl. Phys. 115 , 172615 (2014).
- [53] H. Sato, M. Yamanouchi, S. Ikeda, S. Fukami, F. Matsukura, and H. Ohno,
MgO/CoFeB/Ta/CoFeB/MgO recording structure in magnetic tunnel junctions with perpendicular easy axis, IEEE Tran. Magn. 49 , 4437 (2013).
- [54] Z. C. Wen, H. Sukegawa, S. Kasai, M. Hayashi, S. Mitani, and K. Inomata, Magnetic Tunnel Junctions with Perpendicular Anisotropy Using a Co2FeAl Full-Heusler Alloy, Appl. Phys. Express 5 , 063003 (2012).
- [55] K. Chae, D. Lee, T. Shim, J. Hong, and J. Park, Correlation of the structural properties of a Pt seed layer with the perpendicular magnetic anisotropy features of full Heusler-based Co2FeAl/MgO/Co2Fe6B2 junctions via a 12-inch scale Si wafer process, Appl. Phys. Lett. 103 , 162409 (2013).
- [56] M. S. Gabor, T. Petrisor Jr., C. Tiusan, and T. Petrisor, Perpendicular magnetic anisotropy in Ta/Co2FeAl/MgO multilayers, J. Appl. Phys. 114 , 063905 (2013).
## Figures and captions:
FIG. 1. (Color online) (a) Out-of-plane (2 θ-ω scan) XRD patterns for polycrystalline CFA fullHeusler alloy films on MgO-buffered Si/SiO2 amorphous substrates with varied MgO buffer thickness, t MgO: 2.5, 5.0, 7.5, and 10.0 nm. (b) Normalized integrated intensity of CFA(002) peak, and (c) Ratios of (002) to (004) peaks as a function of t MgO for as-deposited (as-dep.) samples and those annealed at 400 ° C and 480 ° C.
FIG. 2. (Color online) Surface morphology of polycrystalline CFA full-Heusler alloy films on MgO-buffered Si/SiO2 amorphous substrates with respect to the thickness of the MgO buffer. The CFA films were 30-nm-thick and were deposited at RT and post-annealed at 400 ° C and 480 ° C, respectively. Inset: AFM image of the surface of the CFA film annealed at 400 ° C.
FIG. 3. (Color online) TMR ratios as a function of MgO buffer thickness t MgO for polycrystalline CFA/MgO/CoFe MTJs with different thicknesses of the MgO barrier, t barr : 1.5 nm, 1.8 nm, and 2.0 nm. The 30-nm-thick CFA bottom electrodes were directly deposited on the MgO buffer layer and
post-annealed at 400° C and 480° C after deposition at RT. The MTJ stacks were annealed at 370 ° C before the CIPT measurement.
FIG. 4. (Color online) The thickness of MgO barrier, t barr , dependence of (a) TMR ratios and (b) RA at RT for polycrystalline CFA/MgO/CoFe MTJs with as-deposited, 400 ° C and 600 ° C annealed MgO(7.5 nm)/Cr(40 nm) buffer layers on SiO2 amorphous substrates, characterized using CIPT measurement.
FIG. 5. (Color online) RA dependence of TMR ratios for polycrystalline CFA-MTJs with a 2-nmthick CFA layer ('thin-CFA') as a bottom electrode. The squared symbol indicates the TMR ratio for a polycrystalline CFA-MTJ with a 30-nm-thick CFA layer ('thick-CFA').
0.0
FIG. 6. (a) Schematic illustration of the structure of polycrystalline CFA-MTJ nanopillar. (b) R H -loops for a polycrystalline CFA-MTJ nanopillar. Wide arrows show the magnetic configurations of bottom (free) and top (reference) electrodes of the MTJ, and narrow arrows indicate sweep direction of the applied magnetic field. (c) R-I loops for the MTJ at magnetic fields of 0, -11 and -20 Oe, respectively. Arrows indicate sweep direction of the applied current. (d) Representative R I - loops of the CFA-MTJ at applied magnetic field of -11 Oe. (e) and (f) Switching probabilities for I c,P->AP and
I c,AP->P obtained by repeating R I - measurements for 300 times. Solid lines are fitting curves given by Eq. (3). All measurements were performed at RT.
FIG. 7. The H ext dependence of (a) resonant frequency f 0 , (b) demagnetization field H d , and magnetic damping parameter H estimated by fitting the FMR spectrum at each magnetic field for a sample of a 18-nm-thick polycrystalline CFA film. Inset of (a) is typical FMR spectra at H ext of 500, 1000, and 1500 Oe. (c) CFA thickness t dependence of saturation magnetization M s and damping constant 0 of the polycrystalline CFA film.
 | 10.1103/PhysRevApplied.2.024009 | [
"Zhenchao Wen",
"Hiroaki Sukegawa",
"Shinya Kasai",
"Koichiro Inomata",
"Seiji Mitani"
] | 2014-08-02T05:38:17+00:00 | 2014-08-02T05:38:17+00:00 | [
"cond-mat.mtrl-sci"
] | Tunnel magnetoresistance and spin-transfer-torque switching in polycrystalline Co2FeAl full-Heusler alloy magnetic tunnel junctions on Si/SiO2 amorphous substrates | We studied polycrystalline B2-type Co2FeAl (CFA) full-Heusler alloy based
magnetic tunnel junctions (MTJs) fabricated on a Si/SiO2 amorphous substrate.
Polycrystalline CFA films with a (001) orientation, a high B2 ordering, and a
flat surface were achieved using a MgO buffer layer. A tunnel magnetoresistance
(TMR) ratio up to 175% was obtained for an MTJ with a CFA/MgO/CoFe structure on
a 7.5-nm-thick MgO buffer. Spin-transfer torque induced magnetization switching
was achieved in the MTJs with a 2-nm-thick polycrystalline CFA film as a
switching layer. Using a thermal activation model, the intrinsic critical
current density (Jc0) was determined to be 8.2 x 10^6 A/cm^2, which is lower
than 2.9 x 10^7 A/cm^2, the value for epitaxial CFA-MTJs [Appl. Phys. Lett.
100, 182403 (2012)]. We found that the Gilbert damping constant evaluated using
ferromagnetic resonance measurements for the polycrystalline CFA film was
~0.015 and was almost independent of the CFA thickness (2~18 nm). The low Jc0
for the polycrystalline MTJ was mainly attributed to the low damping of the CFA
layer compared with the value in the epitaxial one (~0.04). |
1408.0345v2 | ## Exfoliating pristine black phosphorus down to the monolayer: photo-oxidation and quantum confinement
Alexandre Favron , Etienne Gaufrès , Fréderic Fossard , Pierre L. Lévesque , Anne-Laurence 1 2 3 2 Phaneuf-L'Heureux , Nathalie Y-W. Tang , Annick Loiseau , Richard Leonelli , Sébastien 4 2 3 1 Francoeur and Richard Martel * 4 2
1 Regroupement Québécois sur les Matériaux de Pointe (RQMP) and Département de physique,
Université de Montréal, Montréal QC H3C 3J7, Canada
2 RQMP and Département de chimie, Université de Montréal, Montréal QC H3C 3J7, Canada
3 Laboratoire d'Etude des Microstructures, UMR 104 CNRS-Onera, Châtillon, France
4 RQMP and Département de génie physique, Polytechnique Montréal, Montréal QC H3C 3A7, Canada
*Correspondence to: [email protected]
Thin layers of black phosphorus have recently raised interest for their two-dimensional (2D) semiconducting properties, such as tunable direct bandgap and high carrier mobilities. This lamellar crystal of P atoms stacked together by weak van der Waals forces can be exfoliated down to the stratophosphane monolayer (also called phosphorene) using procedures similar to those used for graphene. Properties of this 2D material are however challenging to probe due to a fast and ubiquitous degradation upon exposure to ambient conditions. Herein, we investigate the crystal degradation using in-situ Raman and transmission electron spectroscopies and highlight a process involving a photo-induced oxidation reaction with adsorbed oxygen in water. The experimental conditions to prepare and preserve mono-, bi- and multilayers of stratophosphane in their pristine states were determined. Study on these 2D layers provides new insights on the effect of confinement on the chemical reactivity and the vibrational modes of black phosphorus.
## Introduction
Black Phosphorus (P(black)) stands out from the other allotropes of elemental phosphorus, such as the amorphous red phosphorus used to ignite matchsticks by friction and the notoriously unstable white phosphorus polymorph, by its good stability up to 550°C and its anisotropic lamellar structure similar to that of graphite 1,2 . Composed of puckered layers of P
atoms stacked together by weak van der Waals forces, P(black) crystals can be exfoliated down to ultrathin layers using techniques similar to those used to prepare graphene from graphite and dichalcogenide monolayers from bulk crystals . Bulk P(black) is a direct bandgap semiconductor 3 (Eg=0.35 eV) showing intrinsic electron and hole mobilities at room temperature of 220 and 350 cm 2 V -1 s -1 , respectively 4-6 . Few years ago, the properties of exfoliated P(black) have raised interest because of their potential use as anode materials in Lithium-ion batteries 7,8 . In the context of two-dimensional (2D) materials such as graphene and chalcogenides, thickness-dependent carrier mobility in few layers transistors has been reported , giving impressive hole mobilities (up 9 to 1000 cm V 2 -1 s -1 ) and ON/OFF current ratios (up to ~10 ). Moreover, a significant blue shift of 5 the bandgap energy with decreasing layer thickness was obtained by ab-initio calculations and observed in preliminary luminescence results 10 . Another study combining Raman spectroscopy and atomic force microscopy (AFM) of exfoliated P(black) by mechanical cleavage and plasma treatments has given Raman signatures and optical contrasts that can be used to determine the layer thickness 11 , albeit the signature of quantum confinement with the number of layers remains to be further clarified.
It appears then that exfoliated P(black) offers new and exciting prospects for applications in electronics and optoelectronics. However, Bridgman's first synthesis of P(black) dates back as far as 100 years ago and yet the literature on the topic is surprisingly sparse and commercial use of the material is inexistent, which acknowledges major difficulties ahead for its synthesis, processing and manipulation 12,7 . The notoriously poor chemical and structural stability of elemental phosphorus 13 raises then important concerns for all processing steps carried out in ambient conditions. Hence, an important milestone towards the establishment of phosphorusbased electronics is the better understanding of the chemical and physical properties of phosphorus thin layers.
Here, we present a systematic investigation of the chemical reactivity of exfoliated layers of P(black) with oxygen and water in different conditions and provide guidelines on how to prevent oxidation and prepare pristine (oxygen free) materials down to the monolayer. A thorough spectroscopic investigation was first used to determine the origin of the irreversible oxidation of thin layers of P(black) occurring immediately after exfoliation. The experiments revealed a thickness dependent photo-activated charge transfer process with adsorbed oxygen in
water under ambient conditions. Control over this ubiquitous photo-oxidation reaction was achieved and provided access to pristine layers for further studies using transmission electron microscopy (TEM), Electron Energy Loss (TEM-EELS) and polarized Raman spectroscopies. TEM images of mono-, bi- and multilayers along with unambiguous signatures in the Raman spectra of electronic confinement effects are shown and discussed using group symmetry analysis.
Because the monolayer is composed of tervalent phosphorus atoms, it belongs, according to the IUPAC nomenclature, to the phosphane group 14,15 . Therefore, a monolayer of exfoliated P(black) is hereafter designated stratophosphane (2D-phosphane), as opposed to the name 'phosphorene' used in previous papers, in order to avoid confusion with other chemical derivatives of the materials and other elemental monolayers having sp - or sp -hybridization (e.g. 2 3 graphene, silicene, graphane, germanane, etc.).
Figure 1 presents the main evidences of degradation acquired by AFM and Raman spectroscopy on exfoliated layers produced using a technique derived from the scotch-tape exfoliation technique 16 in order to quantify the morphology and structural changes, respectively. Layers produced by direct exfoliation were first identified using optical contrast in the microscope and imaged only few minutes after exfoliation in ambient conditions. A typical AFM image of a multilayer 2D-phosphane is shown in Figure 1a. The flake appears in AFM with a relative thickness of about 2.8 nm, as measured from the folded-back layer. Considering that the layer thickness of a 2D-phosphane monolayer is about 0.53 nm, the apparent height in AFM indicates the presence of roughly five layers . Consistent with recent reports 2 13,17 , degradation is noted by the homogeneous distribution of small bumps or defects at the surface of the layer in AFM images taken shortly after exfoliation. These bumps appear to grow and then collapse from the top of the layer rather than from the edges. The signs of degradation become obvious in AFM (Figure 1b) or under the optical microscope after few days. A drastic transformation is observed along with new droplets distributed in the vicinity of the initial position of the flakes. Obviously, degradation took place in those experimental conditions. Surprisingly, it was noted that degradation slowed down significantly when the samples was kept in the dark.
Figure 1 | Photo-oxidation of thin layers of 2D-phosphane. a-b, AFM images taken immediately after exfoliation on a SiO2/Si substrate ( a ) and after a few days under ambient conditions ( b ). c, Raman spectra at λ = 532 nm measured in air 24, 48, 96 and 120 min after exfoliation. d, Time dependence of the integrated intensity of the A 2 g Raman mode in different conditions: Air, vacuum, mixture of O2 and H2O, and in air but with a 20 nm capping layer of parylene C. Curves were vertically shifted for clarity. e, Time evolution of the intensity of A 2 g mode for different laser fluences for a sample immersed in an aqueous solution in air. Inset: Intensity decay time as a function of laser fluence. The dashed line shows a slope of minus one.
Raman spectra of a 2D-phosphane multilayer are shown in Figures 1c at 24, 48, 96 and 120 minutes after exfoliation and continuous exposure to a fluence of 2x10 Wcm 4 -2 at λ = 532 nm. Using established vibrational modes of bulk P(black), the observed Raman modes centered at 361 cm -1 , 438 cm -1 and 466 cm -1 are assigned to Ag 1 , B2g and Ag 2 , respectively 6,18-20 . As Raman scattering intensity is very sensitive to crystal quality, the quenching of the intensity within two hours is another indication of a fast degradation of the multilayer 2D-phosphane under light exposure.
An experiment was carried out in vacuum (< 5 x10 -6 Torr) in order to verify if the mere presence of visible light suffices to induce degradation. To minimize the influence of the hydrophilicity of the Si/SiO2 substrate, a single crystal flake of multilayer 2D-phosphane was mechanically exfoliated and transferred onto a SiO2/Si substrate coated with 20 nm of parylene C
21
. Raman spectra were recorded at room temperature as a function of time with and without gas exposure. No change in the Raman signal is observed for samples kept in vacuum (Figure 1d) or
exposed separately to oxygen and water (at a partial pressure of 400 and 13 Torr, respectively, which is comparable to their partial pressure in air, Fig. S7). Experiments in vacuum using a higher flux of photons (up to 6 x10 Wcm 4 -2 ) also provided no evidence of degradation. The lightinduced degradation of the multilayer 2D-phosphane is however instantly activated by introducing air or alternatively a mixture of pure oxygen and water vapor (Figure 1d). Moreover, the integrated Raman intensity as a function of time indicates that degradation in air is monoexponential, ~ e -t/τ , where t and τ represent the time and a decay time, respectively. Note that different decay times are measured between experiments in air and in O2/H2O (5.5 and 36 min -1 , respectively) due to the different photon flux used (i.e. 1.7x10 Wcm 4 -2 and 1.8x10 Wcm 3 -2 , respectively). Measurements of this kinetics in a buffer solution at pH between 5.8 and 7.8 indicated no dependency with pH. Finally, prolonged exposures to O2, H2O and then a mix O2/H2O were carried out in the dark and no degradation was noted, even after several hours (Fig. S7). All of these results reveal that three major environmental parameters are simultaneously required for the degradation of multilayer 2D-phosphane: water, oxygen and visible light.
Figure 2 | Chemical analysis by hyperspectral TEM-EELS spectroscopy of a multilayer 2Dphosphane. a High Angle Annular Dark Field (HAADF) contrast image taken at 80 kV. b-d
EELS images extracted from the data cube at the energy of phosphorus L-2,3 edge (130.2 and 136 eV) and oxygen K-edge (534 eV). e EELS spectra corresponding to the selected areas in (a). Note that Pt 4 is acquired from another flake that is more oxidized, as described in Supplemental Information, Figure S10a.
Partially degraded samples were prepared for further TEM analysis of this chemistry using a combination of High Angle Annular Dark Field (TEM-HAADF) and hyperspectral electron energy loss spectroscopy (TEM-EELS) in imaging modes. HAADF contrast is directly linked to the number of atoms interacting with the electron beam, thus to the layer thickness, while EELS provides a local chemical analysis of the sample using signals at 130.2 eV, 136 eV and 534 eV that can be attributed to the presence of P, PxOy, and O, respectively (see Supplementary). In Figure 2a, the HAADF image of the multilayer 2D-phosphane flake indicates regions with for instance ~ 7 (Pt1), and ~ 2-3 (Pt2 and Pt3) 2D-phosphane layers (Fig. S11). The EELS spectra in these regions and the intensity map of the entire flake are displayed in Figure 2e and 2b-d, respectively. Pt1 region exhibits weak concentrations of oxygen and PxOy, species comparatively to the pristine P concentration. On the contrary, the presence of a significant amount of oxygen atoms preferentially localized in the thinnest regions of the flake is noted in Figure 2c. These regions are also characterized by the presence of pristine P and PxOy species (Figure 2d), indicating partial oxidation. Recorded on another flake (Fig S10), the spectrum Pt4 in Figure 2e clearly demonstrates that the signature of pristine P at 130.2 eV can completely vanish at the cost of an increase of the P Oy peak loss, indicating a complete degradation by x oxidation reaction. These experimental results demonstrate that thin layers of 2D-phosphane are oxidized and the extent of the oxidation is thickness dependent and faster for ultra-thin layers.
From the results presented in Figures 1 and 2, a light-induced oxidative reaction mechanism can be sketched using the two main reaction steps involving the coverage, θ , of a pristine and intrinsic 2D-phosphane:
$$\theta + h v \neq \theta ^ { * }, \quad & \text{Eq.1} \\ \theta ^ { * } + O _ { 2 ( a q ) } \rightarrow O _ { 2 ( a q ) } ^ { \dots - } + \theta \rightarrow \theta _ { o x }. \quad & \text{Eq.2}$$
In the first step, the optical excitation of the 2D-phosphane ground state produces a coverage of excited 2D-phosphane ( θ * ), whose steady-state population depends on the photon flux, the
recombination rate and the absorption cross section (Eq. 1). θ * is implicitly assumed here to involve precursors of free electron-hole pairs. In the second step (Eq. 2), the excited state of population θ * undergoes a charge transfer reaction toward the aqueous oxygen molecules adsorbed at the surface of the layer, leading to p-doped 2D-phosphane. Finally, the charge transfer reaction with the oxygen-water redox couple generates reactive intermediate species (i.e. strong Brønsted bases), such as superoxide anions, that react spontaneously with the surface atoms of the 2D-phosphane layer and etch them in the process leading to oxidized species of coverage θ ox .
In analogy with a model explaining the ubiquitous air doping of carbon nanotubes and graphene layers deposited on an SiO2 surfaces 22-24 , the Marcus-Gerischer theory 25 (MGT) was used to model the reaction rate glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=2,font=/IFMIOK+CambriaMath> , driven by the charge transfer from 2D-phosphane to oxygen/water, leading oxidation and etching (see supplementary for details and Fig. S12). This rate is given by:
$$\frac { d \theta } { d t } \in - \theta J _ { p h } [ O _ { 2 } ] \cdot e x p \left [ - \frac { \left ( \frac { | E _ { g, n } | } { 2 } + E _ { i } - E _ { F, r e d o x } ^ { 0 } + \lambda \right ) ^ { 2 } } { 4 k _ { B } T \lambda } \right ],$$
where 𝜃𝜃 is the coverage of pristine 2D-phosphane, 𝐽𝐽 glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=1,font=/IFMIOK+CambriaMath> is the laser fluence, [ O2 ] is the oxygen concentration in water, 𝐸𝐸 glyph<c=1,font=/IFMIOK+CambriaMath> is the intrinsic Fermi level, 𝐸𝐸 glyph<c=1,font=/IFMIOK+CambriaMath> ,glyph<c=1,font=/IFMIOK+CambriaMath> is the thickness dependent bandgap energy for a n-layer 2D-phosphane, 𝐸𝐸 glyph<c=1,font=/IFMIOK+CambriaMath> ,glyph<c=4,font=/IFMIOK+CambriaMath> glyph<c=2,font=/IFMIOK+CambriaMath> glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=3,font=/IFMIOK+CambriaMath> glyph<c=5,font=/IFMIOK+CambriaMath> glyph<c=1,font=/IFMIOK+CambriaMath> is the energy level of the oxygen acceptor state, and 𝜆𝜆 is the renormalization energy of oxygen in water. This expression highlights the key role of the three main parameters on the reaction rate ([O2], 𝜃𝜃 and 𝐽𝐽 glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=1,font=/IFMIOK+CambriaMath> ) and of the exponential dependency with the overlap between the 2D-phosphane density of states and the energy levels of the oxygen acceptor states in water.
As shown in Figure 1e, this model of the degradation process is tested using measurements on multilayer 2D-phosphane immersed in water (see Fig. S5) by recording the Raman signal as a function of time for a laser fluence between 1.2 x10 and 6.4 x10 Wcm 4 4 -2 . As expected, the results show that the higher the photon flux, the faster the reaction. Indeed, on a log-log scale, the decay time as a function of fluence shows an approximate slope of -1 (see inset Figure 1e). Furthermore, the simple MGT explains the thickness dependence of the multilayer
2D-phosphane reactivity in terms of electronic confinement. When a thin sample is prepared, the bandgap shifts towards higher energies and closer to energy levels of the oxygen acceptor states, thereby strongly increasing the rate of charge transfer and hence the oxidation rate. This is especially true for the thinnest flakes, as observed in Figure 2. Other experiments (not shown) determined also that the reaction rate is independent of the pH of the solution because the reaction is driven by the 𝑂𝑂 glyph<c=1,font=/IFMIOK+CambriaMath> /𝑂𝑂 glyph<c=1,font=/IFMIOK+CambriaMath> ∙glyph<c=1,font=/IFMIOK+CambriaMath> as the P(black) is consumed before reaching equilibrium. MGT seems to fit quite well all of the observed behaviors and therefore proves that rigorous conditions are required in order to prepare thin layers of pristine (oxygen free) 2D-phosphane.
To prepare pristine layers, we devised an exfoliation procedure in the dark or inside a nitrogen-filled glove box in order to remove at least one of the components leading to degradation. The 2D-phosphane samples are transferred in those conditions onto a 300 nm SiO2/Si substrate and the thinnest flakes are identified in the glove box by AFM and optical microscopy using a bandpass filter (λ >580 nm) to provide a more accurate value of the optical contrast. Finally, samples were directly transferred from the glove box to a vacuum chamber and then pumped down to 2.10 -6 Torr for Raman experiments. Another strategy for experiemnts outside the glove box consists in coating the samples with a thin layer of a protecting insulator material. As an example in Figure 1d, 300 nm of parylene C is an excellent barrier against water adsorption and diffusion.
Figure 3 | HRTEM images at 200kV of 2D-phosphane layers viewed in the crystallographic projection along the c - axis ((110) zone axis corresponding to the edge-on view and slightly tilted (17°) top view) in ( a ) - ( b ) and along the b -axis ((201) zone axis) in ( c ) - ( d ): HRTEM images are shown in ( b ) and ( d ). Fast Fourier Transforms of the white dashed areas highlighted in images ( b ) and ( d ) are shown in ( a ) and ( c ) respectively together with corresponding simulated images of the structure models depicted in blue. e Evolution of the number of 2D-phosphane along the cross line in the inset extracted from the High Angle Annular Dark Field (HAADF) contrast image (inset) taken at the edge of a thin flake.
Pristine 2D-phosphane layers were prepared in dim light and quickly transferred on a TEM grid (see Supplementary Information) for structural analysis. Figure 3 presents TEM images of mono-, bi- and multilayer 2D-phosphane, Fast Fourier Transforms (FFT) of selected areas, and image simulations of the layered structure projected along different orientations. The exfoliation provided micrographs with top and side views. The puckered structure of 2Dphosphane is clearly revealed in Figure 3d, thanks to a fold in the layers that provides a view along their b -axis. The structure is confirmed by both the analysis of the FFT pattern of the
selected area and the image simulation of the layered structure viewed along the (201) zone axis. The same can be said for the mono- and multilayer 2D-phosphane in Figure 3b and viewed along the c -axis ((110) zone axis). From the image simulations shown in Figure 3a, the stacking of the layers can be distinguished and monolayer 2D-phosphane identified. Furthermore, TEM-EELS core-loss analysis at the phosphorus K and L edges revealed that the sample essentially consists of pure phosphorus. Scanning TEM High Angle Annular Dark Field (STEM-HAADF, Figure 3e), which intensity is proportional to the number of phosphorus planes, was used to determine the thicknesses and the structural integrity of the 2D-phosphane samples. Overall, these TEM images demonstrate that pristine (with oxygen content of less than 1%) 2D-phosphane (composed of 𝑛𝑛 = 1 to 8 layers) can be readily produced when appropriate precautions are taken.
A large selection of samples produced in a glove box was characterized by AFM, optical contrast and Raman spectroscopy. As shown in Figure 4, the energy position and linewidth of Raman modes are consistent with those previously reported for bulk P(black) 18 . However, significant variations are observed as a function of thickness, from the bulk ( 𝑛𝑛 = ∞ ) down to monolayer 2D-phosphane ( 𝑛𝑛 = 1 ). The AFM image of this monolayer is shown in Fig. S6. The central frequencies of Ag 1 and B2g do not significantly shift, but a noticeable relatively symmetric broadening develops for 𝑛𝑛 ≤ 5 . Down to the bilayer, widths increased by 1.5 cm -1 with respect to the bulk. In contrast, the behavior of Ag 2 is both qualitatively and quantitatively different. As shown in Figure 5b, the central frequency of the Ag 2 shifts towards higher energies, reaches a maximum value for the bilayer (+2.6 cm -1 ), and then red shifts such that the frequency of the monolayer is only +2 cm -1 above that of the bulk. The full-width at half-maximum (FWHM) of Ag follows a similar non-monotonic behavior. As shown in Figure 5a, the width increases 2 continuously from the bulk value of 3.3 to 6.8 cm -1 for the bilayer, and it then drops to 5.3 cm -1 for the monolayer. These two non-monotonic dependencies strongly suggest the presence of more than one Raman allowed vibrational modes in the vicinity of Ag 2 . This hypothesis is confirmed by the low-temperature bilayer spectra shown in the inset of Figure 4, where the presence of two modes can be unambiguously identified at 77 K.
Figure 4 | Raman spectroscopy of n-layer 2D-phosphane and bulk P(black). Raman spectra at λ = 532 nm on layers of different thicknesses. Spectra are normalized to the Ag 2 intensity and vertically shifted for clarity (scaling factor on right is only for the Si peak region near 520 cm -1 ). Inset : expanded view of the Ag 2 mode at 469 cm -1 of the bilayer 2D-phosphane at 300K and 77K.
Figure 5 | Raman characteristics of the Ag 2 mode of n-layer 2D-phosphane as a function of AFM thickness, optical contrast and laser polarization. a and b, Evolution of the Ag 2 fullwidth-at-half-maximum (FWHM) (a) and of the spectral position (b) as a function of optical contrast. Open blue and solid red circles in (a) and (b) correspond to the monolayer 2Dphosphane and bulk P(black), respectively. c, Polarization-resolved integrated Raman intensities of Ag 1 , B 2g and Ag 2 from the trilayer (optical contrast of 0.22). The flake was excited with the polarization corresponding to the Ag 2 intensity maximum.
These Raman results can be understood on the basis of an analysis of the effect of the number of monolayers on the symmetry of multilayer 2D-phosphane. Bulk P(black) is orthorhombic (space group Cmce; no. 64) with the main crystallographic b axis normal to the layers. The unit cell is composed of 8 P atoms, disposed in two layers, each containing 4 P atoms that are covalently bonded through p-orbitals to three neighbors 26 . Consistent with the results in Figure 1c and Figure 4, the D2h factor group symmetry of bulk P(black) predicts three out of a total of six Raman active modes for an excitation along the crystal b axis and perpendicular to the two main ( a and c) axes of the 2D-phosphane plane. The preparation of thin layers of 2Dphosphane breaks the translational invariance along b and reduces the number of symmetry operations. For samples with an odd or even number of monolayers, the symmetry is Pmna and Pbcm, respectively. Irrespective of the number of layers involved, the factor group always remains D2h and, conveniently, the same 8 one-dimensional representations can be used to label all vibrational modes. The broadening of the modes observed in Figure 4 can be associated with the Davydov splitting of monolayer vibrational modes and the appearance of new Raman-
allowed modes 27,28 . Both effects result from the larger number of phosphorus atoms composing the primitive cell of bilayers, trilayers, and other few-monolayer samples compared to that of the monolayer 2D-phosphane and bulk P(black) primitive cells, which both contain only four atoms. Accordingly, multilayers are characterized by 6𝑛𝑛 Raman modes, half of which can be observed for scattered light along the b axis. Therefore, singly degenerate bulk modes and their corresponding monolayer modes are replaced, in few layers 2D-phosphane, by a number of neardegenerate modes corresponding to in- and out-of-phase stacking of monolayer vibrational motions 27 . Interestingly, some Raman modes of the multilayer are built from the combination of infrared (IR) modes in symmetrically opposed monolayers. Therefore, bulk and monolayer forbidden Raman modes can appear at energies close to that of bulk IR modes.
Davydov splitting requires at least 𝑛𝑛 = 3 , for which at least two Raman modes coexist in a narrow frequency range 27 . It therefore does not explain the broadening observed for the bilayer. For Ag 1 and B2g, this broadening is weak and could be induced by various effects, from interactions with the substrate to inhomogeneous broadening on the scale of the diffractionlimited optical probe. For Ag 2 however, the degeneracy is clearly lifted, suggesting the presence of another mechanism. As described above, the presence of a second mode could be explained by combining two monolayer IR modes of B2u 1 representation to produce a Raman-allowed Ag 2 mode. In the bulk, B2u 1 is found at 469 cm -1 , which is very close to that of Ag 2 at 470 cm -1 . This symmetry analysis also predicts that all these near-generate modes should be absent in the monolayer because its primitive cell contains only 4 atoms. Indeed, we find that the width of Ag 2 of the monolayer is narrower than that of the bi- and trilayer, as was found for example for MoS2 29 .
The shift of Ag 2 , a vibrational mode associated to atomic movements in the plane of the monolayer, bears resemblance to the hardening of E2g 1 in MoS2 with decreasing thickness 29 . Interestingly, the positions of Ag 1 , assigned to atomic displacements perpendicular to the monolayer, and B2g do not significantly shift. Compared to transition metal dichalcogenides, interlayer interactions and their effects on vibrational energies appear to be qualitatively different. The frequency dependence of Raman modes is often used in 2D materials to reliably determine sample thickness with a monolayer precision. For 2D-phosphane, a similar procedure can be exploited, but only if Davydov-related effects highlighted above are properly taken into
account. The monotonic dependence of intensity ratios on 𝑛𝑛 could be used to determine the number of monolayers involved, but the analysis must also carefully consider crystal anisotropy 30 . As highlighted in Figure 5c, Raman intensities are very sensitive to polarization of the excited and scattered beam. These polarization responses were obtained from the trilayer 2Dphosphane and respect the expected polarization selection rule.
In conclusion, a photo-induced oxidation process of 2D-phosphane samples was identified and shown to involve oxygen species in water present in air at the surface of the layer. The rate of the reaction was found to vary with the layer thickness, with the oxygen concentration and linearly with the light intensity. The oxidation rate is also predicted to depend exponentially on the square of the energy gap of the layer. A procedure was devised to exfoliate black phosphorus while preventing degradation in ambient conditions. The Raman signatures of mono-, bi- and multilayer 2D-phosphane were obtained and several thickness-related effects were identified and discussed using symmetry arguments. This work paves the way to a systematic investigation of the electrical and optical properties of pristine 2D-phosphane down to the monolayer.
## Materials and Methods:
The black phosphorous (99.998%) was bought from Smart Elements (Vienna, Austria). Polydimethylsiloxane (PDMS) stamps were made using the PDMS solutions (Sylgard 184) bought from Dow Corning. Parylene C 19 was purchased from Cookson Electronics Specialty Coating Systems.
The 2D-phosphane layers were obtained by a modified scotch tape exfoliation technique 16 performed in a glove box under a clean nitrogen flow. First, the P(black) crystal is exfoliated onto a flat PDMS and then transferred onto a curved PDMS. The 2nd PDMS stamp covered with flakes is then rolled onto a 300 nm SiO2/Si substrate (depicted in Supplemental, Figure S1).
For the degradation study, the exfoliation was done in a relatively dark room, and keeping the light very low, the flakes were characterized by optical contrast using a 580 nm low pass filter, followed by Raman spectrometry. For measurements under water, the pH was adjusted to 5.8 using phosphate buffer and the flakes were kept under water with a homemade liquid cell that is air tight for Raman measurements (See Supplemental Figure S5). For the optical contrast, AFM thickness and Raman correlation measurements, the degradation was avoided by placing an optical microscope and an atomic force microscope in a glove box and under continuous nitrogen
flow. The exfoliation procedure of the n-layer 2D-phosphane and their AFM and optical characterizations were all performed in the same glove box. Sample transfers from the glove box to a cryostat for Raman scattering measurements were performed in-situ in order to avoid air exposure. To protect thin layers of 2D-phosphane during Raman characterization, the cryostat was continuously maintained at a pressure of < 5.10 -6 Torr.
## ACKNOWLEDGEMENTS
The authors would like to acknowledge the assistance of Andreas Dietrich with early steps of sample preparation and characterization and discussions with Thomas Szkopek. The authors also acknowledge Patricia Moraille, from the Laboratoire de Caracterisation des Matériaux (LCM) at the Université de Montreal for her expertise on controlled atmosphere AFM measurements. This work was possible because of financial support from the Natural Sciences and Engineering Research Council of Canada (NSERC), the Fonds de Recherche du QuébecNature et Technologie (FRQNT), and the Canada Research Chair (CRC) programmes. The research leading to these results has also received partial funding from the European Union Seventh Framework Program under grant agreement n°604391 Graphene Flagship.
## REFERENCES
- 1. Bridgeman, P. W. Two new modifications of Phosphorus. J. Am. Chem. Soc. , 36 , 1344-1363 (1914).
- 2. Hultgren, R. Gingrich, N. S. & Warren, B. E. The atomic distribution in red and black phosphorus and the crystal structure of black phosphorus, J. of Chem. Phys . 3 , 351 (1935).
- 3. Reich, E. S. Phosphorene excites materials scientists, Nature 506 , 19 (2014).
- 4. Keyes, R. The electrical properties of black phosphorus. Phys. Rev. 92 , 580 (1953).
- 5. Warschauer, D. Electrical and optical properties of crystalline black phosphorus. J. Appl. Phys. 34 , 1853 (1963).
- 6. Appalakondaiah, S., Vaitheeswaran, G., Lèbegue, S., Christensen, N. E. & Svane, A. Effect of van der Waals Interactions on the structural and elastic properties of black phosphorus. Phys. Rev. B 86 , 035105 (2012).
- 7. Park C.-M. & Sohn, H.-J. Black phosphorus and its composite for lithium rechargeable batteries. Adv. Mat. 19 , 2465 (2007).
- 8. Sun, L.-Q., et al . Electrochemical activity of black phosphorus as an anode material for lithium-ion batteries . J. Phys. Chem C , 116, 14772 (2012).
- 9. Li, L., et al . Black phosphorus field-effect transistors. Nat. Nanotech . Published on line on March 2 nd (2014).
- 10. Liu, H., et al. Phosphorene: An Unexplored 2D Semiconductor with a High Hole Mobility. ACS Nano , online, DOI 10.1021/nn501226z. (2014)
- 11. Lu, W., et al. Plasma-assisted fabrication of monolayer phosphorene and its Raman characterization. Nano Research , online ASAP, DOI 10.1007/s12274-014-0446-7 (2014).
- 12. Asahina H. and Morita A., Band structure and optical properties of black phosphorus. J. Phys. C: Solid State Phys . 17 1839-1852 (1984).
- 13. Koenig, S. P., Doganov, R. A., Schmidt, H., Castro Neto, A. H. & Özyilmaz, B. Electric field effect in ultrathin black phosphorus. Appl. Phys. Lett . 104 , 103106 (2014).
- 14. IUPAC. Compendium of Chemical Terminology, 2nd ed. (the "Gold Book"). Compiled by A. D. McNaught and A. Wilkinson. Blackwell Scientific Publications , Oxford (1997). XML on-line corrected version: http://goldbook.iupac.org (2006-) created by M. Nic, J. Jirat, B. Kosata; updates compiled by A. Jenkins. ISBN 0-9678550-9-8.
- 15. Holleman, A.F., Wiberg, N., Inorganic Chemistry , Academic Press, ISBN: 0-12-352651-5 (2001).
- 16. Meitl, M.A., et al . Transfer printing by kinetic control of adhesion to an elastomeric stamp. Nature Mater . 5 , 33-38 (2006).
- 17. Castellanos-Gomez, A. et al., Isolation and characterization of few-layer black phosphorus, 2D Materials 1 , 025001 (2014).
- 18. Sugai, S. & Shirotani, I. Raman and infrared reflection spectroscopy in black phosphorus. Sol. State Comm. 53 , 753 (1985).
- 19. Vanderborgh, C.A. & Schiferl, D. Raman studies of black phosphorus from 0.25 to 7.7 GPa at 15K. Phys. Rev. B , 40 , 9595-9599, (1989).
- 20. Akahama, Y., Kobayashi, M. & Kawamura, H. Raman study of black phosphorus up to 13 GPa. Solid State Commun . 104 , 311 (1997).
- 21. Sari, S. S. et al. Graphene field-effect transistors with parylene gate dielectric. Appl. Phys. Lett . 95 , 242104 (2009).
- 22. Aguirre, C. M. et al . The role of the oxygen/water redox couple in suppressing electron conduction in field-effect transistors. Adv. Mater . 21 , 3087-3091 (2009).
- 23. Levesque, P. L. et al . Probing charge transfer at surface using graphene transistors. Nano Lett . 11 , 132-137 (2011).
- 24. Brunel, D. et al . Control over the interface properties of carbon nanotube-based optoelectronic memory devices. Appl. Phys. Lett . 102, 1, (2013).
- 25. Memming R., Semiconductor electrochemistry , Wiley-VCH Weinheim ISBN: 3-527-30147X (2001).
- 26. Du, Y., Ouyang, C., Shi, S. & Lei, M. Ab initio studies on atomic and electronic structures of black phosphorus. J. Appl. Phys . 107 , 093718 (2010).
- 27. Tonndorf, P. et al . Photoluminescence emission and Raman response of monolayer MoS2, MoSe2, and WSe2. Opt. Express 21 , 4908 (2013).
- 28. Terrones, H. et al. New First Order Raman-active modes in few layered transition metal dichalcogenides. Sci. Rep . 4 , 4215 (2014).
- 29. Lee, C. et al. Anomalous lattice vibrations of single- and few-layer MoS2. ACS Nano , 4 , 2695 (2010).
- 30. Xia, F. Wang, H. & Jia, Y. Rediscovering black phosphorus: A unique anisotropic 2D material for optoelectronics and electronics. arXiv :1402.0270 [cond-mat.mes-hall]
Supplementary file for
## Exfoliating pristine black phosphorus down to the monolayer: photo-oxidation and quantum confinement
Alexandre Favron , Etienne Gaufrès , Fréderic Fossard , Pierre L. Lévesque , Anne-Laurence 1 2 3 2 Phaneuf-L'Heureux , Nathalie. Y-W. Tang , Annick Loiseau Richard Leonelli , Sébastien 4 2 3 1 Francoeur and Richard Martel * 4 2
1 Regroupement québécois sur les matériaux de pointe (RQMP) and Département de physique, Université de Montréal, Montréal PQ H3C 3J7, Canada
2 RQMP and Département de chimie, Université de Montréal, Montréal PQ H3C 3J7, Canada
3
Laboratoire d'Etude des Microstructures, UMR 104 CNRS-Onera, Châtillon, France
4 RQMP and Département de génie physique, École Polytechnique de Montréal, Montréal PQ H3C 3A7, Canada
*Correspondence to: [email protected]
## A. METHODS
## A1- Exfoliation of Black Phosphorous
Stratophosphane (2D-phosphane) mono-, bi- and multilayers were exfoliated from black phosphorus (P(black)) using a technique derived from the scotch-tape exfoliation technique . The 1 modified exfoliation process involved two polydimethylsiloxane (PDMS) stamps, as depicted in Figure S1. The first PDMS stamp was reticulated in a petri dish, while the second one was reticulated in a cylindrical tube of 1 cm diameter. The exfoliation was done in a dark room or in a glove box under nitrogen flux.
Figure S1 | Three step exfoliation technique of P(black). Step 1: Exfoliation done on the flat PDMS-1. Step 2: The flakes were reported on semi-spherical PDMS-2 stamp. Step 3: the stamp was rolled on the substrate (300 nm SiO2 on a Si wafer) with an estimated speed of 0.1 cm/s.
## A2- Optical contrast measurements of 2D-phosphane layers
The flakes were localized and characterized under an optical microscope (Olympus BX51) placed in a glove box using red light (glass colored optical filter 580 nm) to limit the photo-oxidation of the flakes. The optical contrast, Cred , was measured preferentially at the center of the image and using the following formula:
$$C _ { r e d } = \ \frac { R _ { s u b } - R _ { f l a k e } } { R _ { s u b } },$$
where Rsub and Rflake are the reflection signals on the substrate and flake respectively. Some examples of 2D-phosphane layers thus produced are presented in Figure S2.
Figure S2 | Examples of optical micrograph of the localization of flakes of 2D-phosphane multilayers using different optical contrasts.
## A3- Measurements of the layer thickness using atomic force microscopy (AFM)
The substrate used is a 300 nm SiO2 grown on a n ++ Si wafer that was previously patterned with metallic alignment marks (crosses done by photolithography). Short time before exfoliation, the substrate was annealed at 300°C for 3 hours and then quickly placed in a glove box equipped with an optical microscope fitted with a glass colored optical filter 580 nm and an AFM instrument (Auto Probe CP, ThermoMicroscopes). The exfoliation procedure as well as optical and AFM characterizations of the exfoliated P(black) flakes were all done in the same glove box and under a continuous flow of nitrogen (i.e. without contact to the atmosphere). The thicknesses of the flakes was measured using intermittent imaging mode at 85% damping and using silicon cantilever probes for tapping mode (ACTA from AppNano, Si probes with Al coating, tip radius < 10 nm, spring constant 25-75 N/m). The experimental procedure was done in the glove box to avoid degradation and to establish a correlation between the optical contrast and the average thickness of different flakes. The results of many different 2D-phosphane layers are presented in Figure S3.
Figure | S3 Evolution of the optical contrast of flakes as a function of the thickness . The thickness of many 2D-phosphane thin layers with their different optical contrast values as measured by AFM and optical microscopy. The flakes are exfoliated in a glove box on 300 nm SiO2/Si substrates.
## A4 - Photo-oxidation monitoring
A simple way to follow the progression of the photo-oxidation on the flakes was to compare a sequence of optical microscopic images at different exposure time to illumination in air. Several bumps appear after several minutes of exposition as shown in Fig. S4. AFM measurements in similar conditions also revealed the presence of bumps even before they can be seen in the optical microscope (See for instance Figure 1a in the main text)
Figure S4 | Optical microscopy images of a 2D-phosphane multilayer on a SiO2/Si wafer taken just after exfoliation (left) and a few hours in air with light exposure (right).
To quantify the photo-oxidation, we followed the integrated intensity of the Raman modes of P(black) modes, which is directly related to the coverage of 2D-phosphane multilayers in their pristine state. Hence, the oxidation of the surface will cause the Raman modes to decrease in intensity. The Raman spectrometer used is a custom built instrument with an excitation laser line at 532 nm (Centenia laser), a 50X objective and a nitrogen cooled charged-coupled device camera (JY Symphony). The spectral region probed was between 200-550 cm -1 with a precision of ± 0.2 cm -1 .
## Raman measurements on multilayer 2D-phosphane flakes for degradation experiments:
The sample stage was equipped with a cryostat (Janis ST500) coupled to a gas manifold and a pump. The vacuum pressure inside the cryostat chamber was below 2 x 10 -5 Torr. The gases introduced in the chamber were oxygen (leak of ~0.5 Torr) and water vapor at partial pressures of 475 Torr and 13 Torr, respectively. The deionized water used in this experiment was previously deoxygenated using several freeze thaw cycles during pumping.
Raman measurements on samples immersed in water:
As shown in the Figure S5, the samples were placed in a buffer solution maintained at pH of 5.8 using phosphoric acid and sodium hydroxide. To minimize the laser path length through the solution, a cover glass slide was placed close to the sample with a PDMS donut on top. The slide was isolated from the solution using a PDMS donut as shown in Figure S5.
Figure S5 | Scheme of the experimental setup for the Raman experiments of the 2Dphosphane multilayers in deionized water.
## A5 - Raman measurement of the monolayer 2D-phosphane
The figure S6a presents the AFM image of a monolayer 2D-phosphane. The small dimension of the flake prevented measurement of optical contrast and the presence of a multilayer 2Dphosphane in contact with the monolayer imposed stringent conditions for the Raman measurements. The Figure S6b shows the position of the laser beam spot on the sample (532 nm, NA 0.55). It was placed carefully on the upper corner of the monolayer in order to avoid contamination of the signal by the multilayers. The very small part of the monolayer under the beam spot implied a long acquisition time (20 min) at a fluence of 100 ∝ W. m -2 ∝ .
Figure S6 | Experiments on the monolayer 2D-phosphane. a: AFM image of a monolayer 2Dphosphane. b: Zoom of (a) and representation of the beam spot position on the monolayer 2Dphosphane for the Raman measurement in Figure 4, main text.
## B. ADDITIONAL EXPERIMENTAL RESULTS
## B1 - Raman measurements
Figure S7 | Evolution of the Ag 2 integrated Raman intensity of multilayer 2D-phosphane under different gas exposures. The laser illumination is at a wavelength of 532 nm light source of 200 ∝ W. m -2 ∝ intensity.
Figure S8 | Evolution of the full width at half maximum (FWHM) of the Ag 1 , B 2g and Ag 2 Raman modes of 2D-phosphane layers of different thicknesses as a function of optical contrast.
## B2- Transmission electron microscopy (TEM) imaging and spectroscopies
## i) Microscope characteristics
TEM and high-angle annular dark-field (HAADF) images were recorded with a Libra 200 MC Zeiss operating at 200 kV and at 80 kV. The microscope is equipped with a monochromatized Schottky FEG source delivering an energy resolution down to 150 meV and a Gatan Ultrascan camera. Point resolution is 2.3 Å and information limit is 1.2 Å at 200kV. The TEM can operate in scanning mode (STEM) with a spatial resolution under 1 nm and is equipped with Bright Field (BF) and High Angular Dark Field (HADF) detectors. Finally the machine is equipped with an in-column energy filter making possible energy filtered TEM, EELS measurements in Scanning TEM mode and acquisition of spectrum imaging datacubes (or hyperspectral EELS).
## ii) Exfoliation of P(black) on the TEM grid
P(black) exfoliation was done using PDMS stamps in a low-light 'dark' room. After three exfoliation steps, the grid was delicately placed in contact with the PDMS stamp to transfer thin layers on the carbon thin film of the grid for TEM analysis.
## iii) Simulations of High Resolution TEM (HRTEM) images
Experimental HRTEM images of the 2D-phosphane layers were simulated using the JEMS software with the following TEM parameters:
| | ZA [110] | ZA [201] |
|----------|-------------------|--------------------|
| α (mrad) | 0.1 | 0.1 |
| Δ f (nm) | Scherzer - 67 ± 5 | Scherzer - 500 ± 5 |
| HV (kV) | 200 | 200 |
| C S (mm) | 1.2 | 1.2 |
| C C (mm) | 1.2 | 1.2 |
| Δ E (eV) | <0.7 | <0.7 |
## iv) EELS measurements
Figure S9 displays a core loss spectrum recorded at 80 kV on an almost pure sample used for TEM imaging in Figure 2. Assignments of the peaks are based on Refs (2) and (3). Phosphorous states are identified by inspecting the P-L2,3 edge. The main sharp P-L2,3 edge at 130.2 eV is the main signature of the P(black) structure. This peak is followed by a very weak peak at 136 eV assigned to an oxidized state corresponding to the P2O5 stoichiometry. Correlated with the peak at 136 eV, the spectrum displays a very weak peak at 534 eV corresponding to the K-O edge.
Figure S9 | Core loss spectrum recorded at 80 kV and an energy resolution of 0.8 eV on a multilayer 2D-phosphane sample at P-L2,3 and O - K edges.
Figure S10 | HAADF image (a) and core loss images at 136 eV (b) and 534 eV (c), recorded at 80 kV with an energy resolution of 0.8 eV. The Pt 4 mark in (a) corresponds to the location of the EELS spectrum displayed in Figure 2e of the main text.
Figure S11 | Evolution of the number of 2D-phosphane layers along the cross-section extracted from the High Angle Annular Dark Field (HAADF) contrast image (a) and used for the experiment in Figure 2 of the main text.
## C. MARCUS-GERISCHER THEORY OF PHOTO-OXIDATION
We have developed a simple model based on Marcus-Gerischer theory (MGT) in order to better explain the photo-induced oxidation process in 2D-phosphane layers and its dependency with layer thickness.
## C1 - Charge transfer doping
Charge transfer between the 2D-phosphane layers and the environment is here assumed to take place through partial reaction steps of the well-known oxygen-water redox couple that is naturally present in air, in accordance with the following reaction:
$$O _ { 2 ( a q ) } + 4 e ^ { - } + 4 H ^ { + } \hookrightarrow 2 H _ { 2 } O.$$
It was previously established that this oxygen-water redox equilibrium reaction, which is pH dependent, can displace the Fermi level of a semiconductor, such as graphene, towards the Nerstian potential of the aqueous (aq) solution of oxygen through direct charge transfer (CT). 4 This reaction is driven by the potential difference between the two systems. In the case of P(black), the driving force is significant because the Nerstian potential of the redox reaction at pH 6 is -5.3 V (i.e. -5.3 eV potential energy relative to the vacuum level) whereas the workfunction of intrinsic P(back) is about 4.2 eV . In the case of graphene, which workfunction 5 is about 4.5 eV, CT reaction leads to p-doping because equilibrium is only reached when the Fermi level equals that of the Nerstian potential of the solution - i.e. at ~-5.3 eV - because there is an infinite reservoir of oxygen in air (i.e. oxygen concentration is constant due to Henry's law). In the case of P(black), this charge transfer reaction appears however shortcut at an intermediate step by the local oxidation of P atoms (see main text), which reaction leads to etching rather than doping.
## C2 - Rate limiting step
For completion, the oxygen/water redox reaction above involves at least 4 intermediate steps (i.e. a total of 4 electrons transferred) and each step presents different kinetics and intermediate species. The first is an electron transfer from P(black) to the solvated oxygen acceptor state according to:
$$O _ { 2 ( a q ) } + e ^ { - } \hookrightarrow O _ { 2 ( a q ) } ^ { - }.$$
Because this reaction does not required neither 𝐻𝐻 glyph<c=1,font=/IFMIOK+CambriaMath> nor 𝑂𝑂𝐻𝐻 glyph<c=1,font=/IFMIOK+CambriaMath> , it is pH independent. Also, its reverse direction is not really favored since 𝑂𝑂 glyph<c=1,font=/IFMIOK+CambriaMath> (glyph<c=1,font=/IFMIOK+CambriaMath>glyph<c=2,font=/IFMIOK+CambriaMath> ) ∙glyph<c=1,font=/IFMIOK+CambriaMath> species are reactive intermediates against P(black). In the MGT framework, this CT step is expected to be very slow because aqueous O2 acceptor state is located at about -3.1 eV relative to vacuum , which is well above that of the 6 P(black) band edges (See Figure S12). As a result of poor overlap, this CT step appears as the rate limiting step for the oxidation of P(black) in air. Two experimental facts described in the main text support this hypothesis: 1) CT rate as determined by Raman is very slow in the dark; 2) the photo-oxidation reaction is found to be independent of the pH of the solution, at least in the range between 5.8 and 7.8.
## C3 - Reaction kinetics based on MGT
## i) General expression and considerations about energy levels
To access the kinetics of the photo-induced CT reaction, we use the MGT describing the interaction of the solvent (water) with 𝑂𝑂 glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=2,font=/IFMIOK+CambriaMath> and 𝑂𝑂 glyph<c=1,font=/IFMIOK+CambriaMath> (glyph<c=1,font=/IFMIOK+CambriaMath>glyph<c=2,font=/IFMIOK+CambriaMath> ) ∙glyph<c=1,font=/IFMIOK+CambriaMath> . Because of fast water rearrangements due the presence of the charge, the level fluctuates in energy over a range of ± 1 eV, which is 7 expressed in the MGT using a total distribution (TD) function of the state that follows a Gaussian shape (See Figure S12). This CT rate is given by:
$$D _ { O _ { 2 } } ( E ) = [ 0 _ { 2 } ] \cdot b _ { o } e x p \left [ \frac { - \left ( E - E _ { F, r e d o x } ^ { 0 } + \lambda \right ) ^ { 2 } } { 4 k T \lambda } \right ],$$
where 𝑂𝑂 glyph<c=1,font=/IFMIOK+CambriaMath> is the concentration of oxygen in water, 𝐸𝐸 is the potential energy (in eV), 𝐸𝐸 glyph<c=1,font=/IFMIOK+CambriaMath> ,glyph<c=4,font=/IFMIOK+CambriaMath> glyph<c=2,font=/IFMIOK+CambriaMath> glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=3,font=/IFMIOK+CambriaMath> glyph<c=5,font=/IFMIOK+CambriaMath> glyph<c=1,font=/IFMIOK+CambriaMath> is the electrochemical potential energy (or Fermi level of the solution), λ is the reorganization energy, and 𝑏𝑏 glyph<c=1,font=/IFMIOK+CambriaMath> is a normalizing parameter such that 𝐷𝐷 glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=1,font=/IFMIOK+CambriaMath> 𝐸𝐸 𝑑𝑑𝐸𝐸 = 𝑂𝑂glyph<c=1,font=/IFMIOK+CambriaMath> . In principle, the CT kinetics may be related to the overlap between the density of states of the n-layer 2D-phosphane and the solution, but the levels in the solution are not equivalent, in a conventional sense, to that a band. That is, an electron cannot be transferred to aqueous oxygen through a simple physical process, such as band-to-band tunneling, because molecular rearrangements are taking place during the course of the reaction . It remains however possible, energetically, to transfer an 4 electron from P(black) or n-layers 2D-phosphane with excess electrons to the oxygen state because the energy levels are close.
Because 𝐸𝐸 glyph<c=1,font=/IFMIOK+CambriaMath> ,glyph<c=4,font=/IFMIOK+CambriaMath> glyph<c=2,font=/IFMIOK+CambriaMath> glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=3,font=/IFMIOK+CambriaMath> glyph<c=5,font=/IFMIOK+CambriaMath> glyph<c=1,font=/IFMIOK+CambriaMath> =-4.1 eV for 𝑂𝑂 glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=2,font=/IFMIOK+CambriaMath> / 𝑂𝑂 glyph<c=1,font=/IFMIOK+CambriaMath> (glyph<c=1,font=/IFMIOK+CambriaMath>glyph<c=2,font=/IFMIOK+CambriaMath> ) ∙glyph<c=1,font=/IFMIOK+CambriaMath> and workfunction of P(back) is about 4.2 eV (bandgap, Eg=0.35 eV), it appears clear in Figure S12 that intrinsic (undoped) P(black) should be roughly stable in normal condition, which is consistent with experimental observations of a relatively good stability of the crystal in air, as reported by Bridgeman more than 100 years ago . 8
Figure | S12 Sketch of the densities of states (DOS) of aqueous oxygen acceptor and of nlayers 2D-phosphane, as predicted by the Marcus-Gerisher theory . On the left : DOS of mono-, bi-, tri- 2D-phosphane layers and bulk P(Black) extracted from Ref. 5 and 10. On the right : the energy diagram of the DOS of aqueous oxygen acceptor from Ref. (4) with the enlargement of the DOS due to solvent reorganization. The figure also illustrates by the blue arrows the proposed CT reaction induced by photo-excitation of n-layer 2D-phosphane.
## ii) Detail on the kinetics of charge transfer to bulk P(black)
The kinetic for the oxidation of the P(black) can be described by the following set of reactions.
The overall reaction is:
$$\theta + O _ { 2 ( a q ) } + h \nu \rightarrow \theta _ { o x } \,. \quad ( S 3 )$$
S3 can be subdivided into the following mechanism:
$$\begin{array} { c } \text{uuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu
} } \\ \theta + h v \stackrel { k _ { 1 } } { \to } \theta ^ { * }, & \quad ( S 4 ) \\ \theta ^ { * } \stackrel { k _ { 2 } } { \to } \theta + \Gamma _ { x }, & \quad ( S 5 ) \\ \theta ^ { * } + O _ { 2 ( a q ) } \stackrel { k _ { 3 } } { \to } O _ { 2 ( a q ) } ^ { - } + \theta + h ^ { + }, & \quad ( S 6 ) \\ \quad & \quad O _ { 2 ( a q ) } ^ { - } + \theta + h ^ { + } \stackrel { k _ { 4 } } { \to } \theta _ { o x } \,. & \quad ( S 7 ) \\ \quad. \quad \, \omega \cdot \quad. \quad \omega \cdot \quad \omega \cdot \quad \omega \cdot \quad \omega ^ { * } \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \end{array}$$
$$( S 6 )$$
$$O _ { 2 ( a q ) } ^ { - } + \theta + h ^ { + } \stackrel { \ddot { \mu } _ { 4 } } { \to } \theta _ { o x } \,.$$
In these equations, 𝜃𝜃 is the coverage of pristine P(black); 𝜃𝜃 ∗ is the coverage of excited P(black) by light - note that 𝜃𝜃 ∗ is also linked to the carrier concentration of photo-generated electrons, 𝛿𝛿𝛿𝛿 , at steady state because of the formation of electron-hole pairs; 𝜃𝜃 glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=2,font=/IFMIOK+CambriaMath> is the coverage of oxidized P(black) in the form of 𝑃𝑃𝑂𝑂 glyph<c=1,font=/IFMIOK+CambriaMath> ; ℎ𝜈𝜈 is a photon count; Γglyph<c=1,font=/IFMIOK+CambriaMath> regroups all decay channels leading to the relaxation of P(black) excited state to its ground state (phonon generation, fluorescence, etc.); 𝑘𝑘 glyph<c=1,font=/IFMIOK+CambriaMath> , 𝑘𝑘 glyph<c=1,font=/IFMIOK+CambriaMath> , 𝑘𝑘 glyph<c=1,font=/IFMIOK+CambriaMath> , 𝑘𝑘 glyph<c=1,font=/IFMIOK+CambriaMath> are the rate constants.
Reaction S4 describes the direct photoexcitation of P(black). In this equation, we consider that electron-hole pairs are produced from the excitation and their relaxation towards the CB edge is very fast. Thus, carriers are generated and characterized by the energy of the band edges. Reaction S5 represents all the decay channels, radiative or not, of P(black) excited states. Driven by the CT to aqueous O2, CT reaction S6 is responsible for the formation of the aqueous oxygen radial anion at the surface of P(black). This reaction is slow and assumed to be the limiting step with a rate constant, 𝑘𝑘 glyph<c=1,font=/IFMIOK+CambriaMath> . As detailed below, 𝑘𝑘 glyph<c=1,font=/IFMIOK+CambriaMath> can be readily derived by the MGT. The last reaction S7 is a fast oxidation of P(black) by the oxygen anion oxidative specie or related intermediates, leading to etching.
Note that additional reactions are required in order to complete the reduction of 𝑂𝑂 glyph<c=1,font=/IFMIOK+CambriaMath> towards 𝐻𝐻 glyph<c=1,font=/IFMIOK+CambriaMath> 𝑂𝑂 (i.e. a total of 4 electrons are involved in the complete water/oxygen redox reaction). However, our experiments indicate a high reactivity of 𝑂𝑂 glyph<c=1,font=/IFMIOK+CambriaMath> (glyph<c=1,font=/IFMIOK+CambriaMath>glyph<c=2,font=/IFMIOK+CambriaMath> ) ∙glyph<c=1,font=/IFMIOK+CambriaMath> intermediate species towards P(black) and thus any further electron transfer steps are neglected. To be noted, the reaction S6 and S7 are pdoping and de-doping processes of P(black), respectively. Thus, these steps overall lead to no doping of P(black). Furthermore, 𝑂𝑂 glyph<c=1,font=/IFMIOK+CambriaMath> (glyph<c=1,font=/IFMIOK+CambriaMath>glyph<c=2,font=/IFMIOK+CambriaMath> ) ∙glyph<c=1,font=/IFMIOK+CambriaMath> species can attack any pristine sites on the P(black) surface because the positive charges in reaction S6 is delocalized in the P(black).
To simplify the reaction kinetics, we assume that S4 and S5 are characterized by extremely fast kinetics compared to the CT processes in S6 and S7. Thus, these reactions can be considered as static, while solving S4 and S5 for the photo-stationary equilibrium situation, which sets the carrier population available for CT, as:
$$\frac { d \theta ^ { * } } { d t } = k _ { 1 } \theta \cdot h \nu - k _ { 2 } \cdot \theta ^ { * } = 0 \. \quad \text{(S8)}$$
This yields:
$$\theta ^ { * } = \frac { k _ { 1 } } { k _ { 2 } } \theta \cdot h \nu \,. \quad \text{ (S9)}$$
From this equation, an excess of electron in the conduction band at steady state, 𝛿𝛿𝛿𝛿 , can be obtained for CT, corresponding to a coverage 𝜃𝜃 ∗ . This excess is therefore directly proportional to the photon flux and surface coverage 𝜃𝜃 . The concentration of thermally generated carriers is neglected in the actual process, which is justified by the significant bandgap of the material (of more than 0.35 eV in n-layer 2D-phosphane). This is consistent with the good stability of our samples in ambient conditions when kept in the dark.
One can develop further S9 for a thin layer of P(black) to get: 𝛿𝛿𝛿𝛿 = 𝑔𝑔glyph<c=1,font=/IFMIOK+CambriaMath>glyph<c=2,font=/IFMIOK+CambriaMath>glyph<c=3,font=/IFMIOK+CambriaMath> 𝜏𝜏 glyph<c=1,font=/IFMIOK+CambriaMath> = (𝐽𝐽 glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=1,font=/IFMIOK+CambriaMath> 𝐴𝐴𝜎𝜎 glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=2,font=/IFMIOK+CambriaMath> glyph<c=3,font=/IFMIOK+CambriaMath> 𝜌𝜌 glyph<c=1,font=/IFMIOK+CambriaMath> 𝑛𝑛 𝜃𝜃) 𝜏𝜏 glyph<c=1,font=/IFMIOK+CambriaMath> , where 𝑔𝑔 glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=2,font=/IFMIOK+CambriaMath> glyph<c=3,font=/IFMIOK+CambriaMath> is the optical generation, corresponding to 𝐽𝐽 glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=1,font=/IFMIOK+CambriaMath> 𝐴𝐴𝜎𝜎 glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=2,font=/IFMIOK+CambriaMath> glyph<c=3,font=/IFMIOK+CambriaMath> 𝜌𝜌 glyph<c=1,font=/IFMIOK+CambriaMath> 𝑛𝑛 𝜃𝜃 , and 𝜏𝜏 glyph<c=1,font=/IFMIOK+CambriaMath> = glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=1,font=/IFMIOK+CambriaMath> is the effective carrier lifetime (i.e. electron lifetime in the CB). In this expression, 𝐽𝐽 glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=1,font=/IFMIOK+CambriaMath> is the laser fluence, A the unit area, 𝜎𝜎 glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=2,font=/IFMIOK+CambriaMath> glyph<c=3,font=/IFMIOK+CambriaMath> is the absorption cross-section, 𝜌𝜌 glyph<c=1,font=/IFMIOK+CambriaMath> is the phosphorus surface density, 𝑛𝑛 is the number of layer.
In the presence of the aqueous oxygen condensed at the surface of P(black), the electron transfers from the CB edge to the oxygen acceptor state of the solution proceeds according to reaction S6. This process generates aqueous superoxide anions, 𝑂𝑂 glyph<c=1,font=/IFMIOK+CambriaMath> (glyph<c=1,font=/IFMIOK+CambriaMath>glyph<c=2,font=/IFMIOK+CambriaMath> ) ∙glyph<c=1,font=/IFMIOK+CambriaMath> , that react spontaneously with the surface atoms of the 2D-phosphane layer and etch them in the process. We can then assume that the oxidation reaction will be mainly limited by the S6 reaction step.
In the MGT framework, the electron transfer rate between the conduction band of P(black) and the acceptor states of the solution, described by S6 , is explicitly given by:
$$\frac { d O _ { 2 ( a q ) } ^ { - } } { d t } = \int _ { - \infty } ^ { \infty } b _ { i } \theta ^ { * } ( 1 - f ( E ) ) D _ { O _ { 2 } } ( E ) d E. \quad ( S 1 0 )$$
Where 𝐷𝐷 glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=1,font=/IFMIOK+CambriaMath> is given by S2, 𝑏𝑏 glyph<c=1,font=/IFMIOK+CambriaMath> regroups several prefactors related to the geometry of the molecular system and f(E) is the Fermi-Dirac distribution. Assuming now that all electrons are distributed at the minimum of the CB of a n-layer 2D-phosphane, 𝐸𝐸 glyph<c=1,font=/HXCCAM+CambriaMath> ,glyph<c=1,font=/IFMIOK+CambriaMath> , by a Dirac delta function, we obtain after integrating:
$$\frac { d O _ { 2 ( a q ) } ^ { ; - } } { d t } = b _ { i } \theta ^ { * } D _ { O _ { 2 } } \Big ( E _ { c, n } \Big ) \,, \quad ( S 1 1 )$$
or more explicitly using 𝛿𝛿𝛿𝛿 = 𝜃𝜃 ∗ and equation S2:
$$\frac { d O _ { 2 ( a q ) } ^ { \cdot - } } { d t } = \, b _ { i } b _ { o } [ O _ { 2 } ] \, \delta \eta \, e x p \left [ - \frac { ( E _ { c, n } - E _ { F, r e d o x } ^ { 0 } + \lambda ) ^ { 2 } } { 4 k T \lambda } \right ]. \quad ( S 1 2 )$$
Because the rate equation for reaction S6 is: glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=1,font=/IFMIOK+CambriaMath>( glyph<c=1,font=/HXCCAM+CambriaMath>glyph<c=2,font=/HXCCAM+CambriaMath> ) ∙ glyph<c=1,font=/HXCCAM+CambriaMath> glyph<c=1,font=/HXCCAM+CambriaMath> glyph<c=2,font=/HXCCAM+CambriaMath> = 𝑘𝑘 glyph<c=1,font=/IFMIOK+CambriaMath> 𝜃𝜃 ∗ 𝑂𝑂 glyph<c=1,font=/IFMIOK+CambriaMath> , the reaction constant k3 therefore become:
$$k _ { 3 } = b _ { i } b _ { o } e x p \left [ - \frac { \left ( E _ { c, n } - E _ { F, r e d o x } ^ { 0 } + \lambda \right ) ^ { 2 } } { 4 k _ { B } T \lambda } \right ]. \quad \quad ( S 1 3 )$$
Last, we can also write Equation S12 by taking into account that glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=1,font=/IFMIOK+CambriaMath>( glyph<c=1,font=/HXCCAM+CambriaMath>glyph<c=2,font=/HXCCAM+CambriaMath> ) ∙ glyph<c=1,font=/HXCCAM+CambriaMath> glyph<c=1,font=/HXCCAM+CambriaMath> glyph<c=2,font=/HXCCAM+CambriaMath> = glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=1,font=/HXCCAM+CambriaMath>glyph<c=2,font=/HXCCAM+CambriaMath> glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=2,font=/IFMIOK+CambriaMath> = -glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=1,font=/IFMIOK+CambriaMath> glyph<c=2,font=/IFMIOK+CambriaMath> . By renaming the variable 𝑏𝑏 ≡ 𝑏𝑏 glyph<c=1,font=/IFMIOK+CambriaMath> 𝑏𝑏 glyph<c=1,font=/IFMIOK+CambriaMath> , the oxidation rate can be formulated by the following kinetic equation:
$$\frac { d \theta } { d t } = - b \delta \eta \, [ O _ { 2 } ] e x p \left [ - \frac { \left ( E _ { c, n } - E _ { F, r e d o x } ^ { 0 } + \lambda \right ) ^ { 2 } } { 4 k _ { B } T \lambda } \right ].$$
The above expression can also be rewritten in terms of the energy gap of a n-layer 2D-phosphane, 𝐸𝐸 glyph<c=1,font=/IFMIOK+CambriaMath> ,glyph<c=1,font=/IFMIOK+CambriaMath> , assuming an intrinsic Fermi level, 𝐸𝐸 glyph<c=1,font=/IFMIOK+CambriaMath> , i.e. in the middle of the gap. With the definitions above, we can develop more explicitly the expression for the oxidation rate of the P(black):
$$\frac { d \theta } { d t } = - b ( A \sigma _ { a b s } \rho _ { A } \tau _ { n } n \, k _ { 0 } ) \, \theta J _ { p h } \, [ O _ { 2 } ] \cdot e x p \left [ - \frac { \left ( \frac { | E _ { g, n } | } { 2 } + E _ { i } - E _ { F, r e d o x } ^ { 0 } + \lambda \right ) ^ { 2 } } { 4 k _ { B } T \lambda } \right ]. \quad ( S 1 5 )$$
As discussed in the main text, the above expression describes well the observed behavior for the photo-oxidation of the P(black) and of n-layer 2D-phosphane. That is, the Raman intensity is
proportional to the coverage 𝜃𝜃 , which then decays exponentially with time. The oxidation rate is also directly proportional to the photon flux, the 𝑂𝑂 glyph<c=1,font=/IFMIOK+CambriaMath> concentration (or partial pressure) and more strikingly, as evidenced by TEM-EELS results, the rate is fastest for the thinnest n-layer 2Dphosphane.
## C5 -On the consequences of electronic confinement on CT kinetics
The situation of n layer 2D phosphane is different than for bulk P(black) --because electronic confinement in nlayer 2D phosphane leads -to an enlargement of the band gap energy 9,10 . According to MGT, larger bandgaps should indeed improve the energy level alignment between the 2D phosphane conduction band and the solvated oxygen acceptor state (see illustration in -Figure S12). Hence, MGT predicts faster kinetics for electron transfer reaction toward the solution acceptor states (i.e. O2(aq)) with decreasing thickness of the layer. In other words, the rate of transfer from the conduction band increases exponentially with the square of the bandgap energy, as indicated in Equation S15 . Note that this is true up to the maximum level of the acceptor states, which is roughly 3.1eV. This situation is illustrated in Figure S12 by the relative -length of the arrows in blue .
For intrinsic 2D phosphane layers kept in the dark, the electron population in the conduction -band is rather limited at room temperature and mainly given by the intrinsic carrier concentration. This population decreases with increasing bandgap. This explains the relatively good stability of thin 2D phosphane layers in the dark in ambient condition. As discussed in the main text, n layer --2Dphosphane exposure to light will generate a significant carrier population that is photo -generated by the process. The electrons and holes accumulate at the conduction and valence band edges and become available for transfer to the aqueous oxygen acceptor states .
## REFERENCES
- et al . Transfer printing by kinetic control of adhesion to an elastomeric stamp.
1. Meitl, M.A., Nature Mater . 5 , 33-38 (2006)
- 2. Wagner, C.D. et al. NIST, Standard Reference Database 20, Version 3.4 (http:/srdata.nist.gov/xps/) (2003)
- 3. Moulder, J.F., Stickle, W.F., Sobol, P.E., Bomben, K.D., Handbook of X-ray Photoelectron Spectroscopy, Perkin-Elmer Corp., Eden Prairie, MN, (1992)
- 4. Memming R., Semiconductor electrochemistry , Wiley-VCH Weinheim ISBN: 3-527-30147-X (2001)
- 5. Goodman. N.B., Ley, L. & Bullett, D., Valence-band structures of phosphorus allotropes W. Phy. Rev. B 27, 12 (1983)
- 6. Levesque, P. L. et al . Probing charge transfer at surface using graphene transistors. Nano Lett . 11 , 132-137 (2011).
- 7. Ignazak, A., Schickler, W. & Bartensclager, S., J. Electroanal. Chem . 586 (2006)
- 8. Bridgeman, P. W. Two new modifications of Phosphorus. J. Am. Chem. Soc. , 36 , 1344-1363 (1914).
- 9. Liu, H., Neal, A. T., Zhu, Z., Luo, Z., Xu, X., Tomanek D. & Ye, P. D. Phosphorene: An Unexplored 2D Semiconductor with a High Hole Mobility. ACS Nano , online, DOI 10.1021/nn501226z. (2014).
- 10. Zhang, S. et al., Extraordinary Photoluminescence and Strong Temperature/Angle-dependent Raman Responses in Few-layer Phosphorene, arXiv:1407.0502, 2014 | null | [
"A. Favron",
"E. Gaufrès",
"F. Fossard",
"P. L. Lévesque",
"A-L. Phaneuf-L'Heureux",
"N. Y-W. Tang",
"A. Loiseau",
"R. Leonelli",
"S. Francoeur",
"R. Martel"
] | 2014-08-02T06:49:40+00:00 | 2014-08-21T14:13:19+00:00 | [
"cond-mat.mes-hall"
] | Exfoliating pristine black phosphorus down to the monolayer: photo-oxidation and electronic confinement effects | Thin layers of black phosphorus have recently raised interest for their
two-dimensional (2D) semiconducting properties, such as tunable direct bandgap
and high carrier mobilities. This lamellar crystal of P atoms stacked together
by weak van der Waals forces can be exfoliated down to the stratophosphane
monolayer (also called phosphorene) using procedures similar to those used for
graphene. Properties of this 2D material are however challenging to probe due
to a fast and ubiquitous degradation upon exposure to ambient conditions.
Herein, we investigate the crystal degradation using in-situ Raman and
transmission electron spectroscopies and highlight a process involving a
photo-induced oxidation reaction with adsorbed oxygen in water. The
experimental conditions to prepare and preserve mono-, bi- and multilayers of
stratophosphane in their pristine states were determined. Study on these 2D
layers provides new insights on the effect of confinement on the chemical
reactivity and the vibrational modes of black phosphorus. |
1408.0346v1 | ## Synthesis and post-annealing effects of alkaline-metal-ethylenediamine-intercalated superconductors Ax (C2H8N2) Fe2- Se2 ( y z A = Li, Na) with T c = 45 K
Takashi Noji, Takehiro Hatakeda, Shohei Hosono, Takayuki Kawamata, Masatsune Kato, Yoji Koike
Department of Applied Physics, Graduate School of Engineering, Tohoku University,
6-6-05 Aoba, Aramaki, Aoba-ku, Sendai 980-8579, Japan
New iron-based intercalation superconductors Ax (C2H8N2) Fe2y z Se2 ( A = Li, Na) with T c = 45 K have successfully been synthesized via intercalation of dissolved alkaline metal in ethylenediamine. The c -axis lengths of Ax (C2H8N2) Fe2y z Se2 ( A = Li, Na) are 20.74(7) Å and 21.9(1) Å, respectively, and are about 50 % larger than that of K Fe2 x Se2, indicating that not only alkaline metal but also ethylenediamine is intercalated between the Se-Se layers of FeSe. It seems that the highT c of Ax (C2H8N2) Fe2y z Se2 ( A = Li, Na) is caused by the possible two-dimensional electronic structure due to the large c -axis length. Through the post-annealing in an evacuated glass tube, it has been found that T c decreases with increasing post-annealing temperature and that deintercalation of EDA from the as-intercalated sample takes place at low temperatures below 250 ℃ .
Keywords: superconductivity, intercalation, FeSe, etylenediamine, alkaline metal
## 1. Introduction
The superconducting transition temperature, T c , of the iron-based chalcogenide superconductor FeSe is only 8 K [1], but dramatically increases up to ~37 K by the application of high pressure [2]. The successful synthesis of potassium-intercalated KxFe2-ySe2 with T c ~31 K at ambient pressure was an exciting breakthrough [3]. Recently, it has been reported that the intercalation into FeSe of alkaline and alkaline-earth metals in liquid ammonia yields a variety of compounds with significantly enhanced T c 's of 40 - 46 K [4 - 6]. The c-axis length of Li0.9Fe2Se2(NH3)0.5 with T c = 44 K is 16.518 Å [5] and larger than 14.0367 Å in K Fe2Se2 [3], indicating x that not only lithium but also ammonia is intercalated. Moreover, it has been reported that lithium- and pyridine-intercalated Li (C5H5N) Fe2x y z Se2 exhibits superconductivity with T c = 45 K and that post-annealing of the intercalated sample drastically expands the c -axis length from 16.0549 Å to 23.09648 Å and increases the superconducting shielding volume fraction [7].
Here, we report on the successful synthesis of new superconductors Ax (C2H8N2) Fe2y z Se2 ( A = Li, Na) with T c = 45 K via intercalation of dissolved alkaline metal in ethylenediamine (EDA), C2H8N2. Post-annealing effects on the crystal structure and superconductivity are also discussed.
## 2. Experimenta l
Polycrystalline samples of FeSe were prepared by the solid-state reaction method. Starting materials were powders of Fe and Se, which were weighted stoichiometrically, mixed thoroughly and pressed into pellets. The pellets were sealed in an evacuated quartz tube and heated at 800 ℃ for 40 h. The obtained pellets of FeSe were pulverized into powder to use for the intercalation. Dissolved alkaline metal in EDA was intercalated into the powdery FeSe as follows. An appropriate amount of the powdery FeSe was placed in a beaker filled with 0.2 M solution of pure lithium or sodium metal dissolved in EDA. All the process was performed in an argon-filled glove box. Post-annealing of as-intercalated samples was carried out at 100 - 500 ℃ for 60 h in an evacuated glass tube. Both FeSe and the intercalated samples were characterized by powder x-ray diffraction using Cu Kα radiation. For the intercalated samples, an airtight sample holder was used. In order to observe the superconducting transition, the magnetic susceptibility, χ , was measured using a superconducting quantum interference device (SQUID) magnetometer. Measurements of the electrical resistivity, ρ , were also carried out by the standard dc four-probe method. For the ρ measurements, as-intercalated powdery samples were pressed into pellets. Then, the pellets were sintered at 200 ℃ for 20 h in an evacuated glass tube. Thermogravimetric (TG) measurements were performed in flowing gas of argon, using a commercial analyzer (SII Nano Technology Inc., EXSTAR DSC7020).
## 3. Results and discussion
Figure 1 shows powder x-ray diffraction patterns of as-intercalated samples of Ax (C2H8N2) Fe2y z Se2 ( A = Li, Na). The broad peak around 2 θ = 20° is due to the airtight sample holder. Although there are unknown peaks, most of sharp Bragg peaks are due to the intercalation compound of Ax (C2H8N2) Fe2y z Se2 ( A = Li, Na) and the host compound of FeSe, so that they are able to be indexed based on the ThCr2Si2-type and PbO-type structures, respectively. Therefore, it is found that alkaline metal and EDA are partially intercalated into FeSe, while there remains a non-intercalated region of FeSe in the
Fig. 1. Powder x-ray diffraction patterns of as-intercalated samples consisting of A x (C H N ) Fe 2 8 2 y 2z Se ( 2 A = Li, Na) and FeSe using Cu K radiation. Indexes without and α with asterisk are based on the ThCr Si -type and PbO-type structures, respectively. 2 2 Peaks marked by ▼ are unknown.
samples. The lattice constants of Li (C2H8N2) Fe2x y z Se2 are calculated to be a = 3.458(6) Å and c = 20.74(7) Å. The c-axis length of Na (C2H8N2) Fe2x y z Se2 is 21.9(1) Å. Since the unit cell of Ax (C2H8N2) Fe2y z Se2 ( A = Li, Na) includes two Fe layers, the distance between the neighboring Fe layers is 10.37(4) Å and 10.95(5) Å, respectively, and much larger than 5.515(1) Å of FeSe. Taking into account our previous results that the intercalation of only lithium into Fe(Se,Te) has neither effect on the superconductivity nor crystal structure [8], it is concluded that not only lithium or sodium but also EDA has been intercalated between the Se-Se layers of FeSe.
Figure 2 shows the temperature dependence of χ in a magnetic field of 10 Oe on zero-field cooling (ZFC) and on field cooling (FC) for as-intercalated powdery samples. The first T c is observed at 45 K and the second T c is at 8 K. Taking into account the powder x-ray diffraction results, it is concluded that the first is due to bulk superconductivity of Ax (C2H8N2) Fe2y z Se2 ( A = Li, Na), while the second is due to that of the non-intercalated region of FeSe. The inset shows the temperature dependence of ρ for the sintered (200 ℃ , 20 h) pellet sample. Although the as-intercalated sample pelletized simply at room temperature did not show zero resistivity, it is found that the resistivity of the sintered pellet sample starts to decrease at 43 K with decreasing temperature and reaches zero at 18 K. Since the intercalation of only lithium does not increase T c [8], it seems that the highT c of Ax (C2H8N2) Fe2y z Se2 ( A = Li, Na) is caused by the possible two-dimensional electronic structure due to the large c-axis length.
Figure 3 shows the TG curve on heating up to 900 ℃ at the rate of 1 ℃ /min for the as-intercalated powdery sample. Three steps of mass loss are observed; 2.9 % loss below 100 ℃ , 11.3 % loss between 100 and 200 ℃ , and large loss above 450 ℃ .
Fig. 3. Thermogravimetric (TG) curve on heating at the rate of 1 ℃ /min for the as-intercalated powdery sample consisting of Li (C2H8N2) Fe2x y z Se2 and FeSe.
Fig. 2. Temperature dependence of the magnetic susceptibility, χ , in a magnetic field of 10 Oe on zero-field cooling (ZFC) and field cooling (FC) for as-intercalated powdery samples consisting of Ax (C2H8N2) Fe2y z Se2 ((a) A = Li, (b) A = Na) and FeSe. The inset shows the temperature dependence of the electrical resistivity, ρ , for the sintered (200 ℃ , 20 h) pellet sample consisting of Li (C2H8N2) Fe2x y z Se2 and FeSe.
Figure 4 shows powder x-ray diffraction patterns of the as-intercalated sample and samples post-annealed at various temperatures for 60 h. For samples post-annealed below 200 ℃ , diffraction peaks of Li (C2H8N2) Fe2x y z Se2 remain. For samples post-annealed at 250 ℃ and 300 ℃ , only peaks due to FeSe are observed, and these peaks disappear for the sample post-annealed at 500 ℃ . Accordingly, the first step of the TG curve below 100 ℃ may be due to desorption of EDA on the surface of grains. The second step between 100 and 220 ℃ is due to deintercalation of EDA. The third step above 500 ℃ may be due to unknown products including iron.
Figure 5 shows the temperature dependence of χ in a magnetic field of 10 Oe on ZFC and FC for as-intercalated and post-annealed (100 - 500 ℃ , 60 h) powdery samples. The T c of Li (C H8N2) Fe2x 2 y z Se2 decreases with increasing post-annealing temperature. For samples post-annealed at 250 ℃ and 300 ℃ , only the superconducting transition of FeSe is observed, and the transition also disappears for the sample post-annealed at 500 ℃ . These results are in good correspondence to the results of x-ray diffraction.
Fig. 4. Powder x-ray diffraction patterns of as-intercalated and post-annealed (100 - 500 ℃ , 60 h) samples using Cu K radiation. Indexes without and α with asterisk are based on the ThCr Si -type and 2 2 PbO-type structures, respectively. Peaks marked by▼ are unknown.
Fig. 5. Temperature dependence of the magnetic susceptibility, χ , in a magnetic field of 10 Oe on zero-field cooling (ZFC) and field cooling (FC) for as-intercalated and post-annealed (100 - 500 ℃ , 60 h) powdery samples consisting of Li (C2H8N2) Fe2x y z Se2 and FeSe.
Finally, Fig. 6 shows the maximum Tc 's of FeSe-based superconductors obtained so far at ambient pressure as a function of the distance between the neighboring Fe layers. Interestingly, Tc increases monotonically with increasing such distance and is saturated at about 45 K above 9 Å.
Fig. 6. Relation between the maximum T c and the distance of the neighbouring Fe layers of FeSe-based superconductors.
## 4. Summary
We have succeeded in synthesizing new intercalation compounds Ax (C2H8N2) Fe2y z Se2 ( A = Li, Na) via intercalation of dissolved alkaline metal in EDA. Although the samples include non-intercalated regions of FeSe, the c -axes of Ax (C2H8N2) Fe2y z Se2 ( A = Li, Na) have drastically expanded to be 20.74(7) and 21.9(1) Å, respectively, indicating that not only alkaline metal but also EDA is intercalated between the Se-Se layers of FeSe. Bulk superconductivity has been observed below 45 K in the χ measurements of as-intercalated samples of Ax (C2H8N2) Fe2y z Se2 ( A = Li, Na). Moreover, zero resistivity has been observed below 18 K for the sintered pellet sample of Li (C2H8N2) Fe2x y z Se2. It seems that the highT c of Ax (C2H8N2) Fe2y z Se2 ( A = Li, Na) is caused by the possible two-dimensional electronic structure due to the large c-axis length. Through the post-annealing in an evacuated glass tube, it has been found that T c decreases with increasing post-annealing temperature and that deintercalation of EDA from the as-intercalated sample takes place at low temperatures below 250 ℃ and that FeSe remains at temperatures around 400 ℃ . It has also turned out that the maximum T c 's of FeSe-based superconductors rise with increasing distance between the neighboring Fe layers and are saturated at about 45 K above 9 Å.
## References
- [1] F.-C. Hsu, J.-Y. Luo, K.-W. Yeh, T.-K. Chen, T.-W. Huang, P. M. Wu, Y.-C. Lee, Y.-L. Huang, Y.-Y. Chu, D.-C. Yan, and M.-K. Wu, Proc. Natl. Acad. Sci. U.S.A 105 (2008) 14262 - 14264.
- [2] S. Margadonna, Y. takabayashi, Y. Ohishi, Y. Mizuguchi, Y. Takano, T. Kagayama. T. Nakagawa, M. Takata, and K. Prassides, Phys. Rev. B 80 (2009) 064506/1 - 6.
- [3] J. Guo, S. Jin, G. Wang, S. Wang, K. Zhu, T. Zhou, M. He, and X. Chen, Phys. Rev. B 82 (2010) 180520(R)/1 - 4.
- [4] T. P. Ying, X. L. Chen, G. Wang, S. F. Jin, T. T. Zhou, X. F. Lai, H. Zhang, and W. Y. Wang, Sci. Rep. 2 (2012) 426/1 - 7.
- [5] E.-W. Scheidt, V. R. Hathwar, D. Schmitz, A. Dunbar, W. Scherer, F. Mayr, V. Tsurkan, J. Deisenhofer, and A.Loidl, Eur. Phys. J. B 85 (2012) 279/1 - 5.
- [6] M. Burrard-Lucas, D. G. Free, S. J. Sedlmaier, J. D. Wright, S. J. Cassidy, Y. Hara, A. J. Corkett, T. Lancaster, P. J. Baker, S. J. Blundell, and S. J. Clarke, Nat. Mater. 12 (2013) 15 - 19.
- [7] A. Krzton-Maziopa, E. V. Pomjakushina, V. Y. Pomjakushin, F. Rohr, A. Schilling, and K. Conder, J. Phys.: Condens. Matter 24 (2012) 382202/1 - 6.
- [8] H. Abe, T. Noji, M. Kato, and Y. Koike, Physica C 470 (2010) S487 - S488.
- [9] A. Krzton-Maziopa, Z. Shermadini, E. Pomjakushina, V. Pomjakushin, M. Bendele, A. Amato, R. Khasanov, H. Luetkens, and K. Conder, J. Phys.: Condens. Matter 23 (2011) 052203/1 - 4.
- [10] A. F. Wang, J. J. Ying, Y. J. Yan, R. H. Liu, X. G. Luo, Z. Y. Li, X. F. Wang, M. Zhang, G. J. Ye, P. Cheng, Z. J. Xiang, and X. H. Chen, Phys. Rev. B 83 (2011) 060512(R)/1 - 4.
- [11] K.-W. Yeh, T.-W. Huang, Y.-L. Huang, T.-K. Chen, F.-C. Hsu, P. M. Wu, Y.-C. Lee, Y.-Y. Chu, C.-L. Chen, J.-Y. Luo, D.-C. Yan, and M.-K. Wu: Europhys. Lett. 84 (2008) 37002/1 - 4.
- [12] A. -M. Zhang, T. -L. Xia, K. Liu, W. Tong, Z. -R. Yang, and Q. -M. Zhang, Sci. Rep. 3 (2013) 1216/1 - 5. | 10.1016/j.physc.2014.01.007 | [
"Takashi Noji",
"Takehiro Hatakeda",
"Shohei Hosono",
"Takayuki Kawamata",
"Masatsune Kato",
"Yoji Koike"
] | 2014-08-02T07:04:16+00:00 | 2014-08-02T07:04:16+00:00 | [
"cond-mat.supr-con"
] | Synthesis and post-annealing effects of alkaline-metal-ethylenediamine-intercalated superconductors Ax(C2H8N2)yFe2-zSe2 (A = Li, Na) with Tc = 45 K | New iron-based intercalation superconductors Ax(C2H8N2)yFe2-zSe2 (A = Li, Na)
with Tc = 45 K have successfully been synthesized via intercalation of
dissolved alkaline metal in ethylenediamine. The c-axis lengths of
Ax(C2H8N2)yFe2-zSe2 (A = Li, Na) are 20.74(7) {\AA} and 21.9(1) {\AA},
respectively, and are about 50 % larger than that of KxFe2Se2, indicating that
not only alkaline metal but also ethylenediamine is intercalated between the
Se-Se layers of FeSe. It seems that the high-Tc of Ax(C2H8N2)yFe2-zSe2 (A = Li,
Na) is caused by the possible two-dimensional electronic structure due to the
large c-axis length. Through the post-annealing in an evacuated glass tube, it
has been found that Tc decreases with increasing post-annealing temperature and
that deintercalation of EDA from the as-intercalated sample takes place at low
temperatures below 250 {\deg}C. |
1408.0347v1 | ## AN INVARIANT REGION FOR THE COLLISIONAL DYNAMICS OF TWO BODIES ON KEPLERIAN ORBITS
Dario Benedetto
Dipartimento di Matematica, Sapienza Universit` a di Roma [email protected]
Flavia Lenti
Dipartimento di Scienza e di alta Tecnologia Universit` a dell'Insubria [email protected]
ABSTRACT. We study the dynamics of two bodies moving on elliptic Keplerian orbits around a fixed center of attraction and interacting only by means of elastic or inelastic collisions. We show that there exists a bounded invariant region: for suitable values of the total energy and the total angular momentum (explicitly computable) the orbits of the bodies remain elliptic, whatever are the number and the details of the collisions. We show that there exists a bounded invariant region even in the case of two bodies interacting by short range potential.
Planetary systems, Planetary rings, Elastic collisions, Inelastic collisions MSC 70F15, MSC 37N05
## 1. INTRODUCTION
The interest in the collisional dynamics in a planetary system goes back to Poincar´ e. In particular, in [12] he studies the planetary three-body problem, with one body of large mass (the Sun) and two bodies of small mass (the planets). He indicates how to find periodic solutions (of deuxi` eme esp` ece ), as perturbations of periodic collisional solutions he can construct when the mass of the planets are infinitely small and the distance between them becomes infinitely small. In this approximation, the two bodies are on Keplerian ellipses until the 'choc' (i.e. the interaction), which moves the bodies on two other Keplerian ellipses. During the interaction, only the total energy and the total momentum are conserved, and the choc acts as an elastic collision. It should be noted that the collision can moves the bodies also on hyperbolic orbits but Poincar´ e is only interested in elliptic case.
In this work, we prove that, for this collisional dynamics, there exists an invariant bounded region of of positive measure in the phase space. More precisely, we consider two bodies moving on elliptic Keplerian orbits around a center, interacting only by means of collisions. A collision changes the orbital parameters of the bodies, and a sequence of collisions can move one of the bodies out of the system, on parabolic or hyperbolic orbits. Weshow that for suitable values of the total energy and the total angular momentum (easily computable), the bodies remain on elliptic colliding orbits. Moreover, we extend this result to the case of two point particles interacting by means a bounded short range potential.
We are neglecting the gravitational interaction between the bodies and the influence on the attractive center. These approximations can be only justified for very light particles and for a few revolution times (the time between two consecutive collisions can be very long with respect to the revolution period of the particles). Considering this, our result has the following interpretation for real systems: the particles can not leave the system due to the collisions, unless other perturbations change enough the orbital parameters.
Our result seems not known in the literature, despite its simplicity and despite the great interest in research on planetary systems. We run in the result while we were studying a numerical models for the dynamics of inelastic particles of planetary rings. It is known (see [13], [1], and [2]) that the inelasticity of the collisions is sufficient to guarantee the persistence of the rings. In contrast, in the case of elastic collisions, almost all the particles leave the system on hyperbolic orbits. Here we prove that in the case of only two particles,
the inelasticity is not needed to avoid that the orbits become hyperbolic or parabolic; in this sense, two colliding particles are a stable subsystem of a ring. This observation and its consequences can be interesting for the study of various models of planetary rings. In particular, in some models it is assumed that the collisions are elastic for small relative velocities (see [5] for experimental result on the inelasticity of ice balls).
From the mathematical point of view, it can be interesting to study how the particles moves on the invariant region. Preliminary two-dimensional numerical simulations, in which the impact parameter is randomly chosen, show that the two orbits are approximately tangent, for most of the time. It is difficult to implement a numerical simulation for the more realistic three-dimensional case. The problem is that it is not easy to find efficiently the values of the anomalies which correspond to collisional configurations. Some useful suggestion and some technical insight can be obtained analyzing the solutions to the problem of finding the critical points of the distance from two elliptic orbits (see [4]).
The paper is organized as follows. In section 2, we establish the mathematical notation and the exact nature of the problem. In section 3, we analyze a simplified model considering a two-dimensional dynamics of the orbits instead of the dynamics of the to bodies. We consider more general cases in section 4 (two bodies in R 2 ) and in section 5, in which we analyze the case of two bodies in R 3 ; here we also analyze the case of point particles interacting by means a short range potential.
## 2. THE PROBLEM
The system we analyze consists in two spherical bodies, of mass m 1 and m 2 and radii R 1 and R 2 respectively, which are attracted by a fixed gravitational center and interact by means of elastic or inelastic collisions. We indicate with x i and v i the position of the center and the velocity of the body i = 1 2, with , M = m 1 + m 2 the total mass and with m i = mi / M the fraction of the total mass carried by the body i . We choose the units of measure in such that the gravitational potential energy of the body i is mi / ri , where ri = | x i | is the distance from the attracting center.
In the case of hard spheres, inelastic collisions can be modeled supposing that only a part of the normal impulse is transferred, while the tangential one is conserved. Let us denote with n =( x 1 -x 2 ) / ( R 1 + R 2 ) the direction of the relative position at the moment of the impact, and with w = v 1 -v 2 the relative velocity. The particles are in the incoming configuration iff n w · < 0. The relative velocity after the collision is
$$( 1 ) \quad \ \ w ^ { \prime } = ( I - n \times n ) \, w - ( 1 - 2 \varepsilon ) ( n \times n ) \, w \quad \ \.$$
where I is the identity matrix, ( x × y ) i j = xi y j is the tensor product, n × n is the projector on the direction of n , I -n × n is the projector on the orthogonal plane to n , and finally e ∈ [ 0 0 5 , . ] is the parameter of inelasticity, which is 0 in the case of the elastic collision.
The outgoing velocities v ′ 1 , v ′ 2 can be obtained from w ′ using the conservation of the velocity of the center of mass
$$\mathfrak { v } ^ { \prime } = \mu _ { 1 } \mathfrak { v } _ { 1 } ^ { \prime } + \mu _ { 2 } \mathfrak { v } _ { 2 } ^ { \prime } = \mu _ { 1 } \mathfrak { v } _ { 1 } + \mu _ { 2 } \mathfrak { v } _ { 2 } = \mathfrak { v }$$
If the collision is inelastic, the normal component of w is reduced in modulus:
$$w ^ { \prime } \cdot n = - ( 1 - 2 \varepsilon ) w \cdot n, \quad | w ^ { \prime } \cdot n | \leq | w \cdot n |, \\ \cdot \quad \, \cdot \quad \, \cdot \quad \, \cdot \quad \, \cdot \quad \, \cdot \quad \, \cdot \quad \, \cdot \quad \, \cdot \quad \, \cdot \quad \,.$$
where ( 1 -2 e ) ∈ [ 0 1 , ) is the coefficient of restitution which is 1 in the elastic case. The kinetic energy T = m 1 | v 1 | 2 / 2 + m 2 | v 2 | 2 / 2 decreases and becomes
$$T ^ { \prime } = \frac { 1 } { 2 } \left ( m _ { 1 } | \mathbf v _ { 1 } ^ { \prime } | ^ { 2 } + m _ { 2 } | \mathbf v _ { 2 } ^ { \prime } | ^ { 2 } \right ) = T - 2 M \mu _ { 1 } \mu _ { 2 } \varepsilon ( 1 - \varepsilon ) ( \mathbf w \cdot \mathbf n ) ^ { 2 }.$$
We remark that in some model it is assumed that also the tangential component ( I -n × n ) w is reduced (see [2] and [7]). Moreover the restitution coefficient can depend on the
relative velocity, as shown in a huge number of theoretical and experimental studies (see e.g. [3], [6], and references therein). In all these models the kinetic energy decreases.
Let us describe what it can be happen to the orbits after a collision of the two bodies. Before the collision both energies are negative:
$$\bar { \frac { m _ { i } } { 2 } } \mathbf v _ { i } ^ { 2 } - \frac { \bar { m _ { i } } } { | \mathbf x _ { i } | } < 0, \ \ i \in 1, 2. \\ \text{alent to}$$
These conditions are equivalent to
$$& ( 3 ) \quad \ \ \ \ \ \ \ | \mathfrak { v } + \mu _ { 2 } \mathfrak { w } | ^ { 2 } < 2 | \mathfrak { x } _ { 1 } |, \ \ | \mathfrak { v } - \mu _ { 1 } \mathfrak { w } | ^ { 2 } < 2 | \mathfrak { x } _ { 2 } |, \\ & \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
with | x 1 -x 2 | = R 1 + R 2. If
$$| \mathbf w | < \min \left ( ( \sqrt { 2 | \mathbf x _ { 1 } | } - | \mathbf v | ) / \mu _ { 2 }, ( \sqrt { 2 | \mathbf x _ { 2 } | } - | \mathbf v | ) / \mu _ { 1 } \right ) \\ \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \end{array}$$
the inequalities (3) are satisfied, then, after the collision, the orbits remain elliptic because | w ′ | ≤ | w | , as follows from eq.s (1), (2). The hypothesis (4) is not sufficient to avoid that one of the particles leaves the system, because the next collision can take place at other points with very different values of w v x , , 1 and x 2. More in general, a sequence of collisions on different points can end with a particle which leaves the system on an hyperbolic orbit.
In the next section we show how to control the condition of ellipticity, regardless the collision history. Let us first note that, for bodies with vanishing radii, the possible points of collision are at most two, which is the maximum number of intersection of two non identical co-focal Keplerian orbits. As noted by Poincar´ e in [12], if two orbits have two points of intersections the points and the center of gravity are on a line or the two orbits are in a plane. Then we start our analysis in section 3 with the two dimensional case, also simplifying the dynamics considering two point particles.
## 3. THE INVARIANT REGION FOR TWO POINT-PARTICLES IN R 2
In this section we consider a simplified bi-dimensional mathematical model. We suppose that the bodies are point particles moving on co-focal Keplerian orbits in a plane. In order to allow the particles collide, we have to assume that the orbits intersect (note that this can happen at most in two points). Although this condition is satisfied, the particles can not collide because the cross section is zero for dimensionless bodies. In order to avoids this problem, we considered the dynamics of the orbits : we choose as variables the parameters of two orbits o 1 and o 2, and we evolve the system with the following procedure: we choose one of the points of intersection, we consider in that point two fictitious particles on the two orbits, we choose an impact parameter n and we consider the resulting orbits o ′ 1 and o ′ 2 after the collision of the two particles. We will show, in Theorem 3.1, that for sufficiently low value of the total energy, the new orbits o ′ 1 and o ′ 2 remain intersecting ellipses, for any choice between the two possible collision points and for any choice of the impact parameter n . The keys points of the proof are the conservation of the total momentum, the decrease of the energy, and the fact that the condition of intersection of the orbits is a feature preserved by the dynamics.
For the orbit oi of the particle i , w i is the angle between the x axis and the position of the periapsis (the point of the orbit of the minimal distances from the attractive center); J i is the true anomaly, i.e. the angle in the orbital plane between the particle and the periapsis of the orbit; Ei = v 2 i / 2 -1 / ri is the specific energy (i.e. the energy for unit of mass); Li = r 2 i ˙ J i is the specific angular momentum; ei = √ 1 + 2 EiL i 2 is the eccentricity. The position of the particle i in orbits is given by
$$\mathbf x _ { i } = \frac { L _ { i } ^ { 2 } } { 1 + e _ { i } \cos \vartheta _ { i } } \binom { \cos ( \vartheta _ { i } + \omega _ { i } ) } { \sin ( \vartheta _ { i } + \omega _ { i } ) }$$
/C8/CB/CU/D6/CP/CV
All the quantities Ei and Li are conserved during the Keplerian motion but only the combinations m E 1 1 + m E 2 2 (the total energy) and m L 1 1 + m L 2 2 (the total angular momentum) are conserved in the elastic collisions, moreover m E 1 1 + m E 2 2 decreases if the collisions are inelastic. We fix the initial values of the specific energy and the specific angular momentum of the whole system:
$$\begin{smallmatrix} L = \mu _ { 1 } L _ { 1 } + \mu _ { 2 } L _ { 2 } \\ E = \mu _ { 1 } E _ { 1 } + \mu _ { 2 } E _ { 2 } \end{smallmatrix}$$
We can assume L ≥ 0 without loss of generality, and we consider E < 0, which is the case of a couple of elliptic orbits. We rewrite the orbital parameters in terms of the differences of the energy and angular momentum:
$$& ( 7 ) \quad \delta E = E _ { 1 } - E _ { 2 } \quad \text{ from which } \quad E _ { 1 } = E + \mu _ { 2 } \delta E \quad \text{ and } \quad L _ { 1 } = L + \mu _ { 2 } \delta L \\ & \quad E _ { 2 } = E - \mu _ { 1 } \delta E \quad \text{ and } \quad L _ { 2 } = L - \mu _ { 1 } \delta L. \\ & \text{ from those an CONTINUE} \, \text{ and } \text{the dimension}$$
From these quantities we can obtain the shapes (i.e. the eccentricities) and the dimensions of the orbits. Theirs positions in the framework are specified by the angles w i , but we are only interested in the relative position of the two orbits, which is given by dw = w 1 -w 2.
L
1
e
2
= 0
= 0
1
= 0
L
2
= 0
-
2
/B4/CQ/B5
EL
=
-
0 4
.
FIGURE 1. m 1 = 0 45: . the region of the admissible values for EL 2 = -0 6 and . EL 2 = -0 4. .
Not all the values of d E and d L correspond to couple of orbits: namely the energy Ei and the angular momentum Li must satisfy the condition 0 ≤ e 2 i = 1 + 2 EiL i 2 , i.e.
$$\frac { 1 } { \mu _ { 2 } } \left ( - E - \frac { 1 } { 2 L _ { 1 } ^ { 2 } } \right ) \leq \delta E \leq \frac { 1 } { \mu _ { 1 } } \left ( E + \frac { 1 } { 2 L _ { 2 } ^ { 2 } } \right )$$
These inequalities define, in the space d L , d E , the region of admissibility
A
= (
d
L
,
d
E
)
eq.s (8) hold
,
which we show in fig. 1. The boundary of A corresponds to e 1 = 0, i.e. d E = -1 m 2 ( E + 1 2 L 2 1 ) ,
{
|
}
and e 2 = 0 i.e. d E 1 m 1 ( E + 1 2 L 2 2 ) . As follows from easy calculations, the topology of the set A depends on the value of EL 2 . If EL 2 < -1 / 2, as in fig. 1 (a), the region is not connected (this condition is equivalent to the non existence of the 'mean orbit', i.e. the orbit of energy E and angular momentum L , whose eccentricity is √ 1 + 2 EL 2 ). If -1 / 2 ≤ EL 2 < 0, as in fig. 1 (n), the region is connected.
If d E ∈ ( E / m 1 , -E / m 2 ) the orbits are both elliptic, while if d E > -E / m 2 (i.e. E 1 > 0) the first orbit is hyperbolic and if d E < E / m 1 (i.e. E 2 > 0 ) the second orbit is hyperbolic. If d L ∈ -( L / m 1 , L / m 2 ) both the particles move counterclockwise, while if d L = L / m 2 (i.e. L 1 = 0) or d L = -L / m 1 (i.e. L 2 = 0) one of the orbits degenerates.
We fix m 1 ≤ m 2 without loss of generality, and we consider the case E < 0.
δE
E
E
1
2
= 0
= 0
/D6/CT/D4/D0/CP
/CT/D1/CT/D2/D8/D7
/C8/CB/CU/D6/CP/CV
/D6/CT/D4/D0/CP
/CT/D1/CT/D2/D8/D7
e
δL
## Theorem 3.1. Let be
(9)
where
(10)
$$\sigma = \sigma ( \mu _ { 1 }, \mu _ { 2 } ) = - \frac { ( 1 - e ^ { 2 } ) ( \mu _ { 1 } ^ { 2 } + \mu _ { 2 } ^ { 2 } e ) ^ { 2 } } { 2 \mu _ { 2 } e ^ { 2 } }$$
$$e = \left ( \sqrt { ( \mu _ { 1 } / \mu _ { 2 } ) ^ { 4 } + 8 ( \mu _ { 1 } / \mu _ { 2 } ) ^ { 2 } } - ( \mu _ { 1 } / \mu _ { 2 } ) ^ { 2 } \right ) / 4.$$
If initially
$$E L ^ { 2 } < \sigma$$
the orbits remain elliptic for all times.
Moreover, | L 1 | and | L 2 | are bounded and ei ≤ ci < 1 for suitable constants c 1 , c 2 .
## Remarks.
- i. As we will see in the proof, if the orbits intersect, the value of EL 2 is bounded from below: EL 2 ≥-1 / 2.
- ii. According to the spatial scale invariance of the problem, the behavior of the system depends only on the product EL 2 of the two invariant quantities E and L .
- iii. In the proof, we study the condition of intersection in terms of the orbital parameters; for a similar analysis see [8].
- iv. If M is the mass of the central body, G the Newton constant and k = GM the standard gravitational parameter of the system, the major semi-axes of the orbit i is L i 2 / ( k ( 1 -ei )) where the eccentricity is ei = √ 1 + 2 L i 2 Ei / k 2 . The condition (11) must be rewritten in term of EL 2 / k 2 .
- v. The invariant region is large, from two point of view: it contains couples of orbits which can be very different, and the orbits live in a huge region on the configuration space. For instance, in the case m 1 = m 2 = 0 5, the critical value is . EL 2 = -27 64, / and, for this value, if one of the particles has the same orbit of the Earth, the other particle can intersect the Jupiter orbit, and can arrive at 6 95 U.A. from the Sun. If we . consider two orbits with L 1 = L 2 = L and E 1 = E 2 = E , their eccentricity is √ 5 / 32 ≈ 0 40 and the ratio between the major and the minor semi-axis is approximately 2 33. . .
- vi. We have fixed the attractive center, therefore we are not considering here a three body problem, in which we can fix only the center of mass. The planetary three body system is usually described as a perturbation of the system obtained in the canonical heliocentric variables neglecting the terms of order m m 1 2 (see e.g. [9]). This unperturbed system is a system of two particles moving independently on Keplerian orbits, with respect the position of the large body. The standard gravitational parameters are G M ( + mi ) , and can be different for the two particles. Our results also hold in this case, with minor modifications.
- vii. We have done some preliminary numerical simulations for this model, with n > 2 particles. For n = 3, if initially all the particles can collide with the others, in the elastic case one of the particle leaves the system after few collisions; in the inelastic case one of the particle stops to interact with the others after few collisions, and the orbits of the others two particles converge. A system of a large amount of colliding particles exhibits a complex behavior: a certain number of particles (decreasing with the parameter of inelasticity e ) leaves the system, the others particles asymptotically separate in non interacting clusters of one particle or two colliding particles. Let us note that for some different inelastic model (somewhat artificial), it can be proved the existence of 'ringlets' i.e. the existence of a state of n particle which does not cease to interact, and whose orbits converge (see [7]).
Proof. We prove the theorem in the case of elastic collisions, for which E L , and EL 2 are conserved quantities, and then the condition (11) is invariant for the dynamics. In the case
of inelastic collisions, the thesis follows from the fact that L is conserved and E and EL 2 can only decrease, then the condition (11) is invariant for the dynamics also in this case.
The proof follows from the fact that if EL 2 is sufficiently close to -0 5, and the orbits . intersect, then the two orbits are elliptic, as we now show.
The two orbits intersect if x 1 = x 2 for some value of the anomalies J 1 and J 2. Using eq. (5), this condition is expressed by the equalities
$$\vartheta _ { 2 } - \vartheta _ { 1 } = \omega _ { 1 } - \omega _ { 2 } = \delta \omega \quad \text{and} \ \ L _ { 1 } ^ { 2 } ( 1 + e _ { 2 } \cos \vartheta _ { 2 } ) = L _ { 2 } ^ { 2 } ( 1 + e _ { 1 } \cos \vartheta _ { 1 } ). \\ \intertext { r \dots \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \colon \end{matrix}$$
Inserting in the last equation that J 2 = J 1 + dw , we obtain an equation in the unknown J 2 that can be solved if and only if
$$& ( 1 2 ) & & e _ { 1 } e _ { 2 } \cos \delta \omega \leq 1 + L _ { 2 } ^ { 2 } E _ { 1 } + L _ { 1 } ^ { 2 } E _ { 2 } \\ & \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \end{array}$$
Let us define the set of the values of ( d L , d E ) for which two orbits of parameters L 1 , E 1 and L 2 , E 2 intersect if the angle between the periapsides is dw = h :
$$& ( 1 3 ) & & I _ { \eta } = \{ ( \delta L, \delta E ) \in A | e _ { 1 } e _ { 2 } \cos \eta \leq 1 + L _ { 2 } ^ { 2 } E _ { 1 } + L _ { 1 } ^ { 2 } E _ { 2 } \} \\ & \text{n....... }$$
By definition
$$I _ { \eta _ { 1 } } & \subset I _ { \eta _ { 2 } } \text{ if } \eta _ { 1 } < \eta _ { 2 }, \\ [ \Omega \pi ] \text{ This immediately that if } \text{w}.$$
and in particular I h ⊂ I p if h ∈ [ 0 , p ] . This implies that, if we rotate two intersecting orbits, in such a way that the two periapsides become in opposition ( dw = p ), we obtain two orbits which intersect. The intersection condition is invariant for the dynamics, then all the values ( d L , d E ) during the evolution are in the set
$$& ( 1 4 ) & I _ { \pi } = \{ ( \delta L, \delta E ) \in A | e _ { 1 } e _ { 2 } \geq - ( 1 + L _ { 2 } ^ { 2 } E _ { 1 } + L _ { 1 } ^ { 2 } E _ { 2 } ) \} \\ & \tau _ { 1 } \dots \dots \alpha _ { 2 } \dots \tau _ { 2 } \colon \colon \dots \dots \alpha _ { 2 } \colon \colon \dots$$
Therefore the set I p is invariant for the dynamics.
Now we show that I p is contained in the region in which both the orbits are elliptic, if EL 2 is sufficiently small. The set I p as defined in (14) is the union of the set of values of d L and d E in A which solve
$$( 1 5 ) & & e _ { 1 } ^ { 2 } e _ { 2 } ^ { 2 } = ( 1 + 2 E _ { 1 } L _ { 1 } ^ { 2 } ) ( 1 + 2 E _ { 2 } L _ { 2 } ^ { 2 } ) \geq ( 1 + L _ { 2 } ^ { 2 } E _ { 1 } + L _ { 1 } ^ { 2 } E _ { 2 } ) ^ { 2 } \\ & & \dots \cdots \cdots \cdots$$
provided that
$$& \quad 1 + L _ { 2 } ^ { 2 } E _ { 1 } + L _ { 1 } ^ { 2 } E _ { 2 } \leq 0, \\ & \quad \dots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \cdots \, \end{cases}$$
and the set of values which solve
$$& 1 + L _ { 2 } ^ { 2 } E _ { 1 } + L _ { 1 } ^ { 2 } E _ { 2 } \geq 0, \\ & \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \end{cases}$$
Note that this last equation identifies the region I p / 2 , i.e. the region of intersecting orbits with perpendicular semi-major axis, and it is equivalent to
$$& ( 1 8 ) & \delta E ( \mu _ { 1 } \mu _ { 2 } \delta L + ( \mu _ { 1 } - \mu _ { 2 } ) ) \leq 1 + 2 E L ^ { 2 } - ( \mu _ { 1 } - \mu _ { 2 } ) E \delta L, \\ & \tau _ { - } \, \nu \, \varepsilon \, \vdots \, \dots \, \dots \, \dots \, \dots \,.$$
Eq. (15) is equivalent to
$$& ( 1 9 ) \quad \delta E ^ { 2 } ( \mu _ { 1 } L _ { 1 } ^ { 2 } + \mu _ { 2 } L _ { 2 } ^ { 2 } ) ^ { 2 } - 2 \delta E ( L _ { 1 } ^ { 2 } - L _ { 2 } ^ { 2 } ) ( 1 + E ( \mu _ { 1 } L _ { 1 } ^ { 2 } + \mu _ { 2 } L _ { 2 } ^ { 2 } ) ) + E ^ { 2 } ( L _ { 1 } ^ { 2 } - L _ { 2 } ^ { 2 } ) ^ { 2 } \leq 0 \\ & \dots \colon \colon \dots \colon \dots \colon \colon \colon$$
which can be solved if
$$( 2 0 ) & & ( L _ { 1 } ^ { 2 } - L _ { 2 } ^ { 2 } ) ^ { 2 } ( 1 + 2 E ( \mu _ { 1 } L _ { 1 } ^ { 2 } + \mu _ { 2 } L _ { 2 } ^ { 2 } ) ) \geq 0 \\ \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau - \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau + \tau } ) \times 0.$$
$$& \delta L ^ { 2 } \leq \frac { 1 - 2 | E | L ^ { 2 } } { 2 | E | \mu _ { 1 } \mu _ { 2 } } \\ & \text{The boundary of the region identified by the inequality (19) is $\varepsilon$}$$
If | L 1 | glyph[negationslash] = | L 2 | this condition is equivalent to 1 + 2 E L ( 2 + m m d 1 2 L 2 ) ≥ 0 then the set I p is non void if and only if 1 + 2 EL 2 ≥ 0, and it is bounded by the condition
The boundary of the region identified by the inequality (19) is given by the functions
$$( 2 2 ) \quad \delta E = \frac { L _ { 1 } ^ { 2 } - L _ { 2 } ^ { 2 } } { ( \mu _ { 1 } L _ { 1 } ^ { 2 } + \mu _ { 2 } L _ { 2 } ^ { 2 } ) ^ { 2 } } \left ( 1 + E ( \mu _ { 1 } L _ { 1 } ^ { 2 } + \mu _ { 2 } L _ { 2 } ^ { 2 } ) \pm \sqrt { 1 + 2 E ( \mu _ { 1 } L _ { 1 } ^ { 2 } + \mu _ { 2 } L _ { 2 } ^ { 2 } ) } \right )$$
/C8/CB/CU/D6/CP/CV
/D6/CT/D4/D0/CP
/CT/D1/CT/D2/D8/D7
/C8/CB/CU/D6/CP/CV
/D6/CT/D4/D0/CP
/CT/D1/CT/D2/D8/D7
-
-
FIGURE 2. m 1 = 0 45: the set . I p for EL 2 = -0 41 and . EL 2 = -0 445. . Over the line E 1 = 0, the orbit 1 is hyperbolic.
which have the sign of L 2 1 -L 2 2 .
In figure 2 we show the region I p , identified by the inequality (18) (region I p / 2 ⊂ I p ) and the inequality (19) (region I p \ I p / 2 ). In figure 2 (a), the value of EL 2 is -0 41, and the . region I p intersects the region E 1 ≥ 0 (i.e. d E > | E | / m 2). Then, after a collision, one of the outgoing orbits can become hyperbolic. In figure 2 (b), the value of EL 2 is smaller and the invariant region I p is completely contained in the region d E ∈ -| ( E | / m 2 , | E | / m 1 ) in which both the orbits are elliptic. Then, whatever are the details of the collisions, the two orbits remain elliptic, whit ei ≤ ci < 1 for some constants c 1 , c 2, and | L 1 | , | L 2 | are bounded via (21) and (6).
Now we will show that the behavior of the system is driven by EL 2 : there exists a critical value which separates the two cases. Let us define
$$& ( 2 3 ) & & \bar { d } ( \delta L, \delta E ) = \frac { L _ { 1 } ^ { 2 } } { 1 + e _ { 1 } } - \frac { L _ { 2 } ^ { 2 } } { 1 - e _ { 2 } }. \\ & \text{This quantity is the distance from the neriansis of the orbits 1 and the a}$$
This quantity is the distance from the periapsis of the orbits 1 and the apoapsis of the orbit 2, in the case of w = p . The value of ¯ is 0 on the boundary of d I p in the first quadrant, out of I p / 2 (see fig. 2). Then the critical value of E e L is such that
$$E _ { 1 } = 0, \, e _ { 1 } = 1, \, \bar { d } ( \delta E, \delta L ) = 0, \, \frac { \partial \bar { d } } { \partial \delta L } ( \delta E, \delta L ) = 0$$
(the gradient of ¯ d ( d E , d L ) is vertical in the point of tangency to the line E 1 = 0). By deriving ¯ with respect to d d L we obtain
$$& ( 2 4 ) \quad \partial _ { \delta _ { L } } \bar { d } = 2 \mu _ { 2 } \frac { L _ { 1 } } { 1 + e _ { 1 } } + 2 \mu _ { 1 } \frac { L _ { 2 } } { 1 - e _ { 2 } } - 2 \mu _ { 2 } E _ { 1 } L _ { 1 } \frac { L _ { 1 } ^ { 2 } } { e _ { 1 } ( 1 + e _ { 1 } ) ^ { 2 } } + 2 \mu _ { 1 } E _ { 2 } L _ { 2 } \frac { L _ { 2 } ^ { 2 } } { e _ { 2 } ( 1 - e _ { 2 } ) ^ { 2 } } \\ & \text{black} \quad \underset { 1 } { \lambda } \ \in \ \mathcal { N } \ \mathcal { N } ^ { 2 } - 1 \ \mathcal { N } ^ { 2 } \ \text{black} \quad \underset { 1 } { \lambda }$$
which, using -2 EiL i 2 = 1 -e 2 i , becomes (25)
$$\partial _ { \delta L } \bar { d } = 2 \mu _ { 2 } \frac { L _ { 1 } } { 1 + e _ { 1 } } + 2 \mu _ { 1 } \frac { L _ { 2 } } { 1 - e _ { 2 } } \mu _ { 2 } L _ { 1 } \frac { 1 - e _ { 1 } ^ { 2 } } { e _ { 1 } ( 1 + e _ { 1 } ) ^ { 2 } } - \mu _ { 1 } L _ { 2 } \frac { 1 - e _ { 2 } ^ { 2 } } { e _ { 2 } ( 1 - e _ { 2 } ) ^ { 2 } } = \frac { \mu _ { 2 } } { e _ { 1 } } L _ { 1 } - \frac { \mu _ { 1 } } { e _ { 2 } } L _ { 2 }. \\ \intertext { T h a c o n d i t i o n } \Delta _ { \bar { A } } - \Delta \, \partial \, \text{th a d i f i t i o n } \Delta _ { \bar { I } } \, \text{ in as a l u m o t o s o l u m o t o } \Delta _ { \bar { I } } \, \text{ in }$$
The condition ¶d L d ¯ = 0 and the definition of L in eq. (6) allow us to calculate L 1, L 2 in terms of e 1, e 2:
$$\begin{array} { c } L _ { 1 } = \mu _ { 1 } e _ { 1 } L / ( \mu _ { 1 } ^ { 2 } e _ { 1 } + \mu _ { 2 } ^ { 2 } e _ { 2 } ) \\ L _ { 2 } = \mu _ { 2 } e _ { 2 } L / ( \mu _ { 1 } ^ { 2 } e _ { 1 } + \mu _ { 2 } ^ { 2 } e _ { 2 } ) \end{array}$$
Using these expressions in ¯ d = 0 with e 1 = 1 we obtain the following equation for e 2:
$$2 \mu _ { 2 } ^ { 2 } e _ { 2 } ^ { 2 } = \mu _ { 1 } ^ { 2 } ( 1 - e _ { 2 } ),$$
/C8/CB/CU/D6/CP/CV
/D6/CT/D4/D0/CP
/CT/D1/CT/D2/D8/D7
0 3
.
L
2
/B4/CQ/B5
EL
δE
=
0 52
.
-
E
2
δL
= 0
-
FIGURE 3. m 1 = 0 45: . the values of ¯ in grayscale, in the case d EL 2 = -0 445 and . EL 2 = -0 52. .
which has only one solution in ( 0 1 , given by eq. , ) (10). Substituting this value in the expression of L 2, we obtain the critical values of EL 2 as in (11), imposing 1 + E L 2 2 2 = e 2 2 , with E 2 = E / m 1 and e 1 = 1:
$$E L ^ { 2 } = - ( 1 - e _ { 2 } ^ { 2 } ) ( \mu _ { 1 } ^ { 2 } + \mu _ { 2 } ^ { 2 } e _ { 2 } ) ^ { 2 } / ( 2 \mu _ { 2 } e _ { 2 } ^ { 2 } ). \\ \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \end{array}$$
Note that we can find a similar condition for which the region I p is tangent to the line E 2 = 0, but in the case m 1 ≤ m 2 this second critical value of EL 2 is greater then the previous, then it can be ignored. glyph[square]
## 4. THE INVARIANT REGION FOR TWO BODIES IN R 2
In this section, the case of two bodies and the case of point particles which interact by means of a short range potential are analyzed.
Theorem 4.1. We consider two circular bodies in R 2 of radii R 1 and R , interacting by 2 means of elastic or inelastic collisions. If EL 2 < s m ( 1 , m 2 ) , with s as defined in eq. (9) , and D = R 1 + R 2 is sufficiently small, then the two bodies remain on elliptic orbits.
Proof. We show that if two orbits have points whose distance is less then or equal to D , and EL 2 < s and D is sufficiently small, then the orbits are elliptic. We remark that the condition on the distance is preserved by the dynamics.
Fixed L 1 , L 2 , E 1 , E 2, assuming that the two bodies can collide, we need to distinguish two situations: it exists dw such that the orbits intersect, and then ( d L , d E ) ∈ I p , or the orbits do not intersect for any dw . One of them, named orbit 2, is contained in the other. But min dw min J 1 , J 2 | x 1 -x 2 | ≤ D , and the minimum is reached for dw = p , then 0 < ¯ d ( d L , d E ) ≤ D (note that if the orbit of the particle 1 is contained in the orbit of the particle 2, we have to define ¯ as in (23), but exchanging the indexes 1 d ↔ 2).
If EL 2 < s , the distance between the level set ¯ d = 0 (the boundary of I p ) and the critical lines E 1 = 0, E 2 = 0, is strictly positive. Then, if D is sufficiently small, the set ¯ d ≤ D does not intersect the region in which E 1 ≥ 0 or E 2 ≥ 0, and this proves the theorem. glyph[square]
In figure 3 we show the level sets of ¯ . d In figure 3 (a), EL 2 = -0 445 and the invariant . set I p and the values of ¯ in the complementary region are shown: the black color corred sponds to ¯ d = 0, the white color corresponds to ¯ d ≥ L 2 / 10, while the grays correspond to the values ¯ d ∈ ( 0 , L 2 / 10 . The critical value of ) D is approximately 0 034 . L 2 . Figure 3 (b) is relative to the case EL 2 = -0 52, in which the set . I p is void. Nevertheless, the set ¯ d ≤ D is invariant and is contained in the region of elliptic orbits if D is sufficiently small. In the graphics, the black color corresponds to ¯ d = 0 2 . L 2 , the white to ¯ d ≥ 0 3 . L 2 . In this case, the critical value for D is approximately 0 25 . L 2 .
We are not able to give a simple expression for the critical values of EL 2 and D , but for the case m 1 = m 2 = 1 / 2, in which the particles have the same mass.
/C8/CB/CU/D6/CP/CV
/D6/CT/D4/D0/CP
/CT/D1/CT/D2/D8/D7
E
1
0 2
.
= 0
L
2
2
-
3 0
.
FIGURE 4. m 1 = m 2 = 0 5: the critical value of . EL 2 in function of D L / 2 .
Theorem 4.2. If m 1 = m 2 = 1 / 2 , the conditions on EL 2 and D are
$$E L ^ { 2 } < - \frac { ( 1 - e ^ { 2 } ) ( 1 + e ) ^ { 2 } } { 1 6 e ^ { 2 } } \ w h e r e \, e = \gamma - \sqrt { \gamma ^ { 2 } - \gamma + 1 } \, \text{ with } \gamma = 2 L ^ { 2 } / D > 1$$
Proof. We proceed as in the proof of Theorem 3.1. Assuming m 1 = m 2 = 1 / 2 and EL 2 < -27 64, the critical condition / ¶d L d ¯ ( d E , d L ) = 0 allows us to obtain the value of L 1 and L 2 as in eq. (26), which becomes
$$\begin{array} { c } L _ { 1 } = 2 e _ { 1 } L / ( e _ { 1 } + e _ { 2 } ) \\ L _ { 2 } = 2 e _ { 2 } L / ( e _ { 1 } + e _ { 2 } ) \end{array}$$
Using these values in ¯ d = D with e 1 = 1, we obtain the following equation for e 2
$$2 L ^ { 2 } ( 1 - 2 e _ { 2 } ) = D ( 1 - e _ { 2 } ^ { 2 } ),$$
which is solved in ( 0 1 , ) by
$$e _ { 2 } = \gamma - \sqrt { \gamma ^ { 2 } - \gamma + 1 } \quad \text{where} \quad \gamma = 2 L ^ { 2 } / D \quad \text{with} \quad D < 2 L ^ { 2 }. \\ \int _ { \dots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \\ \int _ { \dots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cd� \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \colon \quad \end{array}$$
The corresponding value of EL 2 is given by EL 2 = -( 1 -e 2 2 )( 1 + e 2 ) 2 / ( 16 e 2 2 ) . glyph[square]
In figure 4 we plot these values in function of D L / 2 . Let us recall that for EL 2 < -0 5 . the two orbits do not intersect.
The last extension we consider in the two dimensional case is that of two point particles interacting by means of a force of symmetric potential energy V ( | x 1 -x 2 | ) , with a compact support.
Theorem 4.3. Let us consider two point particles interacting by means of a force of potential energy V = V ( | x 1 -x 2 | ) , such that V ( r ) = 0 if r ≥ D for some D > 0 and V ( r ) ≥-U with U ≥ 0 .
If EL 2 < s m ( 1 , m 2 ) , with s as in eq. (9) , and and D and U are sufficiently small, then the particles remains on bounded orbits.
Proof. The specific angular momentum L = m 1 L 1 + m 2 L 2 is conserved at any time also in this case, because the potential energy is symmetric. If | x 1 -x 2 | ≥ D the energy of the interaction is zero and the specific energy of the system is exactly
$$E = \mu _ { 1 } \left ( \frac { \mathbf v _ { 1 } ^ { 2 } } { 2 } - \frac { 1 } { | \mathbf x _ { 1 } | } \right ) + \mu _ { 2 } \left ( \frac { \mathbf v _ { 2 } ^ { 2 } } { 2 } - \frac { 1 } { | \mathbf x _ { 2 } | } \right ) \\ \text{he twoarticles leave the region of the interaction. we c}.$$
Therefore, when the two particles leave the region of the interaction, we can apply Theorem 4.1, concluding that the particles remain on elliptic orbits until the next interaction.
To achieve the proof, we have to discuss the motion of the particles during the interaction, i.e. when their distance is less than D . If | x 1 -x 2 | < D , we can consider the two 'osculating' Keplerian orbits, i.e. the Keplerian orbits which correspond to the two couple position-velocity ( x 1 , v 1 ) , and ( x 2 , v 2 ) . The specific energies of this two orbits are
$$E _ { 1 } = \mathbf v _ { 1 } ^ { 2 } / 2 - 1 / | \mathbf x _ { 1 } |, \quad E _ { 2 } = \mathbf v _ { 2 } ^ { 2 } / 2 - 1 / | \mathbf x _ { 2 } |. \\ \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \, \dots \end{cases}$$
These quantities are not the specific energies of the two particles (because the contribution of the interaction is not zero), but verify
$$\tilde { E } = \mu _ { 1 } E _ { 1 } + \mu _ { 2 } E _ { 2 } = E - V ( | \mathbf x _ { 1 } - \mathbf x _ { 2 } | ) / ( m _ { 1 } + m _ { 2 } ) \leq E + U / ( m _ { 1 } + m _ { 2 } ) \\ \intertext { a t i n d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d o d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d e d s e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n s e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i in s e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i s e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i m d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i in d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i s t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e t i n d e s t i n d e t i n d e t i n d e t i n d e t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t e n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s e t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s s e t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i in d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n d e s t i n$$
by the conservation of the total energy. If EL 2 < s and U is sufficiently small, we have also that ˜ EL 2 < s , then we can apply Theorem 4.1, using ˜ E instead of E . Therefore, if D is sufficiently small, as in the hypothesis of Theorem 4.1, the two osculating orbits are elliptic, with bounded value of L 1, L 2, and ei ≤ ci < 1, as follows from the compactness of the set I p ∪{ ¯ d ( d L , d E ) ≤ D } ⊂ ( -L / m 1 , L / m 2 ) × ( E / m 1 , -E / m 2 ) . glyph[square]
## 5. THE INVARIANT REGION FOR TWO BODIES IN R 3
Here we discuss the three dimensional case. We indicate with L i = x i ∧ v i the vector which express the specific angular momentum of the particle i in the positi'on x i whit velocity v i , and with L the specific angular momentum of the whole system L = m 1 L 1 + m 2 L 2, which is a conserved vector. In this case, they hold the analogous of Theorems 3.1, 4.1, 4.3, where the role of L is played by | L | . We summarize these results in the following theorem.
Theorem 5.1. (1) The invariant region for the orbital dynamics in R 3 , defined as in section 3, is given by
$$E | \mathbf L | ^ { 2 } < \sigma \\ e i n \, e a \, \mathbf L \mathbf 0$$
with s = s m ( 1 , m 2 ) defined as in eq. (9) .
- (2) The invariant region for the collisional dynamics of two hard spheres in R 3 , of radii R 1 and R , with D 2 = R 1 + R , is given by E 2 | L | 2 < s with D sufficiently small.
- (3) The invariant region for two point particles interacting by means a potential energy V as in Theorem 4.3, is given by E | L | 2 < s with D and U sufficiently small.
Proof. We prove the theorem staring from the last case, which includes the others as particular ones. As in the proof of Theorem 4.3, we define
$$E _ { 1 } = 1 / 2 \mathbf v _ { 1 } ^ { 2 } - 1 / | \mathbf x _ { 1 } |, \ \ E _ { 2 } = 1 / 2 \mathbf v _ { 2 } ^ { 2 } - 1 / | \mathbf x _ { 2 } |, \ \tilde { E } = \mu _ { 1 } E _ { 1 } + \mu _ { 2 } E _ { 2 }$$
and we note that
$$\tilde { E } & = E \quad \text{ if } \quad | \mathbf x _ { 1 } - \mathbf x _ { 2 } | \geq D \\ \tilde { E } & \leq E + U / ( m _ { 1 } + m _ { 2 } ) \quad \text{ if } \quad | \mathbf x _ { 1 } - \mathbf x _ { 2 } | < D. \\ \text{ while } \tilde { T } - | \mathbf x _ { 1 } | \text{ that modulo of the amount of the output of}$$
Weindicate whit ˜ Li = | L i | the modulus of the angular momentum of the orbit of the particle i , and we define
$$\tilde { L } = \mu _ { 1 } \tilde { L } _ { 1 } + \mu _ { 2 } \tilde { L } _ { 2 }$$
which verifies
$$| \mathbf L | = | \mu _ { 1 } \mathbf L _ { 1 } + \mu _ { 2 } \mathbf L _ { 2 } | \leq \tilde { L } \text{ and } \, E \tilde { L } ^ { 2 } \leq E | \mathbf L | ^ { 2 }$$
( E is negative). Moreover
$$\tilde { E } \tilde { L } ^ { 2 } \leq \left ( E + \frac { U } { m _ { 1 } + m _ { 2 } } \right ) | \tilde { L } ^ { 2 } |$$
then, if E L 2 < s as in the hypothesis, for U
$$\tilde { E } \tilde { L } ^ { 2 } < \sigma$$
| | sufficiently small, it also holds
Let us indicate whit oi the Keplerian orbit identified by the position x i and the velocity v i ; its energy is Ei and its angular momentum is L i . If the particles can interact, o 1 and o 2 have points at distance less then D . We consider the orbits ˜1 o and ˜2 o we obtain rigid rotating in R 3 , around the center, o 1 and o 2, in such that ˜1 o and ˜2 o are in the same plane, and the periapsides of ˜1 o and ˜2 o are in opposition. The energy and the eccentricity of the orbits ˜ oi are the same of the orbits oi , while we can identify the angular momentum of ˜ oi with the positive scalar quantity ˜ Li . This two planar orbits intersect or have points at distance less that D . In the first case, we can apply Theorem 3.1 using (28), and we can conclude that the ˜1, o ˜2, and then o o 1 and o 2 are elliptic. In the second case, we can apply Theorem 4.1, and we can again conclude that, if D is sufficiently small, the orbits are elliptic.
The case of spherical bodies can be considered as a particular case, in which V =+ ¥ , if | x 1 -x 2 | ≤ D . Now U = 0 and we have only to require a sufficiently small value of D . Finally, the case of the orbital dynamics can be considered as the particular case in which D = 0. The hypothesis E | L | 2 < s is then sufficient to achieve the thesis. glyph[square]
## Remarks.
- i. The values of D in Theorem 4.1 can be very large with respect to the scale L 2 of the semi-axis of the orbits, but, if it is so, the system is in the region EL 2 < -1 / 2, in which the orbits do not intersect. Therefore, the system is made of two particle which can interact only by means of grazing collisions.
- ii. It can be interesting to analyze the case of Theorem 4.3 when V is unbounded from below. The particles can leave the system only if their distance remains less then D , and, in this case, we can expect that the center of mass of the two particles moves on an approximately elliptic orbit. On the other hand, there are no a priori bounds on the kinetic energy or on the position of this center of mass.
- iii. It can be also interesting to analyze the case of n > 2 particles, with positive radii. In particular it can be expected that there exist stable ringlets in which the collisions are grazing.
## Acknowledgments
The authors thank L. Biasco, E. Caglioti, G. F. Gronchi, P. Negrini, for usefull suggestions on this subject.
## REFERENCES
- [1] Braich A.: Systems of colliding bodies in a gravitational field. I - Numerical simulation of the standard model , Astron. Astroph. 54 no. 3, pp. 895-907 (1977).
- [2] Braich A., H´ enon M.: Systems of colliding bodies in a gravitational field: II. Effect of transversal viscosity , Astron. Astroph. 59 no. 1, pp. 1-7 (1977).
- [3] Djerassi S.: Collision with friction; Part A: Newton's hypothesis , Multibody Syst. Dyn. 21 , pp. 37-54 (2009) doi:10.1007/s11044-008-9126-2.
- [4] Gronchi G.F.: On the stationary points of the squared distance between two ellipses with a common focus , SIAM Jour. Sci. Comp. 24 1, pp. 61-80 (2002).
- [5] Hatzes A.P., Briges F.G., Lin D.N.C.: Collisional properties of ice spheres at low impact velocities , Royal Astr. Soc. 231 , pp. 1091-1115 (1988).
- [6] Hertzsch J.-M., Scholl H., Spahn F., Katzorke I.: Simulation of collisions in planetary rings , Astron. Astrophys. 320 , pp. 319-324 (1997).
- [7] Kawai T., Shida K.: An inelastic collision model for the evolution of 'Planetary Rings' , Jour. Phys. Soc. Jap. 59 no. 1, pp. 381-388 (1990).
- [8] Laskar J.: On the spacing of planetary systems , Phy. Rev. Let. 84 no. 15, pp. 3240-3243 (2000) .
- [9] Laskar J., Robutel P.: Stability of the planetary three-body problem I , Cel. Mech. Dyn. Astr. 62 , pp. 193-217 (1995).
- [10] Marsden J.E., Ross S.D.: New method in celestial mechanics and mission design , Bull. Am. Math. Soc. 43 1, pp. 43-73 (2005).
- [11] Papaloizou J.C.B.: Planetary system formation , Science 321 pp. 777-778 (2008).
- [12] Poincar´ e H.: Les m´ ethodes nouvelles de la m´ ecanique c´ eleste (1889), tome III, chapitre XXII, Lib. Sci. Tech. A. Blanchard, Paris (1987)
- 12 AN INVARIANT REGION FOR THE COLLISIONAL DYNAMICS OF TWO BODIES ON KEPLERIAN ORBITS
- [13] Poincar´ e H., Vergne H.: Lecons sur les hypotheses cosmogoniques , A. Hemann et fils, Paris (1911). | null | [
"Dario Benedetto",
"Flavia Lenti"
] | 2014-08-02T07:08:11+00:00 | 2014-08-02T07:08:11+00:00 | [
"math.DS",
"70F15, 37N05"
] | An invariant region for the collisional dynamics of two bodies on Keplerian orbits | We study the dynamics of two bodies moving on elliptic Keplerian orbits
around a fixed center of attraction and interacting only by means of elastic or
inelastic collisions. We show that there exists a bounded invariant region: for
suitable values of the total energy and the total angular momentum (explicitly
computable) the orbits of the bodies remain elliptic, whatever are the number
and the details of the collisions. We show that there exists a bounded
invariant region even in the case of two bodies interacting by short range
potential. |
1408.0351v2 | 3
## Primordial Magnetic Helicity Constraints from WMAP Nine-Year Data
Tina Kahniashvili, 1, 2, 3, ∗ Yurii Maravin, 4, 3, † George Lavrelashvili, 5,6, ‡ and Arthur Kosowsky 7, 8, §
1 McWilliams Center for Cosmology and Department of Physics,
Carnegie Mellon University, 5000 Forbes Ave, Pittsburgh, PA 15213, USA
2 Department of Physics, Laurentian University, Ramsey Lake Road, Sudbury, ON P3E 2C,Canada
Abastumani Astrophysical Observatory, Ilia State University, 3-5 Cholokashvili St., 0194 Tbilisi, Georgia
4
Department of Physics, Kansas State University, 116 Cardwell Hall, Manhattan, KS 66506, USA
5 Department of Theoretical Physics, A. Razmadze Mathematical Institute,
I. Javakhishvili Tbilisi State University, 0177 Tbilisi, Georgia
6 Max Planck Institute for Gravitational Physics,
Albert Einstein Institute, 14476 Potsdam, Germany
7 Department of Physics and Astronomy, University of Pittsburgh, 3941 O'Hara Street, Pittsburgh, PA 15260 USA 8 Pittsburgh Particle Physics, Astrophysics, and Cosmology Center (Pitt-PACC), Pittsburgh, PA 15260 USA
(Dated: October 17, 2014)
If a primordial magnetic field in the universe has non-zero helicity, the violation of parity symmetry results in non-zero correlations between cosmic microwave background temperature and B-mode polarization. In this paper we derive approximations to the relevant microwave background power spectra arising from a helical magnetic field. Using the cross-power spectrum between temperature and B-mode polarization from the WMAP nine-year data, we set a 95% confidence level upper limit on the helicity amplitude to be 10 nG 2 Gpc for helicity spectral index n H = -1 9, for a cosmological . magnetic field with effective field strength of 3 nG and a power-law index n B = -2 9 near the scale-. invariant value. Future microwave background polarization maps with greater sensitivity will be able to detect the helicity of an inflationary magnetic field well below the maximum value allowed by microwave background constraints on the magnetic field amplitude.
PACS numbers: 98.70.Vc, 98.80.-k, 98-62.En
## I. INTRODUCTION
A challenging question of modern astrophysics is the origin of observed magnetic fields in galaxies and clusters [1]. Generally, fields observed today began as small seed fields and then were amplified via either adiabatic compression or through turbulent plasma dynamics. One mechanism for seed field generation is generic plasma instabilities and vorticity perturbations [2]. In this causal model, the correlation length of the resulting fields is limited by the horizon, which generically corresponds to comoving galaxy scales. A second possibility is larger seed fields generated during inflation spanning a wide range of correlation lengths up to the horizon today, and amplified through the process of cosmological structure growth [3, 4].
The evolution and amplification of a primordial seed field is strongly influenced by the helicity, or local handedness, of the seed field. Magnetic helicity is a manifestation of parity symmetry violation. While the level of parity violation observed in fundamental physical interactions is small, parity violation is widespread in various astrophysical systems with significant magnetic dynamics, such as one-sided jets from active galactic nuclei and helical magnetic fields in the solar magnetosphere [5]. A seed field with helicity is restructured at large scales by plasma turbulence: the decay of the magnetic field leads to an increase in the relative magnetic helicity until the helicity saturates at the maximum value allowed by the realizability condition for the field strength. The magnetic field correlation length of a helical magnetic field will also increase more quickly than for a non-helical field due to the inverse cascade mechanism. Magnetic fields with maximal helicity are a generic outcome of any extended period of turbulence [6].
Helical magnetic fields can be generated during the electroweak phase transition or during inflation [7-18]. Such a helical cosmological magnetic field might be the source of magnetic helicity needed in galactic dynamo amplification models [19]. Thus testing the helicity of any primordial magnetic field is important for understanding the origin of observed astrophysical magnetic fields [1]. Magnetic helicity in strong local magnetic fields like astrophysical jets can
∗ Electronic address: [email protected]
† Electronic address: [email protected]
‡ Electronic address: [email protected]
§ Electronic address: [email protected]
be deduced from the polarization of synchrotron radiation [20, 21]. For cosmological magnetic helicity the detection issue is more difficult, because the field strengths are much lower and the observational effects more subtle.
The most direct probe of any cosmological magnetic fields is their effect on the cosmic microwave background radiation, and particularly its polarization. The microwave background linear polarization is conventionally decomposed into E-mode (parity-even) and B-mode (parity-odd) components [22, 23]. A non-helical magnetic field contributes to all of the parity-even power spectra, those correlating E with itself and B with itself, in addition to E with the microwave temperature T and the temperature with itself. These contributions were explicitly calculated in Ref. [24], and have been used to constrain the amplitude of a primordial magnetic field [25-32]. However, if a parity-violating helical magnetic field component is present, then it will contribute to the remaining parity-odd power spectra, namely EB or TB [33-36], which are identically zero for magnetic fields with zero helicity. Note that Faraday rotation by magnetic fields [37] imprints itself on the power spectrum and frequency spectrum of microwave background polarization, but is insensitive to helicity for a given magnetic field power spectrum [38-41].
Helical magnetic fields are perhaps the most natural parity-violating source of TB or EB correlations in the microwave background polarization [42-49], but other more speculative parity-violating sources can also induce them. These include a Chern-Simons coupling of photons to another field [42-44, 50-52], a homogeneous magnetic field [53-56], Lorentz symmetry breaking [57-70], or non-trivial cosmological topology [71-74]. If some non-zero TB or EB correlation is detected, the corresponding angular power spectrum must be measured sufficiently well to distinguish between these possibilities.
In this paper we obtain upper limits on the helicity of a primordial magnetic field, using the nine-year WMAP constraints on any cross correlation between microwave background temperature and B-polarization [75]. Current polarization data is consistent with zero cosmological TB signal, as expected in the standard cosmological model. We compute the theoretical estimates of cross correlation given in Ref. [35] and compare with the measured upper limits [75-77]. Since we obtain only upper limits, we assume that magnetic helicity is the only possible parity-violating source present, which gives the most conservative helicity upper limits. For simplicity of calculation, we consider only the vector (vorticity) perturbations sourced by the magnetic field and neglect the tensor (gravitational wave) perturbations. This is a good approximation for angular multipoles l > 50 [34], and for this reason we use measured C TB l constraints only for l > 50; the neglected large angular scales contain little total statistical weight in our constraints.
The outline of the paper is as follows: in Sec. II we review the main characteristics of a helical magnetic field and derive the vorticity perturbations. Section III gives the expression for C TB l due to these vorticity perturbations, and these are compared with the WMAP 9-year upper limits in Sec. IV. Implications and future experimental prospects are discussed in Sec. V. We employ natural units with /planckover2pi1 = c = 1 and gaussian units for electromagnetic quantities.
## II. PROPERTIES OF A COSMOLOGICAL MAGNETIC FIELD
We assume that a cosmological magnetic field was generated during or prior to the radiation-dominated epoch, with the energy density of the field being a first-order perturbation to the standard Friedmann-Lemaˆ ıtre-Robertson-Walker homogeneous cosmological model. We also assume that primordial plasma is a perfect conductor and thus the spatial and temporal dependence of the field separates: B x ( , t ) = B x ( ) /a t ( ) 2 with a t ( ) the cosmological scale factor. The mean helicity density of the magnetic field is given by
$$\mathcal { H } _ { B } = \frac { 1 } { V } \int _ { V } d x \mathbf A ( x ) \cdot \mathbf B ( x ) = \frac { 1 } { V } \int _ { V } d x \mathbf A ( x ) \cdot \nabla \times \mathbf A ( x ), \quad \ \ ( 1 ) \\ \cdot \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots$$
with A the vector potential, in the limit that the integral is over an infinite volume. An integral over a finite but large volume will approximate this helicity density. In general, magnetic helicity is a gauge-dependent quantity, because the vector potential A can be redefined by adding a gradient to it. However, the magnetic helicity is gauge invariant for periodic systems without a net magnetic flux, as shown in Ref. [78]. We assume that our universe can be well approximated by a large box with periodic boundary conditions, provided the dimension of the box is large compared to the Hubble length today. In this case, the magnetic helicity is a well-defined quantity.
A Gaussian random magnetic field is described by the two-point correlation function in wavenumber space as
$$\langle B _ { m } ^ { * } ( k ) B _ { n } ( k ^ { \prime } ) \rangle = ( 2 \pi ) ^ { 3 } \delta ^ { ( 3 ) } ( k - k ^ { \prime } ) \left [ ( \delta _ { m n } - \hat { k } _ { m } \hat { k } _ { n } ) P _ { B } ( k ) + i \epsilon _ { m n } \hat { k } _ { l } P _ { H } ( k ) \right ]. \quad \quad ( 2 ) \\ \hat { t } _ { 1 } = 1 + \frac { 1 } { 2 } + \frac { 1 } { 2 } + \frac { 1 } { 2 } + \frac { 1 } { 2 } + \frac { 1 } { 2 } + \frac { 1 } { 2 } + \frac { 1 } { 2 } + \frac { 1 } { 2 } + \frac { 1 } { 2 } + \frac { 1 } { 2 } + \frac { 1 } { 2 } + \frac { 1 } { 2 } + \frac { 1 } { 2 } + \frac { 1 } { 2 } + \frac { 1 } { 2 } + \frac { 1 } { 2 } + \frac { 1 } { 2 } + \frac { 1 } { 2 } + \frac { 1 } { 2 } + \frac { 1 } { 2 } + \frac { 1 } { 2 } + \frac { 1 } { 2 } + \frac { 1 } { 2 } + 0 }$$
Here ˆ k m = k m /k are the unit wavenumber components, /epsilon1 mnl is the antisymmetric tensor, and δ (3) ( k -k ′ ) is the Dirac delta function. We use the Fourier transform convention B j ( k ) = ∫ d xe 3 i k x · B j ( x ). The symmetric power spectrum
P B ( k ) is related to the mean magnetic energy density by
$$\mathcal { E } _ { B } & = \frac { 1 } { ( 2 \pi ) ^ { 3 } } \int _ { 0 } ^ { k _ { D } } d k k ^ { 2 } P _ { B } ( k ), \\ - & \dots \dots \dots \dots \dots \dots \dots.$$
while the antisymmetric power spectrum P H ( k ) is related to the magnetic helicity density as
$$\mathcal { H } _ { B } & = \frac { 1 } { ( 2 \pi ) ^ { 3 } } \int _ { 0 } ^ { k _ { D } } d k k \frac { 1 } { 2 } P _ { H } ( k ), \\. \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot \quad \cdot$$
where k D is a characteristic damping scale for the magnetic field.
The total energy density and helicity of the magnetic field satisfy the realizability condition
$$\mathcal { H } _ { B } & \leq 2 \xi _ { M } \mathcal { E } _ { B },$$
where
$$\xi _ { M } \equiv \frac { 2 \pi \int _ { 0 } ^ { k _ { D } } d k k P _ { B } ( k ) } { \int _ { 0 } ^ { k _ { D } } d k k ^ { 2 } P _ { B } ( k ) } & & ( 6 ) \\ \text{ngth. The power spectra } P _ { B } ( k ) \text{ and } P _ { H } ( k ) \text{ are generically constrained by } P _ { B } ( k ) \geq \\ \text{me can not mean } D _ { L } \text{ } D _ { L } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{\ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text{ } \text$$
is the magnetic field correlation length. The power spectra P B ( k ) and P H ( k ) are generically constrained by P B ( k ) ≥ | P H ( k ) . | We assume that these power spectra are given by simple power laws, P B ( k ) = A k B n B and P H ( k ) = A H k n H . The constraint on their relative amplitudes implies n H > n B [79]; in addition, finiteness of the total magnetic field energy requires n B > -3 if the power law extends to arbitrarily small values of k . For physical transparency, instead of describing the magnetic field amplitude by the proportionality factors A B and A H , we will use the effective magnetic field amplitude B eff ≡ (8 π E B ) 1 / 2 [80] and the helicity density H B . Using these quantities is convenient because they do not depend on the power law indices n B and n H and are independent of any smoothing scale.
Often, cosmological magnetic fields are characterized by a smoothed value on some comoving length scale λ > λ D = 2 π/k D . Convolving with a Gaussian smoothing kernel, the smoothed magnetic field amplitude B λ is [24]
$$B _ { \lambda } ^ { \, 2 } \equiv | \langle \mathbf B ( x ) \cdot \mathbf B ( x ) \rangle | _ { \lambda } = \frac { 2 } { ( 2 \pi ) ^ { 2 } } A _ { B } \Gamma \left ( \frac { n _ { B } + 3 } { 2 } \right ) \lambda ^ { - n _ { B } - 3 }, \quad \ \ \lambda > \lambda _ { D }. \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
We also introduce a smoothed quantity H λ (the so-called helicity measure or current helicity [36]) related to the magnetic helicity having the same units as B λ and depending on the antisymmetric part of the magnetic field spectrum:
$$H _ { \lambda } ^ { 2 } \equiv \lambda | ( \mathbf B ( x ) \cdot [ \nabla \times \mathbf B ( x ) ] ) | _ { \lambda } = \frac { 2 } { ( 2 \pi ) ^ { 2 } } A _ { H } \Gamma \left ( \frac { n _ { H } + 4 } { 2 } \right ) \lambda ^ { - n _ { H } - 3 }, \quad \quad \lambda > \lambda _ { D }.$$
See Ref. [40] for a more detailed discussion. Then the transformation between the smoothed quantities B λ and H λ and the effective quantities B eff and H B is simply and
$$B _ { \text{eff} } & = \frac { B _ { \lambda } ( k _ { D } \lambda ) ^ { \frac { n _ { B } + 3 } { 2 } } } { \sqrt { \Gamma \left ( \frac { n _ { B } + 5 } { 2 } \right ) } } \,,$$
$$\mathcal { H } _ { B } = \frac { 1 } { 8 \pi } \frac { \lambda H _ { \lambda } ^ { 2 } ( k _ { D } \lambda ) ^ { n _ { H } + 2 } } { ( n _ { H } + 2 ) \Gamma \left ( \frac { n _ { H } + 4 } { 2 } \right ) } \,. \quad & ( 1 0 ) \\ \text{ic field cutoff scale } k _ { D } \text{ is determined by the Alfvén wave damping scale, } \lambda _ { D } \simeq v _ { A } L _ { S } \\ \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \eta \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \phi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \Xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xeta \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi\xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi \xi
\text{$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathgrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ } $\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfra$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathffrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$,\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathftrackfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{}$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathr$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathcfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ \mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak}$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\matfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{ }$\mathfrak{$$
We assume that the magnetic field cutoff scale k D is determined by the Alfv´ en wave damping scale, λ D /similarequal v A L S [81, 82], where v A is the Alfv´n velocity set by the total magnetic energy density [24]. e Since v A /lessmuch 1 the Alfv´n e damping scale will always be much smaller scale than the Silk damping scale (the thickness of the last scattering surface) for standard cosmological models. On the other hand, the CMB fluctuations are determined by the Silk damping scale, and presence of the magnetic field source at smaller scales will not significantly affect the resulting spectra.
## III. MICROWAVE BACKGROUND FLUCTUATIONS FROM A HELICAL MAGNETIC FIELD
A cosmological magnetic field induces Alfv´ en waves sourced by the Lorentz force in the cosmological plasma (see [24, 81-85]), which generically produce non-zero vorticity perturbations. In the case of a stochastic magnetic field the average Lorentz force 〈 L x ( ) 〉 = -〈 B × ∇ × [ B ] 〉 / (4 π ) vanishes, while the root-mean-square Lorentz force 〈 L x ( ) · L x ( ) 〉 1 / 2 is non-zero and acts as a source in the vector perturbation equation. If the magnetic field spectrum Eq. (2) has a helical part P H ( k ), then the Lorentz force two-point correlation function will have both symmetric and antisymmetric pieces. Both contribute to the symmetric piece of the vorticity perturbation spectrum, but only the antisymmetric piece of the Lorentz force, determined entirely by P H ( k ), will contribute to the antisymmetric part of the vorticity perturbation spectrum [35].
In the tight-coupling limit between photons and baryons, the fluid vorticity is sourced by the transverse and divergence-free piece of the Lorentz force. The fluid vorticity at last scattering then translates into temperature and polarization fluctuations in the microwave background radiation [24]. The microwave temperature and E-polarization components are both parity-symmetric, while the B-polarization component is parity-antisymmetric [86]. This implies that the cross-power spectra C TB l and C EB l from stochastic magnetic fields will be nonzero only if P H ( k ) is nonzero [22, 33-36]. In other words, the TB and EB power spectra provide a way to measure whether a primordial magnetic field has a helical component. (A constant magnetic field component also gives non-zero C TB l and C EB l through Faraday rotation [53, 54], but the two distinct contributions can be distinguished by their different power spectra, and by the frequency dependence of a Faraday rotation signal.)
Detailed computations of the various CMB angular power spectra induced by helical and nonhelical magnetic fields have been presented elsewhere [24, 35]. Here we focus on the TB power spectrum, because current data does not put a significant constraint on the much smaller EB power spectrum. For l > 50 where the TB power spectrum has significant power, we neglect tensor contributions, which are smaller. Here we derive an analytic approximation to the TB angular power spectrum, based on the second-order approximation technique from Ref. [87]; this approximate solution is simple and accurate enough for deriving upper limits on the helical magnetic field.
The multipoles of the temperature perturbation from a vector mode in Fourier space are given by
$$\frac { \Theta _ { l } ^ { ( \pm 1 ) } ( k, \eta _ { 0 } ) } { 2 l + 1 } \simeq \sqrt { \frac { l ( l + 1 ) } { 2 } } \ \Omega ^ { ( \pm 1 ) } ( k, \eta _ { \text{dec} } ) \frac { j _ { l } ( k \eta _ { 0 } ) } { k \eta _ { 0 } }, \quad l \geq 2$$
where Ω ± 1 ( k, η ) are the two helicity components of the gauge-invariant vorticity perturbations, constructed from the fluid velocity field and the vector component of the metric perturbations [35]. Here we have made the approximation η 0 -η dec /similarequal η 0 in Eq. (11). For vorticity perturbations sourced by the magnetic field, the l = 1 moment of temperature fluctuation is well approximated by the vorticity perturbation, Θ ( ± 1) ( k, η 0 ) /similarequal Ω ± 1 ( k, η dec ) [24]. For the B-mode polarization perturbation, we have [24]
$$\frac { B _ { l } ^ { ( \pm 1 ) } ( k, \eta _ { 0 } ) } { 2 l + 1 } \simeq \mp \frac { \sqrt { 6 } } { 2 } \sqrt { ( l - 1 ) ( l + 2 ) } \int _ { 0 } ^ { \eta _ { 0 } } d \eta \tilde { \tau } ( \eta ) e ^ { - \tau } P ^ { ( \pm 1 ) } ( k, \eta ) \frac { j _ { l } ( k \eta _ { 0 } - k \eta ) } { k \eta _ { 0 } - k \eta } \,, \\ \text{ere the polarization source is defined by } [ 8 6 ]$$
where the polarization source is defined by [86]
$$P ^ { ( \pm 1 ) } & = \frac { 1 } { 1 0 } \left [ \Theta _ { 2 } ^ { ( \pm 1 ) } - \sqrt { 6 } E _ { 2 } ^ { ( \pm 1 ) } \right ]. & & ( 1 3 ) \\ \text{ization quadruples satisfy the evolution equations}$$
The temperature and polarization quadrupoles satisfy the evolution equations
$$\text{pera and panarization quaaruposes sausiy the evolution equations} \\ \dot { \Theta } _ { 2 } ^ { ( \pm 1 ) } \, & = \, k [ \frac { \sqrt { 3 } } { 3 } \Theta _ { 1 } ^ { ( \pm 1 ) } - \frac { 2 \sqrt { 2 } } { 7 } \Theta _ { 3 } ^ { ( \pm 1 ) } ] + \dot { \tau } \left [ \Theta _ { 2 } ^ { ( \pm 1 ) } - P ^ { ( \pm 1 ) } \right ] \,, \\ \dot { E } _ { 2 } ^ { ( \pm 1 ) } \, & = \, k [ \mp \frac { 1 } { 2 } B _ { 2 } ^ { ( \pm 1 ) } - \frac { 2 \sqrt { 1 0 } } { \alpha _ { 1 } } E _ { 3 } ^ { ( \pm 1 ) } ] + \dot { \tau } \left [ E _ { 2 } ^ { ( \pm 1 ) } + \sqrt { 6 } P ^ { ( \pm 1 ) } \right ] \,.$$
Here the optical depth τ ( η ) = ∫ η 0 η dη τ ′ ˙ ( η ′ ) to photon scattering from conformal time η until today satisfies dτ/dη ≡ -˙ ( τ η ) = σ T n e ( η a η ) ( ), σ T is the Thomson scattering cross section, and n e ( η ) the comoving number density of free electrons. The vector mode of the CMB temperatureB polarization angular power spectrum is given by [86]
$$\varpi _ { 2 } \quad & = \kappa _ { \lfloor } \frac { 1 } { 3 } \varpi _ { 1 } \quad t - \frac { 1 } { 7 } \varpi _ { 3 } \quad \rfloor + \tau \, \lfloor \varpi _ { 2 } \quad t - \tau \quad t \rfloor \, \\ \dot { E } _ { 2 } ^ { ( \pm 1 ) } \, & = \, k [ \mp \frac { 1 } { 3 } B _ { 2 } ^ { ( \pm 1 ) } - \frac { 2 \sqrt { 1 0 } } { 2 1 } E _ { 3 } ^ { ( \pm 1 ) } ] + \dot { \tau } \left [ E _ { 2 } ^ { ( \pm 1 ) } + \sqrt { 6 } P ^ { ( \pm 1 ) } \right ] \. \\ \text{tical depth } \tau ( \eta ) = \int _ { \eta } ^ { \eta _ { 0 } } d \eta ^ { \prime } \dot { \tau } ( \eta ^ { \prime } ) \text{ to photon scattering from conformal time } \eta \, \text{ until today satisfies } d \tau / d \eta \equiv$$
$$C _ { l } ^ { T B \ ( V ) } = \frac { 2 } { \pi } \int d k k ^ { 2 } \left [ \frac { \Theta _ { l } ^ { ( - 1 ) * } ( k, \eta _ { 0 } ) } { 2 l + 1 } \frac { B _ { l } ^ { ( - 1 ) } ( k, \eta _ { 0 } ) } { 2 l + 1 } + \frac { \Theta _ { l } ^ { ( + 1 ) * } ( k, \eta _ { 0 } ) } { 2 l + 1 } \frac { B _ { l } ^ { ( + 1 ) } ( k, \eta _ { 0 } ) } { 2 l + 1 } \right ] \,. \quad \ \ ( 1 6 )$$
In the rest of this section, we approximate this power spectrum in a given cosmological model, for comparison with limits on this power spectrum from temperature and polarization sky maps.
The visibility function ˙ τe -τ is sharply peaked at the time of decoupling, so to determine the B -polarization signal, Eq. (12), we need to know the polarization source P ( ± 1) at the time of decoupling. Differentiating Eq. (13) with respect to conformal time and substituting Eqs. (14) and (15), at leading order we get
$$\lambda _ { \ } ^ { \ }. \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \cdot \, \\ \dot { P } ^ { ( \pm 1 ) } - \frac { 3 } { 1 0 } \dot { \tau } P ^ { ( \pm 1 ) } \simeq \frac { k \sqrt { 3 } } { 3 0 } \Theta _ { 1 } ^ { ( \pm 1 ) } \,,$$
where we have dropped terms containing Θ ( ± 1) 3 , E ( ± 1) 3 and B ( ± 1) 2 (see also Ref. [87]).
In our previous work [24, 35], we assumed the first term of Eq. (17) is small to obtain the approximate solution P ( ± 1) = √ 3 k Θ ( ± 1) 1 / 9 ˙. τ While usually valid, this approximation fails during recombination because k/τ ˙ varies rapidly: inserting P ( ± 1) = √ 3 k Θ ( ± 1) 1 / 9 ˙ τ into the integral of Eq. (12), the integrand becomes proportional to e -τ , and is not anymore peaked at the time of decoupling. Instead, we employ a more precise second-order approximate solution to the source equation, following the technique in Refs. [87, 88]. Details are given in the Appendix; the solution for the temperature-B polarization power spectrum is,
$$C _ { l } ^ { T B } \, \simeq \, - \frac { 3 \pi } { 1 4 } \ln \left ( \frac { 1 0 } { 3 } \right ) \sqrt { \frac { ( l + 2 )! } { ( l - 2 )! } } \frac { ( n _ { B } + 3 ) ( n _ { H } + 2 ) } { n _ { B } + n _ { H } + 2 } \frac { \eta _ { \text{dec} } ^ { 2 } } { \eta _ { 0 } ^ { 2 } } k _ { D } \Delta \eta _ { \text{dec} } \frac { 1 } { ( 1 + R _ { \text{dec} } ) ^ { 2 } } \\ & \times \frac { \mathcal { E } _ { B } \mathcal { H } _ { B } k _ { D } } { \rho _ { \gamma } ^ { 2 } 0 } \int _ { 0 } ^ { x _ { S } } d x x ^ { 4 } D _ { E } ( x ) \left [ 1 + \frac { n _ { H } - 1 } { n _ { B } + 3 } x ^ { n _ { B } + n _ { H } + 2 } \right ] j _ { l } ^ { 2 } ( x k _ { D } \eta _ { 0 } ) \\ \intertext { h the chance of visualises $r = k / k \, $ in the internal $ \text{We have defined a function which models the effect of $Silk}$}$$
with the change of variables x = k/k D in the integral. We have defined a function which models the effect of Silk damping for polarization [88],
$$D _ { E } ( x ) \equiv 0. 2 \left ( e ^ { - c _ { E } ( a _ { 1 } x k _ { D } \eta _ { 0 } ) ^ { b _ { E } } } + e ^ { - c _ { E } ( a _ { 2 } x k _ { D } \eta _ { 0 } ) ^ { b _ { E } } } \right ) & & ( 1 9 ) \\ \ n s t a n s { \, c _ { E } = 0. 2 7. \, b _ { E } = 2. 0. \, a, \, = 0. 0 0 1 1. \, and \, a _ { \sigma } = 0. 0 0 1 9. \, The \, \text{amltude of the annroximate} }$$
with the fitting constants c E = 0 27, . b E = 2 0, . a 1 = 0 0011, and . a 2 = 0 0019. . The amplitude of the approximate solution Eq. (18) differs from that in Ref. [24] by roughly a factor of two.
## IV. CONSTRAINTS FROM WMAP
We obtain constraints on primordial magnetic helicity by comparing the temperatureB -polarization cross correlation function in Eq. (18) with WMAP nine-year data. We assume a standard ΛCDM model. We take the Silk damping scale to be the thickness of the last scattering surface, L S /similarequal ∆ η dec , which is determined by the function D E ( x ), so k S = 0 3 Mpc . -1 . The WMAP C TB l measurement is consistent with a null signal, as expected in the standard cosmological model. We follow a Feldman-Cousins prescription [89] to set 68% and 95% confidence level (CL) upper limits on the primordial magnetic field [75]. We only consider multipoles with /lscript > 50 to simplify the analysis; in this range the measured values of C TB /lscript are uncorrelated between different /lscript values. This restriction does not significantly impact sensitivity to the magnetic field, since most signal is for larger values of multipole number.
A comparison between our model for C TB /lscript and the nine-year WMAP data is given in Fig. 1 for two magnetic field helicity models: one with power law n H = -1 9 and amplitude . H B = 10 5 nG 2 Mpc, and one with power law n H = -0 6 and amplitude . H B = 10 8 nG 2 Mpc. For both cases, we set the value of the effective magnetic field B eff to 1 nG and the spectral index to its inflationary value of n B = -2 99, which is somewhat below current cosmological . limits [90]. These models both produce a helical magnetic field which is just at the level which can be ruled out from the WMAP 9-year microwave background polarization power spectra. The helicity amplitude H B varies strongly with spectral index, because for larger values of n H the helicity is more concentrated on small scales, close to the damping scale, which contribute little to the microwave background signal.
The upper limits on the H B as functions of n H are given in Fig. 2 for three scenarios: n B = -2 99, . n B = -2 0, . and n B = n H -1. We also present the limits in terms of H λ for the same three scenarios in Fig. 3, using a smoothing scale of λ = 1 Mpc which is commonly used in the magnetic field literature.
The results are relatively insensitive to the systematic uncertainty in the cross-correlation signal due to modeling of the cutoff scale; a plausible range of cutoff scales gives a signal difference which is smaller than the measurement uncertainties in the WMAP data. Systematic uncertainties with a size up to 20% of the predicted values of C TB l have only small effects on the magnetic field limits obtained here.
FIG. 1: A comparison between the temperatureB -polarization cross correlation model for B eff = 1 nG, n B = -2 99; the solid . red line is for a helicity amplitude of H B = 10 5 nG 2 Mpc and helicity spectral index n H = -1 9 while the dotted blue line . is for H B = 10 8 nG 2 Mpc and n H = -0.6. Also shown are the nine-year WMAP data (solid gray dots with bars indicating uncertainties).
H
FIG. 2: The 95% upper limits on H B as a function of n H for n B = -2 99 (black solid), . n B = -2 0 (red short dash), and . n B = n H -1 (blue long dash). For all three cases, B eff = 1 nG.
H
FIG. 3: The 95% upper limits on H λ for a smoothing scale λ = 1 Mpc, as a function of n H for n B = -2 99 (black solid), . n B = -2 0 (red short dash), and . n B = n H -1 (blue long dash). For all three cases, B eff = 1 nG.
## V. CONCLUSIONS
The results presented here are the first direct constraint on a helical primordial magnetic field by its contribution to the parity-odd temperature-B polarization cross-power spectrum C TB l of the microwave background. No experiment to date has detected a non-zero value for this power spectrum; we use the WMAP 9-year measurement which is consistent with zero to place upper limits on the combined mean field strength and helicity of a primordial magnetic field. The primordial magnetic field amplitude constraint of around B eff = 3 nG from the microwave background temperature and E-polarization power spectra [32, 91-93] gives an upper limits on magnetic helicity H B less than around 10 nG 2 Gpc for a nearly scale-invariant power spectrum with n B = -2 99. . The helicity limits become weaker for larger values of n B . Recent work has argued for more stringent upper limits of B eff < 1 nG from constraints on the trispectrum induced by magnetic fields, rather than the power spectrum [94]. If magnetic fields from inflation are produced with a magnetic curvature mode as advocated by Ref. [95], then the trispectrum constraint is even stronger, pushing the magnetic field amplitude down to B eff < . 0 05 nG.
The mean helicity amplitude over a given volume is constrained by the realizability condition Eq. (5). The smaller the value of the magnetic field B eff , the lower the helicity H B which can be supported by the field. Any cosmological field will have physical effects measured over an effective volume which is at most the Hubble volume, so the effective comoving correlation length of this field is limited by the Hubble length H -1 0 . For a given microwave background constraint on H B and assuming a magnetic field strength equal to some current upper limit, the maximal magnetic helicity which saturates the realizability condition must have a correlation length ξ M = 4 π H B /B 2 eff . If this correlation length is larger than the Hubble length, then a magnetic field of the given amplitude cannot support helicity as large as the measured limit. For a magnetic field with B eff = 3 nG and the corresponding helicity equal to the limiting value H B = 10 nG 2 Gpc , the correlation length for maximal helicity is around 10 Gpc: current measurements provide a helicity constraint which is just at the level of the maximum possible helicity for the magnetic field strength. If the field strength is significantly lower, then the helicity limits derived in this paper are substantially above the maximum helicity allowed by Eq. (5).
Upcoming polarization data from the Planck satellite, as well as high-resolution ground-based experiments like ACTPol [96] and SPTPol [97], will strengthen limits on both the magnetic field amplitude and helicity, for two reasons: first, the signal increases for larger l values beyond those probed by WMAP, and second, upcoming experiments will
produce polarized maps over large portions of the sky with much greater sensitivity than WMAP. Interest in B-mode polarization has exploded due to the recent results from the BICEP2 collaboration [98]. Experiments searching for B-polarization from primordial tensor modes (at large angular scales) and gravitational lensing (at small angular scales) will drive continual increases in sensitivity over the coming decade. Planck's maps have a sensitivity (around 85 µ K-arcmin for the SMICA map) which is a factor of 4 lower than WMAP (around 360 µ K-arcmin), corresponding to errors in C TB l smaller by a factor of 16. The recent PRISM satellite proposal [99] envisions full-sky polarization maps with sensitivity of 3 µ K-arcmin, which would give C TB l errors smaller than the WMAP errors used here by a factor of 10 4 .
Limits on the magnetic field amplitude B eff from the microwave background power spectra will not improve substantially, because they are limited by cosmic variance in the power spectra from other non-magnetic sources of fluctuations. In contrast, sensitivity improvements in polarization will continue to improve helicity limits from C TB l because this signal is not limited by cosmic variance: it is zero for standard-cosmology primary perturbations which do not violate parity. (At least this is the case until extreme sensitivities are reached where the cosmic variance in C TB l from the residual gravitational lensing contribution to de-lensed maps dominates over the map noise). So future measurements may provide constraints on magnetic field helicity which are much below the maximal helicity allowed by Eq. (5) and the magnetic field amplitude limits.
The TB power spectrum of cosmic microwave background polarization, and its lower-amplitude counterpart EB , provide a valuable opportunity to probe unconventional physics which violates cosmological parity. Of contributors to these power spectra, gravitational lensing and helical magnetic fields are the two sources which rely only on standard, demonstrated physical effects. The microwave background lensing spectrum can be calculated to high accuracy within the standard model of cosmological structure formation, so any departures from this signal would be a good bet for revealing the existence of significant helical magnetic fields in the universe. In turn, the detection of helicity would give valuable information about the still-mysterious origin of magnetic field in the cosmos.
## Acknowledgments
It is our pleasure to thank A. Brandenburg, L. Campanelli, K. Kunze, H. Tashiro, A. Tevzadze, and T. Vachaspati for useful discussions. T.K. and G.L. acknowledge partial support from the Swiss NSF SCOPES grant IZ7370152581. T.K. and A.K. were supported in part through NASA Astrophysics Theory program grant NNXl0AC85G and NSF Astrophysics and Astronomy Grant Program grants AST-1109180 and AST-1108790. T.K. acknowledges partial support from Berkman Foundation. A.K. is partly supported by NSF Astrophysics and Astronomy grant AST-1312380. This work made use of the NASA Astrophysical Data System for bibliographic information.
## Appendix A: Derivation of C TB ( V ) l
The solution of Eq. (17) can be written in the form
$$\intertext { n e s u u i o n o f e q. ( 1 \, r ) \, c a n \, e b e \, w r u e n \, m \, e n e \, \text{or} } P ^ { ( \pm 1 ) } & = \frac { \sqrt { 3 } } { 3 0 } k \int _ { 0 } ^ { \eta } d \eta ^ { \prime } e ^ { - \frac { 3 } { 0 0 } \tau ( \eta, \eta ^ { \prime } ) } \Theta _ { 1 } ^ { ( \pm 1 ) } ( \eta ^ { \prime } ) = \frac { \sqrt { 3 } } { 3 0 } k \int _ { 0 } ^ { \eta } d \eta ^ { \prime } e ^ { + \frac { 3 } { 0 0 } \tau ( \eta ) } e ^ { - \frac { 3 } { 0 0 } \tau ( \eta ^ { \prime } ) } \Theta _ { 1 } ^ { ( \pm 1 ) } ( \eta ^ { \prime } ) \,, \quad ( A 1 ) \\ \text{whara } \tau ( n \, n ^ { \prime } ) & - \tilde { \eta } ^ { \eta } \, d \eta ^ { \prime \prime } \tilde { \eta } ( n ^ { \prime \prime } ) \, - \tau ( n ^ { \prime } ) \, - \tau ( n ) \, \text{and } \text{that visibility from action } V ( n ) & - \tilde { \eta } ^ { - \tau ( \eta ) } - \tilde { \eta } ^ { - \tau } \, c a n \, h o a n v o r u n t e d }$$
where τ ( η, η ′ ) = ∫ η η ′ dη ′′ ˙ ( τ η ′′ ) = τ ( η ′ ) -τ ( η ) and the visibility function V ( η ) = d dη e -τ ( η ) = ˙ τe -τ can be approximated by the asymmetric Gaussian function [88, 100]
$$V ( \eta ) & = V ( \eta _ { \text{dec} } ) \exp [ - \frac { ( \eta - \eta _ { \text{dec} } ) ^ { 2 } } { 2 \Delta \eta _ { \text{dec} } ^ { 2 } } ] \, & ( \text{A2} ) \\ _ { \text{1ec} } - \eta ) + \Delta \eta _ { \text{dec} } _ { 2 } \Theta ( \eta - \eta _ { \text{dec} } ), \, \Delta \eta _ { \text{dec} } _ { 1 } = 0. 0 0 1 1 \eta _ { 0 } \text{ and } \Delta \eta _ { \text{dec} } _ { 2 } = 0. 0 0 1 9 \eta _ { 0 } \text{ and } \Theta ( \eta ) \text{ is the}$$
where ∆ η dec = ∆ η dec 1 Θ( η dec -η ) + ∆ η dec 2 Θ( η -η dec ), ∆ η dec 1 = 0 0011 . η 0 and ∆ η dec 2 = 0 0019 . η 0 and Θ( ) is the η usual step function. The prefactor V ( η dec ) = 1 √ 2 π ∆ η dec is calculated from the normalization condition ∫ η 0 0 V ( η dη ) = 1. Substituting the solution Eq. (A1) into Eq. (12) we obtain
$$\frac { \mu } { 2 l + 1 } = \mp \frac { \mu } { 1 0 \sqrt { 2 } } \sqrt { ( l - 1 ) ( l + 2 ) } \int _ { 0 } ^ { \eta _ { 0 } } d \eta V ( \eta ) \frac { j _ { l } ( k ( \eta _ { 0 } - \eta ) ) } { k ( \eta _ { 0 } - \eta ) } \int _ { 0 } ^ { \eta } d \eta ^ { \prime } e ^ { + \frac { 3 } { 1 0 } \tau ( \eta ) } e ^ { - \frac { 3 } { 1 0 } \tau ( \eta ^ { \prime } ) } \Theta _ { 1 } ^ { ( 0 - 1 ) } ( \eta ^ { \prime } ) \. \quad ( A 3 ) \\ \text{Since the visibility function } V ( \eta ) \text{ is sharply peaked around } \eta = \eta _ { \text{dec} } \text{ and } e ^ { - \frac { 3 } { 2 } \tau ( \eta ^ { \prime } ) } \text{ behaves like a step function, the}$$
Since the visibility function V ( η ) is sharply peaked around η = η dec and e -3 10 τ ( η ′ ) behaves like a step function, the Θ ( ± 1) 1 ( η ′ ) factor can approximately be pulled out from the η ′ integration and we get
$$\frac { B _ { l } ^ { \pm 1 } ( k, \eta _ { 0 } ) } { 2 l + 1 } = \mp \frac { k } { 1 0 \sqrt { 2 } } \sqrt { ( l - 1 ) ( l + 2 ) } \int _ { 0 } ^ { \eta _ { 0 } } d \eta V ( \eta ) \frac { j _ { l } ( k ( \eta _ { 0 } - \eta ) ) } { k ( \eta _ { 0 } - \eta ) } \Theta _ { 1 } ^ { ( \pm 1 ) } ( \eta ) \int _ { 0 } ^ { \eta } d \eta ^ { \prime } e ^ { + \frac { 3 } { 4 } \tau ( \eta ) } e ^ { - \frac { 3 } { 4 0 } \tau ( \eta ^ { \prime } ) } \,.$$
Noticing that V ( η ) ∝ exp( -γ η ( -η dec ) 2 ) and j l ( k η ( 0 -η )) contains a mixture of oscillating modes e ipη and e -ipη with p ∝ k , the formula ∫ ∞ -∞ e -γη 2 e ipη dη = e -p / γ 2 4 ∫ ∞ -∞ e -γη 2 dη gives the approximation [88]
$$\int _ { 0 } ^ { \eta _ { 0 } } d \eta V ( \eta ) \frac { j _ { l } ( k ( \eta _ { 0 } - \eta ) ) } { k ( \eta _ { 0 } - \eta ) } \Theta _ { 1 } ^ { ( \pm 1 ) } ( \eta ) \approx \frac { j _ { l } ( k ( \eta _ { 0 } - \eta _ { \text{dec} } ) } { k \eta _ { 0 } - k \eta _ { \text{dec} } } \Theta _ { 1 } ^ { ( \pm 1 ) } ( \eta _ { \text{dec} } ) D _ { E } ( k ) \int _ { 0 } ^ { \eta _ { 0 } } d \eta V ( \eta ) \, \quad \ ( A 5 ) \\. \quad \ \sim \ \ a \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \colon \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \cdot \, \end{matrix}$$
where D E ( k ) is the Silk damping factor for polarization [88], Eq. (19). ′ ′
Introducing a new variable x ≡ τ ( η ) /τ ( η ), approximating dη = -∆ η dec dx/x and noticing that
$$\int _ { 0 } ^ { n _ { 0 } } d \eta \, V ( \eta ) \int _ { 1 } ^ { \infty } \frac { d x } { x } e ^ { - \frac { 3 } { n } x \tau ( \eta ) } e ^ { \frac { 3 } { n } \tau ( \eta ) } = \Delta \eta _ { d e c } \cdot \int _ { 0 } ^ { \infty } d \tau e ^ { - \frac { 7 } { n } \tau ( \eta ) } \int _ { 1 } ^ { \infty } \frac { d x } { x } e ^ { - \frac { 3 } { n } x \tau ( \eta ) } = \Delta \eta _ { d e c } \frac { 1 0 } { 7 } \ln \frac { 1 0 } { 3 } \,, \quad ( A 6 )$$
we get
$$\L _ { \Theta \bullet \bullet } ^ { \bullet 1 } \\ \frac { B ^ { \pm 1 } ( k, \eta _ { 0 } ) } { 2 l + 1 } \, & = \, \mp \frac { \sqrt { 2 } } { 1 4 } \ln ( \frac { 1 0 } { 3 } ) \sqrt { ( l - 1 ) ( l + 2 ) } D _ { E } ( k ) \frac { j _ { l } ( k \eta _ { 0 } ) } { k \eta _ { 0 } } k \Delta \eta _ { d e c } \Theta _ { 1 } ^ { ( \pm 1 ) } ( k, \eta _ { d e c } ) \\ & = \, \mp \frac { \sqrt { 2 } } { \ln ( \frac { 1 0 } { 2 } ) \sqrt { ( l - 1 ) ( l + 2 ) } D _ { e } ( k ) k \Delta n _ { a _ { a _ { a _ { a _ { a _ { b _ { a _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { b _ { - b _ { - b _ { - b _ { - b _ { - b _ { - b _ { - b _ { - b _ { - b _ _ { - b _ { - b _ - b _ - b _ - b _ - b _ - - b _ - - b _ - - b - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - } - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -. - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -$$
Making use of Eq. (A8) and Eq. (11), we finally obtain for the temperatureB -polarization cross correlation function
$$2 l + 1 & \quad - \, ^ { \top } 1 4 \, ^ { \dots } \, 3 \, ^ { \prime \, \vee } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, _ { \ } ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \top } \, ^ { \left ( A 8 ) } \\ \quad = \, \mp \frac { \sqrt { 2 } } { 1 4 } \ln ( \frac { 1 0 } { 3 } ) \sqrt { ( l - 1 ) ( l + 2 ) } D _ { E } ( k ) k \Delta \eta _ { \text{dec} } \frac { j _ { l } ( k \eta _ { 0 } ) } { k \eta _ { 0 } } \Omega ^ { ( \pm 1 ) } ( k, \eta _ { \text{dec} } ) \,. \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \ \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \ \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \, \quad \,
q. \, ( A 8 ) \text{ and } E q. \, ( 1 1 ), \, \text{ we finally obtain for the temperature-B-polarization cross correlation function}$$
$$C _ { l } ^ { T B \ ( V ) } = - \frac { 2 } { 7 \pi } \ln ( \frac { 1 0 } { 3 } ) \sqrt { \frac { ( l + 2 )! } { ( l - 2 )! } } \int d k k ^ { 2 } D _ { E } ( k ) \omega ( k, \eta _ { d e c } ) \frac { j _ { l } ^ { 2 } ( k \eta _ { 0 } ) } { ( k \eta _ { 0 } ) ^ { 2 } } k \Delta \eta _ { d e c } \,, & & ( A 9 )$$
where ω k ( ) is the helical part of the power spectrum which can be expressed as [35]
$$\omega ( k, \eta ) & = \left [ \frac { k \eta } { ( \rho _ { \gamma, 0 } + p _ { \gamma, 0 } ) ( 1 + R _ { \text{dec} } ) } \right ] ^ { 2 } g ( k ). & & ( A 1 0 ) \\. &. &. &. &. &. &. &. &. &.$$
Here p γ 0 and ρ γ 0 are the radiation pressure and energy density today, R dec is the baryon-photon energy density at decoupling, and g k ( ) can be expressed in terms of the spectral indices n B and n H , values of B λ , H λ , and the smoothing scale λ as
$$g ( k ) = \mathcal { G } \lambda k ( \lambda k _ { D } ) ^ { n _ { B } + n _ { H } + 2 } \left [ 1 + \frac { n _ { H } - 1 } { n _ { B } + 3 } \left ( \frac { k } { k _ { D } } \right ) ^ { n _ { B } + n _ { H } + 2 } \right ] & & ( A 1 1 )$$
with
$$\mathcal { G } = \frac { \lambda ^ { 3 } B _ { \lambda } ^ { 2 } H _ { \lambda } ^ { 2 } } { 2 4 ( n _ { B } + n _ { H } + 2 ) \Gamma \left ( \frac { n _ { B } + 3 } { 2 } \right ) \Gamma \left ( \frac { n _ { H } + 4 } { 2 } \right ) } & & ( A 1 2 ) \\ ( A 1 1 ), \, \text{and} \, ( A 1 2 ) \, \text{in} \, E q. \, ( A 9 ) \, \text{we arrive at} \, E q. \, ( 1 8 ).$$
Then using Eqs. (A10), (A11), and (A12) in Eq. (A9) we arrive at Eq. (18).
- [1] L. M. Widrow, 'Origin of galactic and extragalactic magnetic fields,' Rev. Mod. Phys. 74 , 775 (2002) [arXiv:astro-ph/0207240].
- [2] R. M. Kulsrud and E. G. Zweibel, 'The Origin of Astrophysical Magnetic Fields,' Rept. Prog. Phys. 71 , 0046091 (2008) [arXiv:0707.2783 [astro-ph]].
- [3] A. Kandus, K. E. Kunze and C. G. Tsagas, 'Primordial magnetogenesis,' Phys. Rept. 505 , 1 (2011). [arXiv:1007.3891 [astro-ph.CO]].
- [4] R. Durrer and A. Neronov, 'Cosmological Magnetic Fields: Their Generation, Evolution and Observation,' Astron. Astrophys. Rev 21 , 62 (2013). [arXiv:1303.7121 [astro-ph.CO]].
- [5] A. Brandenburg and K. Subramanian, 'Astrophysical magnetic fields and nonlinear dynamo theory,' Phys. Rept. 417 , 1 (2005) [arXiv:astro-ph/0405052].
- [6] A. G. Tevzadze, L. Kisslinger, A. Brandenburg and T. Kahniashvili, 'Magnetic Fields from QCD Phase Transitions,' Astrophys. J. 759 , 54 (2012) [arXiv:1207.0751 [astro-ph.CO]].
- [7] J. M. Cornwall, 'Speculations on primordial magnetic helicity,' Phys. Rev. D 56 , 6146 (1997) [arXiv:hep-th/9704022].
- [8] M. Giovannini and M. E. Shaposhnikov, 'Primordial hypermagnetic fields and triangle anomaly,' Phys. Rev. D 57 , 2186 (1998) [arXiv:hep-ph/9710234].
- [9] G. B. Field and S. M. Carroll, 'Cosmological magnetic fields from primordial helicity,' Phys. Rev. D 62 , 103008 (2000) [arXiv:astro-ph/9811206].
- [10] T. Vachaspati, 'Estimate of the primordial magnetic field helicity,' Phys. Rev. Lett. 87 , 251302 (2001) [arXiv:astro-ph/0101261].
- [11] H. Tashiro, T. Vachaspati and A. Vilenkin, 'Chiral Effects and Cosmic Magnetic Fields,' Phys. Rev. D 86 , 105033 (2012) [arXiv:1206.5549 [astro-ph.CO]].
- [12] G. Sigl, 'Cosmological Magnetic Fields from Primordial Helical Seeds,' Phys. Rev. D 66 , 123002 (2002) [arXiv:astro-ph/0202424].
- [13] K. Subramanian and A. Brandenburg, 'Nonlinear current helicity fluxes in turbulent dynamos and alpha quenching,' Phys. Rev. Lett. 93 , 205001 (2004) [arXiv:astro-ph/0408020].
- [14] L. Campanelli and M. Giannotti, 'Magnetic helicity generation from the cosmic axion field,' Phys. Rev. D 72 , 123001 (2005) [arXiv:astro-ph/0508653].
- [15] V. B. Semikoz and D. D. Sokoloff, 'Magnetic helicity and cosmological magnetic field,' arXiv:astro-ph/0411496; 'Largescale cosmological magnetic fields and magnetic helicity,' Int. J. Mod. Phys. D 14 , 1839 (2005).
- [16] A. Diaz-Gil, J. Garcia-Bellido, M. Garcia Perez and A. Gonzalez-Arroyo, 'Magnetic field production during preheating at the electroweak scale,' Phys. Rev. Lett. 100 , 241301 (2008) [arXiv:0712.4263 [hep-ph]].
- [17] L. Campanelli, 'Helical Magnetic Fields from Inflation,' Int. J. Mod. Phys. D 18 , 1395 (2009) [arXiv:0805.0575 [astro-ph]].
- [18] L. Campanelli, 'Origin of Cosmic Magnetic Fields,' Phys. Rev. Lett. 111 , no. 6, 061301 (2013) [arXiv:1304.6534 [astroph.CO]].
- [19] R. Banerjee and K. Jedamzik, 'The evolution of cosmic magnetic fields: From the very early universe, to recombination, to the present,' Phys. Rev. D 70 , 123003 (2004) [arXiv:astro-ph/0410032].
- [20] T. A. Ensslin, 'Does circular polarisation reveal the rotation of quasar engines?,' Astron. Astrophys. 401 , 499 (2003) [arXiv:astro-ph/0212387].
- [21] J.P. Vall´e, New Astron. Rev. e 48 , 763 (2004).
- [22] M. Kamionkowski, A. Kosowsky, and A. Stebbins, 'Statistics of Cosmic Microwave Background Polarization,' Phys. Rev. D 55 , 7368 (1997) [arXiv:astro-ph/9611125].
- [23] M. Zaldarriaga and U. Seljak, 'An all sky analysis of polarization in the microwave background,' Phys. Rev. D 55 , 1830 (1997) [arXiv:astro-ph/9609170].
- [24] A. Mack, T. Kahniashvili and A. Kosowsky, 'Vector and Tensor Microwave Background Signatures of a Primordial Stochastic Magnetic Field,' Phys. Rev. D 65 , 123004 (2002) [arXiv:astro-ph/0105504].
- [25] J. R. Shaw and A. Lewis, 'Massive Neutrinos and Magnetic Fields in the Early Universe,' Phys. Rev. D 81 , 043517 (2010) [arXiv:0911.2714 [astro-ph.CO]].
- [26] D. G. Yamazaki, K. Ichiki, T. Kajino and G. J. Mathew, 'Primordial Magnetic Field Effects on the CMB and Large Scale Structure,' Adv. Astron. 2010 , 586590 (2010) [arXiv:1112.4922 [astro-ph.CO]].
- [27] D. G. Yamazaki, K. Ichiki, T. Kajino and G. .J. Mathews, 'Constraints on the neutrino mass and the primordial magnetic field from the matter density fluctuation parameter σ 8 ,' Phys. Rev. D 81 , 103519 (2010) [arXiv:1005.1638 [astro-ph.CO]].
- [28] D. G. Yamazaki, K. Ichiki, T. Kajino and G. J. Mathews, 'New Constraints on the Primordial Magnetic Field,' Phys. Rev. D 81 , 023008 (2010) [arXiv:1001.2012 [astro-ph.CO]].
- [29] D. Paoletti and F. Finelli, 'Constraints on a Stochastic Background of Primordial Magnetic Fields with WMAP and South Pole Telescope data,' Phys. Lett. B 726 , 45 (2013) [arXiv:1208.2625 [astro-ph.CO]].
- [30] D. Paoletti and F. Finelli, 'CMB Constraints on a Stochastic Background of Primordial Magnetic Fields,' Phys. Rev. D 83 , 123533 (2011) [arXiv:1005.0148 [astro-ph.CO]].
- [31] K. E. Kunze, 'CMB anisotropies in the presence of a stochastic magnetic field,' Phys. Rev. D 83 , 023006 (2011) [arXiv:1007.3163 [astro-ph.CO]].
- [32] P. A. R. Ade et al. [Planck Collaboration], 'Planck 2013 results. XVI. Cosmological parameters,' preprint (2013) [arXiv:1303.5076 [astro-ph.CO]].
- [33] L. Pogosian, T. Vachaspati and S. Winitzki, 'Signatures of kinetic and magnetic helicity in the CMBR,' Phys. Rev. D 65 , 083502 (2002) [arXiv:astro-ph/0112536].
- [34] C. Caprini, R. Durrer and T. Kahniashvili, 'The cosmic microwave background and helical magnetic fields: The tensor mode,' Phys. Rev. D 69 , 063006 (2004) [arXiv:astro-ph/0304556].
- [35] T. Kahniashvili and B. Ratra, 'Effects of cosmological magnetic helicity on the cosmic microwave background,' Phys. Rev. D 71 , 103006 (2005) [arXiv:astro-ph/0503709].
- [36] K. E. Kunze, 'Effects of helical magnetic fields on the cosmic microwave background,' Phys. Rev. D 85 , 083004 (2012) [arXiv:1112.4797 [astro-ph.CO]].
- [37] A. Kosowsky and A. Loeb, 'Faraday rotation of microwave background polarization by a primordial magnetic field,' Astrophys. J. 469 , 1 (1996); [arXiv:astro-ph/9601055].
- [38] T. A. Ensslin and C. Vogt, 'The Magnetic Power Spectrum in Faraday Rotation Screens,' Astron. Astrophys. 401 , 835 (2003) [arXiv:astro-ph/0302426].
- [39] L. Campanelli, A. D. Dolgov, M. Giannotti and F. L. Villante, 'Faraday Rotation of the CMB Polarization and Primordial Magnetic Field Properties,' Astrophys. J. 616 , 1 (2004) [arXiv:astro-ph/0405420].
- [40] A. Kosowsky, T. Kahniashvili, G. Lavrelashvili and B. Ratra, 'Faraday Rotation of the Cosmic Microwave Background Polarization by a Stochastic Magnetic Field,' Phys. Rev. D 71 , 043006 (2005) [arXiv:astro-ph/0409767].
- [41] T. Kahniashvili, Y. Maravin and A. Kosowsky, 'Primordial Magnetic Field Limits from WMAP Five-Year Data,' Phys. Rev. D 80 , 023009 (2009) [arXiv:0806.1876 [astro-ph]].
- [42] P. Cabella, P. Natoli and J. Silk, 'Constraints on CPT Violation from WMAP Three-Year Polarization Data: A Wavelet Analysis,' Phys. Rev. D 76 , 123014 (2007) [arXiv:0705.0810 [astro-ph]].
- [43] J. Q. Xia, H. Li, X. l. Wang and X. m. Zhang, 'Testing CPT symmetry with CMB measurements,' arXiv:0710.3325 [hep-ph].
- [44] B. Feng, M. Li, J. Q. Xia, X. Chen and X. Zhang, 'Searching for CPT violation with WMAP and BOOMERANG,' Phys. Rev. Lett. 96 , 221302 (2006) [arXiv:astro-ph/0601095].
- [45] J. -Q. Xia, 'Cosmological CPT Violation and CMB Polarization Measurements,' JCAP 1201 , 046 (2012) [arXiv:1201.4457 [astro-ph.CO]].
- [46] J. -Q. Xia, H. Li and X. Zhang, 'Probing CPT Violation with CMB Polarization Measurements,' Phys. Lett. B 687 , 129 (2010) [arXiv:0908.1876 [astro-ph.CO]].
- [47] M. Li, Y. -F. Cai, X. Wang and X. Zhang, ' CPT Violating Electrodynamics and Chern-Simons Modified Gravity,' Phys. Lett. B 680 , 118 (2009) [arXiv:0907.5159 [hep-ph]].
- [48] M. Li and X. Zhang, 'Cosmological CPT-Violating Effect on CMB Polarization,' Phys. Rev. D 78 , 103516 (2008) [arXiv:0810.0403 [astro-ph]].
- [49] A. Gruppuso, P. Natoli, N. Mandolesi, A. De Rosa, F. Finelli and F. Paci, 'WMAP 7 Year Constraints on CPT Violation from Large Angle CMB Anisotropies,' JCAP 1202 , 023 (2012) [arXiv:1107.5548 [astro-ph.CO]].
- [50] A. Lue, L. M. Wang and M. Kamionkowski, 'Cosmological Signature of New Parity-Violating Interactions,' Phys. Rev. Lett. 83 , 1506 (1999) [arXiv:astro-ph/9812088].
- [51] J. Q. Xia, H. Li, G. B. Zhao and X. Zhang, 'Testing CPT Symmetry with CMB Measurements: Update After WMAP5,' arXiv:0803.2350 [astro-ph].
- [52] S. Saito, K. Ichiki and A. Taruya, 'Probing polarization states of primordial gravitational waves with CMB anisotropies,' JCAP 0709 , 002 (2007) [arXiv:0705.3701 [astro-ph]].
- [53] E. S. Scannapieco and P. G. Ferreira, 'Polarization - temperature correlation from primordial magnetic field,' Phys. Rev. D 56 , 7493 (1997) [arXiv:astro-ph/9707115].
- [54] C. Scoccola, D. Harari and S. Mollerach, 'B polarization of the CMB from Faraday rotation,' Phys. Rev. D 70 , 063003 (2004) [arXiv:astro-ph/0405396].
- [55] M. Demianski and A. G. Doroshkevich, 'Possible extensions of the standard cosmological model: anisotropy, rotation, and magnetic field,' Phys. Rev. D 75 , 123517 (2007) [arXiv:astro-ph/0702381].
- [56] J. R. Kristiansen and P. G. Ferreira, 'Constraining primordial magnetic fields with CMB polarization experiments,' Phys. Rev. D 77 , 123004 (2008) [arXiv:0803.3210 [astro-ph]].
- [57] S. M. Carroll, G. B. Field and R. Jackiw, 'Limits on a Lorentz and parity violating modification of electrodynamics,' Phys. Rev. D 41 , 1231 (1990).
- [58] V. A. Kostelecky and M. Mewes, 'Lorentz-violating electrodynamics and the cosmic microwave background,' Phys. Rev. Lett. 99 , 011601 (2007) [arXiv:astro-ph/0702379].
- [59] Y. -F. Cai, M. Li and X. Zhang, 'Testing the Lorentz and CPT Symmetry with CMB polarizations and a non-relativistic Maxwell Theory,' JCAP 1001 , 017 (2010) [arXiv:0912.3317 [hep-ph]].
- [60] W. -T. Ni, 'From Equivalence Principles to Cosmology: Cosmic Polarization Rotation, CMB Observation, Neutrino Number Asymmetry, Lorentz Invariance and CPT,' Prog. Theor. Phys. Suppl. 172 , 49 (2008) [arXiv:0712.4082 [astroph]].
- [61] R. Casana, M. M. Ferreira, Jr. and J. S. Rodrigues, 'Lorentz-violating contributions of the Carroll-Field-Jackiw model to the CMB anisotropy,' Phys. Rev. D 78 , 125013 (2008) [arXiv:0810.0306 [hep-th]].
- [62] R. R. Caldwell, V. Gluscevic and M. Kamionkowski, 'Cross-Correlation of Cosmological Birefringence with CMB Temperature,' Phys. Rev. D 84 , 043504 (2011) [arXiv:1104.1634 [astro-ph.CO]].
- [63] H. J. Mosquera Cuesta and G. Lambiase, 'Nonlinear electrodynamics and CMB polarization,' JCAP 1103 , 033 (2011) [arXiv:1102.3092 [astro-ph.CO]].
- [64] M. Kamionkowski and T. Souradeep, 'The Odd-Parity CMB Bispectrum,' Phys. Rev. D 83 , 027301 (2011) [arXiv:1010.4304 [astro-ph.CO]].
- [65] V. Gluscevic and M. Kamionkowski, 'Testing Parity-Violating Mechanisms with Cosmic Microwave Background Experiments,' Phys. Rev. D 81 , 123529 (2010) [arXiv:1002.1308 [astro-ph.CO]].
- [66] M. Mewes, 'Optical-cavity tests of higher-order Lorentz violation,' Phys. Rev. D 85 , 116012 (2012) [arXiv:1203.5331 [hep-ph]].
- [67] V. Gluscevic, D. Hanson, M. Kamionkowski and C. M. Hirata, 'First CMB Constraints on Direction-Dependent Cosmological Birefringence from WMAP-7,' Phys. Rev. D 86 , 103529 (2012) [arXiv:1206.5546 [astro-ph.CO]].
- [68] W. -T. Ni, 'Constraints on pseudoscalar-photon interaction from CMB polarization observation,' arXiv:0910.4317 [gr-qc].
- [69] N. J. Miller, M. Shimon and B. G. Keating, 'CMB Polarization Systematics Due to Beam Asymmetry: Impact on Cosmological Birefringence,' Phys. Rev. D 79 , 103002 (2009) [arXiv:0903.1116 [astro-ph.CO]].
- [70] W. -T. Ni, 'Cosmic Polarization Rotation, Cosmological Models, and the Detectability of Primordial Gravitational Waves,' Int. J. Mod. Phys. A 24 , 3493 (2009) [arXiv:0903.0756 [astro-ph.CO]].
- [71] E. A. Lim, 'Can we see Lorentz-violating vector fields in the CMB?,' Phys. Rev. D 71 , 063504 (2005) [arXiv:astro-ph/0407437].
- [72] S. M. Carroll and E. A. Lim, 'Lorentz-violating vector fields slow the universe down,' Phys. Rev. D 70 , 123525 (2004) [arXiv:hep-th/0407149].
- [73] S. H. S. Alexander, 'Is cosmic parity violation responsible for the anomalies in the WMAP data?,' Phys. Lett. B 660 , 444 (2008) [arXiv:hep-th/0601034].
- [74] M. Satoh, S. Kanno and J. Soda, 'Circular polarization of primordial gravitational waves in string-inspired inflationary cosmology,' Phys. Rev. D 77 , 023526 (2008) [arXiv:0706.3585 [astro-ph]].
- [75] C. L. Bennett et al. [WMAP Collaboration], 'Nine-Year Wilkinson Microwave Anisotropy Probe (WMAP) Observations: Final Maps and Results,' Astrophys. J. Suppl. 208 , 20 (2013) [arXiv:1212.5225 [astro-ph.CO]];
- [76] G. Hinshaw et al. [WMAP Collaboration], 'Nine-Year Wilkinson Microwave Anisotropy Probe (WMAP) Observations: Cosmological Parameter Results,' Astrophys. J. Suppl. 208 , 19 (2013); [arXiv:1212.5226 [astro-ph.CO]].
- [77] D. Larson, J. Dunkley, G. Hinshaw, E. Komatsu, M. R. Nolta, C. L. Bennett, B. Gold and M. Halpern et al. , 'SevenYear Wilkinson Microwave Anisotropy Probe (WMAP) Observations: Power Spectra and WMAP-Derived Parameters,' Astrophys. J. Suppl. 192 , 16 (2011) [arXiv:1001.4635 [astro-ph.CO]].
- [78] M. A. Berger, 'Introduction to magnetic helicity'. Plasma Physics and Controlled Fusion 41 167 (1999).
- [79] R. Durrer and C. Caprini, 'Primordial Magnetic Fields and Causality,' JCAP 0311 , 010 (2003) [arXiv:astro-ph/0305059].
- [80] T. Kahniashvili, A. G. Tevzadze and B. Ratra, 'Phase Transition Generated Cosmological Magnetic Field at Large Scales,' Astrophys. J. 726 , 78 (2011) [arXiv:0907.0197 [astro-ph.CO]].
- [81] K. Jedamzik, V. Katalinic and A. V. Olinto, 'Damping of Cosmic Magnetic Fields,' Phys. Rev. D 57 , 3264 (1998) [arXiv:astro-ph/9606080].
- [82] K. Subramanian and J. D. Barrow, 'Microwave Background Signals from Tangled Magnetic Fields,' Phys. Rev. Lett. 81 , 3575 (1998) [arXiv:astro-ph/9803261].
- [83] R. Durrer, T. Kahniashvili and A. Yates, 'Microwave Background Anisotropies from Alfven waves,' Phys. Rev. D 58 , 123004 (1998) [arXiv:astro-ph/9807089].
- [84] T. R. Seshadri and K. Subramanian, 'CMBR Polarization Signals from Tangled Magnetic Fields,' Phys. Rev. Lett. 87 , 101301 (2001) [arXiv:astro-ph/0012056].
- [85] K. Subramanian, T. R. Seshadri and J. D. Barrow, 'Small-scale CMB polarization anisotropies due to tangled primordial magnetic fields,' Mon. Not. Roy. Astron. Soc. 344 , L31 (2003) [arXiv:astro-ph/0303014].
- [86] W. Hu and M. J. White, 'CMB Anisotropies: Total Angular Momentum Method,' Phys. Rev. D 56 , 596 (1997) [arXiv:astro-ph/9702170].
- [87] M. Zaldarriaga and D. D. Harari, 'Analytic approach to the polarization of the cosmic microwave background in flat and open universes,' Phys. Rev. D 52 , 3276 (1995) [arXiv:astro-ph/9504085].
- [88] Z. Cai and Y. Zhang, 'Analytic Spectra of CMB Anisotropies and Polarization Generated by Scalar Perturbations in Synchronous Gauge,' Class. Quant. Grav. 29 , 105009 (2012) [arXiv:1204.6683 [astro-ph.CO]].
- [89] G. J. Feldman and R. D. Cousins, 'A Unified approach to the classical statistical analysis of small signals,' Phys. Rev. D 57 , 3873 (1998) [arXiv:physics/9711021 [physics.data-an]].
- [90] P. A. R. Ade et al. [Planck Collaboration], 'Planck 2013 results. XVI. Cosmological parameters,' arXiv:1303.5076 [astroph.CO].
- [91] J. R. Shaw and A. Lewis, 'Constraining Primordial Magnetism,' Phys. Rev. D 86 , 043510 (2012) [arXiv:1006.4242 [astro-ph.CO]].
- [92] D. G. Yamazaki, K. Ichiki and K. Takahashi, 'Constraints on the multi-lognormal magnetic fields from the observations of the cosmic microwave background and the matter power spectrum,' Phys. Rev. D 88 , 103011 (2013) [arXiv:1311.2584 [astro-ph.CO]].
- [93] T. Kahniashvili, Y. Maravin, A. Natarajan, N. Battaglia and A. G. Tevzadze, 'Constraining primordial magnetic fields through large scale structure,' Astrophys. J. 770 , 47 (2013) [arXiv:1211.2769 [astro-ph.CO]].
- [94] P. Trivedi, K. Subramanian, and T.R. Seshadri, 'Primordial Magnetic Field Limits from CMB Trispectrum Scalar Modes and Planck Constraints,' Phys. Rev. D 89 , 043523 (2014) [arXiv:1312.5308 [astro-ph.CO]].
- [95] C. Bonvin, C. Caprini, and R. Durrer, 'Magnetic Fields from Inflation: The CMB Temperature Anisotropies,' Phys. Rev. D 88 , 083515 (2013) [arXiv:1308.3348 [astro-ph.CO]].
- [96] M. Niemack et al., 'ACTPol: A Polarization-Sensitive Receiver or the Atacama Cosmology Telescope,' Proc. SPIE 7741 , 51 (2010) [arXiv:1006.5049 [astro-ph.IM]].
- [97] J.E. Austermann et al., 'SPTPol: An Instrument for CMB Polarization Measurements with the South Pole Telescope,' Proc. SPIE 8452 , 84520E (2012) [arXiv:1210.4970 [astro-ph.IM]].
- [98] P.A.R. Ade et al., 'BICEP2 I: Detection of B-Mode Polarization at Degree Angular Scales,' Phys. Rev. Lett. 112 , 241101 (2014) [arXiv:1403.3985 [astro-ph.CO]].
- [99] P. Andre et al., 'PRISM: A White Paper on the Ultimate Polarimetric Spectro-Imaging of the Microwave and Far Infrared Sky' [arXiv:1306.2259 [astro-ph.CO]].
- [100] D. S. Gorbunov, V. A. Rubakov, Introduction To The Theory of The Early Universe: Cosmological Perturbations and Inflationary Theory , (World Scientific Publishing Company, 2011). | 10.1103/PhysRevD.90.083004 | [
"Tina Kahniashvili",
"Yurii Maravin",
"George Lavrelashvili",
"Arthur Kosowsky"
] | 2014-08-02T08:55:00+00:00 | 2014-10-16T18:41:45+00:00 | [
"astro-ph.CO"
] | Primordial Magnetic Helicity Constraints from WMAP Nine-Year Data | If a primordial magnetic field in the universe has non-zero helicity, the
violation of parity symmetry results in non-zero correlations between cosmic
microwave background temperature and B-mode polarization. In this paper we
derive approximations to the relevant microwave background power spectra
arising from a helical magnetic field. Using the cross-power spectrum between
temperature and B-mode polarization from the WMAP nine-year data, we set a 95\%
confidence level upper limit on the helicity amplitude to be 10 nG$^2$ Gpc for
helicity spectral index $n_H = -1.9$, for a cosmological magnetic field with
effective field strength of 3 nG and a power-law index $n_B = -2.9$ near the
scale-invariant value. Future microwave background polarization maps with
greater sensitivity will be able to detect the helicity of an inflationary
magnetic field well below the maximum value allowed by microwave background
constraints on the magnetic field amplitude. |
1408.0352v1 | Evolution and Consequences of Interacting CMEs of 2012 November 9-10 using STEREO/SECCHI and In Situ Observations
Wageesh Mishra 1 · Nandita Srivastava 1 · D. Chakrabarty 2
c © Springer
••••
Abstract Understanding of the kinematic evolution of Coronal Mass Ejections (CMEs) in the heliosphere is important to estimate their arrival time at the Earth. It is found that kinematics of CMEs can change when they interact or collide with each other as they propagate in the heliosphere. In this paper, we analyze the collision and post-interaction characteristics of two Earth-directed CMEs, launched successively on 2012 November 9 and 10, using white light imaging observations from STEREO/SECCHI and in situ observations taken from WIND spacecraft. We tracked two density enhanced features associated with leading and trailing edge of November 9 CME and one density enhanced feature associated with leading edge of November 10 CME by constructing J-maps. We found that the leading edge of November 10 CME interacted with the trailing edge of November 9 CME. We also estimated the kinematics of these features of the CMEs and found a significant change in their dynamics after interaction. In in situ observations, we identified distinct structures associated with interacted CMEs and also noticed their heating and compression as signatures of CMECME interaction. Our analysis shows an improvement in arrival time prediction of CMEs using their post-collision dynamics than using pre-collision dynamics. Estimating the true masses and speeds of these colliding CMEs, we investigated the nature of observed collision which is found to be close to perfectly inelastic. The investigation also places in perspective the geomagnetic consequences of the two CMEs and their interaction in terms of occurrence of geomagnetic storm and triggering of magnetospheric substorms.
Keywords: Coronal mass ejections, STEREO, Heliospheric Imagers
1 Udaipur Solar Observatory, Physical Research Laboratory,
Udaipur, India -313001
email:
[email protected] email:
[email protected]
Space and Atmospheric Sciences Division, Physical
2
Research Laboratory, Ahmedabad-380009, India email:
[email protected]
## 1. Introduction
Coronal Mass Ejections (CMEs) are huge magnetized plasma expulsions from the Sun into the heliosphere and are considered as primary drivers of geomagnetic storms and various other space weather effects (Gosling et al. , 1990; Echer et al. , 2008). It is therefore required to improve the physics and our understanding of evolution of CMEs in the solar wind, which is found to be dependent on heliospheric environment (Kilpua et al. , 2012; Vrˇnak s et al. , 2013). Various techniques based on stereoscopic observations which estimate the 3D kinematics of CMEs in the heliosphere have been developed (Mierla et al. , 2008; Thernisien, Vourlidas, and Howard, 2009; Thompson, 2009; Liu et al. , 2010a; Lugaz et al. , 2010; Davies et al. , 2013). Using these techniques, association of remote observations of CMEs with in situ observations has been carried out by Liu et al. (2010b); Temmer et al. (2011); Kilpua et al. (2012); M¨stl o et al. (2012); Mishra and Srivastava (2013); Mishra et al. (2014). Also, several models have been developed for predicting the arrival time of CMEs at 1 AU (Gopalswamy et al. , 2000; Vrˇ snak et al. , 2013). Sometimes, CMEs are observed to erupt in quick succession which under certain favorable initial conditions, can interact or merge with each other during their propagation in the heliosphere. In such a scenario, space weather prediction schemes may not be successful without taking into account the post-interaction CMEs characteristics. The compound streams (interaction of CME-CIR or CME-CME) were first inferred by Burlaga, Behannon, and Klein (1987) using observations from Helios and ISEEE 3 spacecraft. Their study suggested that a large geomagnetic storm may be associated with compound streams. The observational evidence for merging of CMEs (or cannibalism) was first given by Gopalswamy et al. (2001) based on the longwavelength radio and SOHO/LASCO observations. Thus, these interactions are of great importance from space weather point of view. Before the era of wide angle imaging far from the Sun, the understanding of involved physical mechanisms in CME-CME or CME-shock interaction was achieved mostly from magnetohydrodynamic (MHD) numerical simulations of the interaction of a shock wave with an magnetic cloud (MC) (Vandas et al. , 1997; Vandas and Odstrcil, 2004; Xiong et al. , 2006), the interaction of two ejecta (Gonzalez-Esparza, Santill´n, a and Ferrer, 2004; Lugaz, Manchester, and Gombosi, 2005; Wang et al. , 2005), and the interaction of two MCs (Xiong et al. , 2007; Xiong, Zheng, and Wang, 2009). Wang, Ye, and Wang (2003) have shown that a forward shock can cause an intense southward magnetic field of long duration in the preceding magnetic cloud. Such modifications in the preceding cloud are important for space weather prediction. Therefore, events involving CME-CME and CME-shock interactions are important candidates to investigate their kinematics and arrival time at near Earth.
After the launch of Solar TErrestrial RElation Observatory (STEREO) in 2006 (Kaiser et al. , 2008), it is possible to continuously image the CMEs and to determine the 3D locations of their features in the heliosphere and hence have direct evidence of CME interaction using images from Heliospheric Imager (HI) on Sun Earth Connection Coronal and Heliospheric Investigation (SECCHI) package (Howard et al. , 2008). However, immediately after the launch of
STEREO, during deep extended solar minimum, not many interacting CMEs were observed. As the current solar cycle progressed, interaction of CMEs now appears a fairly common phenomenon, in particular around solar maximum. The interacting CMEs of 2010 August 1 have been studied extensively by using primarily the STEREO white light imaging, near Earth in situ and, S-Waves radio observations (Harrison et al. , 2012; Liu et al. , 2012; M¨ ostl et al. , 2012; Temmer et al. , 2012; Mart´ ınez Oliveros et al. , 2012; Webb et al. , 2013). Also, Lugaz et al. (2012) have reported a clear deflection of 2010 May 23 - 24 CMEs after their interaction. As mentioned above, therefore, only a few studies on CME interaction have been reported. Several key questions regarding CME interaction need to be addressed, which are not well understood quantitatively,viz. (1) How the dynamics of CMEs change after interaction? What is the regime of interaction, i.e. elastic, inelastic or super-elastic? (Lugaz et al. , 2012; Shen et al. , 2012; Mishra and Srivastava, 2014) (2) What are the consequences of the interaction of CME-shock structure? How the overtaking shock changes the plasma and the magnetic field properties into preceding magnetic cloud? (Lugaz, Manchester, and Gombosi, 2005; Lugaz et al. , 2012; Liu et al. , 2012) (3) What are the favorable conditions for the CME cannibalism and the role of magnetic reconnection in it? (Gopalswamy et al. , 2001) (4) What is the possibility for production of reverse shock at CME-CME interaction site? (Lugaz, Manchester, and Gombosi, 2005) (5) Whether these interacted structures produce different geomagnetic consequences than individual CMEs, on their arrival to magnetosphere? (Farrugia et al. , 2006). (6) What are the favorable conditions for the CME's deflection and enhanced radio emission during CME-CME interaction? (Lugaz et al. , 2012; Mart´ ınez Oliveros et al. , 2012). In light of aforementioned questions and only a few studies reported, it is clear that the prediction of arrival time of interacting CMEs and association of remote observations of such CMEs with in situ is challenging.
In the present study, we focus on the identification, evolution and propagation of two CMEs launched on 2010 November 9 and 10, as they travel from the corona to the inner heliosphere. This study, in which the launch time of the selected CMEs are separated by about 14 hr, gives an opportunity to understand the CME-CME interaction unambiguously for simpler scenario, contrary to complex events around 2010 August 1, wherein four CMEs were launched in quick succession. We find that CMEs of November 9 and 10 are directed towards the Earth and their initial characteristics indicate a probability of their interaction. In Section 2.1, observations and estimated kinematics of these CMEs are presented to investigate if they interacted during their propagation in the heliosphere. In Section 2.2, nature of collision, momentum and energy transfer during collision phase of CMEs are described using their estimated true mass and kinematics. In Section 3.1, in situ observations of remotely tracked CME features are described. Using the derived kinematics combined with Drag Based Model (DBM) of Vrˇ snak et al. (2013), the arrival time and transit speed of tracked features of CMEs at L1 is predicted in Section 3.2. The geomagnetic response of interacted CMEs structures is described in Section 3.3. In Section 4, the results and discussions of the present study are emphasized and conclusions are given in Section 5.
## 2. Observations
For the present study of 2012 November 9-10, CME-CME interaction, we analyzed the white light images from SECCHI suite on-board NASA's twin STEREO (A & B) mission. SECCHI suite consists of an Extreme Ultraviolet Imager (EUVI), two coronagraphs (COR1 and COR2) and, two Heliospheric Imagers (HI1 and HI2). These instruments can image the heliosphere from near the Sun to the Earth and beyond. The EUVI (Field-of-view (FOV): 1-1.7 R glyph[circledot] ), COR1 (FOV: 1.5-4 R glyph[circledot] ) and COR2 (FOV: 2.5-15 R glyph[circledot] ) FOV are centered on the Sun. The HI1 and HI2 have their optical axis aligned in the ecliptic plane and are off-centered from the Sun at solar elongation of 14 ◦ and 53 7 , respectively. . ◦ HI1 and HI2 have wide FOV of 20 ◦ and 70 , respectively. With this identical ◦ imaging package (SECCHI) on twin spacecraft, a CME can be tracked from 0.4 to 88 7 . ◦ elongations. At the time of observations around November 9, the STEREO A & B were 127 ◦ westward and 123 ◦ eastward from the Sun-Earth line at a distance of 0.96 AU and 1.08 AU from the Sun, respectively. The in situ properties were investigated through data collected by instruments on-board Wind spacecraft (Lepping et al. , 1995; Ogilvie et al. , 1995).
## 2.1. Interaction of 2012 November 9-10 CMEs
In order to determine the true direction and speed of CMEs launched on November 9 and 10 in quick succession, 3D reconstruction technique is applied to CME features close to the Sun. We find that the CME of November 9 (hereinafter referred as, CME1) has a slow speed than fast speed November 10 CME (hereinafter referred as, CME2) and both move in a similar direction towards the Earth. These observations indicate the probability of their interaction in the heliosphere. Using SECCHI/HI observations, we attempt to investigate if these CMEs indeed interacted with each other in the heliosphere. We also attempt to identify various CME structures in the in situ observations taken at 1 AU.
## 2.1.1. 3D reconstruction in COR2 Field-of-view
Apartial halo CME1 (angular width of 276 ◦ ) was observed in SOHO/LASCO-C2 FOV at 15:12 UT on 2012 November 9 ( http://cdaw.gsfc.nasa.gov/CME list/ ). In the SECCHI/COR1-B and A, this CME was observed at 14:25 UT in the SW and SE quadrants, respectively. Another partial halo CME2 (angular width ∼ = 210 ) was observed in LASCO C2 FOV at 05:12 UT on November 10. This ◦ CME was detected by SECCHI/COR1-B and A in the SW and SE quadrants, respectively, at 05:05 UT on November 10, 2012.
Figure 1 shows true kinematics of CME1 and CME2 from 3D reconstruction of selected features along the leading edges of corresponding CMEs using tiepointing procedure (scc measure: Thompson 2009) on SECCHI/COR2 data. This tie-pointing procedure uses the concept of epipolar geometry (Inhester, 2006) and due to epipolar constraints, the identification of same feature in two images (from STEREO-A and B) is always possible along the same epipolar line in both images. Before implementing this technique, the processing of
SECCHI/COR2 images and creation of minimum intensity images and then its subtraction from the sequence of processed COR2 images was carried out following the approach described in Mierla et al. (2008); Srivastava et al. (2009). As the separation angle of the STEREO spacecraft is large, it is better to use two independent methods to confirm the results of 3D reconstruction. Therefore, we have also used the visual fitting of both CMEs in the SECCHI/COR2 FOV using the Graduated Cylindrical Shell (GCS) model (Thernisien, Vourlidas, and Howard, 2009; Thernisien, 2011) for the 3D reconstruction of these CMEs. This model represents the flux rope structure of CMEs with two shapes: the conical legs and the curved (tubular) fronts, also known as the 'hollow croissant'. We used contemporaneous image triplets of CME1 around 17:39 UT on November 9 from STEREO-A/COR2, STEREO-B/COR2 and SOHO/LASCO-C3. The best visual fit for CME1 is found in the direction of W02S14, with a half angle of 19 , tilt angle of 9 ◦ ◦ around the axis of symmetry of the model (i.e. rotated 9 ◦ anticlockwise out of the ecliptic plane), and an aspect ratio of 0.52. At this time, CME1 was at a distance of about 9.6 R glyph[circledot] from the Sun. Using the obtained fitted values of half angle and aspect ratio of CME1, its 3D face-on and edge-on angular width is estimated as 100 ◦ and 62 , respectively. We also carried out ◦ the visual fitting using GCS model for CME2 around 06:39 UT on November 10 by exploiting the concurrent image triplets of STEREO-A/COR2, STEREOB/COR2 and SOHO/LASCO-C3. The best fit for CME2 is obtained in the direction of W06S25, at a tilt angle of 9 ◦ around the axis of symmetry of the model, with a half angle of 12 ◦ , and an aspect ratio of 0.19. At this time CME2 was at a distance of 8.2 R glyph[circledot] from the Sun. Using the GCS model fitted values, the 3D face-on and edge-on angular width of CME2 is estimated as 46 ◦ and 22 , respectively. The GCS model fit for CME1 and CME2 is shown in the top ◦ and bottom panel of Figure 2, respectively. We find that the estimated latitude and longitude of both CMEs from GCS model and tie-pointing procedure are in agreement within few degrees. This shows that the results of both tie-pointing and GCS model are reliable and can be used for estimating kinematics of CMEs in the coronagraphic FOV.
From a study of 3D kinematics of the two CMEs using tie-pointing approach (scc measure: Thompson 2009), we found that CME1 is slow with a true speed of 620 km s -1 around 15 R glyph[circledot] while CME2 is faster with a speed of 910 km s -1 at a distance of approximately 15 R glyph[circledot] (Figure 1). Latitude and longitude of reconstructed features for both CMEs suggest that these were Earth-directed and could possibly interact in the interplanetary (IP) medium. If the speed of CMEs is assumed to be constant beyond the outer edge of COR2 FOV, then these CMEs should have collided at approximately 130 R glyph[circledot] on November 11 around 02:30 UT. However, previous studies have shown that the speed of a CMEcan significantly change after the coronagraphic FOV (Lindsay et al. , 1999; Gopalswamy et al. , 2000; Cargill, 2004; Manoharan, 2006; Vrˇ snak et al. , 2010). Therefore, further tracking of CME features in the heliosphere is required to determine the exact location and time of interaction of these CMEs.
## 2.1.2. Tracking of CMEs in HI Field-of-view
Evolution of CME1 and CME2 in the running difference images of COR2, HI1 and HI2 FOV is shown in Figure 3. To track and estimate the arrival times of the CMEs in the heliosphere using HI1 & 2 images, we constructed J-maps based on the method developed by Sheeley et al. (1999); Davies et al. (2009) and derived the variation of elongation of selected features with time. The details about extraction of strip of constant position angle, selection of bin size and adopted procedure for construction of J-maps and derivation of elongation angle from it has been described in section 3.1.1 of Mishra and Srivastava (2013). The positively inclined bright features in the J-maps (Figure 4) correspond to enhanced density structure of the CMEs. We tracked the leading and trailing edge (marked in green and red) of bright features corresponding to slow CME1 and leading edge (blue) of bright feature corresponding to the fast CME2. Prior to interpreting the tracked features of CME1 in J-map as its leading and trailing edge, respectively, the derived elongations corresponding to all the three tracked features were overplotted on the base difference HI1 images. On careful inspection of the sequence of these images, we noticed that the two tracked features of CME1 correspond to first density enhancement at the front and second density enhancement at the rear edge of CME1, respectively. The base difference HI1-A images with overplotted contours of elongation corresponding to tracked features of CME1 and CME2, are shown in Figure 5. It is to be noted that the term 'trailing edge' of CME1 does not correspond to the rear most portion of CME1 but corresponds to the part behind the front and is not an image artifact in J-map constructed from running difference images.
Since, the tracked features of CME2 were not observed well in STEREO B ecliptic J-map, we could not implement the stereoscopic reconstruction technique to estimate the CME kinematics. Instead, we used Harmonic Mean (HM) approximation which requires single spacecraft observation to derive the elongation variations with time for the tracked features which could then be converted into heliocentric distances from the Sun (Lugaz, Vourlidas, and Roussev, 2009). In the HMapproximation, geometry of a CME is considered as a self-similar expanding sphere attached to the Sun and its centre moves along a fixed longitude. The choice of the HM method is based on our earlier study on comparison of various reconstruction techniques involving both stereoscopic and single spacecraft observations (Mishra et al. , 2014). It was found that amongst the single spacecraft techniques, the HM method is the most suitable for arrival time prediction. Implementing the 3D reconstruction technique in Section 2.1.1 (Figure 1), we have shown that the estimated longitude of CME1 and CME2 are approximately 10 ◦ and 2 ◦ east of the Sun-Earth line. We assume that beyond the COR2 FOV, CMEs continue to propagate along the same direction i.e. we ignore any heliospheric longitudinal deflection of CMEs. Therefore, the aforesaid estimated values of longitude were used in the HM approximation to convert the derived elongation from J-maps to radial distance from the Sun.
Figure 6 shows the distance and speed estimated for different tracked features of the two CMEs. The estimated speed is derived from the adjacent distances using a numerical differentiation with three point Lagrange interpolation and
therefore has systematic variations due to slight errors in the estimated distance. It is clear that the leading edge (LE) of CME1 has higher speed ( ≈ 500 km s -1 ) than its trailing edge (TE) speed ( ≈ 350 km s -1 ), averaged over few data points at the entrance of HI1 FOV. Also, the LE and TE of CME1 have lower speeds than LE of CME2. LE of CME2 shows a large radial speed of approximately 1100 km s -1 (ecliptic speed = 950 km s -1 ) in the COR2 FOV i.e.(2.5-15 R glyph[circledot] ) (Figure 1). Beyond 10 R glyph[circledot] distance, the LE of CME2 continuously decelerates for ≈ 10 hours up to 46 R glyph[circledot] where its speed reduces to 430 km s -1 . The extreme fast deceleration of LE of the CME2, starting from COR2 FOV, seems to be due to possible interaction with the preceding CME1. It is likely that the CME1 had large spatial scale due to which the trailing plasma and magnetic fields from CME1 created sufficiently dense ambient medium acting as a huge drag force for CME2 which resulted in its observed deceleration. The extreme fast deceleration of CME2 LE can also be due to closed magnetic structure of CME1 which may act like a magnetic obstacle for the CME2 (Temmer et al. , 2012).
From estimated kinematics of tracked features in heliosphere (top and bottom panel of Figure 6), it is clear that around November 10 at 11:30 UT speed of CME1 TE started to increase from 365 km s -1 with simultaneous decrease of CME2 LE speed from 625 km s -1 . Such an observation of acceleration of CME1 and deceleration of CME2 provides an evidence for the commencement of collision (Temmer et al. , 2012; Lugaz et al. , 2012; Shen et al. , 2012; Mariˇi´ c c et al. , 2014; Temmer et al. , 2014; Mishra and Srivastava, 2014). We define collision phase, as the interval during which the two CMEs come in close contact with each other and show opposite trend of acceleration relative to one another until they attain an approximately equal speed or their trend of acceleration is reversed. Carefully observing the variations of speed of tracked CME1 TE and CME2 LE, the start and end boundary of collision phase is drawn as vertical lines in the bottom panel of Figure 6. We noticed that at the end of the collision phase around November 10 at 17:15 UT, speed of CME1 TE is ≈ 450 km s -1 and speed of CME2 LE is 430 km s -1 . In the beginning of the marked collision phase, CME1 TE is at a distance of 37 R glyph[circledot] and CME2 is at 30 R glyph[circledot] . At the end of the marked collision phase, the CME1 TE and CME2 LE are at a distance of 50 R glyph[circledot] and 46 R glyph[circledot] , respectively. Here, we must highlight that throughout the collision phase, the estimated heliocentric distance of CME1 TE and CME2 LE is never found to be equal. This is because of the large scale structure of CMEs and we suggest that the tracked TE of CME1 and LE of CME2 using J-maps, are not strictly the rearmost trail of CME1 and outermost front of CME2, respectively. Therefore, we had defined 'collision' when tracked features of CMEs (i.e. time-elongation tracks) come in close contact (separation of ≈ 5 R ) with one another. Such small separation between observed collided features glyph[circledot] is possible if forward shock driven by CME2 interacts with the rear most portion of the CME1 as also suggested by Temmer et al. (2012); Mariˇi´ c c et al. (2014).
## 2.2. Momentum, Energy Exchange and Nature of Collision
To understand the momentum exchange during collision of CMEs, we require their true masses. Since, appearance of a CME is due to Thomson scattered
photospheric light from the electrons in the CME (Minnaert, 1930; Billings, 1966; Howard and Tappin, 2009), therefore, the recorded scattered intensity can be converted into the number of electrons and hence mass of a CME can be estimated, if composition of CME is known. But an observer from different vantage points receives different amount of scattered light by the electrons, therefore, the true location of electrons of CME, i.e. propagation direction of CME must be known to estimate the true mass of CME. For a long time, mass of a CME was calculated assuming the CME location in the observer's plane of sky (Munro et al. , 1979; Poland et al. , 1981; Vourlidas et al. , 2000). Although, the propagation direction of CME is calculated in our study using tie-pointing (Thompson, 2009) and Forward modeling (Thernisien, Vourlidas, and Howard, 2009) methods and described in Section 2.1.1, but to avoid any bias of the reconstruction methods, we use the method of Colaninno and Vourlidas (2009) which is based on Thomson scattering theory, to estimate the true propagation direction and true mass of both the CMEs. We used the simultaneous image pair of SECCHI/COR2 to estimate the true mass of CMEs. First, we estimated the mass of CME i.e. M A and M B , from STEREO-A and STEREO-B locations. Then we took the ratio of equations (7) and (8) of Colaninno and Vourlidas (2009) and obtained a new expression as M /M A B = B e ( θ A ) /B e ( θ A +∆), where θ A is the angle of propagation direction of CME measured from the plane-ofsky (POS) of STEREO-A , B ( e θ A ) is the brightness of a single electron at an angular distance of θ A from the POS and ∆ is the separation angle between both STEREO-A and STEREO-B from the Sun-Earth line. Using the measured M A and M B in aforementioned expression, the value of θ A can be estimated. However, we noticed that multiple values of θ A correspond to same ratio of M A and M B but the correct value of θ A among them can be selected by visual inspection of CME in COR2 images. We obtained the true mass of CME using the derived value of θ A in equation (4) of Colaninno and Vourlidas (2009). The estimated mass of CME1 at 18:39 UT on November 9 at ≈ 15 R glyph[circledot] , from STEREO-A and STEREO-B vantage point is 4.60 × 10 12 kg and 2.81 × 10 12 kg, respectively. For CME2, the estimated mass at 07:24 UT on November 10 at ≈ 15 R glyph[circledot] , from STEREO-A and STEREO-B locations is 2.25 × 10 12 kg and 1.31 × 10 12 kg, respectively. The estimated propagation direction of CME1 and CME2 is 19 ◦ and 21 ◦ West from the Sun-Earth line, respectively. We estimated the true mass of CME1 and CME2 as M 1 = 4.66 × 10 12 kg and M 2 = 2.27 × 10 12 kg, respectively.
We assume that after crossing the COR2 FOV and during collision of CME1 and CME2, their estimated true masses (M 1 and M ) remain constant as es2 timated earlier. From the Figure 6, the observed velocity of CME1 and CME2 before the collision is estimated as (u ,u ) = (365,625) km s 1 2 -1 and observed velocity of CME1 and CME2 after the collision is (v ,v ) = (450,430) km s 1 2 -1 . To understand the nature of collision of large scale magnetized plasmoids, we attempt to estimate the coefficient of restitution ( e ) of colliding CME1 and CME2. The coefficient of restitution measures the bounciness (efficiency to rebound) of a pair of objects in collision and is defined as ratio of their relative velocity of separation to relative velocity of approach. Hence, for e < 1, e = 1, and e > 1 the collision is termed as inelastic, elastic and super-elastic and consequently
the kinetic energy of the system after the collision is found to decrease, equal and increase than before the collision, respectively. We restrict ourselves not to estimate the value of e directly by using the pre and post collision velocity of CMEs. This is because the velocities estimated from reconstruction method have some errors and do not guarantee conservation of momentum, a necessary condition of collision, therefore, can lead to erroneous estimation of e value. In our approach, considering the condition of conservation of total momentum for the collision, the velocity of CME1 and CME2 after the collision can be estimated theoretically (v 1 th ,v 2 th ), provided that velocity of CME1 and CME2 before the collision and the coefficient of restitution ( e ) value are known.
Using the velocities (u 1 ,u 2 ) and masses (M ,M ), we iterate for a range of 1 2 e values and estimate the (v 1 th ,v 2 th ) each time. Using this approach, the most suitable value of e , using which the theoretically estimated (v 1 th ,v 2 th ) is found to be closest to the observed (v 1 ,v 2 ) i.e. for which the variance,
σ = √ ( v 1 th -v 1 ) 2 +( v 2 th -v 2 ) 2 , is minimum, can be obtained. The details of the above approach and the equations used, are explained in Mishra and Srivastava (2014). Using the aforementioned approach, the value of e = 0.1 is found with (v 1 th ,v 2 th ) = (458,432) km s -1 and σ = 9. With this theoretically estimated e value, total kinetic energy of CMEs is found to decrease by 6.7% of its value before the collision. Using the estimated velocities of the CMEs before and after the collision (Figure 6), the value for e is calculated equal to 0.08, which is approximately same as obtained from iterations described above. Therefore, our analysis suggests that the observed collision between the CMEs are close to perfectly inelastic in nature. The kinetic energy of CME1 and CME2 before the collision was 3.1 × 10 23 joules and 4.4 × 10 23 joules, respectively. After the collision, based on the observed speeds, we found that the kinetic energy of CME1 increased by 51% to its value before the collision while the kinetic energy of CME2 decreased by 54.5% to its value before the collision. We also noticed that after the collision, momentum of CME1 increased by 23% and momentum of CME2 decreased by 31% to their values before the collision. Such calculations support the claim that significant exchange of kinetic energy and momentum take place during CME-CME collision (Temmer et al. , 2012; Lugaz et al. , 2012; Mariˇ ci´ c et al. , 2014; Shen et al. , 2012).
Considering the uncertainties of ± 100 km s -1 in the speed, we repeated the above analysis. Taking the (v ,v ) = (550,530), we found 1 2 e = 0, i.e. perfectly inelastic. But in this case, (v 1 th ,v 2 th ) is estimated as (450,450) km s -1 which highlight that theoretically estimated velocity is not close to the observed velocity (v 1 ,v 2 ). If we consider the (v ,v ) = (350,330), then it implies that the 1 2 velocities of both CMEs after the collision are less than their values before the collision. Therefore, such velocities also cannot be considered for our case (as defined in Section 2.1.2). As, there are many sources of errors in the estimation of true mass of CMEs, therefore, the effect of uncertainty in mass must also be examined before confirming the nature of observed collision. In our case, the mass ratio (M /M ) is equal to 2.05, varying this ratio between 0.5 to 3.0 in 1 2 a step of 0.25, and repeating the iterative procedure for estimating the most suitable value of e corresponding to each mass ratio, we obtain e close to 0. Figure 7 shows the variation of e and σ with varying mass ratio. It is clear that
despite taking large uncertainties in the mass of CME1 and CME2, the nature of observed collision remains close to perfectly inelastic for the case of interacting CMEs of 2012 November 9-10.
## 3. In Situ observations, Arrival time and Geomagnetic Response of 2012 November 9-10 CMEs
## 3.1. In Situ Identification of Tracked CME Features
We analyzed the in situ data taken from WIND spacecraft located at L1 to identify the tracked density enhanced features of CMEs. Figure 8 shows magnetic field and plasma measurements during November 12, 12:00 UT to November 15, 12:00 UT. The arrival of a forward shock (labeled as S) marked by a sudden enhancement in speed, temperature and density is noticed at 22:20 UT on November 12. The region between the first and second vertical line represents the turbulent sheath region. Based on the CME identification criteria of Zurbuchen and Richardson (2006), the region bounded between second vertical line at 08:52 UT on November 13 and third vertical line at 02:25 UT on November 14, is identified as CME structure. In this region, we observed expansion of CME characterized by a monotonic decrease in proton speed and temperature. Based on the predicted arrival times which is derived using the estimated kinematics of the remotely observed tracked features of CME1 and CME2 as inputs in the DBM (explained in Section 3.2), this region is associated with Earth-directed CME1 launched on November 9. During the passage of CME1, the magnetic field is observed to be high ( ≈ 20 nT), plasma beta is less than unity ( β < 1) with smooth rotation in magnetic field vector. Also, the latitude ( θ ) value of magnetic field vector decreased from 43 ◦ to -43 ◦ and its longitude ( φ ) decreased from 203 ◦ to 74 ◦ . Therefore, this region can be classified as a magnetic cloud (MC) and based on the observed arrival time it is associated with CME1. Due to the interaction of the CMEs, the region associated with CME1 is found to be at higher temperature than found generally for a normal isolated CME. After the third vertical line, magnetic field strength decreased reaching a minimum value of 6 nT around 4:00 UT on November 14. This interval of sudden drop in magnetic field is associated with a sudden rise in density, temperature and plasma β , suggestive of a possible magnetic hole (MH) (Tsurutani et al. , 2006) which is considered as a signature of magnetic reconnection (Burlaga and Lemaire, 1978). Another region of magnetic field depression from 08:05 UT to 10:15 UT on November 14, reaching a minimum value of 3 nT is also noticed. Corresponding to this minimum value of magnetic field, the plasma β and temperature is found to increase. The region during 03:45 UT - 08:05 UT on November 14, bounded between two distinct MH like structures, has enhanced magnetic field ( ≈ 15 nT), and plasma β less than unity. This region, between two MH, seems to be a magnetic field remnant of reconnecting CMEs structures. Based on the extremely elevated interval of plasma beta, temperature, observation of magnetic hole (MH) and probably sudden fast rotation in magnetic field vector during the region bounded between third and fourth vertical line, in Figure 8, this region
is identified as interaction region (labeled as IR) of CME1 TE with CME2 LE. Another structure is identified based on the elevated fluctuating magnetic field and temperature during 12:00 UT - 21:21 UT on November 14, bounded between fourth and fifth vertical line in Figure 8. During this interval, we noticed high magnetic field (9 nT) with no monotonic decrease in temperature and speed profile as well as plasma beta ( β ) is not less than unity. From these observations, i.e. lack of MC signatures and the short duration (9.5 hr) of the structure associated with CME2, we infer that the WIND spacecraft perhaps intersected the flank of CME2 (M¨ ostl et al. , 2010). This is also confirmed by the estimated latitude (-25 ◦ ) of CME2 using GCS model (described in Section 2.1.1). On examining the evolution of CME2 in HI1-A movie, we noticed that CME2 is directed towards the southern hemisphere which is a favorable condition for its flank encounter with in situ WIND spacecraft.
We observed that the magnitude of magnetic field in MC region is constant around ≈ 20 nT which may be due to the passage of shock from CME1. We also noticed that average temperature in first half of CME1 is high ( ≈ 10 5 K) than found in general ( ≈ 10 4 K). The high temperature of the CME1 may occur due to its collision with CME2, thereby resulting in its compression. Another possibility of high temperature of CME1 is due to the passage of forward shock driven by CME2 as reported in earlier studies by Lugaz, Manchester, and Gombosi (2005); Liu et al. (2012); Mariˇi´ c c et al. (2014); Mishra and Srivastava (2014). However, in the present case, the following CME2 is also observed with a high temperature ( ≈ 5 × 10 5 K), which has not been reported in earlier studies of interacting CMEs. We noticed significant high density at the front of CME1, possibly due to overall compression by sweeping of the plasma of CME1 at its leading edge by CME2 driven shock.
## 3.2. Arrival time of Tracked Features
We used the estimated speed at the last point of measurement (up to where CMEs could be tracked unambiguously in HI) and used it as input in the DBM developed by Vrˇ snak et al. (2013) to estimate the arrival time of tracked features at L1. Vrˇnak s et al. (2013) have shown that drag parameter lies in the range of (0.2 - 2.0) × 10 -7 km -1 . Using these values of the drag parameter in the DBM, the arrival time of CME at L1 can be predicted with a reasonably good accuracy (Vrˇ snak et al. , 2013; Mishra and Srivastava, 2013). The estimated speed, time and distance (v , t 0 0 and R glyph[circledot] ) of LE of CME1 (green track in J-map) are used as inputs in the DBM, corresponding to extreme range of drag parameter, to predict its arrival time and transit speed at L1. In situ observations show a peak in density, ≈ 0.5 hr after the shock arrival around 23:00 UT on November 12 with a transit speed of 375 km s -1 , which is expected to be the actual arrival of tracked LE feature corresponding to CME1. The predicted values of arrival time and transit speed of features and errors from the actual values are shown in Table 1.
For TE feature of CME1 and LE of CME2, we assume that they encounter the dense ambient solar wind medium created by the preceding CME1 LE. Therefore, their kinematics with the maximum value of the statistical range of the drag
parameter (2.0 × 10 -7 km -1 ), are used as inputs in the DBM. We consider that the TE of CME1 corresponds to the density enhancement at trailing front of CME1. At the rear edge of CME1, a density enhancement (12 particles cm -3 ) is observed around 23:30 UT in WIND observations on November 13, which is considered as the actual arrival of TE of CME1. The predicted arrival time and transit speed of this feature and errors therein, are given in Table 1. We further noticed that in situ data, an enhancement in density corresponding to LE of CME2 around 12:00 UT on November 14 (marked as arrival of CME2 with fourth vertical line in Figure 8) is observed which can be considered as the actual arrival time of LE of CME2. The predicted values of CME2 LE, using its kinematics with DBM, and errors therein are also listed in Table 1. The error for CME2 LE is large, but, several factors can lead to such large errors which are discussed in Section 4. It must be highlighted that if the 3D speed of CMEs estimated at final height in COR2 FOV is assumed as constant up to L1, then the predicted arrival time of CME1 and CME2 will be ≈ 10-16 hr and 44 hr earlier, respectively, than their predicted arrival times using post-collision speeds combined with DBM. This emphasizes the use of HI observations and post-collision speeds of CMEs as inputs in the DBM for improved arrival time prediction of interacting CMEs. Similarly, using HI observations, Colaninno, Vourlidas, and Wu (2013) have shown that linear fit of deprojected height-time above 50 R glyph[circledot] gives a half a day improvement over CME arrival time estimated using LASCO data.
## 3.3. Geomagnetic consequences of 2012 November 9-10 CMEs
As mentioned in the introduction section, very few studies have been dedicated to the study of the geomagnetic consequences of interacting CMEs in the past. The CMEs of 2012 November 9-10 resulted in a single strong geomagnetic storm with Dst index ≈ -108 nT at 8:00 UT on November 14, therefore it is important to investigate the impact of the interaction of the CME1 of November 9 with the CME2 of November 10 on the terrestrial magnetosphere-ionosphere system (MI) in details.
Figure 9(a-e) reveal the variations in the solar wind parameters in the Geocentric Solar Ecliptic (GSE) coordinate system during 2012 November 12-15. These parameters include solar wind proton density (in cc -1 ), velocity (in km s -1 , negative X-direction), ram pressure (in nPa), Z-component of the interplanetary magnetic field (IMF B z , in nT) and Y-component of the interplanetary electric field ( IEF y , in mV/m) respectively. These data with cadence of 1 minute are taken from the NASA GSFC CDAWeb ( www.cdaweb.gsfc.nasa. gov/istp public/ ). It is also important to note that the solar wind parameters presented in Figure 9(a-e) are corrected for propagation lag till the nose of the terrestrial bow shock. In order to compare the variations of these parameters with the magnetospheric and ionospheric parameters, additional time lags that account for the magnetosheath transit time and the Alfven transit time are calculated (Chakrabarty et al. , 2005). Therefore, the solar wind parameters presented in Figure 9(a-e) are corrected for the propagation lag, point by point, till ionosphere. Figure 9(f) represents the variation in the polar cap (PC) index.
The PC index is shown (Troshichev et al. , 2000) to capture the variations in the ionospheric electric field over polar region efficiently. Figure 9(g) shows the variation in the Sym-H (in nT) index which primarily represents the variation in the magnetospheric ring current (Iyemori and Rao, 1996). The Sym-H index is essentially the high temporal resolution (1 min) version of the Dst index. Figure 9(h) shows variations in westward (midnight sector) auroral electrojet (AL) current (in nT) which captures auroral substorm processes reasonably well. Further, in Figure 9, the arrival of the shock is denoted by S, different vertical lines mark the arrival of different features of CME1 and CME2 and are labeled in the same manner as in Figure 8.
Figure 9(a) reveals the arrivals of two distinctly enhanced density structures, first enhancement occurred during arrival of the shock-sheath region before CME1 LE (on November 12 20:00 UT to November 13 04:00 UT) and second enhancement occurred during arrival of the CME1 TE-IR region (November 13 12:00 UT to November 14 12:00 UT). During the arrival of shock-sheath region before CME1 LE, the peak density reached ≈ 68 cc -1 which is more than two times of the corresponding peak density ( ≈ 30 cc -1 ) observed during the arrival of CME1 TE-IR. Sharp enhancement in the solar wind velocity was also observed vis-a-vis the sharp density enhancement during arrival of shock-sheath region (Figure 9b) when the velocity reached from ≈ 300 km s -1 to ≈ 470 km s -1 . However, solar wind velocity did not change sharply and significantly during arrival of CME1 TE-IR. The changes in the density and the velocity resulted in changes in the solar wind ram pressure shown in Figure 9(c). The peak ram pressure during shock-sheath region before CME LE ( ≈ 18 nPa) was almost twice than that observed during CME1 TE-IR ( ≈ 9 nPa). Figure 9(d) reveals that IMF B z was predominantly southward in both density enhanced intervals although fast fluctuations were observed corresponding to the arrival of the shock-sheath region. The first peak in IMF B z is ≈ -8 nT before the arrival of the shock. Thereafter, IMF B z fluctuated sharply between ≈ -20 nT to ≈ +20 nT during arrival of shock-sheath region before arrival of CME1 LE. During CME1 TEIR, IMF B z reached a peak value of ≈ -18 nT. No significant change in IMF B z is observed during the passage of the CME2 on November 14-15, when its magnitude hovered around the zero line.
Figure 9(e) elicits the variation in IEF y during 2012 November 12-15. It is observed that the peak value of IEF y before the arrival of the shock was ≈ 2 mV/m. However, during the shock event, IEF y fluctuated between ±∼ 8 mV/m. During arrival of CME1 TE-IR, the peak value of IEF y reached ≈ 7.5 mV/m. In fact, similar to IMF B z , a sharp polarity change in IEF y was noticed at ∼ 09:40 UT on November 14. No significant change in IEF y is observed during the passage of the CME2 on November 14-15, when its magnitude hovered around the zero line. Figure 9(f) shows that the PC index increased during both density enhanced intervals and the peak value is ≈ 5 for both the intervals. Figure 9(g) shows variation in the Sym-H index which revealed the development of a geomagnetic storm during CME1 TE region. The Sym-H index reached a value of ≈ -115 nT during the passage of IR on November 14. The ring current activity was not significant during arrival of the shock-sheath region before CME1. Lastly,
Figure 9(h) elicits the variation in the AL index. It is seen that AL reached ≈ -600 nT during the arrival of shock-sheath before CME1 and ≈ -1400 nT during the arrival of CME1 TE-IR. Therefore, significant intensification of westward auroral electrojet occurred during the arrival of CME1 TE-IR.
The above observations (from Figure 9) reveal several interesting points. First, the magnitude of AL seems to remain unaffected by large amplitudes of fluctuations in IMF B z and IEF y during the shock-sheath region. Second and the most important point is that the duration of occurrence of the AL intensification during shock-sheath before CME1 LE and in CME1 TE-IR region were nearly identical with the duration of the southward IMF B z and positive IEF y phases. However, the AL amplitudes seem to be more closely correlated with the IEF y amplitudes during CME1 TE-IR region compared to shock-sheath region. Third, although the peak amplitudes of PC index were nearly same during shock-sheath region before CME1 LE and CME1 TE-IR region, the peak amplitudes of the AL index were significantly different during these two intervals. Fourth, the substorm activity seems to be over during the passage of the CME2, therefore, CME2 did not have any bearing on the triggering of substorms. Figure 9 demonstrates the direct role played by the positive IEF y (or southward IMF B z ) in the storm-time AL intensification particularly when the terrestrial magnetosphere encounters the CME1 TE and IR region. Further, the arrival of CME2 did not affect the terrestrial magnetosphere-ionosphere system.
As aforementioned, major geomagnetic response is noticed during the arrival of trailing edge of the preceding CME (CME1) and the interaction region (IR) of the two CMEs near the Earth. We also conclude that the following CME (CME2) failed to cause a significant geomagnetic activity possibly due to the fact that the in situ spacecraft encountered the flank of this CME as mentioned in Section 3.1. We understand that due to interaction and collision between the trailing edge of the preceding CME1 and the leading edge of the following CME2 as revealed in imaging observations (described in Section 2.1.2), the parameters responsible for geomagnetic activity were significantly intensified at CME1 rear edge and in interaction region found between CME1 and CME2. These results bring out the importance of CME-CME interaction in the formation of interaction region and its role in the significant development of geomagnetic disturbances.
## 4. Results and Discussion
In this paper, we have studied the interactions between two successively launched Earth directed CMEs from the Sun on 2012 November 9 and 10, respectively. The analysis involved remote sensing observations of CME propagation in inner heliosphere from SECCHI/COR2 and HI images and the associated in situ measurements from WIND spacecraft. The first CME (CME1) associated with a filament eruption is observed to be slower (600 km s -1 ) than the successive CME (CME2) launched on November 10 with a higher velocity (900 km s -1 ). Both the CMEs were launched in approximately the same direction towards the Earth, indicating a possibility of interaction because of their relative speeds and the same direction. Different features were tracked as brightness enhancements
in J-maps obtained from COR2 and HI images and subsequently HM technique was applied to estimate the kinematics of the tracked features of the CMEs. From the estimated kinematics, the site of collision of CMEs could be located at a distance of approximately 35 R glyph[circledot] at 12:00 UT on November 10, which is at least 85 R glyph[circledot] before and 15 hours earlier than as predicted by using the initial kinematics of CMEs in COR2 FOV. The estimated kinematics were then used as inputs in the DBM to determine the arrival times of the tracked features. After the interaction, CME2 transferred some of its velocity to CME1 and both were observed to propagate together, showing a distinct deceleration of CME2.
As described earlier, CME1 LE is found to be propagating with higher speed than CME1 TE. Therefore, CME1 LE (green track) did not collide with CME2 LE (blue track). With the start of collision phase, the LE of CME1 is also found to accelerate and after the collision phase, significant acceleration of CME1 LE is noticed. This may occur either because of sudden impact (push) from CME1 TE to CME1 LE during collision or due to the passage of a shock driven by the CME2 or a combined effect of both. Beyond the collision phase, up to an estimated distance of nearly 100 R glyph[circledot] , LE of CME2 was found moving behind the TE of CME1 and both these features propagated together decelerating slowly. For few hours after the observed collision phase, it is noticed that speed of the LE and TE of CME1 is slightly higher than LE of CME2, which can increase the observed separation of these structures. The J-maps show that both these features can be tracked for further elongations after interaction. However, due to the limitation of the HM method for higher elongations (Lugaz, Vourlidas, and Roussev, 2009), we restricted our measurements on the tracked features up to 100 R glyph[circledot] . Our analyses show an evidence that after the interaction, the two features did not merge. It appears that LE of CME2 interacted with TE of CME1 and continued to propagate with a reduced speed of 500 km s -1 . In the SECCHI/HI-B images these features cannot be well distinguished and therefore it is difficult to use any stereoscopic method to infer about a possible longitudinal deflection (based on stereoscopic methods) of these features after interaction.
During the collision between CME1 and CME2, we noticed a large deceleration of fast CME2 while relatively less acceleration of slow CME1, till both approached an equal speed. This is expected to occur if the mass of CME1 is larger than that of CME2, which is indeed the case. In Section 2.2 we have shown that the mass of CME1 is ≈ 2.0 times larger than the mass of CME2. This result is an important finding and is in agreement with the second scenario of interaction described in Lugaz, Vourlidas, and Roussev (2009). It also must be noted that we have estimated the mass of CMEs in COR2 FOV while interaction takes place in HI FOV. Further, we cannot ignore the possibility of increase in mass of CME due to mass accretion at its front via snowplough effect in the solar wind beyond COR FOV (DeForest, Howard, and McComas, 2013). We must also mention that in our calculation of momentum and energy transfer, although we use the total mass of CME1 but only a part of CME1 (i.e. TE of CME1) takes part in collision with CME2. Considering the uncertainties in the derived speeds and mass of the CMEs, our analysis reveals that the nature of collision remains close to perfectly inelastic. However, we also acknowledge that defining collision phase is crucial and seems to play an important role in this calculation. In this
context, it must be emphasized that marking of the start of collision is often difficult. This is because the following fast speed CME2 starts to decelerate, well before the actual merging or collision is revealed in imaging observations. This is also shown in earlier studies for other interacting CMEs (Temmer et al. , 2012; Shen et al. , 2012; Mariˇi´ c c et al. , 2014; Mishra and Srivastava, 2014). The interaction of CME2 with streamers or overlying coronal magnetic field lines (Temmer et al. , 2008) may be responsible for deceleration of following CME2 before its collision with CME1 takes place. However, the contribution of other factors, e.g., the solar wind drag acting on CME2, interaction of CME2 with trailing edge of preceding CME1, and magnetic interaction of CME2 with preceding CME1 cannot be completely ignored. Also, possibly a shock launched by the following CME2 can contribute in acceleration of the preceding CME1 before the actual merging of both CMEs. Therefore, different timing and large time-interval of acceleration of one CME and deceleration of other, prevent us to pinpoint the exact start and end of the collision phase.
We concluded that the collision of interacting CMEs of 2012 November 910 is close to perfectly inelastic in nature. Here, it is noted that we are dealing with large scale magnetized three-dimensional structures, however, in the present study we have reduced the calculations to one-dimensional collision which seems to be accurate as tracked features of both CME1 and CME2 are considered to move along approximately the same trajectory. This implies that there will be a small difference between velocity of tracked feature of one CME and its component along the direction of propagation (i.e. longitude) of tracked feature of other CME. We must admit that possibility of change in longitude of tracked features during their collision can partially contribute to some errors in the estimated speed. In light of earlier reported observations (Shen et al. , 2012; Temmer et al. , 2012; Mishra and Srivastava, 2014), it seems that collision of CMEs can occur in all i.e. elastic, inelastic or super-elastic regimes. An in-depth study is required further for understanding the nature of CME-CME collision which may depend on the characteristics of CMEs, locations and duration of collision phase in the heliosphere.
At the last point of measurement in HI FOV around 120 R glyph[circledot] , the estimated speed of TE of CME1 and LE of CME2 are approximately equal ( ≈ 470 km s -1 ). Therefore, both features are expected to arrive at L1 at approximately the same time. Comparing to in situ measured actual arrival time, the delayed arrival of LE of CME2 is possible due to higher drag force acting on it resulting in its deceleration. However, the propagation direction of CME2 in southward hemisphere, as noticed in HI1-A images and also estimated in COR2 FOV using 3D reconstruction, can also account for flank encounter of CME2 and thus its delayed sampling by WIND spacecraft at L1. This is consistent with our interpretation in Section 3.1. Since, in the HM method, the front of the CME is assumed spherical which may not be the real case for CME2, as it was moving with fast speed and has large probability to distort and flatten its front while interacting with structured solar wind (Odstrcil, Pizzo, and Arge, 2005). In this case, the estimated speed using HM method will lead to some error and can result in the delayed arrival of LE of CME2. Keeping all these issues in mind, we believe that it is quite probable that the tracked feature corresponding to LE of CME2
(blue track in J-map) is not sampled by in situ spacecraft. Therefore, erroneous predicted arrival time for LE of CME2 is related to incorrect identification of tracked CME2 LE in in situ data which is taken as a reference (actual) for remotely observed tracked CME2 LE.
The present study not only estimates the kinematics and arrival times based on HI observations and their application to the DBM model of CME propagation, it also attempts to associate tracked brightness enhancements as observed in remote sensing COR and HI (J-maps) with the in situ observations (M¨stl o et al. , 2010, 2011; Rollett et al. , 2012, 2013). The association of HI observations with in-situ measurements from WIND leads to many interesting results. In spite of two CMEs launched from the Sun in succession in the Earthward direction, we observe only one shock in in situ data which may suggest the merging of shocks driven by CME1 and CME2, if both CMEs would have driven shocks. However, such a claim cannot be made unless we are well familiar with the in situ signatures of merged shock and plasma structure following it. The sweeping of plasma to high density at the front of CME1 and its compressed heating is most likely due to passage of CME2 driven shock through the MC associated with CME1. Based on the predicted arrival time of tracked features, it seems that the CME2 driven shock and CME1-sheath region is tracked as CME1 LE in J-map. Therefore, we infer that CME1 LE is propagating probably into an unperturbed solar wind. Our study also provides a possibility of formation of interaction region (IR) at the junction of trailing edge of preceding CME and leading edge of following CME. We show that during collision of the CMEs, kinetic energy exchange up to 50% and momentum exchange between 23 to 30%. Our study also demonstrate that the arrival time prediction significantly improved using HIs on-board STEREO compared to COR2 observations, and also emphasizes the importance of understanding of post-collision kinematics in further improving the arrival time prediction for a reliable space weather prediction scheme.
Our study reveals clear signatures of interaction of these CMEs in remote and in situ observations and also helps in identification of separate structures corresponding to these CMEs. In spite of the interaction of the two CMEs in the interplanetary medium which generally results in complex structures as suggested by Burlaga, Behannon, and Klein (1987); Burlaga, Plunkett, and St. Cyr (2002), in our case, we could identify interacted CMEs as distinguished structures in WIND spacecraft data. Even after collision of these CMEs, they did not merge which may be possibly because of strong magnetic field and higher density of CME1 than CME2. This needs further confirmation, and therefore, it is worth to investigate what decides the formation of merged CME structure or complex ejecta during CME-CME interaction. Due to single point in situ observations of CME, we acknowledge the possibility of ambiguity in marking the boundaries of CMEs. In the present case, the boundaries for CME1 and IR are distinctly clear. Also, slight ambiguity in the boundary of CME2 (flank encounter) will not change our interpretation because the main geomagnetic response is caused due to enhanced negative Bz in trailing portion of CME1 and its extension in IR. Here, we also point out that temperature in IR is lesser than the temperature for CME2 region but temperatures in both regions are
elevated as compared to a normal non-interacting CME. The observations of unexpected larger temperature in CME2 region than IR region may be due to the possibility that sheath region of CME2 is intersected by in situ spacecraft, as we have interpreted.
The association of geomagnetic storms with isolated single CMEs has been carried out extensively for a long time (Tsurutani et al. , 1988; Gosling et al. , 1991; Gosling, 1993; Gonzalez et al. , 1989, 1994; Echer et al. , 2008; Richardson and Cane, 2011). However, only few studies have been dedicated to understand the role of interacting CMEs in the generation of geomagnetic storms (Burlaga, Behannon, and Klein, 1987; Farrugia et al. , 2006; Wang, Ye, and Wang, 2003; Xiong et al. , 2006). Our study is important as it focuses on the role of interacting CMEs in the generation of geomagnetic storms as well as substorms. In the context of substorms, our study highlights that persistence of IMF Bz in the southward direction is more important than the amplitude in driving the substorm activity as manifested by the AL intensification. Using WINDMI model, Mays et al. (2007) have shown that the IP shock and sheath features for CMEs contribute significantly to the development of storms and substorms. But, in our study, sharp and large southward excursions in the midst of fluctuating IMF Bz associated with shock (shock-sheath region before CME1) were found less effective in producing strong substorm activity. Therefore, further investigations are required regarding characteristics (geometry, intensity) of shock and preceding CME in the context of triggering of substorms, as has been shown in earlier studies (Jurac et al. , 2002; Wang et al. , 2003). Regarding geomagnetic storms, our study suggests that the trailing edge of preceding CME (CME1) and IR formed between two interacting CMEs are efficient candidates for intense geomagnetic storms.
Another interesting aspect regarding substorms noticed in our study is that the nearly equal amplitude responses of the PC index corresponding to the shocksheath region (IMF Bz sharply fluctuating between southward and northward directions) preceding the CME1 interval vis-`-vis the CME1 interval (IMF Bz a steadily turning southward). This is interesting as the responses of Sym-H and AL during these two intervals are quite different in terms of amplitudes of variations. Janzhura, Troshichev, and Stauning (2007) have inferred that the PC index for the sunlit polar cap (summer hemisphere, high polar ionospheric conductance) responds mainly to the geo-effective interplanetary electric field whereas the PC index for the dark winter cap (winter hemisphere, low polar ionospheric conductance) responds better to the particle precipitation in the auroral zone like the AE and AL indices. It is to be noted here that, in the present study, no distinction is made between the variations of PC index in the northern (PCN) and southern (PCS) hemispheres. On the other hand, it is known that AL index is constructed based on magnetometer observations around the northern auroral oval only. The discrepancy between the variations in the PC index and the AL index may, therefore, suggest towards the hemispherical asymmetry in the response of polar ionosphere corresponding to this event. This is not unexpected as asymmetric auroral intensities in the northern and southern hemispheres are reported in literature (Laundal and Østgaard, 2009). Further investigations are needed to understand this aspect.
## 5. Conclusions
The CMEs of November 9 and 10, provide us a rare opportunity to investigate the consequences of CME-CME interaction. A combination of heliospheric imaging and in situ observations are used for improving our understanding of CME kinematics, post-collision characteristics and nature of collision. Main results of our study can be concluded as follows:
- i) The analysis of propagation kinematics as obtained from J-maps provide evidence that the CME1 and CME2 collide at 35 R glyph[circledot] much earlier than by using the estimated kinematics in the COR2 FOV. This emphasizes the importance of heliospheric imaging particularly, for interacting CMEs and ascertaining their impact and arrival at the Earth.
- ii) Our analysis shows that post interaction and collision kinematics is required for a better prediction of CME arrival time at 1 AU. In the present case of November 9 and 10 CMEs, the speeds and momentum of CMEs changed from 23% to 30% compared to their values before the collision. Our results also highlight that estimated kinematics, in particular after collisions are important to combine with DBM for improving the estimation of arrival times of different features of CME which experience different drag forces during their propagation through the heliosphere. Based on this study, we conclude that CMEs cannot be treated as completely isolated magnetized plasma blobs, especially when they are launched in quick succession.
- iii) Using estimated mass and kinematics before and after the collision, we estimated the total kinetic energy of the system before and after the collision. We found that total kinetic energy of the system decreased by 6.7% to its value before the collision and the nature of collision is close to perfectly inelastic. The study also supports the idea that CME interaction or collision can lead to the heating and compression of both preceding and following CMEs.
- iv) It is clear that if the trailing edge of November 9 CME would not have been tracked using imaging observation, one could not have witnessed the collision. Further, trailing edge of this CME had strong negative component of magnetic field for long duration ( ≈ 13 hr) which may be due to its collision with the leading edge of November 10 CME. Therefore, our study reveals that tracking of different features of CME seems to be necessary for better understanding of CME-CME interaction.
- v) It is observed in the present case that the persistence of IMF B z in the southward direction is more important (rather than the amplitude) in driving the substorm activity as manifested in the AL intensification. Sharp and equally large southward excursions in the midst of fluctuating IMF B z associated with shock were found less effective in producing equally strong substorm activity. It is identified that the interaction region (IR), formed due to collision between CME1 TE and CME2 LE, has intensified plasma and magnetic field parameters which are responsible for major geomagnetic activity.
The present case study of interacting CMEs of 2012 November 9-10, highlight the importance of heliospheric imaging for estimating the kinematics of CME features before and after their collision and interactions. Our study support that
interacting CMEs can result in strong geomagnetic storms and also in substorms. Further statistical study of such interacting CMEs is required to understand their nature of collision and to investigate the characteristics of the CMEs (mass, strength and orientation of magnetic field, speed and direction of propagation, duration of collision phase) which are responsible for their interaction.
Acknowledgements We acknowledge the UK Solar System Data Centre for providing the processed Level-2 STEREO/HI data. The in situ measurements of solar wind data from ACE and WIND spacecraft were obtained from NASA CDAweb ( http://cdaweb.gsfc.nasa.gov/ ). We also acknowledge the use of DBM, developed by Bojan Vr˘nak and available at s http://oh.geof. unizg.hr/CADBM/cadbm.php , in this study. We also thank the reviewer for useful comments, which helped improve the manuscript. The work by N.S. partially contributes to the research for European Union Seventh Framework Programme (FP7/2007-2013) for the Coronal Mass Ejections and Solar Energetic Particles (COMESEP) project under Grant Agreement No. 263252.
Table 1. First column shows the tracked features of CMEs and second column shows their estimated kinematics (by HM technique) which is used as input to the DBM. The predicted arrival time and transit speed of the tracked features at L1, corresponding to the extreme range of the drag parameter used in the DBM, is shown in the third and fourth column. Errors in predicted arrival time and speed, based on comparison with in situ arrival time and speeds, are shown in column fifth and sixth. The errors in arrival time with negative (positive) sign indicate that predicted arrival is earlier (later) than the actual arrival time. The errors in transit speed with negative (positive) sign indicate that the predicted transit speed is lesser (greater) than the transit speed measured in situ.
| Tracked Features | Kinematics as in- puts in DBM [t 0 , R 0 (R glyph[circledot] ), v 0 (km s - 1 )] | Predicted arrival time (UT) using kinematics + DBM [ γ = 0.2 - 2.0 (10 - 7 km - 1 )] | Predicted transit speed (km s - 1 ) at L1 [ γ =0.2 - 2.0 (10 - 7 km - 1 )] | Error in predicted ar- rival time (hr) [ γ = 0.2 - 2.0 (10 - 7 km - 1 )] | Error in predicted speed (km s - 1 ) [ γ = 0.2 - 2.0 (10 - 7 km - 1 )] |
|--------------------|------------------------------------------------------------------------------------|----------------------------------------------------------------------------------------|------------------------------------------------------------------------------|----------------------------------------------------------------------------|--------------------------------------------------------------------------|
| CME1 LE | Nov 11 13:42, 545, 137 | Nov 12 18:10 - Nov 13 00:30 | 490 to 380 | -5 to 1.5 | 115 to 5 |
| CME1 TE | Nov 11 18:35, 120, 470 | Nov 13 15:25 | 375 | -8 | -15 |
| CME2 LE | Nov 12 00:30, 124, 455 | Nov 13 19:40 | 375 | -16 | -35 |
Figure 1. From top to bottom panels, true height, velocity, acceleration, longitude and latitude of selected feature along the leading edge as derived from Tie-Pointing method have been plotted as a function of time for CME1 (left) and CME2 (right)
Figure 2. The top and bottom rows show the GCS model wireframe overlaid on the CME1 and CME2 images, respectively. The triplet of concurrent images around 17:39 UT on November 9 and around 06:39 UT on November 10 corresponding to CME1 and CME2, respectively, are from STEREO/COR2-B (left), SOHO/LASCO-C3 (middle) and STEREO/COR2-A (right).
Earth PA:
2012-11-10T02.09.01.003 STEREO Hl1 Spacecraft Longitude (HEEQ): 127.190 Spacecraft Latitudee (HEEQ):
Earth PA:
91.260
2012-11-11722.09.21.003 STEREO A HI2 Longitude (HEEQ) 127,.290 STEREO\_A Latitude (HEEQ): 7.189
7.230
Earth PA: 91.280
2012-11-10710.09.01.004 STEREO HI1 Spacecraft Longitude (HEEQ): 127.209 Spacecraft Latitudee (HEEQ):
Earth PA:
91.300
2012-11-12T04.09.21.004 STEREO HI2 Longitude (HEEQ) 127,31 STEREO HI2 Latitude (HEEQ):
Earth PA:
91.500
Earth PA: 91.530
Figure 3. Evolution of CMEs in running difference images of COR2, HI1 and HI2 FOV is shown in upper, middle and lower panels respectively. Left and right panels show observations from STEREO/SECCHI Ahead spacecraft at two different times. Contours of elongation (green) and position angle (blue) is overlaid on the images. In each image, the postion angle is overlaid in interval of 10 ◦ and the horizontal red line is along the ecliptic at the position angle of Earth. In upper panel (both left and right) the vertical red line marks the 0 ◦ position angle. In each panels, the C1 and C2 corresponds to CME1 and CME2.
## STEREO HI and COR-2 Spacecralt A: Ecliptic
Figure 4. Time-elongation map (J-map) using the COR2 and HI observations of STEREO/SECCHI spacecraft during interval of 2012 November 9-14 is shown. The features corresponding to CME1 leading edge (LE), CME1 trailing edge (TE) and CME2 leading edge are (LE) tracked and over plotted on the J-map with green, red and blue, respectively. The red rectangle (rightmost) is an enlarged plot of the red rectangle (on the left) which clearly shows that the red and blue tracks meet in HI1 FOV.
Figure 5. The contours of derived elongation of tracked CME1 LE (green), CME1 TE (red) and CME2 LE (blue) features from the J-map is overplotted on the base difference HI1-A images. In each image, the position angle (sky blue) is overlaid in interval of 10 ◦ and the horizontal red line is along the ecliptic at the position angle of Earth.
Figure 6. In the top and bottom panels, distance and speed of the tracked features CME1 LE, CME1 TE and CME2 LE as marked in Figure 4 with green, red and blue, respectively, are shown. Speeds of these features are calculated from the differentiation of adjacent distance points using the three point Lagrange interpolation (shown with ∆ symbol in the bottom panel). Speed of these features are estimated by differentiating the first order polynomial fit to estimated distance points for an interval of approximately 5 hr (shown in solid line in the bottom panel). In the top and bottom panels, vertical lines show the error bars. We have assumed a fractional error of 5% in estimated distance which is used to determine the uncertainties in the speeds. From the left, the two vertical lines (black) mark the start and end of the collision phase.
Figure 7. The best suited coefficient of restitution ( e ) corresponding to different mass ratios of CME1 and CME2 are shown with + symbol and corresponding variance ( σ ) in velocity is shown with ∆ symbol.
Figure 8. From top to bottom, total magnetic field magnitude, z-component of magnetic field, proton density, proton temperature, proton speed, latitude and longitude of magnetic field vector, and plasma beta ( β ) is shown for the time interval of 12:00 UT on November 12 to 12:00 UT on November 15. From the left, first, second, third, fourth and fifth vertical (dashed) lines mark the arrival of shock, CME1 leading edge (LE), CME1 trailing edge (TE), CME2 LE and CME2 TE, respectively. Interaction region (IR) is shown during the interval of third and fourth vertical lines.
Figure 9. Plot shows variations of geomagnetic field parameters during the interval 2012 November 12-15. Left panels show, top to bottom: Proton density (n/cc), Velocity (km/s), Ram Pressure (nPa) and B z component (nT). Right panels show, from top to bottom: interplanetary electric field's 'y' component IEF y (mV/m), polar cap (PC) index, Sym-H (nT) and AL index (nT). From the left, vertical lines and their labels are as defined in Figure 8.
## References
Billings, D.E.: 1966, A guide to the solar corona , San Diego, CA: Academic.
Burlaga, L.F., Lemaire, J.F.: 1978, Interplanetary magnetic holes - Theory. J. Geophys. Res. 83 , 5157 - 5160. doi: 10.1029/JA083iA11p05157 .
Burlaga, L.F., Behannon, K.W., Klein, L.W.: 1987, Compound streams, magnetic clouds, and major geomagnetic storms. J. Geophys. Res. 92 , 5725 - 5734. doi: 10.1029/JA092iA06p05725 .
Burlaga, L.F., Plunkett, S.P., St. Cyr, O.C.: 2002, Successive CMEs and complex ejecta. J. Geophys. Res. 107 , 1266. doi: 10.1029/2001JA000255 .
- Cargill, P.J.: 2004, On the Aerodynamic Drag Force Acting on Interplanetary Coronal Mass Ejections. Solar Phys. 221 , 135 - 149. doi: 10.1023/B:SOLA.0000033366.10725.a2 .
- Chakrabarty, D., Sekar, R., Narayanan, R., Devasia, C.V., Pathan, B.M.: 2005, Evidence for the interplanetary electric field effect on the OI 630.0 nm airglow over low latitude. Journal of Geophysical Research (Space Physics) 110 , 11301. doi: 10.1029/2005JA011221 .
- Colaninno, R.C., Vourlidas, A.: 2009, First Determination of the True Mass of Coronal Mass Ejections: A Novel Approach to Using the Two STEREO Viewpoints. Astrophys. J. 698 , 852 - 858. doi: 10.1088/0004-637X/698/1/852 .
Colaninno, R.C., Vourlidas, A., Wu, C.C.: 2013, Quantitative comparison of methods for predicting the arrival of coronal mass ejections at Earth based on multiview imaging. J. Geophys. Res. 118 , 6866 - 6879. doi: 10.1002/2013JA019205 .
Davies, J.A., Harrison, R.A., Rouillard, A.P., Sheeley, N.R., Perry, C.H., Bewsher, D., Davis, C.J., Eyles, C.J., Crothers, S.R., Brown, D.S.: 2009, A synoptic view of solar transient evolution in the inner heliosphere using the Heliospheric Imagers on STEREO. Geophys. Res. Lett. 36 , 2102. doi: 10.1029/2008GL036182 .
Davies, J.A., Perry, C.H., Trines, R.M.G.M., Harrison, R.A., Lugaz, N., M¨ ostl, C., Liu, Y.D., Steed, K.: 2013, Establishing a stereoscopic technique for determining the kinematic properties of solar wind transients based on a generalised self-similarly expanding circular geometry. Astrophys. J. 776 , 1.
DeForest, C.E., Howard, T.A., McComas, D.J.: 2013, Tracking Coronal Features from the Low Corona to Earth: A Quantitative Analysis of the 2008 December 12 Coronal Mass Ejection. Astrophys. J. 769 , 43. doi: 10.1088/0004-637X/769/1/43 .
Echer, E., Gonzalez, W.D., Tsurutani, B.T., Gonzalez, A.L.C.: 2008, Interplanetary conditions causing intense geomagnetic storms (Dst = -100 nT) during solar cycle 23 (1996-2006). J. Geophys. Res. 113 , 5221. doi: 10.1029/2007JA012744 .
Farrugia, C.J., Jordanova, V.K., Thomsen, M.F., Lu, G., Cowley, S.W.H., Ogilvie, K.W.: 2006, A two-ejecta event associated with a two-step geomagnetic storm. J. Geophys. Res. 111 , 11104. doi: 10.1029/2006JA011893 .
Gonzalez, W.D., Gonzalez, A.L.C., Tsurutani, B.T., Smith, E.J., Tang, F.: 1989, Solar windmagnetosphere coupling during intense magnetic storms (1978-1979). J. Geophys. Res. 94 , 8835 - 8851. doi: 10.1029/JA094iA07p08835 .
Gonzalez, W.D., Joselyn, J.A., Kamide, Y., Kroehl, H.W., Rostoker, G., Tsurutani, B.T., Vasyliunas, V.M.: 1994, What is a geomagnetic storm? J. Geophys. Res. 99 , 5771 - 5792. doi: 10.1029/93JA02867 .
Gonzalez-Esparza, A., Santill´n, a A., Ferrer, J.: 2004, A numerical study of the interaction between two ejecta in the interplanetary medium: one- and two-dimensional hydrodynamic simulations. Annales Geophysicae 22 , 3741 - 3749. doi: 10.5194/angeo-22-3741-2004 .
Gopalswamy, N., Lara, A., Lepping, R.P., Kaiser, M.L., Berdichevsky, D., St. Cyr, O.C.: 2000, Interplanetary acceleration of coronal mass ejections. Geophys. Res. Lett. 27 , 145 - 148. doi: 10.1029/1999GL003639 .
Gopalswamy, N., Yashiro, S., Kaiser, M.L., Howard, R.A., Bougeret, J.-L.: 2001, Radio Signatures of Coronal Mass Ejection Interaction: Coronal Mass Ejection Cannibalism? Astrophys. J. Lett. 548 , L91-L94. doi: 10.1086/318939 .
Gosling, J.T.: 1993, The solar flare myth. J. Geophys. Res. 98 , 18937 - 18950. doi: 10.1029/93JA01896 .
- Gosling, J.T., Bame, S.J., McComas, D.J., Phillips, J.L.: 1990, Coronal mass ejections and large geomagnetic storms. Geophys. Res. Lett. 17 , 901 - 904. doi: 10.1029/GL017i007p00901 .
- Gosling, J.T., McComas, D.J., Phillips, J.L., Bame, S.J.: 1991, Geomagnetic activity associated with earth passage of interplanetary shock disturbances and coronal mass ejections. J. Geophys. Res. 96 , 7831 - 7839. doi: 10.1029/91JA00316 .
Harrison, R.A., Davies, J.A., M¨stl, C., Liu, Y., Temmer, M., Bisi, M.M., Eastwood, J.P., de o Koning, C.A., Nitta, N., Rollett, T., Farrugia, C.J., Forsyth, R.J., Jackson, B.V., Jensen, E.A., Kilpua, E.K.J., Odstrcil, D., Webb, D.F.: 2012, An Analysis of the Origin and Propagation of the Multiple Coronal Mass Ejections of 2010 August 1. Astrophys. J. 750 , 45. doi: 10.1088/0004-637X/750/1/45 .
Howard, R.A., Moses, J.D., Vourlidas, A., Newmark, J.S., Socker, D.G., Plunkett, S.P., Korendyke, C.M., Cook, J.W., Hurley, A., Davila, J.M., Thompson, W.T., St Cyr, O.C., Mentzell, E., Mehalick, K., Lemen, J.R., Wuelser, J.P., Duncan, D.W., Tarbell, T.D., Wolfson, C.J., Moore, A., Harrison, R.A., Waltham, N.R., Lang, J., Davis, C.J., Eyles, C.J., Mapson-Menard, H., Simnett, G.M., Halain, J.P., Defise, J.M., Mazy, E., Rochus, P., Mercier, R., Ravet, M.F., Delmotte, F., Auchere, F., Delaboudiniere, J.P., Bothmer, V., Deutsch, W., Wang, D., Rich, N., Cooper, S., Stephens, V., Maahs, G., Baugh, R., McMullin, D., Carter, T.: 2008, Sun Earth Connection Coronal and Heliospheric Investigation (SECCHI). Space Sci. Rev. 136 , 67 - 115. doi: 10.1007/s11214-008-9341-4 .
Howard, T.A., Tappin, S.J.: 2009, Interplanetary Coronal Mass Ejections Observed in the Heliosphere: 1. Review of Theory. Space Sci. Rev. 147 , 31 - 54. doi: 10.1007/s11214-0099542-5 .
Inhester, B.: 2006, Stereoscopy basics for the STEREO mission. arXiv:astro-ph/0612649 .
Iyemori, T., Rao, D.R.K.: 1996, Decay of the Dst field of geomagnetic disturbance after substorm onset and its implication to storm-substorm relation. Annales Geophysicae 14 , 608 - 618. doi: 10.1007/s00585-996-0608-3 .
Janzhura, A., Troshichev, O., Stauning, P.: 2007, Unified PC indices: Relation to isolated magnetic substorms. Journal of Geophysical Research (Space Physics) 112 , 9207. doi: 10.1029/2006JA012132 .
Jurac, S., Kasper, J.C., Richardson, J.D., Lazarus, A.J.: 2002, Geomagnetic disturbances and their relationship to Interplanetary shock parameters. Geophys. Res. Lett. 29 , 1463. doi: 10.1029/2001GL014034 .
Kaiser, M.L., Kucera, T.A., Davila, J.M., St. Cyr, O.C., Guhathakurta, M., Christian, E.: 2008, The STEREO Mission: An Introduction. Space Sci. Rev. 136 , 5 - 16. doi: 10.1007/s11214007-9277-0 .
Kilpua, E.K.J., Mierla, M., Rodriguez, L., Zhukov, A.N., Srivastava, N., West, M.J.: 2012, Estimating Travel Times of Coronal Mass Ejections to 1 AU Using Multi-spacecraft Coronagraph Data. Solar Phys. 279 , 477 - 496. doi: 10.1007/s11207-012-0005-x .
Laundal, K.M., Østgaard, N.: 2009, Asymmetric auroral intensities in the Earth's Northern and Southern hemispheres. Nature 460 , 491 - 493. doi: 10.1038/nature08154 .
Lepping, R.P., Ac˜na, M.H., Burlaga, L.F., Farrell, W.M., Slavin, J.A., Schatten, K.H., Maru iani, F., Ness, N.F., Neubauer, F.M., Whang, Y.C., Byrnes, J.B., Kennon, R.S., Panetta, P.V., Scheifele, J., Worley, E.M.: 1995, The Wind Magnetic Field Investigation. Space Sci. Rev. 71 , 207 - 229. doi: 10.1007/BF00751330 .
Lindsay, G.M., Luhmann, J.G., Russell, C.T., Gosling, J.T.: 1999, Relationships between coronal mass ejection speeds from coronagraph images and interplanetary characteristics of associated interplanetary coronal mass ejections. J. Geophys. Res. 104 , 12515 - 12524. doi: 10.1029/1999JA900051 .
Liu, Y.D., Luhmann, J.G., M¨stl, C., Martinez-Oliveros, J.C., Bale, S.D., Lin, R.P., Harrison, o R.A., Temmer, M., Webb, D.F., Odstrcil, D.: 2012, Interactions between Coronal Mass Ejections Viewed in Coordinated Imaging and in situ Observations. Astrophys. J. Lett. 746 , L15. doi: 10.1088/2041-8205/746/2/L15 .
Liu, Y., Davies, J.A., Luhmann, J.G., Vourlidas, A., Bale, S.D., Lin, R.P.: 2010a, Geometric Triangulation of Imaging Observations to Track Coronal Mass Ejections Continuously Out to 1 AU. Astrophys. J. Lett. 710 , L82-L87. doi: 10.1088/2041-8205/710/1/L82 .
Liu, Y., Thernisien, A., Luhmann, J.G., Vourlidas, A., Davies, J.A., Lin, R.P., Bale, S.D.: 2010b, Reconstructing Coronal Mass Ejections with Coordinated Imaging and in Situ Observations: Global Structure, Kinematics, and Implications for Space Weather Forecasting. Astrophys. J. 722 , 1762 - 1777. doi: 10.1088/0004-637X/722/2/1762 .
Lugaz, N., Manchester, W.B. IV, Gombosi, T.I.: 2005, Numerical Simulation of the Interaction of Two Coronal Mass Ejections from Sun to Earth. Astrophys. J. 634 , 651 - 662. doi: 10.1086/491782 .
Lugaz, N., Vourlidas, A., Roussev, I.I.: 2009, Deriving the radial distances of wide coronal mass ejections from elongation measurements in the heliosphere - application to CME-CME interaction. Annales Geophysicae 27 , 3479 - 3488. doi: 10.5194/angeo-27-3479-2009 .
Lugaz, N., Hernandez-Charpak, J.N., Roussev, I.I., Davis, C.J., Vourlidas, A., Davies, J.A.: 2010, Determining the Azimuthal Properties of Coronal Mass Ejections from MultiSpacecraft Remote-Sensing Observations with STEREO SECCHI. Astrophys. J. 715 , 493 - 499. doi: 10.1088/0004-637X/715/1/493 .
Lugaz, N., Farrugia, C.J., Davies, J.A., M¨ ostl, C., Davis, C.J., Roussev, I.I., Temmer, M.: 2012, The Deflection of the Two Interacting Coronal Mass Ejections of 2010 May 23-24 as Revealed by Combined in Situ Measurements and Heliospheric Imaging. Astrophys. J. 759 , 68. doi: 10.1088/0004-637X/759/1/68 .
Manoharan, P.K.: 2006, Evolution of Coronal Mass Ejections in the Inner Heliosphere: A Study Using White-Light and Scintillation Images. Solar Phys. 235 , 345 - 368. doi: 10.1007/s11207006-0100-y .
Mariˇ ci´, c D., Vrˇ snak, B., Dumbovi´ c, M., Zic, T., Roˇa, D., Hrˇina, D., Luli´, S., Romˇtajn, ˇ s z c s I., Buˇ si´, c I., Salamon, K., Temmer, M., Rollett, T., Veronig, A., Bostanjyan, N., Chilingarian, A., Mailyan, B., Arakelyan, K., Hovhannisyan, A., Muji´, N.: 2014, Kinematics of c Interacting ICMEs and Related Forbush Decrease: Case Study. Solar Phys. 289 , 351 - 368. doi: 10.1007/s11207-013-0314-8 .
Mart´ ınez Oliveros, J.C., Raftery, C.L., Bain, H.M., Liu, Y., Krupar, V., Bale, S., Krucker, S.: 2012, The 2010 August 1 Type II Burst: A CME-CME Interaction and its Radio and White-light Manifestations. Astrophys. J. 748 , 66. doi: 10.1088/0004-637X/748/1/66 .
Mays, M.L., Horton, W., Kozyra, J., Zurbuchen, T.H., Huang, C., Spencer, E.: 2007, Effect of Interplanetary Shocks on the AL and Dst Indices. Geophys. Res. Lett. 34 , 11104. doi: 10.1029/2007GL029844 .
Mierla, M., Davila, J., Thompson, W., Inhester, B., Srivastava, N., Kramar, M., St. Cyr, O.C., Stenborg, G., Howard, R.A.: 2008, A Quick Method for Estimating the Propagation Direction of Coronal Mass Ejections Using STEREO-COR1 Images. Solar Phys. 252 , 385 396. doi: 10.1007/s11207-008-9267-8 .
Minnaert, M.: 1930, On the continuous spectrum of the corona and its polarisation. With 3 figures. (Received July 30, 1930). Z. Astrophys. 1 , 209.
Mishra, W., Srivastava, N.: 2013, Estimating the Arrival Time of Earth-directed Coronal Mass Ejections at in Situ Spacecraft Using COR and HI Observations from STEREO. Astrophys. J. 772 , 70. doi: 10.1088/0004-637X/772/1/70 .
Mishra, W., Srivastava, N.: 2014, Morphological and kinematic evolution of three interacting coronal mass ejections of 2011 february 13-15. Astrophys. J. , Under review.
Mishra, W., Srivastava, N., , Davies, J.A.: 2014, A comparison of reconstruction methods for the estimation of coronal mass ejections kinematics based on secchi/hi observations. Astrophys. J. , in press. doi: 10.1088/0004-637X/784/1/1 .
M¨ ostl, C., Temmer, M., Rollett, T., Farrugia, C.J., Liu, Y., Veronig, A.M., Leitner, M., Galvin, A.B., Biernat, H.K.: 2010, STEREO and Wind observations of a fast ICME flank triggering a prolonged geomagnetic storm on 5-7 April 2010. Geophys. Res. Lett. 37 , 24103. doi: 10.1029/2010GL045175 .
M¨ ostl, C., Rollett, T., Lugaz, N., Farrugia, C.J., Davies, J.A., Temmer, M., Veronig, A.M., Harrison, R.A., Crothers, S., Luhmann, J.G., Galvin, A.B., Zhang, T.L., Baumjohann, W., Biernat, H.K.: 2011, Arrival Time Calculation for Interplanetary Coronal Mass Ejections with Circular Fronts and Application to STEREO Observations of the 2009 February 13 Eruption. Astrophys. J. 741 , 34. doi: 10.1088/0004-637X/741/1/34 .
M¨ ostl, C., Farrugia, C.J., Kilpua, E.K.J., Jian, L.K., Liu, Y., Eastwood, J.P., Harrison, R.A., Webb, D.F., Temmer, M., Odstrcil, D., Davies, J.A., Rollett, T., Luhmann, J.G., Nitta, N., Mulligan, T., Jensen, E.A., Forsyth, R., Lavraud, B., de Koning, C.A., Veronig, A.M., Galvin, A.B., Zhang, T.L., Anderson, B.J.: 2012, Multi-point Shock and Flux Rope Analysis of Multiple Interplanetary Coronal Mass Ejections around 2010 August 1 in the Inner Heliosphere. Astrophys. J. 758 , 10. doi: 10.1088/0004-637X/758/1/10 .
Munro, R.H., Gosling, J.T., Hildner, E., MacQueen, R.M., Poland, A.I., Ross, C.L.: 1979, The association of coronal mass ejection transients with other forms of solar activity. Solar Phys. 61 , 201 - 215. doi: 10.1007/BF00155456 .
Odstrcil, D., Pizzo, V.J., Arge, C.N.: 2005, Propagation of the 12 May 1997 interplanetary coronal mass ejection in evolving solar wind structures. Journal of Geophysical Research (Space Physics) 110 , 2106. doi: 10.1029/2004JA010745 .
Ogilvie, K.W., Chornay, D.J., Fritzenreiter, R.J., Hunsaker, F., Keller, J., Lobell, J., Miller, G., Scudder, J.D., Sittler, E.C. Jr., Torbert, R.B., Bodet, D., Needell, G., Lazarus, A.J., Steinberg, J.T., Tappan, J.H., Mavretic, A., Gergin, E.: 1995, SWE, A Comprehensive Plasma Instrument for the Wind Spacecraft. Space Sci. Rev. 71 , 55 - 77. doi: 10.1007/BF00751326 .
Poland, A.I., Howard, R.A., Koomen, M.J., Michels, D.J., Sheeley, N.R. Jr.: 1981, Coronal transients near sunspot maximum. Solar Phys. 69 , 169 - 175. doi: 10.1007/BF00151264 .
Richardson, I.G., Cane, H.V.: 2011, Geoeffectiveness (Dst and Kp) of interplanetary coronal mass ejections during 1995-2009 and implications for storm forecasting. Space Weather 9 , 7005. doi: 10.1029/2011SW000670 .
Rollett, T., M¨stl, C., Temmer, M., Veronig, A.M., Farrugia, C.J., Biernat, H.K.: 2012, Cono straining the Kinematics of Coronal Mass Ejections in the Inner Heliosphere with In-Situ Signatures. Solar Phys. 276 , 293 - 314. doi: 10.1007/s11207-011-9897-0 .
Rollett, T., Temmer, M., M¨stl, C., Lugaz, N., Veronig, A.M., M¨stl, U.V.: 2013, Assessing o o the Constrained Harmonic Mean Method for Deriving the Kinematics of ICMEs with a Numerical Simulation. Solar Phys. 283 , 541 - 556. doi: 10.1007/s11207-013-0246-3 .
Sheeley, N.R., Walters, J.H., Wang, Y.-M., Howard, R.A.: 1999, Continuous tracking of coronal outflows: Two kinds of coronal mass ejections. J. Geophys. Res. 104 , 24739 - 24768. doi: 10.1029/1999JA900308 .
Shen, C., Wang, Y., Wang, S., Liu, Y., Liu, R., Vourlidas, A., Miao, B., Ye, P., Liu, J., Zhou, Z.: 2012, Super-elastic collision of large-scale magnetized plasmoids in the heliosphere. Nature 8 , 923 - 928. doi: 10.1038/nphys2440 .
Srivastava, N., Inhester, B., Mierla, M., Podlipnik, B.: 2009, 3D Reconstruction of the Leading Edge of the 20 May 2007 Partial Halo CME. Solar Phys. 259 , 213 - 225. doi: 10.1007/s11207009-9423-9 .
Temmer, M., Veronig, A.M., Vrˇ snak, B., Ryb´k, J., G¨m¨ry, P., Stoiser, S., Mariˇi´, D.: 2008, a o o c c Acceleration in Fast Halo CMEs and Synchronized Flare HXR Bursts. Astrophys. J. Lett. 673 , L95-L98. doi: 10.1086/527414 .
Temmer, M., Rollett, T., M¨stl, C., Veronig, A.M., Vrˇ snak, B., Odstrˇil, D.: 2011, Influence o c of the Ambient Solar Wind Flow on the Propagation Behavior of Interplanetary Coronal Mass Ejections. Astrophys. J. 743 , 101. doi: 10.1088/0004-637X/743/2/101 .
Temmer, M., Vrˇ snak, B., Rollett, T., Bein, B., de Koning, C.A., Liu, Y., Bosman, E., Davies, J.A., M¨stl, C., ˇ ic, T., Veronig, A.M., Bothmer, V., Harrison, R., Nitta, N., Bisi, M., Flor, o Z O., Eastwood, J., Odstrcil, D., Forsyth, R.: 2012, Characteristics of Kinematics of a Coronal Mass Ejection during the 2010 August 1 CME-CME Interaction Event. Astrophys. J. 749 , 57. doi: 10.1088/0004-637X/749/1/57 .
Temmer, M., Veronig, A.M., Peinhart, V., Vrˇ snak, B.: 2014, Asymmetry in the CME-CME Interaction Process for the Events from 2011 February 14-15. Astrophys. J. 785 , 85. doi: 10.1088/0004-637X/785/2/85 .
Thernisien, A.: 2011, Implementation of the Graduated Cylindrical Shell Model for the Threedimensional Reconstruction of Coronal Mass Ejections. ApJS 194 , 33. doi: 10.1088/00670049/194/2/33 .
Thernisien, A., Vourlidas, A., Howard, R.A.: 2009, Forward Modeling of Coronal Mass Ejections Using STEREO/SECCHI Data. Solar Phys. 256 , 111 - 130. doi: 10.1007/s11207009-9346-5 .
Thompson, W.T.: 2009, 3D triangulation of a Sun-grazing comet. Icarus 200 , 351 - 357. doi: 10.1016/j.icarus.2008.12.011 .
Troshichev, O.A., Lukianova, R.Y., Papitashvili, V.O., Rich, F.J., Rasmussen, O.: 2000, Polar cap index (PC) as a proxy for ionospheric electric field in the near-pole region. Geophys. Res. Lett. 27 , 3809 - 3812. doi: 10.1029/2000GL003756 .
Tsurutani, B.T., Smith, E.J., Gonzalez, W.D., Tang, F., Akasofu, S.I.: 1988, Origin of interplanetary southward magnetic fields responsible for major magnetic storms near solar maximum (1978-1979). J. Geophys. Res. 93 , 8519 - 8531. doi: 10.1029/JA093iA08p08519 .
Tsurutani, B.T., Gonzalez, W.D., Gonzalez, A.L.C., Guarnieri, F.L., Gopalswamy, N., Grande, M., Kamide, Y., Kasahara, Y., Lu, G., Mann, I., McPherron, R., Soraas, F., Vasyliunas, V.: 2006, Corotating solar wind streams and recurrent geomagnetic activity: A review. Journal of Geophysical Research (Space Physics) 111 , 7. doi: 10.1029/2005JA011273 .
Vandas, M., Odstrcil, D.: 2004, Acceleration of electrons by interacting CMEs. Astron. Astrophys. 415 , 755 - 761. doi: 10.1051/0004-6361:20031763 .
Vandas, M., Fischer, S., Dryer, M., Smith, Z., Detman, T., Geranios, A.: 1997, MHD simulation of an interaction of a shock wave with a magnetic cloud. J. Geophys. Res. 102 , 22295 22300. doi: 10.1029/97JA01675 .
Vourlidas, A., Subramanian, P., Dere, K.P., Howard, R.A.: 2000, Large-Angle Spectrometric Coronagraph Measurements of the Energetics of Coronal Mass Ejections. Astrophys. J. 534 , 456 - 467. doi: 10.1086/308747 .
Vrˇ snak, B., ˇ ic, Z T., Falkenberg, T.V., M¨stl, C., Vennerstrom, S., Vrbanec, D.: 2010, The o role of aerodynamic drag in propagation of interplanetary coronal mass ejections. Astron. Astrophys. 512 , A43. doi: 10.1051/0004-6361/200913482 .
Vrˇ snak, B., Zic, T., Vrbanec, D., Temmer, M., Rollett, T., M¨stl, C., Veronig, A., Calogovi´, ˇ o ˇ c J., Dumbovi´ c, M., Luli´, S., Moon, Y.-J., Shanmugaraju, A.: 2013, Propagation of Interc planetary Coronal Mass Ejections: The Drag-Based Model. Solar Phys. 285 , 295 - 315. doi: 10.1007/s11207-012-0035-4 .
Wang, Y.M., Ye, P.Z., Wang, S.: 2003, Multiple magnetic clouds: Several examples during March-April 2001. Journal of Geophysical Research (Space Physics) 108 , 1370. doi: 10.1029/2003JA009850 .
Wang, Y.M., Ye, P.Z., Wang, S., Xue, X.H.: 2003, An interplanetary cause of large geomagnetic storms: Fast forward shock overtaking preceding magnetic cloud. Geophys. Res. Lett. 30 , 1700. doi: 10.1029/2002GL016861 .
Wang, Y., Zheng, H., Wang, S., Ye, P.: 2005, MHD simulation of the formation and propagation of multiple magnetic clouds in the heliosphere. Astron. Astrophys. 434 , 309 - 316. doi: 10.1051/0004-6361:20041423 .
Webb, D.F., M¨ ostl, C., Jackson, B.V., Bisi, M.M., Howard, T.A., Mulligan, T., Jensen, E.A., Jian, L.K., Davies, J.A., de Koning, C.A., Liu, Y., Temmer, M., Clover, J.M., Farrugia, C.J., Harrison, R.A., Nitta, N., Odstrcil, D., Tappin, S.J., Yu, H.-S.: 2013, Heliospheric Imaging of 3D Density Structures During the Multiple Coronal Mass Ejections of Late July to Early August 2010. Solar Phys. 285 , 317 - 348. doi: 10.1007/s11207-013-0260-5 .
Xiong, M., Zheng, H., Wang, S.: 2009, Magnetohydrodynamic simulation of the interaction between two interplanetary magnetic clouds and its consequent geoeffectiveness: 2. Oblique collision. Journal of Geophysical Research (Space Physics) 114 , 11101. doi: 10.1029/2009JA014079 .
Xiong, M., Zheng, H., Wang, Y., Wang, S.: 2006, Magnetohydrodynamic simulation of the interaction between interplanetary strong shock and magnetic cloud and its consequent geoeffectiveness: 2. Oblique collision. Journal of Geophysical Research (Space Physics) 111 , 11102. doi: 10.1029/2006JA011901 .
Xiong, M., Zheng, H., Wu, S.T., Wang, Y., Wang, S.: 2007, Magnetohydrodynamic simulation of the interaction between two interplanetary magnetic clouds and its consequent geoeffectiveness. Journal of Geophysical Research (Space Physics) 112 , 11103. doi: 10.1029/2007JA012320 .
Zurbuchen, T.H., Richardson, I.G.: 2006, In-Situ Solar Wind and Magnetic Field Signatures of Interplanetary Coronal Mass Ejections. Space Sci. Rev. 123 , 31 - 43. doi: 10.1007/s11214006-9010-4 . | 10.1007/s11207-014-0625-4 | [
"Wageesh Mishra",
"Nandita Srivastava",
"D. Chakrabarty"
] | 2014-08-02T08:57:40+00:00 | 2014-08-02T08:57:40+00:00 | [
"astro-ph.SR"
] | Evolution and Consequences of Interacting CMEs of 2012 November 9-10 using STEREO/SECCHI and In Situ Observations | Understanding of the kinematic evolution of Coronal Mass Ejections (CMEs) in
the heliosphere is important to estimate their arrival time at the Earth. It is
found that kinematics of CMEs can change when they interact or collide with
each other as they propagate in the heliosphere. In this paper, we analyze the
collision and post-interaction characteristics of two Earth-directed CMEs,
launched successively on 2012 November 9 and 10, using white light imaging
observations from STEREO/SECCHI and in situ observations taken from WIND
spacecraft. We tracked two density enhanced features associated with leading
and trailing edge of November 9 CME and one density enhanced feature associated
with leading edge of November 10 CME by constructing J-maps. We found that the
leading edge of November 10 CME interacted with the trailing edge of November 9
CME. We also estimated the kinematics of these features of the CMEs and found a
significant change in their dynamics after interaction. In in situ
observations, we identified distinct structures associated with interacted CMEs
and also noticed their heating and compression as signatures of CME-CME
interaction. Our analysis shows an improvement in arrival time prediction of
CMEs using their post-collision dynamics than using pre-collision dynamics.
Estimating the true masses and speeds of these colliding CMEs, we investigated
the nature of observed collision which is found to be close to perfectly
inelastic. The investigation also places in perspective the geomagnetic
consequences of the two CMEs and their interaction in terms of occurrence of
geomagnetic storm and triggering of magnetospheric substorms. |
1408.0353v4 | ## Superconducting properties variation with A15 composition in Nb3Sn
## Yingxu Li a) and Yuanwen Gao
Key Laboratory of Mechanics on Environment and Disaster in Western China, The Ministry of Education of China, Lanzhou, Gansu 730000, P. R. China
Department of Mechanics and Engineering Science, College of Civil Engineering and Mechanics, Lanzhou University,
Lanzhou, Gansu 730000, P. R. China a)
Corresponding author: [email protected]; Tel: 0086 931 891 4359; Fax: 0086 931 891 4561.
PACS: 74.70.Ad; 74.25.fc; 74.20.Fg.
Abstract : We extend the Ginzburg-Landau-Abrikosov-Gor'kov (GLAG) theory to account for the variation of the upper critical field c2 H with Sn content in A15-type Nb3Sn. c2 H at the vicinity of the critical temperature c T is related quantitatively to the electrical resistivity, specific heat capacity coefficient and c T , with inclusion of the electron-phonon coupling correction, Pauli paramagnetic limiting and martensitic phase transformation of A15 lattices. c2 H near c T is then extrapolated to c2 (0) at temperature 0K, and H c2 (0) versus tin content agrees well with experiment results. We find H that, as Sn content gradually approaches the stoichiometry, Nb3Sn undergoes a transition from the dirty limit to clean limit, split by the phase transformation boundary. The H T -phase boundary and the pinning force show different behaviors in the cubic and tetragonal phase. Applying the theoretical formulas in technical Nb3Sn wires, we obtain the dependence of the composition gradient on the superconducting properties variation in the A15 layer, as well as the curved tail at vicinity of c2 H in the Kramer plot of the Nb3Sn wire. This gives a better understanding of the inhomogeneous-composition inducing discrepancy between the results by the state-of-art scaling laws and experiments.
Keywords: Upper critical field; Composition inhomogeneity; Superconductivity; A15 type Nb3Sn.
## 1. Introduction
At present, conventional low-temperature superconductors such as Nb3Sn have been extensively applied in high-energy and nuclear physics, as well as in magnetic resonance imaging systems [1]. Inhomogeneity of Sn content is inevitable in practical Nb3Sn conductors, due to the high vapor pressure of Sn at the formation temperature of the A15 phase in a solid-state diffusion reaction [2, 3]. The tin variation in a conductor covers nearly the entire A15 phase field of binary Nb 1 Sn β β -with 0.18 ~ 0.255 [4]. In a Nb3Sn conductor, the Sn gradient across the A15 layer has a β = remarkable impact on the local variation of the superconducting properties. Experiments show that, the upper critical field c2 B varies almost linearly at ~5T per at% between 19.5% and ~24 at% Sn, and the transition temperature c T
versus tin concentration β also exhibits a linear relation within nearly the entire A15 phase field [3]. When β approaching stoichiometry of Nb3Sn, 24at% , β ≈ c T (or c2 B ) versus β no longer keeps the linear relationship due to the lattice softening (decreasing in phonon frequency) [5], which means the varying Sn content leads to the spontaneous cubic-tetragonal phase transformation of A15 lattices. Recent experiment [6] also investigates the stress-induced transformation behaviors at low temperatures for polycrystalline Ti-51.8Ni (at.%) specimens. The alloying addition (Ti and/or Ta), which is introduced in most modern high-field Nb3Sn conductors for increasing the electrical resistivity, suppressing the martensitic phase transformation and thus raising c2 B , also be approximately linear with the Sn content β [7]. As for other additives, the ZrO2 precipitates in Nb3Sn wires could refine Nb3Sn grain size such that change the pinning behavior [8]. Recent experiments also demonstrate a strong correlation between the composition and the superconducting properties in YBa2Cu3O7-x films [9].
For Nb3Sn wires, the scaling law and the experiment show a disagreement in the field dependence of the pinning force at high reduced fields [10]. One of the reasons could be the inhomogeneity of microstructure and composition [10]. This also explains the observation that the scaling field lies below the experimental c2 B of Nb3Sn wires. In fact, the scaling field for the critical current reflects the average properties over the wire volume; it thus relates to the local variation of the critical field dependent on the composition gradient [11]. Cooley and the coauthors simulate the effect of Sn composition gradients on the superconducting properties of powder-in-tube (PIT) Nb3Sn strand, with an ideal structure modelled by concentric shells with varying Sn content [7]. They found that different Sn profiles have a pronounced effect on the irreversibility fields defined by the extrapolation of Kramer plots K H , and also that Sn gradients reduce the elementary pinning force, K H and the critical current density c J [7].
The temperature dependence of the upper critical field c2 ( ) in inhomogeneous Nb3Sn conductors, as the H T field-temperature phase boundary, is comprehensively investigated by Godeke et al . [2]. It is concluded that, the complete field-temperature phase boundary can be described with the simplest form of the Maki-DeGennes (MDG) relation, and independent of compositional variation, measuring technique, criterion for the critical field and strain state [2].
Various experiments have recently been conducted to investigate the dependence of the superconductivity and magnetic properties of 3 Nb Sn samples on Sn content and disorder [12-14]. The underlying physics for the superconducting properties variation with the A15 composition in Nb3Sn is however still not very clear. The already-existing physical formulas for this dependence are insufficient for describing the state-of-art experimental results of binary Nb3Sn samples and practical Nb3Sn wires. As for practical engineering significance, describing the superconducting properties dependence on the A15 composition in theory will facilitate the understanding of the optimization for the critical current density c J , since tin composition and possible additives are important for the very high c J now achieved in commercial strands [15, 16].
In this paper, we extend the Ginzburg-Landau-Abrikosov-Gor'kov (GLAG) theory to account for the A15 composition dependence of the superconducting properties in Nb3Sn. In this theory, the occurrence of c2 H is due to the
breaking of orbital pair and the Pauli paramagnetic limiting. Two scattering mechanisms should be considered in type-II superconductors like Nb3Sn: electron-transport scattering by impurity (disorder) and spin-orbit scattering. As for A15 Nb3Sn, composition inhomogeneity and its deviation from stoichiometry may cause defect and site disorder in Nb3Sn lattices [4], which contribute most to the scattering by impurity. Based on this physical picture, c2 H is correlated to the superconductivity parameters (the coherence length, the London penetration depth and etc.) as well as the scattering characteristics (the mean free path of electron transporting). As for the strong-coupling superconductor like Nb3Sn, one should include the correction for electron-phonon interaction to the weak-coupling BCS value. Since the microscopic parameters mentioned above cannot be determined directly, we correlate them to the transition temperature, the normal-state resistivity and the coefficient of electronic heat capacity. The three material parameters have been extensively measured as a function of tin content [3, 4, 17, 18]. In this manner, we can determine the superconducting properties variation with composition concentration. The following section will present the detail.
## 2. Upper critical field variation with composition concentration
## 2.1. Upper critical field at vicinity of superconducting transition temperature
The best quality Nb3Sn samples, with highest transition temperature c T 's and resistance ratios ( (300K) / (20K) ), have very narrow resistive transitions [18]. The transitions tend to broaden in high fields. RRR ρ ρ = Measured ternary PIT wire also exhibits a narrow transition at a wide range of fields and temperatures [2]. At the vicinity of c T , the upper critical fields c2 ( ) are thus nearly the same, and independent of the selected critical-state H T criterion. We are then allowed to determine c2 H near c T uniquely. The resistivities ρ near c T are approximated as c ( ) , measured at temperatures just above T ρ c T , and the coefficient of electronic heat capacity γ remains a constant which satisfies the low-temperature heat capacity formula without undergoing a specific heat jump.
In the microscopic physical concept, the scattering of impurity (Appendix A and Fig. A1) enters into the superconductivity of type II superconductors by changing the Ginzburg-Landau (GL) parameter κ at the vicinity of c T . We have κ in two limiting cases, clean c 1 ( ) ( ) ( ) without scattering effect and T T T κ κ χ = dirty dirty c 2 ( ) ( ) ( ) T T T κ κ χ = relevant to scattering, where clean c L 0 ( ) 0.96 (0) / T κ δ ξ ≈ and dirty c L ( ) 0.72 (0) / T l κ δ = (see Appendix B for GLAG description of the superconductivity parameters). Here, L (0) is the London penetration depth of the magnetic field at δ 0K. 0 ξ is the standard coherence length. l is the mean free path of electron transporting. clean κ and dirty κ refer to the clean limit ( 0 l ξ << ) and dirty limit ( 0 l ξ >> ), respectively. 1 ( ) and T χ 2 ( ) represent the temperature dependence T χ of κ for the clean limit and dirty limit, respectively; calculations show that they vary little with T near c T , 1 2 ( ) ( ) 1 [19]. T T χ χ ≈ ≈
In light of the classic proposal [18, 20], the superconductivity parameters L (0) , δ 0 ξ and the scattering parameter l involved in the GL parameter κ can be linked to three independent material parameters, the transport scattering
resistivity ρ , the low-temperature coefficient of electronic heat capacity γ and the superconducting transition temperature c T . Assuming a spherical Fermi surface and isotropic metal, we are allowed to use the electron conduction formula Eq. (A1) and the electron heat capacity relation Eq. (C1) in the GLAG description of the superconductivity parameters (Appendix C), and then express l , 0 ξ , L (0) , δ clean κ and dirty κ as a function of ρ , γ and c T , Eqs. (C15) -(C19). Including the correction for the anisotropy in Nb3Sn (Appendix D), one must consider the change of Fermi surface shape from the isotropic model, which leads to the corrected expressions, Eqs. (D3)-(D6).
FIG. 1. Variations of material parameters c T , ρ , γ , D θ and 2 B c (0) / ( ) with Sn content k T ∆ β in the A15 range of binary Nb 1 Sn β β -. The experimental data are extracted from [3]. The solid lines represent the global fits to the experimental data in the A15 range.
Let us consider the dependences of c T , ρ and γ with Sn content β . Experiments [3] show that c T variation with β is essentially linear up to ~24 at% ; while between 24 and 25 at% Sn content, c T versus β has a saturation, Fig. 1. The A15 type lattice undergoes a spontaneous cubic-tetragonal transformation in this range. The lattice softening, as one of the consequences, implies a decreasing in the lattice stiffness ( D B D k ω θ = ) such that it reduces c T according to
McMillan c T equation [Eq. (6)]. The normal-state resistivity ρ , as measured just above c T , decreases moderately with increase of β within 18~24 at%. Approaching 25 at% , β = ( ) exhibits an obviously stronger decrease. ρ β Below 24 at%, the coefficient of electronic heat capacity, γ , changes linearly with β by ~1.5mJ -2 1 K mol -⋅ ⋅ per at% Sn. Note that resistivity measurements [21] also show pressure-induced resistivity saturation in Fe17wt%Si. Following the treatment to the experimental data in [3], we use global fits to the experimental c T , γ and ρ versus β in the entire A15 range (Fig. 1 and Table 1),
$$T _ { c }, \ \gamma \text{ or } \rho = \begin{cases} a _ { 1 } \beta + b _ { 1 }, & \text{ linear fit}, \\ a _ { 2 } \beta ^ { 3 } + b _ { 2 } \beta ^ { 2 } + c _ { 2 } \beta + d _ { 2 }, \text{ cubic fit}. \end{cases}$$
TABLE 1. Material parameters variation with Sn content β [at.%].
| | Linear fitting | ( 1 1 a b χ β = + ) | | |
|-----------------------|------------------|-----------------------------------------|---------|--------|
| χ | 1 a | 1 b | | |
| γ | 1.3473 | -20.983 | | |
| D θ | -14.018 | 584.94 | | |
| B c 2 (0) / ( ) k T ∆ | 0.24043 | -1.2087 | | |
| | Cubic fitting | ( 3 2 2 2 2 2 a b c d χ β β β = + + + ) | | |
| χ | 2 a | 2 b | 2 c | 2 d |
| c T | -0.065923 | 4.3157 | -91.612 | 641.31 |
| ρ | -0.021449 | 1.2923 | -26.735 | 196.6 |
We are now ready to calculate the dependence of the superconducting characteristic lengths L (0) , δ 0 ξ and l with the Sn concentration β , by substituting Eq. (1) into Eqs. (D4), (D5) and (D6). The London penetration depth L (0) increases with Sn content, while the coherence length δ 0 ξ is reduced, Fig. 2. L (0) and δ 0 ξ are roughly with the same order of magnitude over the A15 range. By comparing the coherence length 0 ξ and the electronic mean free path l , one may distinguish the clean limit, the dirty limit and the intermediate state at any A15 composition. At lower Sn content, 0 ξ maintains rather high value compared to l ; but this difference is mitigated as 0 ξ continues to decrease and l increase, for rising β until the phase transformation boundary. In the tetragonal phase range, 0 ξ is approximately equal to or even lower than l . This implies that, as Sn content gradually approaches the stoichiometry, Nb3Sn undergoes a transition from the 'dirty' limit ( 0 l ξ >> ) to the 'clean' limit ( 0 l ξ << ), and the phase transformation boundary may be taken as the boundary of this transition. In Fig. 2, we also presents the variation of GL parameter κ with tin content, calculated by Eqs. (C19) and (D3). κ in the two limiting cases, clean κ and dirty κ , have the opposite change with Sn content; their change nearly counteract one another in the cubic phase range. While in the tetragonal phase range, dirty κ exhibits a more severe decrease compared to the increase in clean κ . Thus, clean dirty κ =κ κ + varies little in the cubic phase range, but decreases largely in the tetragonal phase.
FIG. 2. Composition dependence of the London penetration depth L (0) , coherence length δ 0 ξ and electronic mean free path l (Top), and the GL parameter κ (Bottom).
The magnetic properties of type II superconductors gives c2 c 2 H H κ = at the vicinity of c T , where c H is the thermodynamic critical field [22]. Combining this with the expressions κ and c H , Eqs. (C19), (C20) and (D3), renders the upper critical field c2 H in the clean limit and the dirty limit, Eqs. (C14) and (D7) (see Appendix C and D for detail). Thus, c2 H at the intermediate 0 / l ξ can be obtained by summing c2,clean H and c2,dirty H ,
$$H _ { c 2 } = C _ { 1 } \chi _ { 1 } ( T ) T _ { c } ^ { 2 } ( 1 - T / T _ { c } ) \gamma _ { m j } ^ { 2 } + C _ { 2 } \chi _ { 2 } ( T ) T _ { c } ( 1 - T / T _ { c } ) \gamma _ { m j } \rho _ { \Omega m }.$$
Here, 8 1/3 1 2 2/3 2 1 B F 1.979 10 ( / ) C ce k n S S π ---= × and 6 1 1 2 B 1.356 10 C cek π ---= × in which the physical constants are listed in Table A1. The involved three material parameters c T , mJ γ and Ω m ρ as well as their variations with the composition concentration in A15 Nb3Sn have been widely measured [3], Eq. (1) and Fig. 1. Thus, substituting Eq. (1) into Eq. (2), one finds the dependence of the upper critical field c2 H with Sn content. Note that the above formulas are valid at the vicinity of the superconducting transition temperature c T .
## 2.2. Corrections to upper critical field for electron-phonon interaction
A15 type Nb3Sn is a strong-coupling superconductor such that the corrections for electron-phonon (EP) interaction are required. We obtain the upper critical field, Eq. (2), based on the breaking and scattering of cooper pairs
in a weak-coupled interaction. Equation (2) is corrected for EP interaction in two ways: renormalizing the material parameters, and introducing correction parameters to superconductivity itself [18, 23, 24].
We determine which material parameter should be renormalized in terms of Grimvall principle [23]. Both the density of states and the wave function are renormalized by EP interaction, and the electron mass in the absence of EP interactions is replaced by a renormalized electron mass b ep (1 ) , where b refers to band values. The m m λ = + electronic density of states ( ) at Fermi level is renormalized by an enhancement factor 1 υ µ ep λ + where ep λ is the EP interaction parameter, and thus the coefficient of the electronic heat capacity mJ γ is enhanced by (1 ep mJ ) λ γ + . However, mJ γ is not changed for EP renormalization in Nb3Sn, since there are no EP renormalization effects in the change of the Fermi surface dimensions on alloy and the change in ( ) always depends on the Fermi level and υ µ follows almost rigidly any shift in Fermi energy. The electrical resistivity Ω m ρ [Eq. (A1)] is not renormalized, since the renormalization of the electron mass m exactly cancels against the renormalization of the scattering matrix element as it enters the averaged time τ between collisions. We find that, these are implicitly followed by Devantay at el. [25], who do not take renormalizations on mJ γ and Ω m ρ . Thus, we take the only EP correction in Nb3Sn by multiplying a factor c2 ( ) for H T η c2 H ,
$$H _ { c 2 } = \eta _ { H _ { c } } ( T ) [ C _ { 1 } \chi _ { 1 } ( T ) T _ { c } ^ { 2 } ( 1 - T / T _ { c } ) \gamma _ { m } ^ { \ 2 } + C _ { 2 } \chi _ { 2 } ( T ) T _ { c } ( 1 - T / T _ { c } ) \gamma _ { m } \rho _ { \Omega m } ] \,.$$
Here, c2 ( ) is the ratio of the strong-coupled magnetic pair-breaking parameter to the weak-coupled BCS value, and H T η can be evaluated by the detailed EP spectrum [26].
If c T and the energy gap (0) at 0K are determined experimentally, we have the strong-coupling correction ∆ (0) η ∆ to (0) , ∆ 1 (0) B c meas B c BCS B c meas B c meas (2 (0) / ) / (2 (0) / ) (2 / ) (2 (0) / ) 0.283(2 (0) / ) k T k T k T k T η ∆ π η -= ∆ ∆ = ∆ ≈ ∆ in which the BCS equation B c (0) ( / ) k T π η ∆ = with 1.78 has been used for the second equal sign. The characteristic η = (equivalent Einstein) frequency 0 [erg] is then determined using [18] (Appendix E) ω
$$\eta _ { _ { \Delta ( 0 ) } } = 1 + 5. 3 ( \omega _ { 0 } k _ { _ { B } } ^ { - 1 } T _ { _ { c } } ^ { - 1 } ) ^ { - 2 } \ln ( \omega _ { 0 } k _ { _ { B } } ^ { - 1 } T _ { _ { c } } ^ { - 1 } ) \,.$$
The EP correction to c2 H , c2 c ( ) , can thus be obtained, H T η
$$\eta _ { H _ { c 2 } } ( T _ { c } ) & = 1 + \pi ^ { 2 } ( \omega _ { 0 } k _ { B } ^ { - 1 } T _ { c } ^ { - 1 } ) ^ { - 2 } [ 0. 6 \ln ( \omega _ { 0 } k _ { B } ^ { - 1 } T _ { c } ^ { - 1 } ) - 0. 2 6 ] \,.$$
The another correction parameter (not involved in EP correction for Nb3Sn), the EP interaction parameter ep λ , is determined by the McMillan strong-coupled c T equation [24]
$$T _ { c } = & \frac { \theta _ { \mathfrak { D } } } { 1. 4 5 } \exp [ - \frac { 1. 0 4 ( 1 + \lambda _ { c p } ) } { \lambda _ { c p } - \mu ^ { ^ { * } } ( 1 + 0. 6 2 \lambda _ { c p } ) } ] \,,$$
where D θ is the Debye temperature and * µ is the pseudo potential parameter for electron Coulomb repulsion (for Nb3Sn * 0.2 ). Note that to obtain a more accurate µ ≈ ep λ for Nb3Sn one should use the Allen-Dynes c T formula, which extends the application range 0 ep 1.5 of Eq. (6) to a large λ < < ep λ value. Experiments give ep 1.8 for λ ≈
Nb3Sn. However, the available experimental data is D θ and not the characteristic phonon frequency required in the Allen-Dynes c T formula. Thus we still apply Eq. (6) at the cost of some accuracy.
The corrections for EP interaction require the information of the Debye temperature D θ and the energy gap (0) ∆ at 0K, Eq. (3). Unfortunately, the available data are reported at very limited Sn content. We use linear fits to D θ and B c (2 (0) / ) according to the identification [3] of the gradual changes from strong coupling to weak coupling with k T ∆ decreasing β , Fig. 1.
FIG. 3. EP correction parameters, the Debye temperature D θ and energy gap (0) , versus tin content. ∆
Figure 3 presents the calculated EP correction parameters in terms of Eqs. (1), (5) and (6). The result of the strong-coupling correction c2 H η to 2 c H approximately equals to the calculated value [18] of 1.17 at c 17.8K . The T = calculated EP interaction parameter ep λ at the stoichiometry 25at% Sn is slightly lower than the generally accepted value of 1.8 [27]. However, at larger derivation from the stoichiometry, the calculation differs relatively larger from the value of 1.8.
## 2.3. Upper critical field at temperature 0K
In the following we will concern the behavior of c2 H at temperatures far from c T , where the GLAG theory does not apply. This will generalize the above results near c T [e.g. Eq. (3)] to a wide temperature range and up to 0K. The temperature dependence of c2 H in the scaling law for Nb3Sn is determined by the Maki-de Gennes (MDG) relation [28],
$$H _ { c 2 } ( T ) = H _ { c 2 } ( 0 ) [ 1 - ( T / T _ { c } ) ^ { \lambda } ] \,.$$
Recent measurements suggest 1.52 ; this value has a universe applicability to a wide range of off-stoichiometric λ ≈ samples and different methods of determining c2 H [11]. This value of λ is also the power determined from the MDG theory and Eliashberg theory [28, 29]. Godeke et al. demonstrate that the MDG description (7) is universal for c2 ( ) H T relation of Nb3Sn independent of the compositional variation [2, 11]. Recent experiment shows that this relation is applicable for Nb3Sn with or without undergoing the cubic-to-tetragonal transition [12]. We are then allowed to
extrapolate c2 H at vicinity of c T to c2 (0) at temperature 0K for any Sn content. If H c2 0 ( ) with H T 0 c T T → has been obtained using Eq. (3), then one can deduce c2 (0) as H
$$H _ { c 2 } ( 0 ) = H _ { c 2 } ( T _ { 0 } ) / [ 1 - ( T _ { 0 } / T _ { c } ) ^ { \lambda } ] \,.$$
We assume that EP interaction correction to c2 H far from c T is the same as that for c2 H near c T .
WHH equation, derived from Gor'kov superconductivity theory (Green function method) and taking into account electron spin and spin-orbital scattering [30], is capable of giving rather satisfactory descriptions for c2 H behavior of a wide range of commercial and experimental Nb3Sn wires. This equation can be written as a simple form
$$H _ { c 2 } ( 0 ) = 0. 6 9 H _ { c 2 } ^ { \, ^ { \prime } } T _ { c } \,,$$
where c c2 c2 ( ) T H dH dT ′ = -[14]. Substituting the derivative of Eq. (3) into Eq. (9) and then comparing to Eq. (2), one finds c2 c2 0 0 c (0) 0.69 ( ) / (1 / ) . H H T T T = -It is further shown that the only difference between WHH equation and MDG relation is the power of ( 0 c / ) ; Since T T 0 c / 1 this difference has very limited impact on the T T → c2 (0) values. H This is consistent with the viewpoint [30, 31] that MDG relation is a good approximation to Werthamer theory.
FIG. 4. Upper critical field c2 B variation as a function of tin content β at: the vicinity of c T (Top) and 0K temperature (Bottom). MDG: extrapolation of c2 c ( ) with MDG relation, Eqs. (1), (2) and (8); EP+MDG: extrapolation of EP H T corrected c2 c ( ) with MDG, Eqs. (1), (2), (3) and (8); PPL: Pauli paramagnetic limit, Eqs. (1) and (10); EP+PPL: H T EP corrected Pauli paramagnetic limit, Eqs. (1) and (11); □ : Experiment dataset 1 [3]; △ : Experiment dataset 2 [12].
We now calculate the upper critical field c2 B near c T using Eqs. (1), (2) and (3), Fig. 4. The temperature 0 T near c T is designated as 0 c 0.9 T T = . One finds the dependence of c2 0 ( ) on Sn content B T β with/without the EP correction. The results of c2 0 ( ) are then substituted into the MDG relation, Eq. (8), to obtain B T c2 B at 0K, c2 (0) . Since the WHH B relation is equivalent to the MDG relation, we do not implement it redundantly. The calculation results are impressive (Fig. 4): c2 (0) obtained by B c2 0 ( ) with EP correction and then extrapolated by MDG relation (EP+MDG) is in good B T agreement with the experiments. This validates the above GLAG descriptions, and also the MDG description for the temperature dependence of the upper critical field at any composition over the A15 phase field.
We find that, the cubic-tetragonal phase boundary at ~24at% Sn separates the increasing c2 B versus β from the decreasing c2 B versus β . As shown both in the experimental c2 (0) versus tin content and the EP+MDG curve, the B maximum ~29T of c2 (0) appears at ~24at% Sn, and at both sides of the peak, B c2 (0) decreases as tin content deviates B more from 24at% Sn. In the tetragonal phase, c2 (0) has a more serious reduction; however, at the vicinity of 24at% Sn, B c2 (0) have nearly the same values at both sides of 24at%Sn, namely in the cubic phase range and the tetragonal phase B range. This is consistent with the experiment by Zhou et al. [12], their tetragonal phase [ c2 (0.3K) 29.1T at B = 24.6 0.2 at% ] and cubic phase [ β = ± c2 (0.3K) 29.0T at B = 23.7 0.4 at% ] samples exhibiting almost identical β = ± c2 (0) 29 0.2 T . We infer that, this coincidence occurs in a limited range, where the tin content deviates small from B ≈ ± the phase transformation boundary; for a large deviation there is a stronger depression of c2 (0) in the tetragonal phase. B This phenomenon is caused by the underlying relationship between the superconductivity of Nb3Sn and the related material parameters; physical properties of the latter is continually changed by the spontaneous cubic-tetragonal transformation.
FIG. 5. Upper critical field c2 (0) in the clean and dirty limit case. B
We now consider the upper critical field c2 (0) decomposing into the component in the clean limit case [Eqs. B (D7)] and that in the dirty limit case [Eqs. (C14)], c2 c2, clean c2, dirty (0) B B B = + . One finds that c2, clean B increases as β raised, but for contrast the dirty one increases up to 24 at% and then deceases drastically, Fig. 5. As we already know, Nb3Sn undergoes a transition from the 'dirty' limit to 'clean' limit as β increased. By comparing Figs. 5 and 6, we
find that, within 18at%~24at%, the component c2, dirty B dominates in c2 (0) and determines the trend of B c2 (0) curve. B Once crossing over the phase transformation boundary, B takes over (0) variation with B β .
c2, clean c2
## 2.4. Limit for upper critical field: Pauli paramagnetic limit
Preferential Pauli-paramagnetic lowering of normal-state free energy should place a limit on the orbital-pair-breaking c2 H of filamentary high-field superconductor [32],
$$H _ { c 2 } ( 0 ) \leq H _ { p } ( 0 ) \equiv 1. 8 4 \times 1 0 ^ { 4 } T _ { c } \ [ \text{Oe} ] \,.$$
This equation is a rough estimation for the limit field p (0) , since the assumption that the zero temperature difference H between superconducting and normal-state magnetizations be at least equal to the Pauli conduction-electron-spin magnetization, is somewhat contrary to experimental results [33]. The correction to p (0) for EP interaction is given H by
$$H _ { p } ( 0 ) = 1. 8 4 \times 1 0 ^ { 4 } \eta _ { _ { H _ { c } } } ( 0 ) ( 1 + \lambda _ { _ { \infty p } } ) ^ { 1 / 2 } T _ { _ { c } } \ [ \text{Oe} ] \,,$$
where c 2 1 1 2 1 1 c 0 B c 0 B c ( ) 1 ( ) [1.1ln( ) 0.14] assuming H T k T k T η π ω ω -----= + + c ( ) is independent with temperature [18]. H T η
In Fig. 4, PPL (Pauli paramagnetic limit), Eqs. (1) and (10), slightly lowers the EP+MDG curve after ~23at% Sn, and it thus provides a good boundary with the experimental data before the phase transformation. This justifies that, PPL is independent with the EP interaction in Nb3Sn. In Fig. 4, EP corrected PPL [Eqs. (1) and (11)] is much higher than the experimental curve and has no restriction to c2 (0) . This is consistent with the view by Orlando B et al. [18], who demonstrate that EP corrected PPL has nothing to do with Nb3Sn superconductivity, since the strong EP interaction in Nb3Sn increases largely the Pauli limiting field above its BCS value and the spin-orbit scattering is less involved. This explains why we only take the impurity scattering into account while exclude the spin-orbit scattering.
## 3. Temperature dependence of upper critical field
FIG. 6. Temperature dependence c2 ( ) at varying Sn content in the range of the cubic phase at increment of 1at% B T from 18at% to 24at% Sn. The insert figure represents the situation in the tetragonal phase for 0.2at% increment between 24.2at% and 25.4at% Sn.
TABLE 2. Calculated H-T phase boundary and comparison to experiments.
| Samples | [at%] β | c ( )[ μΩ cm] T ρ ⋅ | 2 [mJ / molK ] γ | c [K] T | c2 (0)[T] B |
|-------------------------------------|------------|-----------------------|--------------------|-----------|---------------|
| Thin films by Orlando et al. | -/23.59 a) | 35/35.0 | -/10.8 | 16.0/16.4 | 29.5/30.8 |
| [18] | -/24.94 | 9/9.1 | -/12.6 | 17.4/18.2 | 26.3/25.2 |
| Bulk by Godeke et al. [2] | -/24.33 | 22(50%)/22.0 | -/11.8 | 16.5/17.6 | 27.4/30.1 |
| Bulk by Godeke et al. [2] | -/24.33 | 22(90%)/22.0 | -/11.8 | 16.6/17.6 | 28.3/30.1 |
| Polycrystal by Guritanu et al. [17] | -/24.46 | -/19.4 | 13.7/11.8 | 17.8/17.8 | 25.0/29.4 |
a) Before the oblique line is the experimental value and after that presents the calculated value.
Now, we focus on the temperature dependence of c2 B at any A15 composition concentration, Eqs. (1), (3), (7) and (8). In Fig. 6, the temperature dependence c2 ( ) increases at any temperature as raising the Sn content B T β within the cubic phase range. The tetragonal phase exhibits a reverse behavior, c2 B at 0K and at most of other temperatures increasing with reduced Sn content. Orlando et al. show that in thin films c2 (0) is increased with B ρ rising, however c T is suppressed [2, 18]. For c ( ) 35 μΩ cm T ρ = ⋅ , there exists c2 (0) 29.5T B = and c 16.0K T = ; while c ( ) 9 μΩ cm leads to T ρ = ⋅ c2 (0) 26.3T and B = c 17.4K [18]. This corresponds to the calculated composition T = dependence in the tetragonal phase range, Figs. 1 and 6. The measured bulk needle by Godeke et al. exhibits a similar behavior: c ( ) 22 μΩ cm at T ρ = ⋅ a 50% normal-state resistance criterion corresponds to c2 (0) 27.4T B = and c 16.5K , and at a 90% criterion T = c2 (0) 28.3T and B = c 16.6K [2]. A summary of the calculations and the T = comparisons to experiments is given in Table 2. Through the theoretical expressions, we also figure out the material parameter that is undetermined by experiments, Table 2.
FIG. 7. Temperature dependence c2 ( ) for two homogenous off-stoichiometric Nb3Sn samples, and comparison to B T experiments [12].
We also calculate c2 ( ) for two rather highly homogenous Nb3Sn samples, Fig. 7. Their material parameters, B T reproduced from [12], are listed in the insert of the figure. Note that in the absence of the raw experimental value of γ , we infer it from Eq. (1). Calculations show that, the dirtier sample (25Sn-1800) with the larger resistivity ( 31.3 μΩ cm ) ⋅ always exhibits a higher c2 B at any temperature while c T is suppressed. This follows the above observation by Orlando et al. [18] and Godeke et al. [2]. The little discrepancy from the experiment may be attributed to the absence of the raw data of γ and the greater uncertainty in experimentally determining the composition concentration (fluctuation of ±0.7at% Sn for the dirtier sample compared to ±0.2at% for the cleaner one).
## 4. Flux pinning force and Kramer plot
We now extend the method for the upper critical field c2 B to account for the pinning behavior at different composition concentrations. Using the flux pinning model proposed by Kramer [34], the pinning force per volume for Nb3Sn conductors, p ( ) , is given by [11] F B
$$F _ { p } ( B ) = J _ { c } B = 1 2. 8 \kappa ^ { - 2 } B _ { c 2 } ^ { \ 2. 5 } ( B / B _ { c 2 } ) ^ { 0. 5 } ( 1 - B / B _ { c 2 } ) ^ { 2 } \ [ \text{GN} \cdot \mathbb { m } ^ { - 3 } ] \,,$$
where c J is the critical current density. p ( ) F B is associated with Sn concentration through the composition dependences c2 ( , ) and B T β ( ) , which are formulated by Eqs. (1), (3), (7), (8), (C19) and (D3). The Kramer κ β function 0.5 0.25 K c ( ) f B J B = is linear with the magnetic induction B and can identify c2 B at which K ( ) 0 [11], f B =
$$f _ { _ { K } } ( B ) = J _ { _ { c } } ^ { 0. 5 } B ^ { 0. 2 5 } = 1. 1 \times 1 0 ^ { 5 } \kappa ^ { - 1 } ( B _ { _ { c 2 } } - B ) \, [ A ^ { 0. 5 } \mathbf m ^ { - 1 } T ^ { 0. 2 5 } ] \,.$$
From Fig. 8 we observe a profound influence of the composition concentration on the field dependent pinning force p ( ) . Within the cubic phase range, the increase of Sn content raises the F B p ( ) remarkably and shifts the peak F B in each p ( ) curve to the high field region. This is associated with the composition dependent superconducting F B properties. To be specific, the magnitude of c2 B determines the position of the peak in p ( ) , while the height of F B p ( ) is related to both F B c2 B and κ , Eq. (12). c2 B in the cubic phase range increases as β rising (Fig. 4) while κ varies little (Fig. 2), thus resulting in the shift of the peak and the increase in the height of p ( ) . This case differs from F B that for the tetragonal phase range, where rising β leads to a drastic decrease in both c2 B and κ . As in Fig. 8, the peak shifts to the left and the height is nearly the same (due to 2 2.5 p c2 ( ) F B B κ -∝ ). The experimental data of a Nb3Sn conductor locate roughly between the calculated p ( ) curves for different Sn contents. This implies that, to describe F B a real Nb3Sn conductor one may consider the composition gradient in the conductor which results in a weighted average of local homogenous composition properties. For Kramer plot, we find its variation consistent with p ( ) for the F B similar reasons, Fig. 8.
FIG. 8. Effect of composition concentration on the magnetic field dependence of the flux pinning force (Top) and Kramer plot (Bottom). The experiment data are extracted from [28] [11].
## 5. Composition gradient effect on practical Nb3Sn wires
In the following we will apply our descriptions for the superconducting properties variation with tin content in the concentric shells model proposed by Cooley et al. [7], to arrive at the real composition dependent behavior in practical Nb3Sn conductors. In the Nb3Sn filament of PIT wires, the A15 layer exists between a central Sn-rich core and a coaxial Nb tube, Fig. 9. This structure is convenient for using a series of concentric shells with varying Sn concentration to simulate the composition inhomogeneity in the wire. The Sn content β varies with the position in the A15 layer [7],
$$\beta = 1 8 + 3. 5 [ 1 - r ^ { N } + ( 1 - r ) ^ { 1 / N } ] \, \left [ \text{at} \% \right ],$$
where N indicates the severity and steepness of the overall gradient and r is the normalized radius of the filament cross section. The radius r is set to be 0 at the A15/Sn interface and reaches its maximum 1 at the Nb/A15 interface. The actual position in this area is counted as 10 5 [ μ m], indicating that the inner radius of the A15 layer is R r = + min 10 μ m and the outer radius is R = max 15 μ m. R =
For PIT wires, the prefactor of the flux pinning force differs from the Kramer model [Eq. (13)], 2 0.5 2 3 p c c2 c2 c2 ( ) 0.35 ( / ) (1 / ) [GN m ] F B J B B B B B B -= = -⋅ and 4 0.25 0.5 1 0.25 K c2 c2 ( ) 1.871 10 ( ) [A m T ] f B B B B --= × -. We can then link c2 ( , ) B T β and c J to the radius r , and the local magnetic moment is calculated as
2 2 o c [Am ], where m R tLJ π = t is the thickness of each of the 100 shells, i.e. max min ( ) / 100 and t R R = -L is the sample length.
FIG. 9. Distribution of the Sn content β (Top left), pinning force p F (Top right) and magnetic moment o m (Bottom right) on the cross section of PIT wire filament (Bottom left), at temperature T =4.2K. N indicates the severity of the overall gradient.
Figure 9 presents the Sn content variation with the radius r at different gradients N . It is found that, larger steepness of the overall gradient N has a more drastic decrease near the outer radius (closer to the real situation). The distribution of the magnetic moment o m along the radius r at different applied magnetic fields is also presented in Fig. 9. At any magnetic field B , the regions near the Nb/Nb3Sn ( 1) and/or Nb3Sn/Sn ( r = 0 ) interfaces appear no r = magnetic moment and thus loss of superconductivity. The larger region of the loss occurs for the higher magnetic field, with a suppression in the magnitude of magnetic moment. In fact, the magnetic moment is related to the local critical current density and thus the pinning force p ( ) . One can find the Sn content dependence of F B p ( ) from Fig. 8; in the F B cubic phase range, p ( ) is raised with the Sn content increasing at any field, and F B p F for lower Sn content is more probably turns to disappear at higher fields. These are the underlying reasons for the vanishing of the magnetic moment. The tetragonal phase has a similar corresponding relationship. So, the loss of superconductivity near the boundary of A15 layer is mainly associated with the change of the flux-pinning behavior due to the A15 composition variation.
FIG. 10. Effect of the composition gradient on the critical current density (Left) and Kramer function 0.5 0.25 c J B (Right) of PIT wires. The measurements on SMI ternary PIT wire are duplicated from [2]. The Kramer plot with a composition gradient 20 has a positive curvature approaching the experiment. N =
Summing the contributions o m from each shell results in the total magnetic moment ot m , which can be measured by magnetometry. The critical current density c J of the wire is then expressed as 3 3 2 c ot max min 3 / [ ( )] [A/m ] [7]. In Fig. 10, we present the variation of J m L R R π = -c J as a function of the magnetic field B for different Sn gradients N . The homogenous sample results are presented to emphasize the effect of the tin gradient. The results of c ( ) allow one to depict the Kramer plot J B k ( ) . The Kramer plot [2] for the measurements f B on a SMI ternary PIT wire exhibits an anomalous curvature 'tail' at the vicinity of c2 B , Fig. 10. Two calculated homogenous samples with 21at% and 23at% appear no curvature, however, the inhomogeneous samples have a curved k ( ) when approaching f B c2 B . An acceptable agreement with the experiment is found for the conductor with gradient severity N =20. At this point, we quantitatively predicts the positive curvature in the Kramer plot of the PIT wire, emphasizing the importance of the composition inhomogeneity in the superconducting properties of practical Nb3Sn conductors. The composition dependent pinning force and Kramer function explains the disagreement between the results of the scaling laws and the experiments for practical inhomogeneous conductors.
## 6. Conclusions
Although Nb3Sn has been extensively used in fusion engineering area like ITER, the dependence of superconductivity with inhomogeneous composition is established incompletely in theory. Based on the GLAG theory frame, we derive a series of expressions for the superconductivity parameters and the upper critical field as a function of the three material parameters. These relations have a complete self-consistent theory basis describing the variation of superconductivity of Nb3Sn with Sn content. The correction for EP interaction are included. The theoretical estimation of c2 (0) variation with Sn content, provided with the fits to the material parameters, shows an acceptable agreement H with the experiments.
Nb3Sn undergoes a transition from 'dirty' limit ( 0 l ξ ) to 'clean' limit ( 0 l ξ ) as Sn content gradually approaches the stoichiometry. The change in superconductivity at the vicinity of the critical temperature and in the related material parameters determines the composition dependence of the upper critical field. The MDG description is universal for describing H-T phase boundary over the A15 phase field.
In the cubic phase range, c2 ( ) increases as raising Sn content. There appears an inverse B T c2 ( ) behavior in the B T tetragonal phase range. A significant influence of the composition concentration on p ( ) curves is observed. Within F B the cubic phase range, the increase of Sn content raises the p ( ) remarkably and shifts the peak in each F B p ( ) curve F B to the right side. The peak shifts to the left and the shape is nearly the same in the tetragonal range. This can be explained by the composition dependencies c2 ( , ) and B T β ( ) . κ β
The effect of composition gradient on the superconducting properties of PIT wires is considered by applying the obtained formulas in Cooley's concentric shells model. The loss of superconductivity near the boundary of A15 layer is mainly associated with the change of the flux-pinning behavior due to the A15 composition variation. The inhomogeneous conductor with gradient severity N =20 predicts well the curved 'tail' approaching c2 B in the Kramer plot. This implies that the composition inhomogeneity is an important factor in the unusual phenomenon of the practical Nb3Sn conductors.
However, we cannot yet include the effect of the alloying addition (Ti and/or Ta) and the matrix material in a ternary Nb3Sn wire. The composition gradient is an important factor in the unusual phenomenon of practical Nb3Sn conductors but not the unique determinant. The most possible role of the alloying addition taken in Nb3Sn is the scattering impurity (Appendix A), which changes the electrical resistivity deeply. We will include this effect in the future work to increase the practical value of the present theory.
## ACKNOWLEDGMENTS
The authors appreciate the financial supports from the National Natural Science Foundation of China (11372120, 11032006 and 11121202), National Key Project of Magneto-Restriction Fusion Energy Development Program (2013GB110002) and Fundamental Research Funds for the Central Universities (lzujbky-2014-227).
## Appendix A: Scattering by impurity in normal metal
Let us first introduce the concept of a mean free path l , which may be expressed as l v τ = if v signifies the averaged velocity (approximated as Fermi velocity F v ) and τ the averaged time between collisions. For a gas of free electrons, electrical conduction can be regarded as the diffusion of electrons under an external force eE . The mean free path l is a distance travelled by the electrons without undergoing collisions, Fig. A1. The electrical conductivity σ is thus associated with τ in the form of
$$\sigma = e ^ { 2 } D \upsilon ( \mu ) = n _ { \epsilon } e ^ { 2 } \tau / m \,,$$
where D is the diffusion coefficient / 3 and D lv = ( ) is the electronic density of states at the Fermi surface [35]. υ µ m is the electron mass, e is the unit charge and e n is the number of electrons per unit volume. The collision depends on certain scattering processes. If the scattering is arisen from impurities and this interaction is elastic and weak, one can obtain the scattering possibility of a system of electrons in the field of impurity centers with Born approximation and the collision integral [36]; consequently, the collision time τ is obtained as 1 i n τ λ -= where
$$\lambda = ( 4 \hbar { ) } ^ { - 1 } \upsilon ( \mu ) \oint \left | M _ { \ p p ^ { \prime } } ( \theta ) \right | ^ { 2 } ( 1 - \cos \theta ) d \Omega \,,$$
and i n is the number of impurity atoms per unit volume (i.e. impurity concentration). The integration is taken over the surface of Fermi sphere, Ω is the solid angle and θ is the angle between two momentums 0 p ( 2 0 / 2 p m µ = where 0 p is the Fermi momentum and µ is the Fermi energy) and 0 ′ p on the Fermi surface ( p is fixed and ′ p varies with d Ω during integration). ( ) indicates the matrix element of interaction energy of an electron with the impurity. pp M θ ′
## Appendix B: Coherence length, penetration depth and GL parameter
The coherence length ξ indicates a cooper pair coherence between electrons extending to a certain distance in a pure superconductor. ξ is determined as F ~ / ( ) where v T ξ ∆ ( ) is the energy gap in BCS theory [37]. If T ∆ c T T → , then 2 1/2 B c c 3.06[ ( )] k T T T ∆ ≈ -, where B k is Boltzmann constant (Table A1) [19]. Use is commonly made of the standard coherence length,
$$\xi _ { _ { 0 } } = \hbar { v } _ { _ { F } } / \pi \Delta ( 0 ) = ( \eta / \pi ^ { 2 } ) \hbar { v } _ { _ { F } } / k _ { _ { B } } T _ { _ { c } } \,.$$
The second equal sign holds since the BCS theory gives [19]
$$k _ { _ { B } } T _ { _ { c } } & = ( 2 \eta / \pi ) \hbar { \omega } _ { _ { D } } \exp ( - 2 / \ g \upsilon )$$
and
$$\Delta ( 0 ) = ( \pi / \eta ) k _ { _ { B } } T _ { _ { c } } \,,$$
where 1.78 . We clarify that, the BCS theory holds true for traditional lowη = c T superconductors, with the isotropic model of metal and the weak coupling interaction (i.e. ( ) 1 with the electron-phonon interaction constant g υ µ << 3 0 ~ / g p m ). These conditions are not always fulfilled. For Nb3Sn with c 4.2K and T = D 94.5K , one finds ω = ( ) 0.62 which dose not fulfill g υ µ = ( ) 1 . Thus, the BCS theory and its extending results differ from g υ µ << experiment at some extent. However, the disagreement of the theory is remarkably reduced after an appropriate correction [19]. The merit of this theory is the concise physical concepts and brief mathematical representations. The London penetration depth of the magnetic field in a pure superconductor is formulated as 2 2 1/2 L s ( / 4 ) mc n e δ π = according to the definition 1 L 0 0 H Hdx δ ∞ -= ∫ . Here, s n is known as the number of superconducting electrons per volume in contrast to the total number e n . At the vicinity of c T , there exists s e c c / ( ) / n n T T T ≈ -upon the Ginzburg-Landau equations. From the above discussion, one figures out that both L δ and ξ have a dependence of
1/2 c ( ) T T --at c T T → . Hence, the GL parameter κ , defined as the ratio of δ and ξ , tends to a constant at c T T → [38]. For this purpose, we introduce the GL description of κ ,
$$\kappa = 2 ^ { 3 / 2 } e H _ { c } \delta ^ { 2 } / \hbar { c } \,,$$
and the microscopic descriptions of L ( ) [38], T δ
$$\delta _ { \L } ( T ) = \delta _ { \L } ( 0 ) [ 1 / 2 ( 1 - T / T _ { \epsilon } ) ] ^ { 1 / 2 } \,,$$
where
$$\delta _ { L } ( 0 ) = ( m c ^ { 2 } / 4 \pi n _ { _ { e } } e ^ { 2 } ) ^ { 1 / 2 } \,.$$
TABLE A1. Dimension, unit and constant value of physical quantities (Gaussian units). a)
| Physical quantities | Dimension | Unit | Constant value |
|--------------------------------------------|----------------------|----------------------|------------------|
| Boltzmann constant B k | 2 2 1 E L MT T - - | erg/K | 16 1.381 10 - × |
| Critical temperature c T | E T | K | - |
| Coefficient of electronic heat capacity γ | 1 2 2 E L MT T - - - | 3 2 erg cm K - - ⋅ ⋅ | - |
| Collision time τ | T | s | - |
| Density of states at Fermi surface ( ) υ µ | 5 1 2 L M T - - | 1 3 erg cm - - ⋅ | - |
| Electron mass m | M | g | 28 9.109 10 - × |
| Elementary charge e | 3/2 1/2 1 L M T - | esu | 10 4.803 10 - × |
| Planck constant | 2 1 L MT - | erg s ⋅ | 27 1.055 10 - × |
| Electrons number per unit volume e n | 3 L - | -3 cm | 23 2.59 10 × |
| Electrical resistivity ρ | T | s | - |
| Fermi velocity F v | 1 LT - | cm/s | - |
| Flux quantum 0 Φ | 3/2 1/2 1 L M T - | Maxwell | / c e π |
| Light velocity in vacuum c | 1 LT - | cm/s | 10 2.998 10 × |
| Ratio of Fermi surface F / S S | Unity | Unity | 0.35 |
| Upper critical field c2 H | 1/2 1/2 1 L M T - - | Oe | - |
| Upper magnetic induction intensity c2 B | 1/2 1/2 1 L M T - - | Gauss | - |
a) All the formula derivations use the Gaussian units except for the unit conversion for practical application. L , M ,
T and E T mean the dimension of length, mass, time and temperature, respectively.
Also the thermodynamic critical field c H near c T is formulated as [39]
$$H _ { c } = ( 1 6 \pi p _ { 0 } m / 7 \zeta _ { 3 } h ^ { 3 } ) ^ { 1 / 2 } k _ { B } T _ { c } ( 1 - T / T _ { c } ) \,,$$
where the Riemann zeta function 3 3 1 1.2 . Using Eqs. (B5) and (B7) in Eq. (B4) one finds n n ζ ∞ -= = ∑ ≈
$$\kappa = ( 3 2 / 7 ) ^ { 1 / 2 } \pi ^ { 1 / 2 } \zeta _ { 3 } ^ { - 1 / 2 } \hbar { ^ { - 5 / 2 } } c ^ { - 1 } e k _ { b } m v _ { F } ^ { 1 / 2 } \delta _ { L } ^ { 2 } ( 0 ) T _ { c } \approx 3. 4 5 9 \hbar { ^ { - 5 / 2 } } c ^ { - 1 } e k _ { b } m v _ { F } ^ { 1 / 2 } \delta _ { L } ^ { 2 } ( 0 ) T _ { c } \ \.$$
Substituting 0 ξ for c T by Eq. (B1), we obtain
$$\kappa = ( 8 / 7 ) ^ { 1 / 2 } \eta \pi ^ { - 2 } \zeta _ { 3 } ^ { - 1 / 2 } h ^ { - 3 / 2 } m ^ { 3 / 2 } v _ { F } ^ { 3 / 2 } n _ { e } ^ { - 1 / 2 } \delta _ { L } ( 0 ) / \xi _ { 0 } \,.$$
Recall that the Fermi momentum 0 p is associated with the number e n of electrons per volume as [35]
$$p _ { 0 } & = \hbar { ( } 3 \pi ^ { 2 } n _ { e } ) ^ { 1 / 3 } \,,$$
then we rewrite Eq. (B9) as
$$\kappa = ( 2 4 / 7 ) ^ { 1 / 2 } \gamma \pi ^ { - 1 } \zeta _ { 3 } ^ { - 1 / 2 } \delta _ { \mathbb { L } } ( 0 ) / \xi _ { 0 } \approx 0. 9 5 8 \delta _ { \mathbb { L } } ( 0 ) / \xi _ { 0 } \,.$$
Lattice distortion under EP interaction
A15 type NbsSn, space group Pm3n. Nb atoms occupy 6c positions (chains, in white). Sn atoms take 2a positions (in black).
In normal-state Nb;Sn, collisions between clectrons and ions, viz. scattering by impurity; generate electric resistivity .
In superconducting Nb;Sn;, two electrons are bound to form Cooper due to EP attraction. Scattering vanishes in the cases 5 <<l while exists for pair
FIG. A1. Schematic of the scattering mechanism in normal-state and superconducting A15-type Nb3Sn.
Extending to a wide temperature range, κ is formulated as a temperature-dependent relation c 1 ( ) ( ) ( ) ; T T T κ κ χ = calculations show that 1 ( ) varies little with T χ T [19]. For the case l ξ << , scattering from impurity has little impact on the coherence length and the superconductivity [40]. This implies κ remains Eq. (B11) for this case.
It is natural to address the opposite limiting case l ξ >> . At first sight, the scattering from impurities reduces the coherence length. Indeed, one could use the diffusion process of an electron in the field of impurity (Appendix A and Fig. A1) to justify the speculation. Recall that under the consideration of scattering, D is the diffusion coefficient with / 3 in the diffusion equation D lv = e e j D n = -∇ . Here, e j is the diffusion flux, i.e. the number of electrons passing through 1m 2 plane in 1s. It follows from the diffusion equation that 1/2 ~ ( ) x Dt where x is the distance travelled by a electron in t . Without scattering and in the same period t , an electron travels a distance of ~ vt ξ . Substituting this into x expression, we can obtain the effective coherence length ξ ′ as 1/2 1/2 ~ ~ ( ) ~ ( / ) i x l n ξ ′ ξ ξ . This is the
justification of the above speculation. The scattering of impurity also affects the penetration length δ ; the approximation relation is written as 1/2 1/2 L L ~ ( / ) ~ ( ) i l n δ ′ δ ξ δ ξ [19]. Finally, we arrive at the effective GL parameter κ ′ in the impurity field for l ξ >> , L L / ~ / ~ i l n κ ′ = δ ′ ξ ′ δ δ . Based on a rigorous gauge-invariant solution of the linearized Gor'kov equations [40], one can obtain a precise relation, c L ( ) 0.72 (0) / T l κ ′ δ = and c 2 ( ) ( ) ( ) . T T T κ ′ κ ′ χ = The property of 2 ( ) is similar with T χ 1 ( ) . We are now ready to summary the dependence of GL parameter T χ κ with temperature and impurity concentration i n [with the aid of Eqs. (B8) and (B11)],
$$\kappa ( T ) = \kappa ( T _ { c } ) \chi _ { l } ( T ), \ \xi < < l \ ( \bullet \text{clean} ^ { \bullet } \lim i t ),$$
where c L 0 ( ) 0.96 (0) / T κ δ ξ ≈ or 5/2 1 1/2 2 c B F L c ( ) 3.46 (0) T c ek mv T κ δ --= .
$$\kappa ( T ) = \kappa ( T _ { c } ) \chi _ { 2 } ( T ), \ \xi > > l \ ( \ell \text{dirty} ^ { \prime \prime } \limit ),$$
where -1 c L i F ( ) 0.72 (0) T n v κ δ λ = or c L ( ) 0.72 (0) / T l κ δ = . We have omitted the superscript of κ ′ in Eqs. (B12) and (B13).
The GL theory of magnetic properties of type II superconductors gives that at the vicinity of c2 H [22],
$$H _ { c 2 } & = \sqrt { 2 } \kappa H _ { c } \,.$$
## Appendix C: Relation with material parameters
Now we correlate the electronic density of states, ( ) , to the coefficient of electronic heat capacity, υ µ γ , upon isotropic model of metal,
$$\gamma = \pi ^ { 2 } k _ { _ { B } } ^ { \ 2 } \upsilon ( \mu ) / 3 \,.$$
The use of isotropic model equation here is reasonable since the experiments do not show evidence of a highly anisotropic Fermi surface for Nb3Sn [41]. Note that the anisotropy of the physical properties is experimentally identified in compositionally inhomogeneous single crystals [42], scandium [43] and cuprate superconductors [44]. In fact, the electronic structures of Nb3Sn are very complex, and ( ) exhibits a sharp variation relating to the anomalous υ µ isotope effect in Nb3Sn superconductors [45]. ( ) may be expressed as a function of the Fermi momentum υ µ 0 p :
$$\upsilon ( \mu ) = p _ { 0 } m / ( \pi ^ { 2 } h ^ { 3 } ) \,.$$
Combining Eqs. (C1) and (C2), the Fermi velocity F v is related to γ in a manner of
$$v _ { F } = 3 k _ { B } ^ { - 2 } m ^ { - 2 } \hbar { ^ { 3 } } \gamma \,.$$
Let's consider the electrical resistivity ρ , Eq. (A1), and substituting Eqs. (C1) and (C3) into it, one obtains
$$l = 3 ^ { - 1 } \pi ^ { 2 } e ^ { - 2 } m ^ { 2 } h ^ { - 3 } k _ { \mathrm B } ^ { \, 4 } \gamma ^ { - 2 } \rho ^ { - 1 }.$$
This implies a variation of the electronic average free path l as a function of γ and ρ . We then reformulate the coherence length 0 ξ , the number of electrons per unit volume, e n , and the London penetration depth L (0) by using δ Eqs. (B1), (B6) and (B10) in Eq. (C3), respectively,
$$\xi _ { 0 } = 5. 3 4 \pi ^ { - 2 } m ^ { - 2 } k _ { \mathrm B } ^ { - 3 } h ^ { 4 } \gamma T _ { c } ^ { - 1 } \,,$$
$$n _ { \mathfrak { e } } = 9 \pi ^ { - 2 } m ^ { - 3 } k _ { \mathfrak { B } } ^ { - 6 } \hbar { ^ { 6 } } \gamma ^ { 3 } \,,$$
$$\delta _ { \L } ( 0 ) = 0. 1 6 7 \pi ^ { 1 / 2 } m ^ { 2 } c e ^ { - 1 } k _ { \text{B} } ^ { \text{3} } h ^ { - 3 } \gamma ^ { - 3 / 2 } \,.$$
Now using Eqs. (B12), (B13), (C3), (C4) and (C7), we are allowed to find the dependence of κ on the three independent material parameters γ , ρ and c T for the dirty limit and clean limit near c T :
$$\kappa _ { \text{clean} } = 9. 4 3 \times 1 0 ^ { - 2 } \pi ^ { 3 / 2 } m ^ { 4 } c e ^ { - 1 } k _ { \mathfrak { B } } ^ { 6 } h ^ { - 7 } \chi _ { 1 } ( T ) \gamma ^ { - 5 / 2 } T _ { c } \,,$$
$$\kappa _ { \text{dinty} } = 0. 3 6 1 \pi ^ { - 3 / 2 } c e k _ { \text{B} } ^ { - 1 } \chi _ { 2 } ( T ) \gamma ^ { 1 / 2 } \rho \,.$$
For a preliminary validation we use the dimensional analysis upon four basic quantities which are length L , mass M , time T and temperature E T (Table A1). The LHS of Eq. (C9), for instance, is dimensionless, and the dimension of the RHS is deduced as [RHS] 1 1/2 1 3/2 1/2 1 2 2 1 1 1 2 2 1/2 B E E [ ][ ][ ] [ ] [ ] ( )( )( ) ( ) 1 [LHS] . c e k LT L M T L MT T L MT T T γ ρ ---------= = = = Further, using Eq. (C3) in Eq. (B7), we express the thermodynamic critical field c H as
$$H _ { _ { c } } = 2. 3 9 0 5 \pi ^ { 1 / 2 } T _ { _ { c } } ( 1 - T / T _ { _ { c } } ) \gamma ^ { 1 / 2 } \,.$$
Substituting Eqs. (C8)-(C10) into Eq. (B14), one finds the upper critical field c2 H in the two limiting cases,
$$H _ { c 2, \text{clean} } = 0. 3 1 8 8 \pi ^ { 2 } m ^ { 4 } c e ^ { - 1 } k _ { _ { B } } ^ { \ 6 } \hslash ^ { - 7 } \chi _ { _ { 1 } } ( T ) \gamma ^ { - 2 } T _ { c } ^ { \ 2 } ( 1 - T / T _ { c } ) \left [ \text{Oe} \right ],$$
$$H _ { c 2, \text{dirty} } = 1. 2 2 0 4 \pi ^ { - 1 } c e k _ { \i } ^ { - 1 } \chi _ { 2 } ( T ) T _ { c } ( 1 - T / T _ { c } ) \gamma \rho \left [ \text{o} \right ].$$
For application convenience we convert ρ in Gaussian unit [s] into Ω m [ Ω m] in fashion to experiments, using ρ ⋅ the unit conversion relation 1s=1esu 11 cm=9.0 10 cm and ⋅ × Ω⋅ 1 9 s m 9 10 ρ ρ --Ω = × . For γ there is also a conversion relation 4 erg mJ =1.0 10 γ γ × by the relation 1J 3 2 7 3 2 cm K =1.0 10 erg cm K ----⋅ ⋅ × ⋅ ⋅ . As experiments always measure γ in 1 2 [mJ mol K ] , --⋅ ⋅ 3 2 mJ [mJ cm K ] in the formulas should be further transformed as γ --⋅ ⋅ mJ mJ mol / V γ γ → (since 3 2 1 2 mol V mJ cm K 1mJ mol K ----⋅ ⋅ = ⋅ ⋅ where mol V is the volume occupied by 1mol atoms).
For Nb3Sn, recent specific heat measurements reveal 1 3 mol 11.085 0.005 mol cm at 10K temperature [17]. V -= ± Thus, Eqs. (C11) and (C12) are rewritten as
$$H _ { c 2, \text{clean} } = C _ { 1 } \chi _ { 1 } ( T ) T _ { c } ^ { 2 } ( 1 - T / T _ { c } ) \gamma _ { m } ^ { - 2 } \ [ 0 e ] \,,$$
$$H _ { c 2, \text{dirty} } & = C _ { 2 } \chi _ { 2 } ( T ) T _ { c } ( 1 - T / T _ { c } ) \gamma _ { m } \rho _ { \Omega m } \ [ \text{o} ] \,.$$
where 8 2 4 1 6 7 1 B 0.3188 10 C m ce k π ---= × and 6 1 1 2 B 1.356 10 C cek π ---= × . We also concern the determination of whether the experimental sample is in dirty limit or not. The same unit conversations are applied to the electronic free path l , the penetration depth L (0) , the coherence length δ 0 ξ and the GL parameter κ , and we rewrite Eqs. (C4), (C7), (C5), (C8), (C9) and (C10) into
$$l = 3 0 \pi ^ { 2 } e ^ { - 2 } m ^ { 2 } h ^ { - 3 } k _ { _ { B } } ^ { 4 } \gamma _ { _ { m J } } ^ { - 2 } \rho _ { _ { \Omega m } } ^ { - 1 },$$
$$\xi _ { 0 } = 5. 3 4 \times 1 0 ^ { 4 } \pi ^ { - 2 } m ^ { - 2 } k _ { \text{B} } ^ { - 3 } h ^ { 4 } \gamma _ { \text{m} } T _ { \text{c} } ^ { - 1 } \,,$$
$$\delta _ { \text{L} } ( 0 ) = 0. 1 6 7 \times 1 0 ^ { - 6 } \pi ^ { 1 / 2 } m ^ { 2 } c e ^ { - 1 } k _ { \text{B} } ^ { \ 3 } \hbar { ^ { - 3 } } \gamma _ { \text{m} } ^ { \ \ - 3 / 2 } \,,$$
$$\kappa _ { \text{clean} } = 3. 1 2 7 \times 1 0 ^ { - 1 2 } \pi ^ { 5 / 2 } m ^ { 4 } c e ^ { - 1 } k _ { \text{B} } ^ { \, 6 } h ^ { - 7 } \chi _ { 1 } ( T ) \gamma _ { m } ^ { \, - 5 / 2 } T _ { c } \,,$$
$$\kappa _ { \text{dirty} } = 4. 0 1 1 \times 1 0 ^ { - 9 } \pi ^ { - 3 / 2 } c e k _ { \text{B} } ^ { - 1 } \chi _ { _ { 2 } } ( T ) \gamma _ { _ { m } } ^ { \quad 1 / 2 } \rho _ { \Omega m } \,,$$
$$H _ { _ { \epsilon } } = 2. 3 9 0 5 \times 1 0 ^ { 2 } \pi ^ { 1 / 2 } T _ { _ { \epsilon } } ( 1 - T / T _ { _ { \epsilon } } ) \gamma _ { _ { m } } ^ { \ \ 1 / 2 } \,.$$
If experiments give the magnetic flux density c2 [T] then it is necessary to implement B 4 c2 c2 =( [Oe] 10 ) [T] B H -× in the above formulas (since c2 0 c2 [T]= [A/m] and 1A/m=4 B H µ ⋅ 3 10 Oe with π × -7 0 =4 10 T m/A ). µ π × -⋅
## Appendix D: Effective electron mass accounting for anisotropic metal
Accounting for the change of Fermi surface shape due to the transition to the anisotropic model and/or the condensed matter from the isotropic metal model, we employ the formulation of wave vector for the anisotropic metal model:
$$k _ { F } = 3 ^ { 1 / 3 } \pi ^ { 2 / 3 } ( n _ { \mathrm e } ^ { 2 / 3 } S / S _ { F } ) ^ { 1 / 2 }$$
with the de Broglie relation * 0 F F p m v k = = where * m is the effective electron mass. Here, F / S S is the ratio of the free Fermi surface to the Fermi surface of a free-electron gas of density e n . Substituting this into Eq. (C3) leads to
$$m ^ { ^ { * } } & = 3 ^ { 2 / 3 } \pi ^ { - 2 / 3 } k _ { \text{B} } ^ { - 2 } \hslash ^ { 2 } ( n ^ { 2 / 3 } S / S _ { \text{F} } ) ^ { - 1 / 2 } \gamma \\ & = 3 ^ { 2 / 3 } \times 1 0 ^ { 4 } \pi ^ { - 2 / 3 } k _ { \text{B} } ^ { - 2 } \hslash ^ { 2 } ( n ^ { 2 / 3 } S / S _ { \text{F} } ) ^ { - 1 / 2 } \gamma _ { \text{mJ} }.$$
Substituting this equation into Eq. (C18) one may obtain
$$\kappa _ { \text{clean} } = 5. 8 5 4 \times 1 0 ^ { 5 } \pi ^ { - 1 / 6 } c e ^ { - 1 } k _ { \text{B} } ^ { - 2 } \hbar { ( } n ^ { 2 / 3 } S / S _ { \text{F} } ) ^ { - 2 } \chi _ { 1 } ( T ) T _ { c } \gamma _ { m } ^ { - 3 / 2 } \,.$$
The following equations can be obtained in the same way:
$$l = 9 \times 1 0 ^ { 9 } ( 3 \pi ^ { 2 } ) ^ { 1 / 3 } \hbar { e } ^ { - 2 } ( n ^ { 2 / 3 } S / S _ { F } ) ^ { - 1 } \rho _ { \Omega m } ^ { - 1 } \,,$$
$$\xi _ { 0 } = 1. 2 3 4 \times 1 0 ^ { - 4 } \pi ^ { - 2 / 3 } k _ { _ { B } } ( n ^ { 2 / 3 } S / S _ { _ { F } } ) \gamma _ { _ { m } J } ^ { - 1 } T _ { _ { c } } ^ { - 1 } \,,$$
$$\delta _ { \L } ( 0 ) = 7 2. 2 5 7 \pi ^ { - 5 / 6 } c e ^ { - 1 } \hbar { k } _ { \mathfrak { B } } ^ { - 1 } ( n ^ { 2 / 3 } S / S _ { \mathrm F } ) ^ { - 1 } \gamma _ { _ { m } J } ^ { - 1 / 2 } \,,$$
$$H _ { c 2, \text{clean} } = 1. 9 7 9 \times 1 0 ^ { 8 } \pi ^ { 1 / 3 } c e ^ { - 1 } k _ { \i } ^ { - 2 } h ( n ^ { 2 / 3 } S / S _ { F } ) ^ { - 2 } \chi _ { 1 } ( T ) \gamma _ { m } ^ { \ 2 } T _ { c } ^ { 2 } ( 1 - T / T _ { c } ) \,.$$
## Appendix E: Lambert W(X) function
Equation (4) can be solved via the Lambert ( ) function, which is defined as the solution to W X exp( ) W W X = . ( , ) is the W K X K -th branch of the multi-valued function ( ) . It follows that a solution to W X 2 ln y x x -= is exp[ ( , 2 ) / 2] , where x W K y = --K representing the branch of the multivalued W is selected to ensure positive and real x . For 1 1 0 B c k T ω --in question, we arrive at 1 1 0 B c (0) exp{ 0.5 [ , 0.377( 1)]} with k T W K ω η ∆ --= ---1. K = -
## Reference
- 1 J. V. Minervini, Nature materials 13 , 326 (2014).
- 2 A. Godeke, M. C. Jewell, C. M. Fischer, A. A. Squitieri, P. J. Lee, and D. C. Larbalestier, Journal of Applied Physics 97 , 093909 (2005).
- 3 R. Flükiger, D. Uglietti, C. Senatore, and F. Buta, Cryogenics 48 , 293 (2008).
4
- A. Godeke, Superconductor Science and Technology 19 , R68 (2006).
- 5 L. Vieland, R. W. Cohen, and W. Rehwald, Physical Review Letters 26 , 373 (1971).
- 6 K. Niitsu and R. Kainuma, Phys. Status Solidi B 251 , 2041 (2014).
7
- L. D. Cooley, C. M. Fischer, P. J. Lee, and D. C. Larbalestier, Journal of Applied Physics 96 , 2122 (2004).
- 8 X. Xu, M. Sumption, X. Peng, and E. W. Collings, Applied Physics Letters 104 , 082602 (2014).
- 9 C. A. Reynolds, B. Serin, W. H. Wright, and L. B. Nesbitt, Physical Review 78 , 487 (1950).
- 10 G. D. Marzi, V. Corato, L. Muzzi, A. d. Corte, G. Mondonico, B. Seeber, and C. Senatore, Superconductor Science and Technology 25 , 025015 (2012).
- 11 A. Godeke, B. ten Haken, H. H. J. ten Kate, and D. C. Larbalestier, Superconductor Science and Technology 19 , R100 (2006).
- 12 J. Zhou, Y. Jo, Z. Hawn Sung, H. Zhou, P. J. Lee, and D. C. Larbalestier, Applied Physics Letters 99 , 122507 (2011).
13
C. Senatore and R. Flükiger, Applied Physics Letters
102
, 012601 (2013).
- 14 L. D. Cooley, Y. F. Hu, and A. R. Moodenbaugh, Applied Physics Letters 88 , 142506 (2006).
- 15 P. J. Lee and D. C. Larbalestier, Cryogenics 48 , 283 (2008).
- 16 X. Xu, M. D. Sumption, and X. Peng, Advanced Materials, DOI: http://10.1002/adma.201404335 (2015).
- 17
- V. Guritanu, W. Goldacker, F. Bouquet, Y. Wang, R. Lortz, G. Goll, and A. Junod, Physical Review B 70 , 184526 (2004).
- 18 T. Orlando, E. McNiff, S. Foner, and M. Beasley, Physical Review B 19 , 4545 (1979).
19
- A. A. Abrikosov, Fundamentals of the Theory of Metals (Elsevier, Amsterdam, 1988).
- 20 H. Wiesmann, M. Gurvitch, A. Ghosh, H. Lutz, O. Kammerer, and M. Strongin, Physical Review B 17 , 122 (1978).
- 21 S. Kiarasi and R. A. Secco, Phys. Status Solidi B, DOI: http://10.1002/pssb.201552029 (2015).
- 22 A. Abrikosov, Reviews of modern physics 76 , 975 (2004).
23
G. Grimvall, Physica Scripta
14
, 63 (1976).
- 24 W. McMillan, Physical Review 167 , 331 (1968).
- 25 H. Devantay, J. Jorda, M. Decroux, J. Muller, and R. Flükiger, Journal of Materials Science 16 , 2145 (1981).
- 26 N. Werthamer and W. McMillan, Physical Review 158 , 415 (1967).
- 27 D. F. Valentinis, C. Berthod, B. Bordini, and L. Rossi, Superconductor Science and Technology 27 , 025008 (2014).
- 28 J. W. Ekin, Superconductor Science and Technology 23 , 083001 (2010).
- 29 S. Oh and K. Kim, Journal of Applied Physics 99 , 033909 (2006).
- 30 N. R. Werthamer, E. Helfand, and P. C. Hohenberg, Physical Review 147 , 295 (1966).
- 31 D. M. J. Taylor and D. P. Hampshire, Superconductor Science and Technology 18 , S241 (2005).
- 32 A. M. Clogston, Physical Review Letters 9
- , 266 (1962).
- 33 R. R. Hake, Applied Physics Letters 10 , 189 (1967).
- 34 E. J. Kramer, Journal of Applied Physics 44 , 1360 (1973).
- 35 J. Singleton, Band theory and electronic properties of solids (Oxford University Press, Oxford, 2001).
- 36 E. M. Lifshitz, L. P. Pitaevskii, and L. D. Landau, Physical kinetics (Pergamon press, Oxford, 1981).
- 37 J. Bardeen, L. N. Cooper, and J. R. Schrieffer, Physical Review 108 , 1175 (1957).
- 38 L. P. Gorkov, Sov. Phys. JETP 9 , 1364 (1959).
- 39 A. A. Abrikosov, Sov. Phys.-JETP (Engl. Transl.);(United States) 5 (1957).
- 40 E. Helfand and N. Werthamer, Physical Review 147 , 288 (1966).
- 41 S. Foner and E. McNiff Jr, Physics Letters A 58 , 318 (1976).
- 42 M. Roslova, et al., CrystEngComm 16 , 6919 (2014).
- 43 S. M. Sichkar and V. N. Antonov, Phys. Status Solidi B 249 , 2118 (2012).
- 44 O. M. Auslaender, et al., Nature Physics 5 , 35 (2009).
- 45 G.-L. Zhao, Phys. Status Solidi B 251 , 1531 (2014). | 10.1038/s41598-017-01292-4 | [
"Yingxu Li",
"Yuanwen Gao"
] | 2014-08-02T09:38:17+00:00 | 2015-06-18T11:15:39+00:00 | [
"cond-mat.supr-con"
] | Superconducting properties variation with A15 composition in Nb3Sn | We extend the Ginzburg-Landau-Abrikosov-Gor'kov (GLAG) theory to account for
the variation of the upper critical field Hc2 with Sn content in A15-type
Nb3Sn. Hc2 at the vicinity of the critical temperature Tc is related
quantitatively to the electrical resistivity, specific heat capacity
coefficient and Tc, with inclusion of the electron-phonon coupling correction,
Pauli paramagnetic limiting and martensitic phase transformation of A15
lattices. Hc2 near Tc is then extrapolated to Hc2(0) at temperature 0K, and
Hc2(0) versus tin content agrees well with experiments. We find that, as Sn
content gradually approaches the stoichiometry, Nb3Sn undergoes a transition
from the dirty limit to clean limit, split by the phase transformation
boundary. The H-T phase boundary and the pinning force show different behaviors
in the cubic and tetragonal phase. Applying the theoretical formulas in the
technical Nb3Sn wire, we obtain the dependence of the composition gradient on
the superconducting properties variation in the A15 layer, as well as the
curved tail at vicinity of Hc2 in the Kramer plot of the Nb3Sn wire. This gives
a better understanding of the inhomogeneous-composition inducing discrepancy
between the results by the state-of-art scaling laws and experiments. |
1408.0354v1 | ## PROBING THE GALACTIC CENTER WITH X-RAY POLARIMETRY
F. Marin 1 , V. Karas , D. Kunneriath 1 1 , F. Muleri 2 and P. Soffitta 2
Abstract. The Galactic center (GC) holds the closest-to-Earth supermassive black hole (SMBH), which makes it the best laboratory to study the close environment of extremely massive compact objects. Polarimetry is sensitive to geometry of the source, which makes it a particularly suitable technique to probe the medium surrounding the GC SMBH. The detection of hard X-ray spectra and prominent iron K α fluorescence features coincident with localized gas clouds (e.g. Sgr B2, Sgr C) is known for nearly twenty years now and is commonly associated with a past outburst of the SMBH whose radiation is reprocessing onto the so-called 'reflection nebulae'. Since scattering leads to polarization, the re-emitted signal from the giant molecular clouds in the first 100 pc of the GC is expected to be polarized. X-ray polarization measurement is thus particularly adapted to probe the origin of the diffuse X-ray emission from the GC reflection nebulae and reveal the past activity of the central SMBH. In this research note, we summarize the results from past and current polarimetric simulations in order to show how a future X-ray polarimeter equipped with imaging detectors could improve our understanding of high-energy astrophysics.
Keywords: Galaxy: center, Galaxy: nucleus, Polarization, Radiative transfer, Scattering
## 1 Introduction
Using the ART-P telescope on-board of Granat , Sunyaev et al. (1993) were the first to image the Galactic center (GC) from the soft (2.5 keV) to the hard (22 keV) X-ray band. The discovery of a source characterized by a spherical morphology in the soft band, and by an extended (i.e. elongated along the Galactic plane) shape in the hard X-ray range, lead Sunyaev et al. (1993) to suggest that part of the hard X-ray emission could be due to Compton-scattered photons originating from a nearby compact source and reprocessed by dense molecular clouds. Later, X-ray images and spectra of the GC obtained by Koyama et al. (1996), using the Advanced Satellite for Cosmology and Astrophysics ( ASCA ), revealed fluorescent K α emission lines from cold iron atoms in Sgr B2. Associated with a hard X-ray spectral slope, Koyama et al. (1996) estimated that the diffuse thermal emission could be due to a past intense irradiation from Sgr A ∗ , the central supermassive black hole (SMBH) that lies in the GC. Such an argument is in agreement with the predictions of Sunyaev et al. (1993) and further detections from other extended gas cloudlets tend to corroborate this idea. In particular, Murakami et al. (2001) presented the first ASCA evidence of diffuse hard X-ray emission associated with strong 6.4 keV fluorescence line and large absorption from the Sgr C complex. Hard X-ray continuum and Fe K α emission have also been found in several other GC molecular clouds: Sgr B1 (Koyama et al. 2007), M0.74-0.09 (Koyama et al. 2007), G0.11-0.11 (Ponti et al. 2010), M0.74-0.09 (Nobukawa et al. 2011), M1 and M2 (Ponti et al. 2010), the Arches cluster (Yusef-Zadeh et al. 2002) and the molecular complex called the Bridge (Bamba et al. 2002).
To uncover the nature of the hot diffuse X-ray emission and test the hypothesis of the flaring theory, Churazov et al. (2002) postulated that an X-ray mission equipped with a state-of-the-art polarimeter would be necessary. Indeed, if past radiation from Sgr A ∗ is reprocessed by extended, distant reflection nebulae, the resulting X-ray emission should be polarized. In order to lay the groundwork for a future polarimetric explorer, several studies have been done (Churazov et al. 2002; Matt 2010; Marin et al. 2014), estimating the polarization degree and the orientation of the polarization position angle that a potential mission could detect. Hence, in this research note, we summarize the work accomplished so far to evaluate the net X-ray polarization emerging from the largest and brightest (i.e. easier to detect) GC molecular clouds.
1 Astronomical Institute of the Academy of Sciences, Boˇn´ ı II 1401, CZ-14100 Prague, Czech Republic c
2 INAF/IAPS, Via del Fosso del Cavaliere 100, I-00133 Roma, Italy
Fig. 1. Reflected energy spectrum and polarization degree of a spherical gas cloud illuminated by a distant continuum source. Top : Polarization degree P , bottom : intensity spectra. The left column is representative of a gas clump situated at the same distance from the observer as the source, while the right column presents results for a cloud situated 100 pc away. Spectra are taken from Churazov et al. (2002).
## 2 Evaluating the polarization emerging from the GC
So far, there have been two main publications exploring the resulting X-ray polarization emerging from the GC. The first one, from Churazov et al. (2002), focused on the reprocessing molecular cloud Sgr B2 and estimated the net soft X-ray polarization signature of this cloud. The second paper, by Marin et al. (2014), modeled the 2 ◦ × 2 ◦ ( ∼ 288 × 288 pc; at 8.5 kpc 1 ′′ ≈ 0.04 pc) central region of the Milky Way at energies higher than 8 keV and included polarization maps. We now review the main results from the two complementary publications.
## 2.1 Sgr B2 in the soft X-ray band
Churazov et al. (2002) investigated the polarization emerging from a single cloud of radius 10 pc representative of the reflection nebulae Sgr B2. In their simulation, the spherical gaseous medium was filled with neutral, solar abundance matter and had a Thomson optical depth of 0.5. It was irradiated by a steady beam of unpolarized photons with energy ranging from 2 to 8 keV. Photoelectric absorption, fluorescent emission and Compton scattering by bound electrons were included and the molecular cloud was located at 100 pc from the continuum source, which is representative of the known projected distance from Sgr A ∗ (Murakami et al. 2001). However, since the three-dimensional parametrization of the GC is unknown, the reflecting nebula can be shifted from the Galactic plane but still conserve the same radial projections. To investigate this effect, Churazov et al. (2002) located their cloud model at two different positions: at the same distance and 100 pc away from the observer than the emitting region.
Their work is summarized in Fig. 1 and the general result is that the reflected radiation should be highly polarized ( > 30%) with a direction of polarization normal to the scattering plane. The amount of P is a function of the three-dimensional position of the cloud, decreasing when Sgr B2 departs from the Galactic plane. Local dilution of the polarization degree by fluorescent photons (being unpolarized) is symptomatic of the composition of the cloud, and the slow but steady decrease of P with energy is due to multiple scatterings.
Thus, using a simple model, Churazov et al. (2002) proved that any detection of polarized X-ray emission from the reflection nebulae around Sgr A ∗ would give constraints on the morphology, composition and position of the scattering clouds.
Fig. 2. Integrated 8 - 35 keV model image of the polarized flux, PF/F ∗ , for the 2 ◦ × 2 ◦ region around the GC. PF/F ∗ is color-coded, with the color scale shown on top of the image (in arbitrary units). P and ψ are represented by black bars drawn in the center of each spatial bin. A vertical bar indicates a polarization angle of ψ = 90 ◦ and a horizontal bar stands for an angle of ψ = 0 . The length of the bar is proportional to ◦ P . The polarization map is taken from Marin et al. (2014).
## 2.2 Broadband GC polarization imaging
The results obtained by Churazov et al. (2002) are valid in the 2 - 8 keV band for small hydrogen column densities (n H < 10 22 cm -2 ). At larger n H , the emission below 5 keV is expected to be highly suppressed. In addition to that, past X-ray observations (Koyama et al. 1986, 1989) revealed the presence of a diffuse plasma emission toward the GC that may additionally dilute the polarization signal below 7 keV. Hence, to avoid most of the dilution by the GC plasma emission and extend the soft X-ray simulations achieved by Churazov et al. (2002), 8 - 35 keV modeling has been undertaken by Marin et al. (2014).
In our recent work, Marin et al. (2014) took into account the dense and warm environment around the central few parsecs around Sgr A ∗ (known as the circumnuclear disc, CND); a cold dusty structure in the shape of a continuous chain of irregular clumps representative of the central molecular zone (CMZ, Molinari et al. 2011); the two bright, complex reflection nebulae Sgr B2 and Sgr C (located at the two extrema of the CMZ); and a continuum source displaced by ∼ 22 pc in projection from the center of the model toward the Western galactic longitude (to be consistent with the shifted gas distribution within the CMZ (Molinari et al. 2011).
The polarization map presented in Fig. 2 is extracted from Marin et al. (2014). It is found that the polarized flux, PF/F ∗ , traces the overall shape of the GC, emphasizing the ∞ -shaped CMZ. PF/F ∗ reaches a maximum at the location of the non-axisymmetric CND surrounding Sgr A , but the local, integrated ∗ P is low ( ∼ 1.0%). Sgr B2 and Sgr C show the secondary, brightest polarized flux knots of the map, associated with P = 66.5% for Sgr B2 and P = 47.8% for Sgr C. The difference in P is due to the asymmetrical spatial location of the cloud with respect to Sgr A . ∗ Both the clouds present a polarization position angle ψ perpendicular to the vertical axis of the model (normal to the scattering plane). P is found to vary for each CMZ cloud and reaches a maximum for the Eastern and Western sections of the elliptical twisted ring but it is most likely that a large fraction of the ∞ -shaped ring will be diluted by background, unpolarized emission from both plasma emission and Sgr A . ∗ Finally, when integrating the whole 2 ◦ × 2 ◦ GC polarized emission, the model produces a net polarization degree of 0.9% associated with ψ = -22.8 ◦ . The combined emission from Sgr A ∗ and the CND thus
dominates the whole polarization picture.
Marin et al. (2014) extended the modeling of Churazov et al. (2002) by radiatively coupling the primary source to a large panel of reprocessing targets. They also extend their cloud parametrization to explore the influence of the location of the scattering nebulae on polarization (not shown in this research note), finding that the two reflection nebulae always produce high polarization degrees ( glyph[greatermuch] 10%).
## 3 The need for a dedicated X-ray polarimetry mission
It has been shown by Churazov et al. (2002) and Marin et al. (2014) that the GC is expected to provide a variety of polarization signals where maximum P is expected to arise from reflection nebulae. Marin et al. (2014) also computed the minimum detectable polarization (MDP) that the NHXM (New Hard X-ray Mission, Tagliaferri & NHXM Consortium 2012; Tagliaferri et al. 2012) could have reached and found that such polarization levels are detectable using a 500 ks observation. Errors on ψ being marginal, the detection of ψ normal to the scattering plane would be unambiguous.
Was the GC active a few hundreds years ago? X-ray polarimetry can definitively prove or reject this hypothesis since the main molecular clouds should be highly polarized ( glyph[greatermuch] 10%) with the electric vector perpendicular to the line connecting Sgr A ∗ to the reprocessing nebula. To spatially constrain the three-dimensional location of each GC component with respect to the central SMBH, it will be necessary to observe at least two molecular clouds. An X-ray mission equipped with a polarization imaging detector, such as the Gas Pixel Detector (GPD) based on the photoelectric effect (Costa et al. 2001), is ideally suited since polarization mapping would reveal the complex morphology of the GC, spatially resolving the largest reflection nebulae, differentiating them from neighborhood sources and potentially enabling the investigation of stratified light echoes from the past activity of Sgr A ∗ .
A dedicated space mission for imaging X-ray polarimetry such as the non-selected projects NHXM or IXPE (the Imaging X-ray Polarimeter Explorer, Weisskopf et al. 2008) would have been sufficiently sensitive to measure the polarization emerging from the 2 ◦ × 2 ◦ GC, even below 8 keV where plasma emission is acting like an unpolarized background and can be further substracted from past spectral data. Finally, X-ray polarization measurement would ultimately test the alternative scenario for the origin of X-ray emission from Sgr B2 and Sgr C in which X-ray features are produced by low-energy cosmic-ray electrons rather than by Compton scattering.
The authors would like to acknowledge the financial support from the COST Action MP1104 and COST-CZ LD12010 .
## References
```
References
Bamba, A., Murakami, H., Senda, A., et al. 2002, ArXiv Astrophysics e-prints
Churazov, E., Sunyaev, R., & Sazonov, S. 2002, MNRAS, 330, 817
Costa, E., Soffitta, P., Bellazzini, R., et al. 2001, Nature, 411, 662
Koyama, K., Awaki, H., Kunieda, H., Takano, S., & Tawara, Y. 1989, Nature, 339, 603
Koyama, K., Inui, T., Hyodo, Y., et al. 2007, PASJ, 59, 221
Koyama, K., Maeda, Y., Sonobe, T., et al. 1996, PASJ, 48, 249
Koyama, K., Makishima, K., Tanaka, Y., & Tsunemi, H. 1986, PASJ, 38, 121
Marin, F., Karas, V., Kunneriath, D., & Muleri, F. 2014, MNRAS, 441, 3170
Matt, G. 2010, X-ray polarimetry and radio-quiet AGN, ed. R. Bellazzini, E. Costa, G. Matt, & G. Tagliaferri, 122
Molinari, S., Bally, J., Noriega-Crespo, A., et al. 2011, ApJ, 735, L33
Murakami, H., Koyama, K., Tsujimoto, M., Maeda, Y., & Sakano, M. 2001, ApJ, 550, 297
Nobukawa, M., Ryu, S. G., Tsuru, T. G., & Koyama, K. 2011, ApJ, 739, L52
Ponti, G., Terrier, R., Goldwurn, A., Belanger, G., & Trap, G. 2010, ApJ, 714, 732
Sunyaev, R. A., Markevitch, M., & Pavlinsky, M. 1993, ApJ, 407, 606
Tagliaferri, G., Hornstrup, A., Huouvelin, J., et al. 2012, Experimental Astronomy, 34, 463
Tagliaferri, G. & NHXM Consortium. 2012, Mem. Soc. Astron. Italiana, 83, 360
Weisskopf, M. C., Bellazzini, R., Costa, E., et al. 2008, in Society of Photo-Optical Instrumentation Engineers (SPIE)
Conference Series, Vol. 7011
Yusef-Zadeh, F., Law, C., Wardle, M., et al. 2002, ApJ, 570, 665
```
Yusef-Zadeh, F., Law, C., Wardle, M., et al. 2002, ApJ, 570, 665 | null | [
"F. Marin",
"V. Karas",
"D. Kunneriath",
"F. Muleri",
"P. Soffitta"
] | 2014-08-02T09:58:56+00:00 | 2014-08-02T09:58:56+00:00 | [
"astro-ph.HE",
"astro-ph.GA",
"85-05",
"J.2.3; J.2.9"
] | Probing the Galactic center with X-ray polarimetry | The Galactic center (GC) holds the closest-to-Earth supermassive black hole
(SMBH), which makes it the best laboratory to study the close environment of
extremely massive compact objects. Polarimetry is sensitive to geometry of the
source, which makes it a particularly suitable technique to probe the medium
surrounding the GC SMBH. The detection of hard X-ray spectra and prominent iron
K$\alpha$ fluorescence features coincident with localized gas clouds (e.g.
Sgr~B2, Sgr~C) is known for nearly twenty years now and is commonly associated
with a past outburst of the SMBH whose radiation is reprocessing onto the
so-called "reflection nebulae". Since scattering leads to polarization, the
re-emitted signal from the giant molecular clouds in the first 100~pc of the GC
is expected to be polarized. X-ray polarization measurement is thus
particularly adapted to probe the origin of the diffuse X-ray emission from the
GC reflection nebulae and reveal the past activity of the central SMBH. In this
research note, we summarize the results from past and current polarimetric
simulations in order to show how a future X-ray polarimeter equipped with
imaging detectors could improve our understanding of high-energy astrophysics. |
1408.0355v1 | ## Almost Decouplability of any Directed Weighted Network Topology
Cai, Ning 1 Khan, M. Junaid 2
1
College of Electrical Engineering, Northwest University for Nationalities, Lanzhou, China 2 PN Engineering College, National University of Sciences and Technology, Karachi, Pakistan
Abstract: This paper introduces a conception that any weighted directed network topology is almost decouplable, which can help to transform the topology into a similar form being composed of uncoupled vertices, and thus reduce the complexity of analysis for networked dynamical systems. As an example of its application, the consensus problem of linear multi-agent systems with time-varying network topologies is addressed. As a result, a necessary and sufficient condition for uniform consensus is proposed.
Key Words: Network; Consensus; Multi-Agent Systems
## 1. Introduction
We are entering a networked world. The notion of complex and dynamical networks keeps on influencing nearly all kinds of disciplines, ranging from communication networks, power networks, biological networks, and economic networks, etc. Its impact on modern engineering and science has already been prominent, and will potentially be even far-reaching.
The graph topology of a network corresponds to certain matrix [1]. For dynamical networked systems, it is a fundamental technique to transform the network topology into similar canonical forms which are equivalent to the original network in the sense of system dynamics but are more convenient to analyze. It is usually expected that a complex network could be transformed into a similar diagonal canonical form, being composed of a number of uncoupled vertices [2]. Unfortunately, such a consideration does not always seem applicable, since a directed network is generally not guaranteed to be decouplable.
1 Corresponding author: Cai, Ning (e-mail: [email protected])
This paper will show that any given weighted directed network topology is almost decouplable, despite the diagonalizability of the associated matrix.
The motivation of the current work originated from our study on consensus problems for dynamical multi-agent systems [3-6]. Consensus implicates that a system achieves certain asymptotic stability even if there are no equilibria, and it actually plays the role like equilibrium of isolated dynamical systems [3-5]. Due to its theoretical importance, the consensus problem has been paid extensive attention by scholars in the area of control theory from various perspectives; see [7-11] and references therein. Particularly, a general framework of the consensus problem for first-order systems is initially proposed in [8]. A necessary and sufficient condition for consensus of LTI first-order systems is given in [9]. Consensus problem of high-order systems is earlier addressed in [12-13]. Both observer-based dynamical protocol and robust consensus are dealt with in [14]. A noticeable approach about oblique decomposition of the state space is devised in [15-16]. The consensus condition of a class of systems with switching graph topology is shown in [17]. A method based on relative Lyapunov function is proposed in [6], which could potentially handle both consensus and non-consensus stability problems for systems composed of nonlinear or even heterogeneous agents.
It is a natural tendency to diagonalize the Laplacian matrix of the graph topology while addressing the consensus problems of multi-agent systems with linear high-order inter-agent protocols. However, due to the requirement of diagonalizability, many discussions are limited only to the case of bidirectional networks [13, 17].
We shall endeavor to reveal a fact that it is essentially not a problem whether or not a weighted directed network is decouplable. With this fact known, in many cases one can easily analyze the dynamics of multi-agent systems via the technique of decoupling the directed network through similarity transformation, with no need to consider the actual diagonalizability of associated matrix. The computational complexity might be reduced significantly, since a high-dimensional large-scale system could thus be converted into a series of independent low-dimensional subsystems. So far as we know, this idea is especially advantageous in dealing with the consensus problems of systems with uncertain or time-varying network topologies.
This paper includes two major parts. The first part is purely mathematical, and it theoretically analyzes the concept of almost decouplability. As an example to illustrate
the application of the main theory, the consensus problem of linear high-order multi-agent systems with time-varying graph topologies is concerned in the second part of the paper.
It is worthy to mention that some of the propositions may seem lesser novel alone. Some analogous arguments could also exist in the literature, e.g. [18]. However, by systematically clarifying and integrating the relevant considerations, both from Mathematics and systems theory viewpoints, a novel and effective methodology can be yielded to deal with many problems in dynamical networked systems, especially stability problems. This is the essential contribution of the current paper.
The rest of this paper is organized as follows. Section 2 depicts the almost decouplability concept, providing the main theoretical result. Section 3 deals with the stability problem of general dynamical systems as preparation for the next section. Section 4 discusses the uniform consensus for multi-agent systems, based on the results shown in previous sections. Finally, the concluding remarks form Section 5.
## 2. Almost decouplatility of directed weighted networks
As the most important part of the paper, the current section expounds the almost decouplability conception. We can see that the meaning of this conception is twofold, which will be reflected by Proposition 1 & Theorem 1, respectively: 1) Without considering the limitation of structure, the Lebesgue measure for the set of undiagonalizable square matrices is zero; 2) For any given directed weighted graph, there exists an arbitrarily close graph with diagonalizable Laplacian matrix, which has a prescribed structure limitation.
Proposition 1: Any asymmetric real-valued square matrix N N L R is almost diagonalizable, i.e. the Lebesgue measure for the set of undiagonalizable matrices in N N R is zero.
The subsequent theoretical grounds are essential for proving Proposition 1, which are quoted from linear algebra.
Definition 1 (Resultant) [19]: Suppose that 1 0 1 ( ) m m m f x a x a x a ( 0 0 ) a and 1 0 1 ( ) n n n g x b x b x b ( 0 0 ), b with their roots being 1 2 , , , m and 1 2 , , , n respectively. The resultant of the two polynomials is defined as follows:
$$R ( f, g ) = a _ { 0 } ^ { n } b _ { 0 } ^ { m } \prod _ { i = 1 } ^ { m } \prod _ { j = 1 } ^ { n } ( \alpha _ { i } - \beta _ { j } )$$
With the following formula [19], one can directly compute the resultant of two polynomials by their coefficients:
$$R ( f, g ) = \Delta ( m, n ) = \begin{vmatrix} a _ { 0 } & a _ { 1 } & \cdots & a _ { m } & & & \\ & a _ { 0 } & a _ { 1 } & \cdots & a _ { m } & & \\ & & & \ddots & & \\ & & & & a _ { 0 } & a _ { 1 } & \cdots & a _ { m } \\ b _ { 0 } & b _ { 1 } & \cdots & b _ { n } & & & \\ & b _ { 0 } & b _ { 1 } & \cdots & b _ { n } & & \\ & & & & \ddots & & \\ & & & & b _ { 0 } & b _ { 1 } & \cdots & b _ { n } \end{vmatrix} \\. \ \cdots \cdots \. \ \cdots \. \ \cdots \. \ \cdots \. \ \cdots \. \ \cdots \. \ \cdots \. \ \cdots \. \ \end{vmatrix}$$
Lemma 1 [19]: An irreducible polynomial ( ) is p x a multiple factor of polynomial ( ) if and only if f x ( ) p x is a common factor of ( ) and f x '( ) , f x where '( ) is the derivative of f x ( ) . f x
Lemma 2 [19]: There exists a common factor of polynomials ( ) and f x ( ) if g x and only if the resultant ( , ) 0. R f g
Proof of Proposition 1: Consider any arbitrary matrix N N ij L l R . Suppose that its characteristic polynomial is
$$f ( x ) = | x I - L | = x ^ { N } + a _ { 1 } x ^ { N - 1 } + \cdots + a _ { _ { N } }$$
with the coefficient vector 1 2 T N N a a a R . The derivative of ( ) f x is
$$f \, ^ { \prime } ( x ) = N x ^ { N - 1 } + a _ { 1 } ( N - 1 ) x ^ { N - 2 } + \cdots + a _ { N - 1 }$$
According to Lemmas 1 & 2, there exists a multiple factor of ( ) if and only if f x ( , ') 0, i.e. the coefficients of R f f ( ) must satisfy the following equation: f x
Evidently, the set of variables 1 , , N a a satisfying the above algebraic equation
forms a manifold of dimension 1. N
Equation (2) determines the relationship between the coefficient vector 1 2 T N a a a and the values of entries in L , which are N scalar functions:
$$\begin{cases} a _ { 1 } = \varphi _ { 1 } ( l _ { 1 1 }, l _ { 1 2 }, \cdots, l _ { N N } ) \\ a _ { 2 } = \varphi _ { 2 } ( l _ { 1 1 }, l _ { 1 2 }, \cdots, l _ { N N } ) \\ \vdots \\ a _ { N } = \varphi _ { N } ( l _ { 1 1 }, l _ { 1 2 }, \cdots, l _ { N N } ) \end{cases}$$
Substituting 1 , , N a a in (2) by (3) will result in an equation taking the following form:
$$\Gamma ( l _ { 1 1 }, l _ { 1 2 }, \cdots, l _ { _ { N N } } ) = 0$$
with ( ) being a scalar function. It is evident that the set of matrices that satisfy (4) forms a manifold of lesser dimension in N N R and therefore its Lebesgue measure is zero. All the matrices in N N R except this set have distinct eigenvalues and are surely diagonalizable. □
Example 1: Consider any matrix in 2 2 R , which is
$$\begin{bmatrix} a & b \\ c & d \end{bmatrix}$$
with , , a b c and d as some arbitrary real numbers. Its characteristic polynomial is
$$\begin{vmatrix} x - a & - b \\ - c & x - d \end{vmatrix} = x ^ { 2 } - ( a + d ) x + a d - b c$$
The polynomial has multiple roots if and only if
$$( a + d ) ^ { 2 } = 4 ( a d - b c )$$
which determines a manifold of measure 0 in 2 2 R .□
Remark 1: The zero Lebesgue measure for the set of undiagonalizable matrices also implies that the diagonalizable set is dense. However, the former condition is stronger than the mere density. As a well-known instance, the set of rational numbers is dense in R whereas its measure is still zero.
Remark 2: The analysis in practical scenarios can be much convenient, with the knowledge that the Lebesgue measure of the set of undiagonalizable matrices is zero. Because deviations always exist between models and the real systems, usually a matrix encountered in an engineering scenario, such as an adjacency matrix or Laplacian matrix of a specific network topology, can be regarded as being
diagonalizable according to Proposition 1. In most cases, it is not necessary to consider the undiagonalizable exceptions.
If the graph topology of a network is denoted by its adjacency matrix for the sake of analysis, then there is a one-to-one correspondence between any graph and a matrix in real number domain; thus, Proposition 1 is sufficient to interpret the idea that the graph topology is almost decouplability. However, while concerning the dynamical multi-agent systems with linear inter-agent protocols, Laplacian matrices are more frequently adopted to represent the graph topologies. Laplacian matrix is a specific type of matrix with structure limitation, i.e. the sum of elements in any row is zero. It cannot guarantee that Proposition 1 is still valid if the matrix has certain structure limitation. The next theorem deals with such a situation. The lemmas that follow are theoretical preparations for the theorem.
Definition 2 (Laplacian Matrix) [1]: The Laplacian matrix of a given directed weighted network topology is
L
D
W
where W is the adjacency matrix and D is the in-degree matrix with ( ) . Note that D diag W 1 1 1 T denotes a vector with all entries being 1.
Lemma 3: is an eigenvector of any Laplacian matrix L with the associated eigenvalue being 0.
$$\text{Proo} \colon \text{Because } \ D \phi = d i a g ( W \phi ) \phi = W \phi, \text{ } L \phi = 0 \,. \Box$$
Lemma 3 can be regarded as a structural limitation or a criterion to check whether a matrix is a Laplacian matrix of certain network topology.
Lemma 4 [20]: (Schur's Unitary Triangularization Theorem) With given n n A R , there exists a unitary matrix U such that * UAU is upper triangular.
Lemma 5 [20]: If two matrices [ ] ij A a and [ ] ij B b in n n R are unitarily equivalent, i.e. there is a unitary matrix U such that * UAU B , then
$$\sum _ { i, \, j = 1 } ^ { n } a _ { i j } ^ { 2 } = \sum _ { i, \, j = 1 } ^ { n } b _ { i j } ^ { 2 }$$
Theorem 1: For any directed graph G of N th order and any value 0, there exists a graph ( ) G that is decouplable, i.e. its Laplacian matrix ( ( )) [ ( )] ij L G l is diagonalizable, meanwhile
$$\sum _ { i, j = 1 } ^ { N } \left ( l _ { i j } - l _ { i j } \left ( \varepsilon \right ) \right ) ^ { 2 } < \varepsilon$$
Proof: Suppose that the eigenvalues of L are 2 0, ,..., N . Let N N U R be unitary such that * U LU is upper triangular. Select an upper triangular matrix N N E R in which the elements satisfy
$$| e _ { i j } | < \sqrt { 2 } \varepsilon / N ( N + 1 ) \ ( i = 1, 2, \dots, N \ j = i, \dots, N )$$
such that 2 22 ,..., N NN e e are different from each other. Besides, let the elements in E also jointly satisfy (6):
$$U E U ^ { ^ { * } } \phi = 0$$
Equation (6) is a system of equations with N equations and (1 ) / 2 N N independent variables, so there must exist feasible solutions since ( 1) / 2 . N N N With such an E satisfying (6), let * ( ( )) L G L UEU , then * ( ( ))-= L G L UEU and according to Lemma 5 and (5),
$$\sum _ { i, j = 1 } ^ { N } ( l _ { i j } - l _ { i j } ( \varepsilon ) ) ^ { 2 } = \sum _ { i = 1 } ^ { N } \sum _ { j = i } ^ { N } e _ { i j } ^ { 2 } < \varepsilon$$
Evidently, ( ) L is diagonalizable because its spectrum is {0, 2 22 ,..., }, N NN e e without identical eigenvalues. □
Remark 3: Proposition 1 & Theorem 1 play different roles, and are compensating each other. They jointly constitute the interpretation for the almost decouplability conception proposed in this paper. Proposition 1 implicates a fact that the set of diagonalizable matrices is dense, including most of the matrices; while Theorem 1 concentrates on the diagonalizability of Laplacian matrices, which is the case with structure limitation.
The idea of almost decouplability elaborated in this section may seem obvious, but it can truly help to reduce the complexity of analysis. The next two sections will illustrate the effectiveness of this idea by an example on the consensus problem of dynamical multi-agent systems with time-varying network topologies.
## 3. Uniform asymptotic stability of slightly perturbed time-varying dynamical systems
This section discusses some general problems in control theory, leading to two
main propositions which collectively imply a fact that the uniform asymptotic stability is robust to sufficiently minor uncertainties of system dynamics across the entire time horizon. As a result, it is feasible to study the stability of a dynamical system via replacing it by another system with just slight variation, but with remarkably simplified analysis. The current section is essentially independent of Section 2, and its main purpose is to provide necessary theoretical basis to support the analysis on the consensus problem in the next section.
Definition 3 (Positive Definite Function): A real-valued, continuously differentiable function ( ) is said to be f x positive definite in a neighborhood of the origin D if (0) 0 and f ( ) 0 for any nonzero f x x D . A function ( , ) f t x ([0, ) D R ) is said to be positive definite if ( ,0) 0 and f t ( , ) ( ) in f t x v x [0, ) D for some positive definite function ( ) . v x
Lemma 6 [21]: Let 0 be a uniformly asymptotically stable equilibrium point x for the nonlinear system
$$\dot { x } = f ( t, x )$$
where :[0, ) n f D R ( n D x R x r ) is continuously differentiable both in t and x . Then, there exist a positive real value 0 r and a continuously differentiable function 0 :[0, ) n V D R ( 0 0 n D x R x r ) that satisfies the following series of inequalities
$$\alpha _ { _ { 1 } } ( \| x \| ) \leq V ( t, x ) \leq \alpha _ { _ { 2 } } ( \| x \| )$$
$$\frac { \partial V } { \partial t } + \frac { \partial V } { \partial x } \, f ( t, x ) \leq - \alpha _ { 3 } ( \| x \| )$$
$$\left \| \frac { \partial V } { \partial x } \right \| \leq \alpha _ { _ { 4 } } ( \| x \| )$$
where 1 , 2 , 3 , and 4 are class K functions.
Proposition 2: Suppose that 0 is x a uniformly asymptotically stable equilibrium point of a nonlinear time-varying system ( , ) , where x f t x ( , ) is f t x continuously differentiable both in t and x , then there must exist a positive definite function ( , ) such that t x 0 is also uniformly asymptotically stable for all the x systems:
$$\dot { x } = f ( t, x ) + \delta ( t, x )$$
where ( , ) is any vector function in t x R
$$\| \delta ( t, x ) \| \leq \gamma ( t, x )$$
n that satisfies )
Proof: It is simple to verify that the origin 0 is also an equilibrium point for x any system (10). According to Lemma 6, there exists a continuously differentiable function ( , ) satisfying V t x inequalities (7)~(9). Let ( , ) be V t x the Lyapunov function candidate for system (10). Then its derivative along the trajectory of (10) is
$$& \dot { V } = \frac { \partial V } { \partial t } + \frac { \partial V } { \partial x } \, \dot { x } \\ & = \frac { \partial V } { \partial t } + \frac { \partial V } { \partial x } \, f ( t, x ) + \frac { \partial V } { \partial x } \, \delta ( t, x )$$
In (11) above,
$$\left | \frac { \partial V } { \partial x } \delta ( t, x ) \right | \leq \left \| \frac { \partial V } { \partial x } \right \| \| \delta ( t, x ) \|$$
Let ( , ) t x be a positive definite function satisfying
$$\gamma ( t, x ) < \alpha _ { 3 } ( \| x \| ) / \left \| \frac { \partial V } { \partial x } \right \| \ ( x \neq 0 )$$
Then, because ( , ) ( , ) , t x t x
$$- \alpha _ { 3 } ( \| x \| ) < \frac { \partial V } { \partial x } \, \delta ( t, x ) < \alpha _ { 3 } ( \| x \| )$$
Considering both (13) and (8), it is evident to know that (11) is negative for 0. x Thus, 0 is uniformly asymptotically stable for any system (10). □ x
Proposition 3: Suppose that 0 is x an equilibrium point of a nonlinear time-varying system ( , ) , then there must exist a positive definite function x f t x ( , ) such that t x 0 is uniformly asymptotically stable for system x ( , ) x f t x if and only if 0 is also uniformly asymptotically stable for system x
$$\dot { x } = f ( t, x ) + \delta ( t, x )$$
where ( , ) is any vector function with proper dimension satisfying t x
$$\| \delta ( t, x ) \| \leq \gamma ( t, x )$$
Proof: According to Proposition 2, for any nonlinear system with origin its uniformly asymptotically stable equilibrium point, there is a positive definite function indicating the bound of permitted variation preserving stability. Thus, there is a map between a system and a positive definite function:
$$\dot { x } = f ( t, x ) \rightarrow \gamma ( t, x )$$
Assume that 0 x is uniformly asymptotically stable for a specific system 0 ( , ) , then there exists a positive definite function x f t x 0 ( , ) t x such that 0 is x uniformly asymptotically stable for the set of systems
$$\left \{ \dot { x } = f _ { 0 } ( t, x ) + \delta ( t, x ) \| \| \delta ( t, x ) \| < \gamma _ { 0 } ( t, x ) \right \}$$
Each element in the above set can also induce a positive definite function, e.g.
$$\dot { x } = f _ { 0 } ( t, x ) + \delta _ { 0 } ( t, x ) \rightarrow \hat { \gamma } _ { 0 } ( t, x )$$
with 0 ( , ) t x being a specific vector function satisfying 0 0 ( , ) ( , ) . t x t x As a result, 0 is uniformly asymptotically stable for the set of systems x
$$\left \{ \dot { x } = f _ { 0 } ( t, x ) + \delta _ { 0 } ( t, x ) + \Delta ( t, x ) \left \| \Delta ( t, x ) \right \| < \hat { \gamma } _ { 0 } ( t, x ) \right \}$$
if 0 is uniformly asymptotically stable for the system x 0 0 ( , ) ( , ) . It is x f t x t x simple to construct a positive definite function ( , ) satisfying t x
$$\gamma ( t, x ) \leq \hat { \gamma } _ { 0 } ( t, x )$$
for all such 0 ˆ ( , ) induced by map (16) from elements in the set (15). Evidently, t x 0 x is uniformly asymptotically stable for the specific system 0 ( , ) x f t x if and only if 0 x is uniformly asymptotically stable for all the elements in
$$\left \{ \dot { x } & = f _ { 0 } ( t, x ) + \delta ( t, x ) \left \| \delta ( t, x ) \right \| < \gamma ( t, x ) \right \} & & \Box$$
Combining the theoretical results in Sections 2 & 3, the methodology for stability analysis of networked systems based on almost decouplability is clearly established here. The process can be summarized as follows. First, let an approximate network topology ( ) G with diagonalizable Laplacian matrix or adjacency matrix replace the original G , no matter whether it is actually diagonalizable. Then, decouple the network via a nonsingular transformation and a system composed of independent low-dimensional subsystems is thus derived, which has the same stability with the original system.
## 4. Uniform consensus of multi-agent systems with time-varying network topologies
The major purpose of this section is to provide an example to show the effectiveness of the main theory introduced previously. As the example, the consensus problem will be analyzed for dynamical multi-agent systems with time-varying network topologies, by approaches based on the idea of almost decouplability. A
necessary and sufficient condition will be derived for the uniform consensus.
The dynamic multi-agent system model concerned here comprises of N agents, and each agent has dynamics of d th order. The state of agent i is denoted by 1 2 [ , ,..., ] T d i i i id x x x x R . The communication architecture among agents is represented by a graph topology G of order N , with each agent corresponding to a vertex. The arc weight of G between agents i and j is denoted by 0 , ij w which can be regarded as the strength of information link. The graph G in this section is time-varying, which is sufficiently smooth in t and can be denoted by its adjacency matrix ( ) : W t
$$G \colon W = \begin{bmatrix} w _ { 1 1 } & w _ { 1 2 } & \dots & w _ { 1 N } \\ w _ { 2 1 } & w _ { 2 2 } & \dots & w _ { 2 N } \\ \vdots & \vdots & & \vdots \\ w _ { N 1 } & w _ { N 2 } & \dots & w _ { N N } \end{bmatrix} \\ \dots \dots \end{bmatrix}$$
Suppose the dynamics of each agent can be described as:
$$\dot { x } _ { i } = A x _ { i } + F \sum _ { j = 1 } ^ { N } w _ { i j } ( t ) ( x _ { j } - x _ { i } ) \ \ ( i \in \{ 1, 2, \dots, N \} )$$
where d d A R and d d F R . If a stack of state vectors is defined as 1 2 T T T T N x x x x , then the system dynamics can be described by
$$\dot { x } = ( I _ { _ { N } } \otimes A - L ( t ) \otimes F ) x$$
with ( ) being the time-varying Laplacian matrix of graph L t G .
As the theme of the current section, consensus is essentially a stability problem. The following definitions clearly indicate the relationship between consensus and stability. With these definitions, the two equivalent terms consensus and asymptotic swarm stability will be used indiscriminately later.
Definition 4 [4]: (Swarm Stability) For a time-varying dynamical multi-agent system with 1 ,..., d N x x R as the states of N agents, if for 0 , ( ) 0 , s.t. ( ) ( ) i j x t x t ( 0 ) t as (0) (0) ( ) ( i j x x , {1, 2,..., } ), i j N then the system is uniformly swarm stable . If lim ( ) , the system is globally uniformly swarm stable.
Definition 5 [4]: (Asymptotic Swarm Stability) If a time-varying dynamical multi-agent system is globally uniformly swarm stable and for , 0 , c ( , ) 0 T T c s.t.
$$\left \| x _ { i } ( t ) - x _ { j } ( t ) \right \| < \varepsilon \\ \sim 1 1 \sim$$
as ( , ) t T c and (0) (0) i j x x c ( , {1, 2,..., } ), i j N then the system is globally uniformly asymptotically swarm stable .
Lemma 7 [9]: The Laplacian matrix L of a directed graph G has exactly a single zero eigenvalue 1 0 iff G has a spanning tree, with the corresponding eigenvector 1 1 1 T .
Lemma 8 [4]: For the time-varying dynamic multi-agent system (17), if it is uniformly asymptotic swarm stable, then as t , the trajectory of each agent tends to be regulated by the equation A .
Lemma 9 [4]: For system (17), if A is Hurwitz, then the system is uniformly asymptotically swarm stable iff it is uniformly asymptotically Lyapunov stable. Besides, the consensus state must be zero.
Theorem 2: For the multi-agent system (17), with 1 2 0, ( ), , ( ) N t t C the time-varying eigenvalues of ( ( )) , L G t if A is unstable, the overall system is uniformly asymptotically swarm stable iff both 1) and 2) below are true:
- 1) 0 T s.t. there do not exist any two values 1 t and 2 t ( 2 1 t t T ) while ( ) G t includes no spanning tree during 1 2 [ , ] ; t t t
- 2) All the following d -dimensional dynamic systems
$$\dot { \xi } = ( A - \lambda _ { \tilde { t } } ( t ) F ) \xi \ ( \lambda _ { \tilde { t } } \neq 0 )$$
are uniformly asymptotically stable.
If A is Hurwitz, then the multi-agent system is uniformly asymptotically swarm stable iff 2) is true.
## Proof: Part I ( A is unstable)
Assume that 1) does not hold but the multi-agent system is still uniformly asymptotically swarm stable. Suppose that G has no spanning tree for 1 2 [ , ] , t t t where 1 t and 2 t can be arbitrarily large, then G must contain 2 distinct k subgraphs 1 ˆ G , 2 ˆ G ,…, ˆ k G , each receiving no information during 1 2 [ , ] t t t . According to Lemma 8, as t is sufficiently large, the consentaneous trajectory 1 ( ) t of the agents associated with 1 ˆ G approaches the solution of equation 1 1 A , whereas the consentaneous trajectory 2 ( ) associated with t 2 ˆ G approaches the solution of 2 2 A . The two trajectories are independent because there is no information exchanged. Thus, 1 2 1 2 ( ) . A Since A is unstable, 1 2 tends to increase during 1 2 [ , ] . t t t This contradicts the assumption that the
multi-agent system is asymptotically swarm stable. Therefore, 1) is a necessary condition. Note that the condition ( ) G t possessing no spanning tree is permitted at any particular instant t T due to the continuity of the system.
Suppose that 1 2 0, ( )..., ( ) N t t C are the eigenvalues of ( ) . According to L t Proposition 1 and Theorem 1, let an approximate ( , ) G t ( 0) replace ( ) , G t with ( , ) L t replacing ( ) L t , which is uniformly diagonalizable. Hence, an alternative system of (18) is derived as:
$$\dot { x } = ( I _ { _ { N } } \otimes A - L ( \varepsilon, t ) \otimes F ) x$$
Considering the continuity of the eigenvalues of matrices, one can infer that the eigenvalues of ( , ) also approach L t 1 2 0, ( )..., ( ) as N t t 0.
Now let us consider the swarm stability of (19). Suppose that 1 ( ) ( , ) ( ) ( ) and let T t L t T t t ( ( ) ) d x T t I x , then (19) is transformed into
$$\dot { \tilde { x } } = ( I _ { _ { N } } \otimes A - \Lambda ( t ) \otimes F ) \tilde { x }$$
According to Propositions 2 & 3, the uniform asymptotic stability of system (20) is equivalent to that of system (18), with 2 3 ( ) (0, ( ), ( ), , ( )) . The N t diag t t t structure of ( ) N I A t F is of the form
$$\dots \dots \dots \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N }\dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ { N } \dots _ N }$$
Let the auxiliary variables
$$\eta _ { i } = \sum _ { j = 1 } ^ { N } w _ { i j } ( \varepsilon, t ) ( x _ { j } - x _ { i } ) \ \ ( i = 1, 2, \dots, N )$$
and their stack vector be 1 2 T T T T N . It follows that
$$\eta = ( L ( \varepsilon, t ) \otimes I _ { _ { d } } ) x$$
With condition 1), according to Lemma 7, any vector in the null space of ( , ) d L t I is of the form when t T , where 1 1 1 T N R and d R . Therefore, 1 2 N x x x in (19) iff 0 when t T . Because of (21) and 1 ( ) d x T I x ,
$$( T \otimes I _ { _ { d } } ) \eta = ( T L ( \varepsilon ) \otimes I _ { _ { d } } ) x = ( T L ( \varepsilon ) \otimes I _ { _ { d } } ) ( T ^ { - 1 } \otimes I _ { _ { d } } ) \tilde { x } = ( \Lambda \otimes I _ { _ { d } } ) \tilde { x }$$
According to Lemma 7, all the values 2 , , N are nonzero as t T , thus from the linear equation (23), one can infer that 0 iff 2 0 with N x x considering the structure of ( ) . By (21), it is clearly known that t 2 3 , , , 0 N x x x iff condition 2) holds. According to Propositions 2 & 3, the uniform asymptotic
stability of (19) is identical with that of (18); therefore condition 2) is also appropriate for the system (18).
## Part II ( A is Hurwitz)
As A is Hurwitz, according to Lemma 9, the system is uniformly asymptotically swarm stable iff it is uniformly asymptotically Lyapunov stable. The stability of system (17) is equivalent to that of (21). It is clear that (21) is uniformly asymptotically stable iff condition 2) is true.□
Remark 4: If without the almost decouplability conception introduced in previous sections, it would be difficult or even impossible to handle the proof and obtain the result of Theorem 2, since the network topology here is not only asymmetric but also time-varying.
Remark 5: Admittedly, Theorem 2 is relatively theoretic, since eigenvalues are usually not easy to obtain. Its main role is to demonstrate the effectiveness of the approach based on the idea of almost decouplability. Nonetheless, this theorem itself still has theoretical significance. First, it provides a necessary and sufficient condition without additional technical hypothesis. Second, the consensus of a large-scale system is transformed into the stability of a series of independent low-dimensional systems, and the computational complexity is thus reduced significantly. Besides, this result separates the consideration of network topology from the dynamics of agents.
## 5. Conclusions
This paper introduces a conception that any weighted directed network topology is almost decouplable, i.e. there always exists an arbitrarily close digraph which can be transformed into a similar form, consisting of independent vertices. This conception could help to reduce the complexity of analysis because a large scale networked system can thereby be decoupled into numerous independent low-dimensional systems through similarity transformations. As an example to illustrate the validity of the methodology proposed, the uniform consensus problem of dynamical multi-agent systems with time-varying network topologies is discussed, leading to establishing a necessary and sufficient condition. Evidently, the almost decouplability conception should also be applicable in other problems, e.g. the robust consensus problem, which will be investigated in our future work.
## Acknowledgments
This work is supported by Program for Young Talents of State Ethnic Affairs Commission (SEAC) China (Grant [2013] 231), and by National Natural Science Foundation (NNSF) of China (Grants 61174067, 61263002, & 61374054).
## References
- [1] C. Godsil, G. Royle, Algebraic Graph Theory , Springer-Verlag: Berlin, 2001.
- [2] G. Chen, X. Wang, X. Li, Introduction to Complex Networks: Models, Structure and Dynamics , Higher Education Press: Beijing, 2012.
- [3] N. Cai, Swarm Stability and Controllability of High-Order Swarm Systems , Doctoral Dissertation, Tsinghua University, 2010. (In Chinese)
- [4] N. Cai, J. Xi, Y. Zhong, 'Swarm stability of high order linear time-invariant swarm systems', IET Control Theory Appl. , vol. 5, pp. 402-408, 2011.
- [5] N. Cai, J. Xi, Y . Zhong, 'Asymptotic swarm stability of high order dynamical multi-agent systems: Condition and application', Control and Intelligent Systems , vol. 40, pp. 33-39, 2012.
- [6] N. Cai, J. Cao, H. Ma, C. Wang, 'Swarm stability analysis of nonlinear dynamical multi-agent systems via relative Lyapunov function', Arab. J. Sci. Eng. , vol. 39, pp. 2427-2434, 2014.
- [7] N. Cai, M. Liu, X. Wei, H. Ma, 'On uniform consensus of linear multi-agent systems with time-varying graph topologies', Proc. Chinese Control Conf. , pp. 6896-6899, 2013.
- [8] A. Jadbabaie, J. Lin, A. Morse, 'Coordination of groups of mobile autonomous agents using nearest neighbor rules', IEEE Trans. Autom. Control , vol. 48, pp. 988-1001, 2003.
- [9] R. Olfati-Saber, R. Murray, 'Consensus problems in networks of agents with switching topology and time-delays', IEEE Trans. Autom. Control , vol. 49, pp. 1520-1533, 2004.
- [10] W. Ren, R. Beard, 'Consensus seeking in multiagent systems under dynamically changing interaction topologies', IEEE Trans. Autom. Control , vol. 50, pp. 655-661, 2005.
- [11] H. Hu, W. Yu, Q. Xuan, L. Yu, G. Xie, 'Consensus for second-order agent dynamics with velocity estimators via pinning control', IET Control Theory Appl. , vol. 7, pp. 1196-1205, 2013.
- [12] H. Hu, A. Liu, Q. Xuan, L. Yu, G. Xie, 'Second-order consensus of multi-agent systems in the cooperation-competition network with switching topologies: A time-delayed impulsive control approach', Syst. Control Lett. , vol. 62, pp. 1125-1135, 2013.
- [13] F. Xiao, L. Wang, 'Consensus problems for high-dimensional multi-agent systems', IET Control Theory Appl. , vol. 1, pp. 830-837, 2007.
- [14] J. Wang, D. Cheng, X. Hu, 'Consensus of multi-agent linear dynamic systems', Asian J. Control , vol. 10, pp. 144-155, 2008.
- [15] Z. Li, Z. Duan, G. Chen, L. Huang, 'Consensus of multiagent systems and synchronization of complex networks: A unified viewpoint', IEEE Trans Circuit. Syst. I: Regular Papers , vol. 57, pp. 213-224, 2010.
- [16] J. Xi, Z. Shi, Y. Zhong, 'Consensus analysis and design for high-order linear swarm systems with time-varying delays', Physica A , vol. 390, pp. 4114-4123, 2011.
- [17] J. Xi, Z. Shi, Y. Zhong, 'Output consensus analysis and design for high-order linear swarm systems: Partial stability method', Automatica , vol. 48, pp. 2335-2343, 2012.
- [18] W. Ni, D. Cheng, 'Leader-following consensus of multi-agent systems under fixed and switching topologies', Syst. Control Lett. , vol. 59, pp. 209-217, 2010.
- [19] D.J. Hartfiel, 'Dense sets of diagonalizable matrices', Proc. Am. Math. Soc. , vol. 123, pp. 1669-1672, 1995.
- [20] B. Feng, Polynomials and Irrational Numbers , Harbin Institute of Technology Press: Harbin, 2008. (In Chinese)
- [21] R.A. Horn, C.R. Johnson, Matrix Analysis , Cambridge University Press: Cambridge, 1985.
- [21] H.K. Khalil, Nonlinear Systems (3rd Edition), Prentice Hall: Upper Saddle River, 2002. | null | [
"Ning Cai",
"M. Junaid Khan"
] | 2014-08-02T10:08:43+00:00 | 2014-08-02T10:08:43+00:00 | [
"cs.SY"
] | Almost Decouplability of any Directed Weighted Network Topology | This paper introduces a conception that any weighted directed network
topology is almost decouplable, which can help to transform the topology into a
similar form being composed of uncoupled vertices, and thus reduce the
complexity of analysis for networked dynamical systems. As an example of its
application, the consensus problem of linear multi-agent systems with
time-varying network topologies is addressed. As a result, a necessary and
sufficient condition for uniform consensus is proposed. |
1408.0356v10 | ## SPECIAL ELEMENTS OF THE LATTICE OF EPIGROUP VARIETIES
V.YU.SHAPRYNSKI ˇ I, D.V.SKOKOV, AND B.M.VERNIKOV
Abstract. We study special elements of eight types (namely, neutral, standard, costandard, distributive, codistributive, modular, lower-modular and upper-modular elements) in the lattice EPI of all epigroup varieties. Neutral, standard, costandard, distributive and lower-modular elements are completely determined. A strong necessary condition and a sufficient condition for modular elements are found. Modular elements are completely classified within the class of commutative varieties, while codistributive and upper-modular elements are completely determined within the wider class of strongly permutative varieties. It is verified that an element of EPI is costandard if and only if it is neutral; is standard if and only if it is distributive; is modular whenever it is lower-modular; is neutral if and only if it is lower-modular and upper-modular simultaneously. We found also an application of results concerning neutral and lower-modular elements of EPI for studying of definable sets of epigroup varieties.
## 1. Introduction and summary
1.1. Semigroup pre-history. The main object we examine in this article is the lattice of all epigroup varieties. But our considerations are motivated by some earlier investigations of the lattice of semigroup varieties and closely related with these investigations. To make clearer a context and motivations of our considerations, we start with a brief explanation of the 'semigroup prehistory' of the present work.
One of the main branches of the theory of semigroup varieties is an examination of lattices of semigroup varieties (see the survey [17]). If V is a variety then L ( V ) stands for the subvariety lattice of V under the natural order (the classtheoretical inclusion). The lattice operations in L ( V ) are the (class-theoretical) intersection denoted by X ∧ Y and the join X ∨ Y , i. e., the least subvariety of V containing both X and Y .
There are a number of articles devoted to an examination of identities (first of all, the distributive, modular or Arguesian laws) and some related restrictions (such as semimodularity or semidistributivity) in lattices of semigroup
2010 Mathematics Subject Classification. Primary 20M07, secondary 08B15.
Key words and phrases. Epigroup, variety, lattice, neutral element, standard element, costandard element, distributive element, codistributive element, modular element, lowermodular element, upper-modular element.
Supported by the Ministry of Education and Science of the Russian Federation (project 2248), by a grant of the President of the Russian Federation for supporting of leading scientific schools of the Russian Federation (project 5161.2014.1) and by Russian Foundation for Basic Research (grant 14-01-00524).
varieties, and many considerable results are obtained here. In particular, semigroup varieties with modular, Arguesian or semimodular subvariety lattice were completely classified and deep results concerning semigroup varieties with distributive subvariety lattices (related to a description of such varieties modulo group ones) were obtained. An overview of all these results may be found in [17, Section 11].
The results mentioned above specify, so to say, 'globally' modular or distributive parts of the lattice of semigroup varieties. The following natural step is to examine varieties that guarantee modularity or distributivity, so to say, in their 'neighborhood'. Saying so, we take in mind special elements in the lattice of semigroup varieties. There are many types of special elements that are considered in lattice theory. Recall definitions of some of them. An element x of a lattice 〈 L ; ∨ ∧〉 , is called neutral if ( x ∨ y ) ∧ ( y ∨ z ) ∧ ( z ∨ x ) = ( x ∧ y ) ∨ ( y ∧ z ) ∨ ( z ∧ x ) for all y, z ∈ L ; standard if ( x ∨ y ) ∧ z = ( x ∧ z ) ∨ ( y ∧ z ) for all y, z ∈ L ;
distributive if x ∨ ( y ∧ z ) = ( x ∨ y ) ∧ ( x ∨ z ) for all y, z ∈ L ;
modular if ( x ∨ y ) ∧ z = ( x ∧ z ) ∨ y for all y, z ∈ L with y ≤ z ;
upper-modular if ( z ∨ y ) ∧ x = ( z ∧ x ) ∨ y for all y, z ∈ L with y ≤ x .
Costandard , codistributive and lower-modular elements are defined dually to standard, distributive and upper-modular ones. There is a number of interrelations between types of elements we consider. It is evident that a neutral element is both standard and costandard; a standard or costandard element is modular; a [co]distributive element is lower-modular [upper-modular]. It is well known also that a [co]standard element is [co]distributive (see [2, Theorem 253], for instance). So, eight types of elements defined above form a partially ordered set under class-theoretical inclusion pictured on Fig. 1.
neutral elements
Figure 1. Special elements in abstract lattices
Note that special elements play an essential role in the abstract lattice theory (see [2, Section III.2] or [18, Sections 2.1 and 2.2], for instance). For instance, if an element x of a lattice L is neutral then L is decomposed into a subdirect product of its intervals ( x ] = { y ∈ L | y ≤ x } and [ x ) = { y ∈ L | x ≤ y } (see [2, Theorem 254]). Thus, the knowledge of what elements of a lattice are neutral, gives the important information on a structure of the lattice as a whole.
All types of elements mentioned above are intensively and successfully studied with respect to the lattice SEM of all semigroup varieties. For brevity, a semigroup variety that is a neutral element of the lattice SEM is called a neutral in SEM variety. Analogous convention is applied for all other types of special elements. Results about special elements in SEM are overviewed in the recent survey [29]. In particular,
- · neutral in SEM varieties were completely determined in [37];
- · it is proved that a semigroup variety is costandard in SEM if and only if it is neutral in SEM [27]; thus, in view of the previous result, costandard in SEM varieties are completely classified;
- · distributive in SEM varieties were completely classified in [30];
- · in fact, standard in SEM varieties are completely described in [30] too because the results of this work readily imply that a semigroup variety is standard in SEM if and only if it is distributive in SEM (see comments after Theorem 3.3 in the survey [29]);
- · a strong necessary conditions for modular in SEM varieties were discovered in [7] 1 and [22] (these results are reproved in a simpler way in [12]);
- · a sufficient condition for modular in SEM varieties was found in [32] and rediscovered in [7];
- · commutative modular in SEM varieties were completely determined in [22];
- · lower-modular in SEM varieties were completely classified in [14];
- · commutative upper-modular in SEM varieties were completely classified in [25]; it is noted in [24] that this result may be expanded on wider class of strongly permutative varieties without any change (a definition of strongly permutative varieties see in Subsection 1.2 below);
- · strongly permutative codistributive varieties were completely described in [27].
Note that the articles [26, 27] contain some other results concerning uppermodular and codistributive elements in SEM .
- 1.2. Epigroups. A considerable attention in the semigroup theory is devoted to semigroups equipped by an additional unary operation. Such algebras are said to be unary semigroups . As concrete types of unary semigroups, we mention completely regular semigroups (see [9]), inverse semigroups (see [8]), semigroups with involution etc.
One more natural type of unary semigroups is epigroups. A semigroup S is called an epigroup if, for any element x of S , there is a natural n such that x n is a group element (this means that x n lies in some subgroup of S ). Extensive information about epigroups may be found in the fundamental work [15] by L. N. Shevrin and the survey [16] by the same author. The class of epigroups is very wide. In particular, it includes all periodic semigroups (because some power of each element in such a semigroup lies in some its finite cyclic subgroup)
1 Note that paper [7] deals with the lattice of equational theories of semigroups, that is, the dual of SEM rather than the lattice SEM itself. When reproducing results from [7], we adapt them to the terminology of the present article.
↦
and all completely regular semigroups (in which all elements are group ones). The unary operation on an epigroup is defined by the following way. If S is an epigroup and x ∈ S then some power of x lies in a maximal subgroup of S . We denote this subgroup by G x . The unit element of G x is denoted by x ω . It is well known (see [15], for instance) that the element x ω is well defined and xx ω = x x ω ∈ G x . We denote the element inverse to xx ω in G x by x . The map x -→ x is just the mentioned unary operation on an epigroup S . The element x is called pseudoinverse to x . Throughout this paper, we consider epigroups as algebras with the operations of multiplication and pseudoinversion. In particular, this allows us to consider varieties of epigroups as algebras with the two mentioned operations. An idea to examine epigroups in the framework of the theory of varieties was promoted by L. N. Shevrin in [15] (see also [16]). An overview of first results obtained here may be found in [17, Section 2].
If S is a completely regular semigroup (i. e., the union of groups) and x ∈ S then x is the element inverse to x in the maximal subgroup containing x . Thus, the operation of pseudoinversion on a completely regular semigroup coincides with the unary operation traditionally considered on completely regular semigroups. We see that varieties of completely regular semigroups (considered as unary semigroups) are varieties of epigroups in the sense defined above. Further, it is well known and may be easily checked that in every periodic epigroup the operation of pseudoinversion may be expressed in terms of multiplication (see [15], for instance). This means that periodic varieties of epigroups may be identified with periodic varieties of semigroups.
It seems to be very natural to examine all restrictions on semigroup varieties mentioned in Subsection 1.1 for epigroup varieties. This is realized in [31, 34] for identities and related restrictions to subvariety lattice. In particular, epigroup varieties with modular, Arguesian or semimodular subvariety lattice are completely classified and epigroup analogs of results concerning semigroup varieties with distributive subvariety lattice are obtained there. In the present article, we start with an examination of special elements in the lattice EPI of all epigroup varieties. We consider here elements of all mentioned above eight types in EPI . For brevity, we call an epigroup variety neutral if it is a neutral element of the lattice EPI . Analogous convention will be applied for all other types of special elements. Our main results give:
- · a complete description of neutral, standard, costandard, distributive and lower-modular varieties;
- · a strong necessary condition and a sufficient condition for modular varieties;
- · a description of commutative modular varieties, strongly permutative codistributive varieties and strongly permutative upper-modular varieties.
One can start with formulations of results. We denote by T , SL and ZM the trivial variety, the variety of all semilattices and the variety of all semigroups with zero multiplication respectively. Our first result is the following
Theorem 1.1. For an epigroup variety V , the following are equivalent:
- a) V is a neutral element of the lattice EPI ;
- b) V is a costandard element of the lattice EPI ;
- c) V is simultaneously a lower-modular and upper-modular element of the lattice EPI ;
- d) V coincides with one of the varieties T , SL ZM , or SL∨ZM .
Thus, there are only a few neutral elements in the lattice EPI . In contrast, we note that the lattice of completely regular semigroup varieties contains infinitely many neutral elements including all band varieties, the varieties of all groups, all completely simple semigroups, all orthodox semigroups and some other (this readily follows from [20, Corollary 2.9]).
Let Σ be an identity system written in the language of one associative binary operation and one unary operation. The class of all epigroups that satisfy Σ (where the unary operation is treated as pseudoinversion) is denoted by K Σ . The class K Σ is not obliged to be a variety because it maybe not closed under taking of (infinite) direct product (see [16, Subsection 2.3], for instance). Note that identity systems Σ with the property that K Σ is a variety are completely determined in [4, Proposition 2.15] (see Lemma 2.4 below). If K Σ is a variety then we use for this variety the standard notation var Σ. It is evident that if the class K Σ consists of periodic epigroups then it is a periodic semigroup variety and therefore, is an epigroup variety. Whence, the notation var Σ is correct in this case. A pair of identities wx = xw = w where the letter x does not occur in the word w is usually written as the symbolic identity w = 0. This notation is justified because a semigroup with such identities has a zero element and all values of the word w in this semigroup are equal to zero. We will refer to the expression w = 0 as to a single identity. Such identities and varieties given by them are called 0reduced . Put
$$\{ r ^ { 2 } u = r u r = u r ^ { 2 } =$$
$$\L \cos \L \cos \L \cos \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \sin \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \sin \L \sin \L \cos \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \sin \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \sin \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \sin \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \sin \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \sin \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \sin \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \sin \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \sin \L \cos \L \sin \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \ cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \Sigma \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \Sigma \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L \cos \L.$$
where n is a natural number. We note that Q 1 = R 1 = T . Our second result is the following
Theorem 1.2. For an epigroup variety V , the following are equivalent:
- a) V is a distributive element of the lattice EPI ;
- b) V is a standard element of the lattice EPI ;
- c) V = M∨N where M is one of the varieties T or SL , and N is one of the varieties Q Q , n , R or R n .
The following result gives a complete classification of lower-modular varieties.
Theorem 1.3. An epigroup variety V is a lower-modular element of the lattice EPI if and only if V = M∨N where M is one of the varieties T or SL , while N is a 0 -reduced variety.
We follow the agreement that an adjective indicating a property shared by all semigroups of a given variety is applied to the variety itself; the expressions like
'completely regular variety', 'periodic variety', 'nilvariety', etc. are understood in this sense. Recall that an identity of the form
$$x _ { 1 } x _ { 2 } \cdots x _ { n } = x _ { 1 \pi } x _ { 2 \pi } \cdots x _ { n \pi }$$
/negationslash where π is a non-trivial permutation on the set { 1 2 , , . . . , n } is called permutative ; if 1 π = 1 and nπ = n then this identity is said to be strongly permutative . A semigroup or a variety that satisfies [strongly] permutative identity also is called [ strongly ] permutative . The following two results describe codistributive and upper-modular varieties in the strongly permutative case.
/negationslash
Theorem 1.4. A strongly permutative epigroup variety V is a codistributive element of the lattice EPI if and only if V = G ∨ X where G is an Abelian group variety, while X is one of the varieties T , SL ZM , or SL∨ZM .
Put C m = var { x m = x m +1 , xy = yx } for arbitrary natural m . In particular, C 1 = SL . It will be convenient for us also to assume that C 0 = T .
Theorem 1.5. A strongly permutative epigroup variety V is an upper-modular element of the lattice EPI if and only if one of the following holds:
- (i) V = M∨N where M is one of the varieties T or SL , and N is a nilvariety satisfying the commutative law and the identity
- (1.1) x y 2 = xy 2 ;
- (ii) V = G ∨ C m ∨ N where G is an Abelian group variety, 0 ≤ m ≤ 2 and N satisfies the commutative law and the identity
$$x ^ { 2 } y & = 0.$$
The remaining three results devoted to modular varieties. A word written in the language of multiplication and pseudoinversion is called a semigroup word if it does not include the operation of pseudoinversion. An identity is called a semigroup identity if both its parts are semigroup words. A semigroup identity u = v is said to be substitutive if u and v depend on the same letters and v may be obtained from u by renaming of letters. The following result gives a strong necessary condition for modular varieties.
Theorem 1.6. If an epigroup variety V is a modular element of the lattice EPI then V = M∨N where M is one of the varieties T or SL , and N is a nilvariety given by 0 -reduced and substitutive identities only.
It is easy to see that this theorem completely reduces the problem of description of modular varieties to nilvarieties defined by 0-reduced and substitutive identities only (see Corollary 3.3 below).
We provide the following sufficient condition for modular varieties.
Theorem 1.7. A 0 -reduced semigroup variety is a modular element of the lattice EPI .
Theorems 1.6 and 1.7 show that in order to describe modular varieties we need to examine nil-varieties satisfying substitutive identities. A natural partial case of substitutive identities are permutative ones, while the strongest permutative identity is the commutative law. Modular varieties satisfying this law possess a complete description.
Theorem 1.8. A commutative epigroup variety V is a modular element of the lattice EPI if and only if V = M∨N where M is one of the varieties T or SL and N is a nilvariety that satisfies the commutative law and the identity (1.2) .
Theorems 1.6 and 1.7 provide a necessary and a sufficient condition for an epigroup variety to be modular respectively. The gap between these conditions seems to be not very large. But the necessary condition is not a sufficient one, while the sufficient condition is not a necessary one. This follows from Theorem 1.8. Indeed, this theorem shows that the variety var { x 3 = 0 , xy = yx } is not modular although it is given by 0-reduced and substitutive identities only, while the variety var { x y 2 = 0 , xy = yx } is modular although it is not 0-reduced.
The article is structured as follows. It consists of eleven sections. In Section 2 we collect definitions, notation and auxiliary results used in what follows. In Section 3 we verify two special cases of Theorem 1.1, namely the claims that the varieties SL and ZM are neutral. These facts are used in each of Sections 4-10. After that we prove Theorem 1.5 in Section 4, Theorem 1.4 in Section 5, Theorems 1.6-1.8 in Section 6 and Theorem 1.3 in Section 7. In Section 8 we discuss an interesting application of Theorems 1.1 and 1.3 concerning so-called definable sets of varieties (a corresponding definition see in Section 8). In Sections 9 and 10 we prove Theorems 1.1 and 1.2 respectively. Sections 4-7, 9 and 10 contain also a number of corollaries of main results. In particular, we note that an epigroup variety is modular whenever it is lower-modular (Corollary 7.3). Finally, in Section 11 we formulate several open questions.
## 2. Preliminaries
- 2.1. Some properties of the operation of pseudoinversion. The following three lemmas are well known and may be easily checked.
Lemma 2.1. The identity
(2.1)
x
=
x
holds in an epigroup S if and only if S is completely regular. /square
It is well known (see [15,16], for instance) that if S is an epigroup and x ∈ S then xx = xx = x ω . This permits to write in epigroup identities expressions of the form u ω rather than uu , for brevity.
Lemma 2.2. If an epigroup variety V satisfies the identity x m = x m +1 then the identities x ω = x = x = x m hold in V . /square
Lemma 2.3. The identity
(2.2)
x = 0
holds in an epigroup if and only if is
S S a nil-semigroup. /square
- 2.2. When K Σ is a variety? An identity is called mixed if one of its parts is a semigroup word but another one is not. Further, a semigroup identity is called balanced if each letter occurs in both its parts the same number of times.
Lemma 2.4 ( [4, Proposition 2.15]) . The class K Σ is an epigroup variety if and only if Σ contains either a semigroup non-balanced identity or a mixed identity. /square
- 2.3. Identities of certain varieties. We denote by F the free unary semigroup over a countably infinite alphabet (with the operations · and ). Elements of F are called words . If w ∈ F then we denote by c w ( ) the set of all letters occurring in w and by t ( w ) the last letter of w . A letter is called simple [ multiple ] in a word w if it occurs in w ones [at least twice]. Put
$$\Lambda ^ { \cdot } \, \stackrel { \cdot } { \mathcal { P } } = \text{var} \{ x y = x ^ { 2 } y, \, x ^ { 2 } y ^ { 2 } = y ^ { 2 } x ^ { 2 } \} \quad \text{and} \quad \stackrel { \cdot } { \mathcal { P } } = \text{var} \{ x y = x y ^ { 2 }, \, x ^ { 2 } y ^ { 2 } = y ^ { 2 } x ^ { 2 } \}.$$
The symbol ≡ stands for the equality relation on the unary semigroup F . The first two claims of the following lemma are well-known and may be easily verified, the third one was proved in [1, Lemma 7].
## Lemma 2.5. A non-trivial semigroup identity v = w holds:
- (i) in the variety SL if and only if c v ( ) = c w ( ) ;
- (ii) in the variety C 2 if and only if c v ( ) = c w ( ) and every letter from c v ( ) is either simple both in v and w or multiple both in v and w ;
- (iii) in the variety P if and only if c v ( ) = c w ( ) and either the letters t ( v ) and t ( w ) are multiple in v and w respectively or t ( v ) ≡ t ( w ) and the letter t ( v ) is simple both in v and w . /square
If w is a semigroup word then /lscript ( w ) stands for the length of w ; otherwise, we put /lscript ( w ) = ∞ . We need the following three remarks about identities of nil-semigroups.
## Lemma 2.6. Let V be a nilvariety.
/negationslash
- (i) If the variety V satisfies an identity u = v with c u ( ) = c v ( ) then V satisfies also the identity u = 0 .
- (ii) If the variety V satisfies an identity of the form u = vuw where the word vw is non-empty then V satisfies also the identity u = 0 .
/negationslash
- (iii) If the variety V satisfies an identity of the form x x 1 2 · · · x n = v and /lscript ( v ) = n then V satisfies also the identity x x 1 2 · · · x n = 0 .
Proof. The claims (i) and (ii) are well known and easily verified.
/negationslash
- (iii) If v is a non-semigroup word, it suffices to refer to Lemma 2.3. Let now v be a semigroup word. If /lscript ( v ) < n then c v ( ) = { x , x 1 2 , . . . , x n } , and the desired conclusion follows from the claim (i). Finally, if /lscript ( v ) > n then the claim we prove readily follows from [10, Lemma 1]. /square
- 2.4. Decomposition of some varieties into the join of subvarieties. As usual, we denote by Gr S the set of all group elements of an epigroup S . For an arbitrary epigroup variety X , we put Gr( X ) = X ∧ GR where GR is the variety of all groups. The variety generated by an epigroup S is denoted by var S . Put
$$\mathcal { L Z } = \text{var} \{ x y = x \} \text{ \ and \ } \mathcal { R Z } = \text{var} \{ x y = y \}.$$
The following two facts play an important role in the proof of Theorem 1.5. 'Semigroup prototypes' of Proposition 2.7 and Lemma 2.8 were given in [35, Proposition 1] and [5] respectively.
Proposition 2.7. If V is an epigroup variety and V does not contain the varieties LZ , RZ P , and ← -P then V = M∨N where M is a variety generated by a monoid, and N is a nilvariety.
Proof. It is verified in [35, Lemma 2] that if a semigroup variety does not contain the varieties LZ RZ P , , and ← -P then V satisfies the quasiidentity
$$e ^ { 2 } = e \longrightarrow e x = x e.$$
Repeating literally the proof of this claim (with using a term 'subepigroup' rather than 'subsemigroup'), one can establish that the similar claim is true for epigroup varieties. Thus, V satisfies the quasiidentity (2.3). The rest of the proof is quite similar to the proof of Proposition 1 in [35].
Let S be an epigroup that generates the variety V , x ∈ S and E the set of all idempotents from S . In view of (2.3), ES is an ideal in S . By the definition of an epigroup, there is a natural n such that x n ∈ Gr S . Then x n = x x ω n and x ω ∈ E . We see that x n ∈ ES . Therefore, the Rees quotient semigroup S/ES is a nil-semigroup and therefore, is an epigroup. The natural homomorphism ρ from S onto S/ES separates elements from S \ ES .
Let now e ∈ E . In view of (2.3), we have that eS is a subsemigroup in S . It is well known that every epigroup satisfies the identity x = xx 2 (see [15, 16], for instance). Hence the equality ex = ex ( ex ) 2 holds. We have verified that, for any e ∈ E , the set eS is a subepigroup in S . Put S ∗ = ∏ e ∈ E eS . Then S ∗
is an epigroup with unit ( . . . , e, . . . ) e ∈ E . It follows from (2.3) that the map ε from S into S ∗ given by the rule ε x ( ) = ( . . . , ex, . . . ) e ∈ E is a semigroup homomorphism. As is well known (see [15,16], for instance), an arbitrary semigroup homomorphism ξ from an epigroup S 1 into an epigroup S 2 is also an epigroup homomorphism (i. e., ξ ( a ) = ξ a ( ) for any a ∈ S 1 ). Therefore, ε is an epigroup homomorphism from S into S ∗ . One can verify that ε separates elements of ES . Let e, f ∈ E x,y , ∈ S and ε ex ( ) = ε fy ( ). Then e · ex = e · fy and f · ex = f · fy . Since e, f ∈ E , we have
$$e x = e f y \text{ \ and \ } f e x = f y.$$
Therefore,
$$e x \stackrel { ( 2. 4 ) } { = } e f y \stackrel { ( 2. 4 ) } { = } e f e x \stackrel { ( 2. 3 ) } { = } f e e x \stackrel { e \in E } { = } f e x \stackrel { ( 2. 4 ) } { = } f y.$$
We see that ex = fy whenever ε ex ( ) = ε fy ( ). This means that ε separates elements of ES .
Thus, ε and ρ are homomorphisms from S into S ∗ and S/ES respectively, and the intersection of kernels of these homomorphisms is trivial. Therefore, the epigroup S is decomposable into a subdirect product of the epigroups S ∗ and S/ES , whence V ⊆ M ∨ N where M = var S ∗ is a variety generated by a monoid and N = var( S/ES ) is a nilvariety. On the other hand S , S/ES ∗ ∈ V , whence M∨N ⊆V . We have proved that V = M∨N . /square
Recall that an epigroup is called combinatorial if all its subgroups are trivial.
Lemma 2.8. If an epigroup variety M is generated by a commutative epigroup with unit then M = G ∨ C m for some Abelian group variety G and some m ≥ 0 .
Proof. It is well known that the variety of all Abelian groups is the least nonperiodic epigroup variety. This variety evidently contains the infinite cyclic group. Further, for each natural m , let G m denote the cyclic group of order m .
It is evident that if M is periodic then the set { m ∈ N | G m ∈ M} has the greatest element. We denote by G the infinite cyclic group whenever the variety M is non-periodic, and the finite cyclic group of the greatest order among all cyclic groups in M otherwise. In both the cases G ∈ M . Further, let D m be the finite cyclic combinatorial epigroup of order m and d m is a generator of D m . Put X = { m ∈ N | D m ∈ M} . If the set X has not the greatest element then the semigroup ∏ m X ∈ D m is not an epigroup since, for example, no power of the element ( . . . , d m , . . . ) m X ∈ belongs to a subgroup. Therefore, the set of numbers X contains the greatest number. We denote this number by n . Repeating literally arguments from the proof of Theorem 1 in [5], we have that every epigroup from M is a homomorphic image of some subepigroup of the epigroup G × D n . Therefore, M = G ∨ D where G = var G and D = var D n . Clearly, G is a variety of Abelian groups. The variety D is generated by a finite epigroup, whence it may be considered as a semigroup variety. It is well known and may be easily verified that ( m + 1)-element combinatorial cyclic monoid generates the variety C m . Therefore, D = C m for some m ≥ 0. /square
It is evident that a strongly permutative variety does not contain the varieties LZ RZ P , , and ← -P . Besides that, every monoid satisfying a permutative identity is commutative. Thus, we have the following corollary of Proposition 2.7 and Lemma 2.8.
Corollary 2.9. If V is a strongly permutative epigroup variety then V = G ∨ C m ∨ N where G is an Abelian group variety, m ≥ 0 and N is a nilvariety. /square
- 2.5. A direct decomposition of one varietal lattice. We denote by AG the variety of all Abelian groups. The aim of this subsection is to prove the following
Proposition 2.10. The lattice L ( AG ∨ C 2 ∨ Q ) is isomorphic to the direct product of the lattices L ( AG ) and L ( C 2 ∨ Q ) .
Proof. We need the following auxiliary statement.
Lemma 2.11. If X ⊆ AG ∨ C 2 ∨ Q then X = G ∨ C m ∨ N where G is some Abelian group variety, 0 ≤ m ≤ 2 and N ⊆ Q .
Proof. Being a subvariety of the variety AG ∨ C 2 ∨Q , the variety X satisfies the identity x y 2 = yx 2 . It is evident that this identity fails in the varieties LZ and RZ . Further, Lemma 2.5(iii) and the dual statement imply that this identity is false in the varieties P and ← -P as well. Therefore, none of the four mentioned varieties is contained in X . Besides that, the variety AG ∨ C 2 ∨Q (and therefore, X ) satisfies the identity x yz 2 = x zy 2 . Substituting 1 for x , we have that all monoids in X are commutative. Proposition 2.7 and Lemma 2.8 imply now that X = G ∨ C m ∨ N for some Abelian group variety G , some m ≥ 0 and some nilvariety N . It is evident that G ⊆ AG . Lemmas 2.1, 2.2 and 2.3 imply that AG C , m and Q satisfy the identities (2.1), x = x m and (2.2) respectively. Therefore, the identity x y 2 = x 2 y holds in the variety AG ∨ C 2 ∨ Q . But this identity is false in the variety C m with m > 2. Hence m ≤ 2. This implies
that the variety AG ∨ C m ∨ Q satisfies the identities x y 2 = yx 2 = x 2 y and xyx = xy x . Since N ⊆ X ⊆ AG ∨ C 2 ∨ Q , Lemma 2.3 implies that the variety N satisfies the identities x y 2 = xyx = yx 2 = 0, whence N ⊆ Q . /square
A semigroup word w is called linear if every letter from c w ( ) is simple in w . For convenience of references, we formulate the following observation that will be helpful also in Section 10.
Lemma 2.12. All non-semigroup words and all non-linear semigroup words except x 2 equal to 0 in the variety Q .
Proof. If u is a non-semigroup word then u = 0 in Q by Lemma 2.3. The claim that a non-linear semigroup word differ from x 2 equal to 0 in Q is evident. /square
Now we start with the direct proof of Proposition 2.10. Let V ⊆ AG ∨ C 2 ∨Q . In view of Lemma 2.11, V = G ∨ C m ∨N for some Abelian group variety G , some 0 ≤ m ≤ 2 and some variety N with N ⊆ Q . Put U = C m ∨ N . We have that V = G ∨ U where G ⊆ AG and U ⊆ C 2 ∨ Q . It remains to establish that this decomposition of the variety V into the join of some subvariety of the variety AG and some subvariety of the variety C 2 ∨ Q is unique.
Let V = G ∨ U ′ ′ where G ′ ⊆ AG and U ′ ⊆ C 2 ∨ Q . We need to verify that G = G ′ and U = U ′ . Let u = v be an arbitrary identity satisfied by G . The variety U satisfies the identity x 3 = x 4 . Then the identity u v 4 3 = u v 3 4 holds in the variety G ∨ U . Let us cancel this identity on u 3 from the left and on v 3 from the right, thus concluding that u = v holds in Gr( G ∨ U ). Therefore, Gr ( G ∨ U ) ⊆ G . The opposite inclusion is evident. Thus, Gr ( G ∨ U ) = G . Analogously, Gr ( G ∨ U ′ ′ ) = G ′ . We see that
$$\mathcal { G } = \text{Gr} \left ( \mathcal { G } \vee \mathcal { U } \right ) = \text{Gr} \left ( \mathcal { V } \right ) = \text{Gr} \left ( \mathcal { G } ^ { \prime } \vee \mathcal { U } ^ { \prime } \right ) = \mathcal { G } ^ { \prime }.$$
It remains to check that U = U ′ . Recall that U = C m ∨N where 0 ≤ m ≤ 2 and N ⊆ Q , while U ′ ⊆ C 2 ∨Q . It is evident that the variety C 2 ∨Q (and therefore, U ′ ) is combinatorial. Therefore, Lemma 2.11 implies that U ′ = C k ∨N ′ for some 0 ≤ k ≤ 2 and some variety N ′ with N ⊆ Q ′ .
/negationslash
Suppose that m = k . We may assume without any loss that m < k , i. e., either m = 0, 1 ≤ k ≤ 2 or m = 1, k = 2. Suppose at first that m = 0 and 1 ≤ k ≤ 2. It is evident that any group satisfies the identity x ω = y ω . Lemma 2.3 implies that this identity holds in N and therefore, in V = G ∨ T ∨ N = G ∨ N . But Lemma 2.5(ii) implies that this identity fails in the variety SL . However, this is impossible because SL ⊆ G ∨ C ′ k ∨ N ′ = V . Suppose now that m = 1 and k = 2. Then Lemmas 2.1, 2.2 and 2.3 imply that the identity x y 2 = x 2 y holds in the variety V = G ∨ C m ∨N = G ∨ SL ∨ N . But this identity is false in the variety C 2 (and therefore, in G ∨ C ′ k ∨ N ′ = V ) by Lemmas 2.2 and 2.5(ii). A contradiction shows that m = k .
Note that if X is an arbitrary epigroup variety then the class of all epigroups in X satisfying the identity (2.2) is the greatest nilsubvariety of X . We denote this subvariety by Nil( X ). Put N = Nil ( U ) and N ′ = Nil ( U ′ ). It suffices to verify that N = N ′ because
$$\mathcal { U } = \mathcal { C } _ { m } \vee \mathcal { N } = \mathcal { C } _ { m } \vee \overline { \mathcal { N } } = \mathcal { C } _ { k } \vee \overline { \mathcal { N } ^ { \prime } } = \mathcal { C } _ { k } \vee \mathcal { N } ^ { \prime } = \mathcal { U } ^ { \prime }$$
in this case. Suppose that N /negationslash = N ′ . It suffices to verify that
$$\mathcal { G } \vee \mathcal { C } _ { m } \vee \overline { \mathcal { N } } \neq \mathcal { G } ^ { \prime } \vee \mathcal { C } _ { k } \vee \overline { \mathcal { N } ^ { \prime } }$$
because this contradicts the equalities
$$\mathcal { G } \vee \mathcal { C } _ { m } \vee \overline { \mathcal { N } } = \mathcal { G } \vee \mathcal { C } _ { m } \vee \mathcal { N } = \mathcal { V } = \mathcal { G } ^ { \prime } \vee \mathcal { C } _ { k } \vee \mathcal { N } ^ { \prime } = \mathcal { G } ^ { \prime } \vee \mathcal { C } _ { k } \vee \overline { \mathcal { N } ^ { \prime } }.$$
We will say that varieties X 1 and X 2 differ with an identity u = v if this identity holds in one of the varieties X 1 or X 2 but fails in another one. Since N /negationslash = N ′ , the varieties N and N ′ differ with some identity. We may assume without loss of generality that this identity holds in N but fails in N ′ . Suppose at first that the identity we mention is a 0-reduced identity u = 0. Since N ⊆ Q ′ , this identity fails in Q . In view of Lemma 2.12, either u is a linear word or u ≡ x 2 . Suppose that u ≡ x x 1 2 · · · x n for some n . Then the identity x x 1 2 . . . x n = 0 holds in N but fails in N ′ . If m = 2 then the variety N contains the variety Nil ( C 2 ) = var { x 2 = 0 , xy = yx } . The latest variety does not satisfy the identity x x 1 2 . . . x n = 0. Therefore, m ≤ 1. Then the variety G ∨ C m is completely regular. Lemmas 2.1 and 2.3 imply now that the variety G ∨ C m ∨N satisfies the identity x x 1 2 · · · x n = x 1 x 2 · · · x n . But Lemma 2.3 implies that this identity is false in N ′ and therefore, in G ∨ C ′ k ∨ N ′ . Thus, (2.5) holds. Let now u ≡ x 2 . Then Lemmas 2.1, 2.2 and 2.3 imply that the identity x 2 = x 2 holds in the variety G ∨ C m ∨N but is false in the variety G ∨ C ′ k ∨N ′ . We see again that the inequality (2.5) holds.
/negationslash
It remains to consider the case when N and N ′ differ with some non-0reduced identity u = v . Suppose that c u ( ) = c v ( ). Lemma 2.6(i) then implies that the variety N satisfies both the identities u = 0 and v = 0. Then the variety N ′ does not satisfy at least one of them because N and N ′ do not differ with u = v otherwise. We see that N and N ′ differ with some 0-reduced identity, and we go to the situation considered in the previous paragraph. Let now c u ( ) = c v ( ). Suppose that the identity u = 0 holds in Q . Since N N ⊆ Q , ′ , we have that the identity u = 0 holds in both the varieties N and N ′ . Then N satisfies also the identity v = 0. But v = 0 fails in N ′ because u = v holds in N ′ otherwise. We see that the varieties N and N ′ differ with some 0-reduced identity. This case has been already considered in the previous paragraph. Thus the identity u = 0 fails in Q . Analogously, v = 0 fails in Q . Since the identity u = v is non-trivial and c u ( ) = c v ( ), Lemma 2.12 implies that the words u and v are linear. Using the fact that c u ( ) = c v ( ) again, we have that the identity u = v is permutative. Then it is evident that this identity holds in G ∨ C m ∨N but fails in G ∨ C ′ k ∨ N ′ . We prove that the inequality (2.5) holds.
Thus, (2.5) fulfills always, whence we are done.
/square
Analog of Proposition 2.10 for semigroup varieties was proved in [21, Proposition 2a] (namely it was checked there that L ( G ∨ C 2 ∨Q ∼ ) = L ( G ) × L ( C 2 ∨Q ) where G is an Abelian periodic group variety). The proof of Proposition 2.10 given above is quite similar to the proof of the mentioned result from [21]. But results of [21] was not used directly above. Therefore, the mentioned result from [21] may be considered now as a consequence of Proposition 2.10.
2.6. Varieties of finite degree. If n is a natural number then a variety X is called a variety of degree n if all nil-semigroups in X are nilpotent of degree ≤ n and n is the least number with such a property. If X is not a variety of degree ≤ n , we will say that X is a variety of degree > n . A variety is said to be a variety of finite degree if it is a variety of degree n for some n . If V is a variety of finite degree, we denote the degree of V by deg( V ); otherwise we put deg( V ) = ∞ . We need the following
Proposition 2.13 ([4, Corollary 1.3]) . Let n be an arbitrary natural number. For an epigroup variety V , the following are equivalent:
- 1) deg( V ) ≤ n ;
- 2) V /notsuperseteql var { x 2 = x x 1 2 · · · x n +1 = 0 , xy = yx } ;
- 3) V satisfies an identity of the form
$$x _ { 1 } \cdots x _ { n } = x _ { 1 } \cdots x _ { i - 1 } \cdot \overline { \overline { x _ { i } \cdots x _ { j } } } \cdot x _ { j + 1 } \cdots x _ { n }$$
for some i and j with 1 ≤ i ≤ j ≤ n .
$$\Box$$
Proposition 2.13 readily implies
Corollary 2.14. deg( X ∧ Y ) = min { deg( X ) , deg( Y ) } for arbitrary epigroup varieties X and Y . /square
The following corollary may be proved quite analogously to Corollary 2.13 of [25] with referring to Proposition 2.13 rather than Proposition 2.11 of [25].
Corollary 2.15. If V is an arbitrary epigroup variety and N is a nilvariety then deg( V ∨ N ) = max { deg( V ) , deg( N ) } . /square
Note that the analog of Corollary 2.15 for arbitrary epigroup varieties is wrong even in the periodic case. For instance, it is easy to deduce from Lemma 2.5(iii), the dual fact and Proposition 2.13 that deg( P ) = deg ( ← -P ) = 2 but deg ( P ∨ P ← -) = 3.
Proposition 2.13 and Lemma 2.1 easily imply
Corollary 2.16. If V is an arbitrary epigroup variety and X is a completely regular variety then deg( V ∨ X ) = deg( V ) . /square
2.7. Some properties of special elements of lattices. The following claim is well known.
Lemma 2.17 ([3, Lemma II.1.1]) . If an element of a lattice L is distributive and modular in L then it is standard in L . /square
This fact together with [2, Theorem 255(iii)] imply the following
Lemma 2.18. An element of a lattice L is neutral in L if and only if it is distributive, codistributive and modular in L . /square
Let I be a lattice identity of the form s = t where s and t are lattice terms depending on ordering set of variables x , x 0 1 , . . . , x n . An element x of a lattice L is called an I -element 2 if s x, x ( 1 , . . . , x n ) = t ( x, x 1 , . . . , x n ) for all x , . . . , x 1 n ∈
2 Probably, it would be more correct to say about ( I, σ )-elements where σ = ( x , x 0 1 , . . . , x n ) is a permutation on the set of variables occurring in I .
$$n$$
- L . Special elements of all types mentioned above are I -elements for appropriate identities I and ordering of variables occurring in I (this claim is evident for neutral, [co]standard and [co]distributive elements, and follows from the fact that the modular law may be written in the form of identity for modular, upper-modular and lower-modular elements).
Lemma 2.19 ([11, Corollary 2.1]) . Let I be a non-trivial lattice identity, L a lattice with 0 , x ∈ L and a an atom and a neutral element of the lattice L . Then x is an I -element in L if and only if x ∨ a has the same property. /square
- 2.8. SL and ZM are atoms. It is evident that atoms of the lattice EPI coincide with atoms of the lattice SEM . The list of atoms of the latter lattice is generally known (see [17], for instance). In particular, the following is valid.
Lemma 2.20. The varieties SL and ZM are atoms of the lattice EPI . /square
## 3. The varieties SL and ZM are neutral
Proposition 3.1. The variety SL is a neutral element of the lattice EPI .
Proof. In view of Lemma 2.18, it suffices to verify that the variety SL is distributive, codistributive and modular. Let X and Y be arbitrary epigroup varieties.
Distributivity . We need to verify the inclusion
$$( \mathcal { S L } \vee \mathcal { X } ) \wedge ( \mathcal { S L } \vee \mathcal { Y } ) \subseteq \mathcal { S L } \vee ( \mathcal { X } \wedge \mathcal { Y } )$$
because the opposite inclusion is evident. Suppose that the identity u = v holds in SL∨ X ∧ Y ( ). In particular, it holds in SL , whence c u ( ) = c v ( ) by Lemma 2.5(i). Let u ≡ w , w , . . . , w 0 1 n ≡ v be a deduction of this identity from identities of the varieties X and Y . Further considerations are given by induction on n .
Induction base. If n = 1 then the identity u = v holds in one of the varieties X or Y . Whence, it holds in one of the varieties SL∨X or SL∨Y , and therefore ( SL∨X ∧ SL∨Y ) ( ) satisfies u = v .
/negationslash
/negationslash
Induction step. Let now n > 1. Consider the words w , . . . , w ′ 1 ′ n -1 obtained from the words w , . . . , w 1 n -1 respectively by equating all the letters that are not occur in u , for some letter of c u ( ). Clearly, the sequence of words u, w ′ 1 , . . . , w ′ n -1 , v also is a deduction of the identity u = v from identities of the varieties X and Y . Thus, we may assume that c w ( 1 ) , . . . , c ( w n -1 ) ⊆ c u ( ). If c w ( 0 ) = c w ( 1 ) = · · · = ( c w n ) then the sequence w , w , . . . , w 0 1 n is a deduction of the identity u = v from identities of the varieties SL∨X and SL∨Y , and we are done. Suppose now that c w ( k ) = c w ( k +1 ) for some 0 ≤ k ≤ n -1. Let i be the least index with c w ( i ) = c w ( i +1 ) and j be the greatest index with c w ( j ) = c w ( j -1 ). Suppose that i > 0. Then c w ( i ) = c u ( ) = c v ( ) and consequences of words w , w , . . . , w 0 1 i and w , w i i +1 , . . . , w n are deductions of the identities u = w i and w i = v respectively from the identities of the varieties X and Y . Lemma 2.5(i) implies now that the identities u = w i and w i = v hold in the variety SL∨ X ∧ Y ( ). By induction assumption these identities hold also in the variety ( SL∨X ∧ SL∨Y ) ( ). Whence, the last variety satisfies the
/negationslash
/negationslash
/negationslash identity u = v too. The case when j < n may be considered quite analogously. Thus, we may suppose that i = 0 and j = n . In other words, c u ( ) = c w ( 1 ) and c v ( ) = c w ( n -1 ).
/negationslash
The identity u = w 1 holds in one of the varieties X and Y . Suppose that it holds in X . Since c u ( ) = c w ( 1 ), Lemma 2.5(i) implies that SL /notsubseteql X . Let S be an epigroup in X and ζ a homomorphism from F to S . For a word w , we denote by w ζ the image of w under ζ . It is well known (see [15,16], for instance) that a variety that does not contain SL consists of archimedean epigroups. Further, a set of group elements in an archimedean epigroup is an ideal of this epigroup. In particular, this is the case for the epigroup S . Now we are going to check that u ζ ∈ Gr S . Since c w ( 1 ) ⊂ c u ( ), there is a letter x ∈ c u ( ) \ c w ( 1 ). Substituting x ω for x in the identity u = w 1 , we obtain the identity u ′ = w 1 that holds in X . Therefore, X satisfies the identity u = u ′ . The word u ′ contains a subword x ω . Since ( x ω ζ ) ∈ Gr S and Gr S is an ideal in S , we have that u ζ ∈ Gr S . Therefore, X satisfies the identity u = uu ω . Similar arguments show that the identity u = uu ω holds in Y whenever Y satisfies u = w 1 , and that one of the varieties X and Y satisfies the identity v = vv ω . Therefore, the sequence of words
$$u, u u ^ { \omega }, w _ { 1 } u ^ { \omega }, \dots, w _ { n - 1 } u ^ { \omega }, v u ^ { \omega }, v w _ { 1 } ^ { \omega }, \dots, v w _ { n - 1 } ^ { \omega }, v v ^ { \omega }, v$$
is a deduction of the identity u = v from identities of the varieties SL∨X and SL∨Y . Hence this identity holds in ( SL∨X ∧ SL∨Y ) ( ).
Codistributivity. In view of Lemma 2.20, if W is an arbitrary epigroup variety then either W ⊇ SL or W∧SL = T . We need to verify that
$$\mathcal { S L } \wedge ( \mathcal { X } \vee \mathcal { Y } ) = ( \mathcal { S L } \wedge \mathcal { X } ) \vee ( \mathcal { S L } \wedge \mathcal { Y } ).$$
Clearly, both the parts of this equality coincides with SL whenever at least one of the varieties X or Y contains SL . It remains to verify that if X /notsuperseteql SL and Y /notsuperseteql SL then X ∨ Y /notsuperseteql SL . This claim immediately follows from the fact that there is a non-trivial identity u = v such that an epigroup variety W does not contain the variety SL if and only if W satisfies the identity u = v (in particular, the identity ( x y ω ω x ω ω ) = x ω has such a property, see [16, Corollary 3.2], for instance).
Modularity. Let X ⊆ Y . We need to verify that
## ( SL∨X ∧Y ⊆ SL∧Y ∨X ) ( )
/negationslash because the opposite inclusion is evident. If SL ⊆ Y then both the parts of the inclusion evidently coincides with SL∨X . Let now SL /notsubseteql Y . Then Lemma 2.20 implies that SL∧Y = T , whence ( SL∧Y ∨X ) = X . Suppose that an identity u = v holds in the variety X . It suffices to verify that this identity holds in ( SL∨X ∧Y ) too. If c u ( ) = c v ( ) then u = v holds in SL by Lemma 2.5(i). Whence it satisfies in SL∨X , and we are done. Let now c u ( ) = c v ( ). Since SL /notsubseteql Y , Lemma 2.5(i) implies that Y satisfies an identity s = t with c s ( ) = c t ( ). We may assume without any loss that there is a letter y ∈ c t ( ) \ c s ( ). Moreover, we may assume that c s ( ) = { x } and c t ( ) = { x, y } (if this is not the case then we equate all letters but y to x in the identity s = t and multiply the identity we obtain on x from the right). Let s 1 [respectively t 1 ] be the word obtained
/negationslash
from the word s [respectively t ] by replacing the letters x and y each to other. Clearly, the identity s 1 = t 1 follows from the identity s = . t
Consider the case when the word v may be obtained from u by replacing of one letter to another one. We may assume without loss of generality that we substitute the letter y for the letter x . Let u 1 , u 2 , u 3 and u 4 be the words obtained from u by substitution of the words s , s 1 , t and t 1 respectively for the letter x . Since x / c v ∈ ( ), the identities u 1 = v , u 2 = v , u 3 = v and u 4 = v are obtained from u = v by the same substitutions. Therefore, the words u , v , u 1 , u 2 , u 3 and u 4 are equal each to other in the variety X . Further, c u ( ) = c u ( 1 ), c v ( ) = c u ( 2 ) and c u ( 3 ) = c u ( 4 ), whence the identities u = u 1 , v = u 2 and u 3 = u 4 hold in SL∨X by Lemma 2.5(i). The identities u 1 = u 3 and u 2 = u 4 follow from s = t and s 1 = t 1 respectively. Hence these identities hold in Y . Thus, the sequence of words u, u 1 , u 3 , u 4 , u 2 , v is a deduction of the identity u = v from the identities of the varieties SL∨X and Y .
Finally, consider an arbitrary identity u = v that holds in X . Replacing one by one all the letters from c u ( ) \ c v ( ) by some letters of c v ( ), we obtain the sequence u, w 1 , . . . , w m , in which any adjacent words differ by replacing one letter. In the sequence of identities u = v, w 1 = v, . . . , w m = v , every identity (except the first one) is obtained from the previous one by replacing one letter. Therefore the words u, w 1 , . . . , w m are equal each to other in X . As we have proved above, this implies that the identities u = w 1 = · · · = w m hold in ( SL ∨ X ) ∧ Y . Analogously, if we replace in the identity v = w m all letters from c v ( ) \ c w ( m ) by an arbitrary letter from c w ( m ), then we obtain a sequence of identities v = w ′ 1 = · · · = w ′ n that hold in ( SL∨X ∧Y ) as well. In particular, these identities hold in X . Moreover, since c w ( m ) = c w ( ′ n ), Lemma 2.5(i) implies that the identity w m = w ′ n holds in SL∨X . Therefore, the identity u = v holds in ( SL∨X ∧Y ) . /square
## Proposition 3.2. The variety ZM is a neutral element of the lattice EPI .
Proof. In view of Lemma 2.18, it suffices to check that ZM is distributive, codistributive and modular. Let X and Y be arbitrary epigroup varieties.
Distributivity. We need to verify that
## ( ZM∨X ∧ ZM∨Y ⊆ZM∨ X ∧Y ) ( ) ( )
because the opposite inclusion is evident. Suppose that an identity u = v holds in ZM∨ X ∧Y ( ). We aim to check that this identity is satisfied by the variety ( ZM∨X ∧ ZM∨Y ) ( ). The identity u = v holds in ZM and there is a deduction of this identity from identities of the varieties X and Y . In other words, there are words u , u 0 1 , . . . , u n such that u 0 ≡ u , u n ≡ v and, for each i = 0 1 , , . . . , n -1, the identity u i = u i +1 holds in one of the varieties X and Y . Let u , u 0 1 , . . . , u n be the shortest sequence of words with such properties. If all the words u , u 0 1 , . . . , u n are not letters then u 0 = u 1 = · · · = u n holds in ZM . This means that the sequence of words u , u 0 1 , . . . , u n is a deduction of the identity u = v from identities of the varieties ZM∨X and ZM∨Y , whence u = v holds in ( ZM∨X ∧ ZM∨Y ) ( ). Let now i be an index such that u i ≡ x for some letter x . Clearly, 0 < i < n because the variety ZM satisfies the identity u 0 = u n but does not satisfy any identity of the kind
x = w . The identity u i -1 = x holds in one of the varieties X and Y , say in X . Then Y satisfies the identity x = u i +1 . Since both the identities u i -1 = x and x = u i +1 fail in ZM , we have that ZM is contained neither in X nor in Y . Therefore, the varieties X and Y are completely regular. By Lemma 2.1 each of the identities u 0 = u 0 and u n = u n holds in one of the varieties X and Y . Further, for each i = 0 1 , , . . . , n -1 one of the varieties X and Y satisfies the identity u i = u i +1 . The words u 0 and u n are not letters, whence the variety ZM satisfies the identities u 0 = u 0 = u 1 = · · · = u n = u n . Summarizing all we say, we obtain that the sequence of words u , u 0 0 , u 1 , . . . , u n , u n is a deduction of the identity u = v from the identities of the varieties ZM∨X and ZM∨Y . Therefore, this identity holds in ( ZM∨X ∧ ZM∨Y ) ( ).
Codistributivity. In view of Lemma 2.20, if W is an arbitrary epigroup variety then either W ⊇ ZM or W∧ZM = T . We need to verify that
$$\mathcal { Z } \mathcal { M } \wedge ( \mathcal { X } \vee \mathcal { Y } ) = ( \mathcal { Z } \mathcal { M } \wedge \mathcal { X } ) \vee ( \mathcal { Z } \mathcal { M } \wedge \mathcal { Y } ).$$
Clearly, both the parts of this equality equals ZM whenever at least one of the varieties X or Y contains ZM . It remains to verify that if X /notsuperseteql ZM and Y /notsuperseteql ZM then X ∨ Y /notsuperseteql ZM . This claim immediately follows from the wellknown fact that an epigroup variety W does not contain the variety ZM if and only if W is completely regular.
Modularity. For any epigroup variety X , we put CR( X ) = CR∧X where CR is the variety of all completely regular epigroups. Suppose that X ⊆ Y . We need to prove that
$$( 3. 1 ) & & ( \mathcal { Z M } \vee \mathcal { X } ) \wedge \mathcal { Y } \subseteq ( \mathcal { Z M } \wedge \mathcal { Y } ) \vee \mathcal { X }$$
because the opposite inclusion is evident. If ZM⊆X [respectively ZM⊆Y ] then both the parts of the inclusion (3.1) coincide with X [with ZM∨X ]. Let now ZM /notsubseteql X and ZM /notsubseteql Y . Then the varieties X and Y are completely regular. Therefore,
$$( \mathcal { Z M } \vee \mathcal { X } ) \wedge \mathcal { Y } \subseteq \text{CR} ( \mathcal { Z M } \vee \mathcal { X } ).$$
Further, ZM∧Y = T , whence the right part of the inclusion (3.1) coincides with X . Let u = v be an identity that holds in X . Lemmas 2.1 and 2.3 imply that ZM∨X satisfies the identity u = v . Therefore, u = v in CR( ZM∨X ) by Lemma 2.1. We have proved that CR( ZM∨X ⊆X ) , whence
$$( \mathcal { Z M } \vee \mathcal { X } ) \wedge \mathcal { Y } \subseteq \text{CR} ( \mathcal { Z M } \vee \mathcal { X } ) \subseteq \mathcal { X } = \mathcal { T } \vee \mathcal { X } = ( \mathcal { Z M } \wedge \mathcal { Y } ) \vee \mathcal { X }.$$
Proposition is proved.
$$\Box$$
For convenience of references, we formulate the following fact that immediately follows from Lemmas 2.19 and 2.20 and Propositions 3.1 and 3.2.
Corollary 3.3. Let I be a non-trivial lattice identity and W is one of the varieties SL ZM , or SL∨ZM . An epigroup variety X is an I -element of the lattice EPI if and only if the variety X ∨ W has the same property. /square
## 4. Upper-modular varieties
Here we verify Theorem 1.5. To do this, we need several auxiliary statements.
Lemma 4.1. If a strongly permutative epigroup variety V is an upper-modular element of the lattice EPI then V is commutative.
/negationslash
Proof. In view of Corollary 2.9, V = G ∨ C m ∨ N where G is an Abelian group variety, m ≥ 0 and N is a nilvariety. If deg( V ) ≤ 2 then N ⊆ ZM , and we are done. Let now deg( V ) > 2. By Proposition 2.13 V contains the variety X = var { x 2 = xyz = 0 , xy = yx } . Suppose that V is not commutative. Let G ′ be a non-abelian group variety. Since V is strongly permutative, every group in V is abelian. Therefore, the variety ( G ∧ V ′ ) ∨ X is commutative. Since X ⊆ V and the variety V is upper-modular, we have that ( G ∧ V ′ ) ∨ X = ( G ∨ X ′ ) ∧ V . We see that the variety ( G ∨ X ′ ) ∧ V is commutative. Hence there is a deduction of the identity xy = yx from the identities of the varieties G ∨ X ′ and V . In particular, there is a word v such that v ≡ xy and the identity xy = v holds either in G ∨ X ′ or in V . The claims (i) and (iii) of Lemma 2.6 imply that a variety with the identity xy = v is either commutative or a variety of degree ≤ 2. The variety G ∨ X ′ is neither commutative (because G ′ is non-abelian) nor a variety of degree ≤ 2 (because deg( X ) > 2). Since deg( V ) > 2, we have that V is commutative. /square
A semigroup variety is called proper if it differs from the variety of all semigroups. It is proved in [25, Theorem 1.1] that if V is a proper upper-modular in SEM variety then, first, V is periodic , and, second, every nilsubvariety of 3 V is commutative and satisfies the identity (1.1). As we have already mentioned in Subsection 1.2, the epigroup analog of the first claim is not true. Our next step is the following partial epigroup analog of the second claim.
Proposition 4.2. If a strongly permutative epigroup variety V is an uppermodular element of the lattice EPI then every nil-semigroup in V satisfies the identity (1.1) .
Proof. According to Lemma 4.1 the variety V is commutative. If all nil-semigroups in V are singleton then the desirable conclusion is evident. Suppose now that V contains a non-singleton nil-semigroup N . Let N be the variety generated by N . Clearly, this variety is commutative. It is evident that ZM⊆N . We need to verify that N satisfies the identity (1.1). Put
$$\mathcal { I } = \text{var} \{ x ^ { 2 } y = x y ^ { 2 }, \, x y = y x, \, x ^ { 2 } y z = 0 \}$$
and N ′ = N ∧ I . It is clear that ZM⊆I , whence ZM⊆N ′ .
A semigroup analog of the proposition we verify is proved in [25] (see the last paragraph of Section 3 in that article). The arguments used there are based on the fact that there is a variety X such that the following two claims are valid:
- (i) ( X ∨ N ′ ) ∧ V ⊆ I ;
3 Note that this is a very special case of the following result obtained in [13, Theorem 1]: if I is a non-trivial lattice identity then a proper semigroup variety is periodic whenever it is an I -element of the lattice SEM .
- (ii) if v ∈ { x y, xyx, yx 2 2 } and w ∈ { xy , yxy, y 2 2 x } then the identity v = w fails in X .
In [25] some periodic group variety plays the role of X . Here we should take another X . Namely, put X = LZM∨RZM where
$$\mathcal { L Z } \mathcal { M } = \text{var} \{ x y z = x y \} \text{ \ and \ } \mathcal { R Z } \mathcal { M } = \text{var} \{ x y z = y z \}.$$
The variety X satisfies the identity xyzxy = xy . Therefore, Lemma 2.5(i) implies that SL /notsubseteql X . Further, substituting 1 for x and y in the identity xyzxy = xy , we obtain that all groups in X are singleton. Hence every commutative semigroup in X is a nil-semigroup. Further, X satisfies the identity xy = ( xy ) 2 , whence all nil-semigroups in X lie in ZM by Lemma 2.6(ii). Since the variety X ∧ V is commutative, X ∧ V ⊆ ZM . The variety V is uppermodular and N ⊆ V ′ . Therefore,
$$( \mathcal { X } \vee \mathcal { N } ^ { \prime } ) \wedge \mathcal { V } = ( \mathcal { X } \wedge \mathcal { V } ) \vee \mathcal { N } ^ { \prime } \subseteq \mathcal { Z } \mathcal { M } \vee \mathcal { N } ^ { \prime } = \mathcal { N } ^ { \prime } \subseteq \mathcal { I }.$$
We have proved the claim (i). To verify the claim (ii), we note that if X satisfies a semigroup identity v = w then the words v and w have the same prefix of length 2 and the same suffix of length 2. Clearly, this is not the case whenever v ∈ { x y, xyx, yx 2 2 } and w ∈ { xy , yxy, y 2 2 x } . Now we can complete the proof by the same arguments as in the last paragraph of [25, Section 3]. /square
The proof of the following statement repeats almost literally the 'only if' part of the proof of Theorem 2 in [33].
Proposition 4.3. If a nilvariety of epigroups X satisfies the identities (1.1) and xy = yx then X is an upper-modular element of the lattice EPI .
Proof. It is easy to prove (see [33, Lemma 2.7], for instance) that X satisfies the identity x yz 2 = 0. Thus, X ⊆ I . Put
$$U = \{ x ^ { 2 }, x ^ { 3 }, x ^ { 2 } y, x _ { 1 } x _ { 2 } \cdots x _ { n } \ | \ n \in \mathbb { N } \}.$$
/negationslash
It is evident that any subvariety of I may be given in I only by identities of the type u = v or u = 0 where u, v ∈ U . Lemma 2.6 implies that if u, v ∈ U and u ≡ v then u = v implies in I the identity u = 0. Now it is very easy to check that the lattice L ( I ) has the form shown on Fig. 2 where
$$\mathcal { I } _ { n } = \text{var} \{ x ^ { 2 } y z = x _ { 1 } x _ { 2 } \cdots x _ { n } = 0, \, x ^ { 2 } y = x y ^ { 2 }, \, x y = y x \} \text{ where } n \geq 4,$$
$$\mathcal { J } = \text{var} \{ x ^ { 2 } y z = x ^ { 3 } = 0, \, x ^ { 2 } y = x y ^ { 2 }, \, x y = y x \},$$
J n = var { x yz 2 = x 3 = x x 1 2 · · · x n = 0 , x y 2 = xy , xy 2 = yx } where n ≥ 4 ,
K = var { x y 2 = 0 , xy = yx , }
K n = var { x y 2 = x x 1 2 · · · x n = 0 , xy = yx } where n ≥ 3 ,
- L = var { x 2 = 0 , xy = yx , }
L n = var { x 2 = x x 1 2 · · · x n = 0 , xy = yx } where n ∈ N .
Note that L 1 = T and L 2 = ZM .
Let X ⊆ I . We have to check that if Y ⊆ X and Z is an arbitrary epigroup variety then ( Z ∨ Y ) ∧ X = ( Z ∧ X ) ∨ Y . For a variety M with M⊆I , we denote by M ∗ the least of the varieties I , J , K and L that contains M . Fig. 2
Figure 2. The lattice L ( I )
shows that if M M ⊆I 1 , 2 then M 1 = M 2 if and only if deg( M 1 ) = deg( M 2 ) and M ∗ 1 = M ∗ 2 . Therefore, we have to verify the following two equalities:
$$( 4. 1 ) & & \deg ( ( \mathcal { Z } \vee \mathcal { Y } ) \wedge \mathcal { X } ) = \deg ( ( \mathcal { Z } \wedge \mathcal { X } ) \vee \mathcal { Y } ),$$
$$\left ( ( \mathcal { Z } \vee \mathcal { Y } ) \wedge \mathcal { X } \right ) ^ { * } = \left ( ( \mathcal { Z } \wedge \mathcal { X } ) \vee \mathcal { Y } \right ) ^ { * }.$$
The equality (4.1). Put deg( X ) = k , deg( Y ) = /lscript and deg( Z ) = m . Clearly, /lscript ≤ k because Y ⊆ X . The set of all natural numbers with the operations min and max is a distributive lattice. Now Corollaries 2.14 and 2.15 apply and we conclude that
$$\deg ( ( \mathcal { Z } \vee \mathcal { Y } ) \wedge \mathcal { X } ) & = \min \{ \max \{ m, \ell \}, k \} = \max \{ \min \{ m, k \}, \min \{ \ell, k \} \} \\ & = \max \{ \min \{ m, k \}, \ell \} = \deg ( ( \mathcal { Z } \wedge \mathcal { X } ) \vee \mathcal { Y } ).$$
The equality (4.1) is proved.
The equality (4.2). Clearly, this equality is equivalent to the following claim: if u is one of the words x 3 , x y 2 and x 2 then the variety ( Z ∨ Y ) ∧X satisfies the identity u = 0 if and only if the variety ( Z ∧ X ) ∨ Y does so. It suffices to verify that u = 0 holds in ( Z ∨ Y ) ∧ X whenever it is so in ( Z ∧ X ) ∨ Y because the opposite claim immediately follows from the evident inclusion ( Z ∧ X ) ∨ Y ⊆ ( Z ∨ Y ) ∧X . Further considerations are divided into two cases.
/negationslash
Case 1: u ≡ x n where n ∈ { 2 3 , } . Then x n = 0 in ( Z ∧ X ) ∨ Y . This means that x n = 0 in Y and there is a deduction of the identity x n = 0 from the identities of the varieties Z and X . In particular, there is a word v such that v ≡ x n and x n = v holds in either Z or X . If x n = v in X then the fact that X is a nil-variety together with the claims (i) and (ii) of Lemma 2.6 imply that x n = 0 in X , and moreover in ( Z ∨ Y ) ∧ X . Let now x n = v in Z . The case when v is a semigroup word may be considered by the same arguments as in the Case 1 in the 'only if' part of the proof of Theorem 2 in [33]. Let now v is a non-semigroup word. Then, in view of Lemma 2.3, the identity v = 0 holds in the variety X , and therefore in Y . Recall that x n = 0 in Y . Therefore, the identity x n = v holds in Y . Since this identity holds in Z as well, we obtain that it holds in Y ∨ Z . We see that the sequence x n , v , 0 is a deduction of the identity x n = 0 from the identities of the varieties Y ∨ Z and X , whence x n = 0 holds in ( Y ∨ Z ) ∧ X .
Case 2: u ≡ x y 2 . We have to check that if the variety ( Z ∧ X ) ∨ Y satisfies the identity (1.2) then the variety ( Z ∨ Y ) ∧X also satisfies this identity. Put
$$W = \{ x ^ { 2 } y, x y x, y x ^ { 2 }, y ^ { 2 } x, y x y, x y ^ { 2 } \}.$$
The variety ( Z ∨ Y ) ∧X is commutative. Therefore, it suffices to verify that this variety satisfies an identity w = 0 for some word w ∈ W . By the hypothesis the variety ( Z ∧ X ) ∨ Y satisfies the identity (1.2). This means that this identity holds in Y and there is a deduction of the identity from identities of the varieties X and Z . Let x y 2 ≡ u , u 0 1 , . . . , u n , 0 be an arbitrary such deduction. The case when u n ∈ W may be considered by the same way as in the 'only if' part of the proof of Theorem 2 in [33].
Let now u n / W ∈ . Since u 0 ∈ W , there is an index i > 0 such that u i / W ∈ while u i -1 ∈ W . The identity u i -1 = u i holds in one of the varieties Z and X . If u i -1 = u i holds in X then X satisfies the identity u i -1 = 0 (this follows from [33, Lemma 2.5] whenever u i is a semigroup word and from Lemma 2.3 otherwise). Therefore, u i -1 = 0 holds in ( Z ∨ Y ) ∧X . Since u i -1 ∈ W , we are done.
Finally, suppose that u i -1 = u i holds in Z . If u i is a semigroup word then we may complete the proof by the same arguments as in the 'only if' part of the proof of Theorem 2 in [33]. Suppose now that u i is not a semigroup word. Lemma 2.3 implies then that the variety Y satisfies the identity u i = 0. Hence the identity u i -1 = u i holds in Y , and therefore the variety Y ∨ Z satisfies this identity. Applying Lemma 2.3 again, we conclude that u i = 0 holds in X . Whence, the sequence u i -1 , u i , 0 is a deduction of the identity u i -1 = 0 from the identities of the varieties Y ∨ Z and X . Thus u i -1 = 0 holds in ( Y ∨ Z ) ∧ X , and we are done.
The equality (4.2) is proved. Thus, we have proved Proposition 4.3. /square
Proof of Theorem 1.5. Necessity. Let V be a strongly permutative uppermodular epigroup variety. In view of Corollary 2.9, V = G ∨ C m ∨ N where G is an abelian group variety, m ≥ 0 and N is a nilvariety. Lemma 4.1 and Proposition 4.2 imply respectively that N is commutative and satisfies the identity (1.1). The variety C m contains a nilsubvariety var { x m = 0 , xy = yx } .
Clearly, this variety does not satisfy the identity (1.1) whenever m ≥ 3. Now Proposition 4.2 applies again and we conclude that m ≤ 2. If the variety N satisfies the identity (1.2) then the claim (ii) of Theorem 1.5 holds. Suppose now that the identity (1.2) fails in N . By [35, Lemma 7] this implies that N contains the variety J . We need to verify that G = T and m ≤ 1. Arguing by contradiction, suppose that either G /negationslash = T or m ≥ 2. Then V contains a variety of the form X ∨ J where X is either a non-trivial abelian group variety or the variety C 2 . It is well known and may be easily checked that the variety C 2 is generated by the epigroup
$$C _ { 2 } = \langle e, a \ | \ e ^ { 2 } = e, \ e a = a e = a, \, a ^ { 2 } = 0 \rangle = \{ e, a, 0 \}$$
and e is the unit of C 2 . Thus, M is generated by an epigroup with unit in any case. Suppose that X satisfies the identity (1.1). Substituting 1 for y in this identity, we have that x 2 = x holds in X . But this identity is false both in a non-trivial group variety and in the variety C 2 . As it is verified in the proof of [35, Lemma 8], this implies that (1.1) is false in any nil-semigroup in X ∨ J . But this contradicts Proposition 4.2.
Sufficiency. If V satisfies the claim (i) of Theorem 1.5 then V is uppermodular by Proposition 4.3 and Corollary 3.3. Suppose now that V satisfies the claim (ii). In other words, V = G ∨ C m ∨ N where G is an abelian group variety, 0 ≤ m ≤ 2, and N satisfies the identities xy = yx and (1.2). Let Y ⊆ V and Z an arbitrary epigroup variety. We aim to verify that
$$( \mathcal { Z } \vee \mathcal { Y } ) \wedge \mathcal { V } = ( \mathcal { Z } \wedge \mathcal { V } ) \vee \mathcal { Y }.$$
As we have already mentioned in the proof of Lemma 2.8, the variety C m is generated by the ( m +1)-element combinatorial cyclic monoid C m and the set X = { m ∈ N | C m ∈ X} has the greatest element. For any m ≥ 0, let c m be a generator of C m . Put C = ∏ m X ∈ C m . Then the semigroup C is not an epigroup because no power of the element ( . . . , c m , . . . ) m X ∈ belongs to a subgroup of C . Thus, the set X has the greatest element. We denote this element by m and put C ( X ) = C m . It is clear that the varieties ( Z ∨ Y ) ∧ V and ( Z ∧ V ) ∨Y are commutative. In view of Corollary 2.9, it suffices to verify the following three equalities:
$$\text{(4.3)} & & \text{Gr} ( ( \mathcal { Z } \vee \mathcal { Y } ) \wedge \mathcal { V } ) = \text{Gr} ( ( \mathcal { Z } \wedge \mathcal { V } ) \vee \mathcal { Y } ),$$
$$C \left ( ( \mathcal { Z } \vee \mathcal { Y } ) \wedge \mathcal { V } \right ) = C \left ( ( \mathcal { Z } \wedge \mathcal { V } ) \vee \mathcal { Y } \right ),$$
$$N \text{il} ( ( \mathcal { Z } \vee \mathcal { Y } ) \wedge \mathcal { V } ) = N \text{il} ( ( \mathcal { Z } \wedge \mathcal { V } ) \vee \mathcal { Y } ).$$
The equality (4.3). If G is a periodic group variety then we denote by exp( G ) the exponent of G , that is the least number n such that G satisfies the identity x = x n +1 . For a non-periodic group variety G , we put exp( G ) = ∞ . Put G 1 = Gr ( ( Z ∨ Y ) ∧ V ) and G 2 = Gr ( ( Z ∧ V ) ∨Y ) . Since G 1 , G 2 ⊆ V , we have that G 1 and G 2 are abelian group varieties. To prove the equality (4.3), it suffices to verify that exp( G 1 ) = exp( G 2 ). This claim is verified by the same arguments as the analogous claim in the proof of Theorem 1.2 in [25].
The equality (4.4). Here and below we need the following easy remark. It is evident that if m ≥ 3 then the identity (1.2) fails in the variety Nil( C m ) = var { x m = 0 , xy = yx } . This means that each part of the equality (4.4) coincides with the variety C m for some 0 ≤ m ≤ 2. Then we may complete the proof of equality (4.4) by the same arguments as in the proof of the equality (4.2) in [25].
The equality (4.5). The varieties G , C 2 and N satisfy the identity x y 2 = x 2 y . By Lemma 2.3 the variety Nil( V ) satisfies the identity (1.2). Fig. 2 shows that it suffices to check the following two claims: first, the varieties ( Z ∨ Y ) ∧ V and ( Z ∧ V ) ∨ Y have the same degree, and second, the variety Nil ( ( Z ∨ Y ) ∧ V ) satisfies the identity
$$x ^ { 2 } = 0$$
if and only if the variety Nil ( ( Z ∧ V ) ∨Y ) ) satisfies this identity. The former claim may be verified by the same way as in the proof of the equality (4.3) in [25] with references to Proposition 2.13, Corollary 2.15 and Corollary 2.16 of the present article rather than Proposition 2.11, Lemma 2.13 and Lemma 2.12 of [25] respectively.
/negationslash
It remains to verify that the variety Nil ( ( Z ∨ Y ) ∧V ) satisfies the identity (4.6) whenever Nil ( ( Z ∧ V ) ∨ Y ) does so (because the opposite claim is evident). Suppose that the identity (4.6) holds in Nil ( ( Z ∧ V ) ∨ Y ) . Corollary 2.9 imply that the variety ( Z ∧ V ) ∨ Y is the join of some group variety, the variety C m for some m ≥ 0 and the variety Nil ( ( Z ∧ V ) ∨Y ) . Here m ≤ 2 because ( Z ∧ V ) ∨ Y ⊆ V . This implies that the variety ( Z ∧ V ) ∨ Y satisfies the identity x 2 = x 2 . In particular, this identity holds in both the varieties Y and Z ∧ V . Therefore, there is a sequence of words u , u 0 1 , . . . , u k such that u 0 ≡ x 2 , u k ≡ x 2 and, for each i = 0 1 , , . . . , k -1, the identity u i = u i +1 holds in one of the varieties Z or V . We may assume that u i ≡ u i +1 for each i = 0 1 , , . . . , k -1. Arguments from the proof of the equality (4.3) in [25] show that it suffices to check that the identity (4.6) holds in one of the varieties Nil( Z ) or Nil( V ). This fact follows from Lemma 2.6 whenever u 1 is a semigroup word and from Lemma 2.3 otherwise.
We complete the proof of Theorem 1.5.
/square
Corollary 4.4. A periodic strongly permutative semigroup variety V is an upper-modular element of the lattice SEM if and only if V is an upper-modular element of the lattice EPI .
Proof. Necessity. Let V be a periodic strongly permutative semigroup variety and V is an upper-modular element of SEM . It follows from results of [5] and the proof of Proposition 1 in [35] that V = G ∨ C m ∨N for some Abelian periodic group variety V , some m ≥ 0 and some nilvariety N . By [25, Theorem 1.1], if X is a proper semigroup variety that is an upper-modular element of SEM and Y is a nilsubvariety of X then Y is commutative. Thus N is commutative. This implies that the variety V is commutative too. Now we may apply [25, Theorem 1.2] and conclude that V satisfies one of the claims (i) or (ii) of Theorem 1.5. Therefore V is an upper-modular variety.
Sufficiency. Let V be a periodic strongly permutative semigroup variety and V is an upper-modular variety. By Theorem 1.5, either V satisfies the claim (i) of this theorem or V satisfies the claim (ii) of this theorem and the variety G from this claim is periodic. In both these cases V is an upper-modular element of SEM by [25, Theorem 1.2]. /square
Theorem 1.5 readily implies the following
Corollary 4.5. If a strongly permutative epigroup variety V is an uppermodular element of the lattice EPI then the lattice L ( V ) is distributive.
Proof. If V satisfies the claim (i) of Theorem 1.5 then it is periodic, whence it may be considered as a semigroup variety. In this case, it suffices to take into account a description of commutative semigroup varieties with distributive subvariety lattice obtained in [36]. Suppose now that V satisfies the claim (ii) of Theorem 1.5. Then V ⊆ AG ∨ C 2 ∨ N where N satisfies the commutative law and the identity (1.2). In view of Proposition 2.10, L ( V ) is embeddable into the direct product of the lattices L ( AG ) and L ( C 2 ∨N ). The former lattice is generally known to be distributive. Finally, the variety C 2 ∨ N is periodic, whence it may be considered as a semigroup variety. To complete the proof, it remains to note that the lattice L ( C 2 ∨ N ) is distributive by the mentioned result of [36]. /square
## 5. Codistributive varieties
Here we verify Theorem 1.4.
Necessity. Here and in Section 9 we need the following statement that may be verified by repeating literally arguments from the first paragraph of the proof of Theorem 1.1 in [27].
Lemma 5.1. Let an epigroup variety V be a codistributive element of the lattice EPI . If V does not contain the varieties P and ← -P then V is a variety of degree ≤ 2 . /square
Let now V be a strongly permutative codistributive variety. It is evident that V is upper-modular. Theorem 1.5 implies that the variety V is commutative. Hence P P , ← -/notsubseteql V . Lemma 5.1 implies that V is a variety of degree ≤ 2. It remains to refer to Corollary 2.9 and the observation that the variety C m has a degree > 2 whenever m ≥ 2.
Sufficiency. In view of Corollary 3.3, it suffices to verify that an abelian group variety G is codistributive. Let Y and Z be arbitrary epigroup varieties. Every epigroup variety is either periodic or contains the variety AG . Suppose that at least one of the varieties Y and Z , say Y , contains AG . Then Y ∨ Z ⊇ Y ⊇ AG ⊇ G and therefore,
$$\mathcal { G } \wedge ( \mathcal { Y } \vee \mathcal { Z } ) = \mathcal { G } = \mathcal { G } \vee ( \mathcal { G } \wedge \mathcal { Z } ) = ( \mathcal { G } \wedge \mathcal { Y } ) \vee ( \mathcal { G } \wedge \mathcal { Z } ).$$
Hence we may assume that the varieties Y and Z are periodic. Now we may complete the proof by the same arguments as in the proof of the implication c) -→ a) of Theorem 1.2 in [27]. /square
Comparing Theorem 1.2 in [27] with Theorem 1.4, we obtain the following
Corollary 5.2. A periodic strongly permutative semigroup variety V is a codistributive element of the lattice SEM if and only if V is a codistributive element of the lattice EPI . /square
## 6. Modular varieties
Here we are going to prove Theorems 1.6-1.8. We need several auxiliary facts.
Proposition 6.1. If an epigroup variety V is a modular element of the lattice EPI then V is periodic.
Proof. Let V be a modular epigroup variety. Suppose that V is non-periodic. Being an epigroup variety, V satisfies the identity x n = x x n ω for some natural n . Consider varieties
N 1 = var { x x 1 2 . . . x n +3 = 0 } and N 2 = var { x x 1 2 . . . x n +3 = 0 , x n +1 y = x y n 2 } .
To prove that V is non-modular, we are going to check that the varieties V , N 1 , N 2 , V ∨ N 1 and V ∧ N 1 form the 5-element non-modular sublattice N 5 . Note that N ⊆ N 2 1 . Whence, to achieve our aim, it suffices to verify the equalities V ∨ N 1 = V ∨ N 2 and V ∧ N 1 = V ∧ N 2 .
The inclusion V ∨ N 2 ⊆ V ∨ N 1 is evident. It is evident also that a nontrivial identity u = v holds in N 2 if and only if either /lscript ( u , /lscript ) ( v ) ≥ n +3 or u = v coincides with the identity x n +1 y = x y n 2 . Let u = v be a non-trivial identity that is satisfied by the variety V ∨ N 2 . Substituting y 2 for y in the identity x n +1 y = x y n 2 , we have x n +1 y 2 = x y n 4 that implies x n +3 = x n +4 . Therefore, a variety satisfying x n +1 y = x y n 2 is periodic. Since the identity u = v holds in a non-periodic variety V , it differs from the identity x n +1 y = x y n 2 . Therefore, /lscript ( u , /lscript ) ( v ) ≥ n +3. This implies that u = v holds in N 1 and therefore, in V ∨ N 1 . Thus, V ∨ N 1 ⊆ V ∨ N 2 . The equality V ∨ N 1 = V ∨ N 2 is proved.
The inclusion V ∧ N 2 ⊆ V ∧ N 1 is evident. The variety V ∧ N 1 is a nilvariety and is contained in V . Since V satisfies x n = x x n ω , Lemma 2.6(ii) implies that x n = 0 holds in V ∧ N 1 . Therefore, x n +1 y = 0 = x y n 2 in V ∧ N 1 . We see that V ∧ N 1 ⊆ N 2 , whence V ∧ N 1 ⊆ V ∧ N 2 . The equality V ∧ N 1 = V ∧ N 2 is proved as well. /square
We denote by PER the lattice of all periodic semigroup varieties.
Lemma 6.2. Let V be a nilvariety that is a modular element of the lattice EPI . If V satisfies a non-substitutive identity u = v then it satisfies also the identity u = 0 .
Proof. If the identity u = v is not a semigroup one then Lemma 2.3 is applied with the desirable conclusion. So, we may assume that u = v is a semigroup identity. Note that the variety V is periodic, whence it may be considered as a semigroup variety. Clearly, V is a modular element of the lattice PER . It is verified in [22, Proposition 2.2] that if a semigroup variety is modular in the lattice SEM then it has the property we verify. All varieties that appear in the proof of this claim are periodic. Therefore, the desirable conclusion is true for modular elements of the lattice PER , and we are done. /square
The formulation of the following statement and its proof are closely related with the formulation and proof of Lemma 3.1 of the article [12]. But we need slightly modify some terminology from this article. Lemma 3.1 of [12] deals with the notions of equivalent and non-stable pairs of (semigroup) words defined in [12]. Here we need some modification of the first notion and do not require the second one at all. So, we call semigroup words u and v equivalent if u ≡ ξ v ( ) for some automorphism ξ on F . Clearly, if words u and v are equivalent semigroup words and c u ( ) = c v ( ) then u = v is a substitutive identity.
Lemma 6.3. Let V be an epigroup variety that is a modular element of the lattice EPI and let u , v , s and t be pairwise non-equivalent words of the same length depending on the same letters. If the variety V satisfies the identities u = v and s = t then it satisfies also the identity u = s .
Proof. In view of Proposition 6.1, the variety V is periodic. Whence, it may be considered as a semigroup variety. Clearly, V is a modular element of the lattice PER . The proof of [12, Lemma 3.1] readily implies that if u , v , s and t are pairwise non-equivalent words of the same length depending on the same letters, V satisfies the identities u = v and s = t and V does not satisfy the identity u = s then there are periodic varieties (in actual fact, even nilvarieties) U and W such that U ⊆ W but ( V ∨ U ) ∧W /negationslash = ( V ∧ W ) ∨ U . This contradicts the claim that V is a modular element of the lattice PER . /square
Proof of Theorem 1.6 . Let V be a modular epigroup variety. According to Proposition 6.1, the variety V is periodic. It follows immediately from [12, Proposition 3.3] that if a periodic semigroup variety is a modular element of the lattice SEM then it is the join of one of the varieties T or SL and a nilvariety. Repeating literally arguments from the proof of this statement with references to Proposition 2.7 and Lemma 2.8 of the present work rather than Lemma 2.6 of the article [12] and to Lemma 6.3 of the present work rather than Lemma 3.1 of the article [12], we obtain that the variety V has the same property. Thus, V = M∨N where M is one of the varieties T or SL and N is a nilvariety. It remains to verify that if N satisfies a non-substitutive identity u = v then N satisfies also the identity u = 0. If the identity u = v is not a semigroup one then Lemma 2.3 is applied with the conclusion that N satisfies the identity u = 0. So, we may assume that u = v is a semigroup identity. Note that the variety N is periodic, whence it may be considered as a semigroup variety. In this situation the desirable conclusion directly follows from [22, Proposition 2.2]. Theorem 1.6 is proved. /square
Let USEM denotes the lattice of all varieties of unary semigroups. The following lemma will be helpful here and in Section 7.
Proposition 6.4. An arbitrary 0 -reduced epigroup variety is a modular and lower-modular element of the lattice USEM .
Proof. Let N be a 0-reduced epigroup variety, while ν a fully invariant congruence on F corresponding to N . Then ν has exactly one non-singleton class (this class includes words that are equal to 0 in N and only them). The lattice USEM is antiisomorphic to the lattice of all fully invariant congruences on
F . Further, the latter lattice is embeddable in the lattice Eq( F ) of all equivalence relations on F . If an equivalence relation ρ on a set X has exactly one non-singleton class then ρ is both modular and upper-modular element in the lattice Eq( X ) (this claim is verified in [6, Proposition 2.2] for modular elements and in [32, Proposition 3] for upper-modular ones). Thus, ν is a modular and upper-modular element in the lattice Eq( F ), and moreover in the lattice of all fully invariant congruences on F . Since the notion of a modular element of a lattice is self-dual, we are done. /square
The lattice EPI is a sublattice in USEM . Whence, Proposition 6.4 immediately implies Theorem 1.7. /square
Proof of Theorem 1.8 . Necessity. Let V be a commutative modular epigroup variety. By Theorem 1.6 V = M∨N where M is one of the varieties T or SL and N is a nilvariety. Corollary 3.3 implies that the variety N is modular. Since every commutative variety satisfies the identity x y 2 = yx 2 , Lemma 6.2 implies that the identity (1.2) holds in N .
Sufficiency. In view of Corollary 3.3, it suffices to verify that a commutative epigroup variety satisfying the identity (1.2) is modular. This fact may be verified by the same arguments as in the proof of the 'if' part of Theorem 1 in [33]. Theorem 1.8 is proved. /square
Theorems 1.8 and 1.5 evidently imply
Corollary 6.5. If a commutative epigroup variety is a modular element of the lattice EPI then it is an upper-modular element of this lattice. /square
Theorem 3.1 in [22] and Theorem 1.8 show that the following is true.
Corollary 6.6. A periodic commutative semigroup variety is a modular element of the lattice SEM if and only if V is a modular element of the lattice EPI . /square
## 7. Lower-modular varieties
Here we prove Theorem 1.3. To achieve this aim, we need the following
Proposition 7.1. If an epigroup variety V is a lower-modular element of the lattice EPI then V is periodic.
Proof. If S is an epigroup and x ∈ S then x = xx ω = x · xx . Thus, every epigroup satisfies the identity
## (7.1) x = x · xx.
It is known (see [15] or [16], for instance) that there is a natural number n such that the n th power of any element x in an arbitrary member S of V is a group element. Therefore, V satisfies the identity x n = x n . Let n be the least number with such a property. Put
$$\mathcal { Y } = \text{var} \{ x ^ { n } y ^ { 3 } x = \overline { \overline { x ^ { n } } } y ^ { 3 } x \} \quad \text{and} \quad \mathcal { Z } = \text{var} \{ x ^ { n } y ^ { 2 } x = x ^ { n } y ^ { 3 } x \}.$$
Clearly, V ⊆ Y . Since V is lower-modular, the equality
$$( \mathcal { Z } \vee \mathcal { V } ) \wedge \mathcal { Y } = ( \mathcal { Z } \wedge \mathcal { Y } ) \vee \mathcal { V }$$
holds. Note that
$$\text{holds. Note that} \\ x ^ { n } y ^ { 2 } x & = x ^ { n } y ^ { 3 } x & \text{in the variety $\mathcal{Z}$} \\ & = \overline { x ^ { n } } y ^ { 3 } x & \text{in the variety $\mathcal{Y}$} \\ & = \overline { x ^ { n } } \cdot \overline { x ^ { n } } \, x ^ { n } y ^ { 3 } x & \text{in the variety $\mathcal{Z} \wedge \mathcal{Y}$ by (7.1)} \\ & = \overline { x ^ { n } } \cdot \overline { x ^ { n } } \, x ^ { n } y ^ { 2 } x & \text{in the variety $\mathcal{Z}$} \\ & = \overline { x ^ { n } } \, y ^ { 2 } x & \text{in the variety $\mathcal{Z} \wedge \mathcal{Y}$ by (7.1).} \\ \text{W.co.to that $\mathcal{Z} \wedge \mathcal{V}$ activate for the identity $\mathcal{Z} n_{2}_{2}_{\mathcal{Z} } - \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline { \overline \over { \overline { \ \overline { \ \overline { \ \overline { \ \overline { \ \overline { \ \overline { \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ } } } } } } } } } } } } }$$
We see that Z ∧ Y satisfies the identity x y x n 2 = x n y x 2 . Therefore, the identity x y x n 2 = x n y x 2 holds in ( Z ∧ Y ) ∨ V and moreover, in ( Z ∨ V ) ∧ Y . This means that there is a deduction of this identity from identities of the varieties Z ∨ V and Y . In particular, one of the varieties Z ∨ V or Y satisfies a non-trivial identity of the form x y x n 2 = w for some word w . It is evident that any identity of such a kind is false in Y . Whence, it holds in Z ∨ V . It is obvious that the words x y x n 2 and x y x n 3 do not contain images of each other relatively to endomorphisms of the unary semigroup F . Hence the identity x y x n 2 = x y x n 3 does not imply any identity of the form x y x n 2 = w where w differs from the words x y x n 2 and x y x n 3 . In particular, all identities of this form are false in Z . Therefore, the variety Z ∨ V satisfies the identity x y x n 2 = x y x n 3 . In particular, this identity holds in V . Thus, V satisfies the identity x n +3 = x n +4 , whence it is periodic. /square
Repeating literally arguments from the proof of Lemma 6.3 but referring to Proposition 7.1 rather than Proposition 6.1, we obtain the following
Lemma 7.2. Let V be an epigroup variety that is a lower-modular element of the lattice EPI and let u , v , s and t be pairwise non-equivalent words of the same length depending on the same letters. If the variety V satisfies the identities u = v and s = t then it satisfies also the identity u = s . /square
Proof of Theorem 1.3 . Sufficiency. Proposition 6.4 immediately implies that a 0-reduced epigroup variety is lower-modular. It remains to refer to Corollary 3.3.
Necessity. In view of Proposition 7.1, the variety V is periodic. Now we may complete the proof repeating literally arguments from the proof of Proposition 3.3 in [12] but referring to Proposition 2.7 rather than Lemma 2.6 in [12], and to Lemma 7.2 rather than Lemma 3.1 in [12]. /square
Comparing Theorems 1.7 and 1.3, we obtain the following
Corollary 7.3. If an epigroup variety is a lower-modular element of the lattice EPI then it is a modular element of this lattice. /square
By the way, we note that neither of the five other possible interrelations between properties of being a modular variety, a lower-modular variety or an upper-modular variety holds. For instance:
- · the variety var { x 2 = 0 , xy = yx } is modular by Theorem 1.8 but not lower-modular by Theorem 1.3;
- · the variety var { xyz = 0 } is modular and lower-modular by Theorems 1.7 and 1.3 respectively but not upper-modular by Theorem 1.5;
- · an arbitrary Abelian periodic group variety is upper-modular by Theorem 1.5 but neither modular nor lower-modular by Theorems 1.6 and 1.3 respectively.
Theorem 1.1 in [14] and Theorem 1.3 show that the following is true.
Corollary 7.4. A periodic semigroup variety V is a lower-modular element of the lattice SEM if and only if V is a lower-modular element of the lattice EPI . /square
## 8. An application to definable varieties
Here we discuss an interesting application of Theorems 1.1 and 1.3. A subset A of a lattice 〈 L ; ∨ ∧〉 , is called definable in L if there exists a first-order formula Φ( x ) with one free variable x in the language of lattice operations ∨ and ∧ which defines A in L . This means that, for an element a ∈ L , the sentence Φ( a ) is true if and only if a ∈ A . A set X of semigroup [epigroup] varieties is said to be definable if it is definable in SEM [respectively in EPI ]. In this situation we will say that the corresponding first-order formula defines the set X .
A number of deep results about definable varieties and sets of varieties of semigroups have been obtained in [7] by Jeˇek and McKenzie. z In particular, it was proved there that the set of all 0-reduced varieties is definable. But the article [7] contains no explicit first-order formula that define this set of varieties. The simple first-order formula that define the class of all 0-reduced semigroup varieties was found in [28, Theorem 3.3] or [29, Subsection 3.5]. Here we are going to verify that the same formula defines the set of 0-reduced varieties in the lattice EPI .
Theorem 1.3 shows that an epigroup variety is 0-reduced if and only if it is lower-modular and does not contain the variety SL . Obviously, the set of all lower-modular varieties is definable. It remains to define the variety SL . Evidently, the lattices SEM and EPI have the same set of atoms. Theorem 1.1 together with the well-known description of atoms of the lattice SEM (see [17, Section 1], for instance) imply that the lattice EPI contains exactly two neutral atoms, namely the varieties SL and ZM . Recall that a semigroup variety V is called a chain if the lattice L ( V ) is a chain. Clearly, every chain variety is periodic, whence it may be considered as epigroup variety. It is well known that the variety ZM is properly contained in some chain variety, while the variety SL is not [19]. Combining the mentioned observations, we see that the class of all 0-reduced varieties may be defined as the class K of epigroup varieties with the following properties:
- (i) every member of K is a lower-modular variety;
- (ii) if V ∈ K and V contains some neutral atom A then A is properly contained in some chain variety.
It is evident that properties (i) and (ii) may be written by simple first-order formulas with one free variable. We prove the following
Proposition 8.1. The class of all 0 -reduced epigroup varieties is definable in the lattice EPI . /square
## 9. Neutral and costandard varieties
Here we prove Theorem 1.1. The proof will be given by the following scheme:
The implications a) -→ b) and a) -→ c) are evident, while the implication d) -→ a) immediately follows from Propositions 3.1 and 3.2, and the well-known fact that the set of all neutral elements of a lattice L forms a sublattice in L (see [2, Theorem 259]). It remains to verify the implications b) -→ d) and c) -→ d).
Suppose that V is either costandard or simultaneously lower-modular and upper-modular. We need to check that V coincides with one of the varieties T , SL ZM , or SL∨ZM . In any case, V is modular (this is evident whenever V is costandard, and follows from Corollary 7.3 whenever V is lower-modular and upper-modular). Then we may apply Theorem 1.6 and conclude that V = M∨N where M is one of the varieties T or SL and N is a nilvariety. It remains to verify that N is one of the varieties T or ZM . By Lemma 2.20 we have to check that N ⊆ ZM . In other words, we need to verify that N is a variety of degree ≤ 2. In view of Corollary 3.3, N is either costandard or simultaneously lower-modular and upper-modular.
Suppose at first that N is costandard. Then it is codistributive. Now Lemma 5.1 applies with the desirable conclusion. We have proved the implication b) -→ d).
It remains to consider the case when N is lower-modular and upper-modular. The variety N is periodic, whence it may be considered as a semigroup variety. Therefore, N is a lower-modular and upper-modular element of the lattice PER .
It follows from [23, Theorem 1] that if a nilvariety of semigroups is lowermodular in SEM then it is 0-reduced. All varieties that appear in the proof of this fact are periodic. So, a nilvariety is 0-reduced whenever it is a lowermodular element in PER . In particular, the variety N is 0-reduced.
Suppose that N is a variety of degree > 2. It follows from [26, Theorem 1] that if a proper semigroup variety of degree > 2 is upper-modular in SEM then it is commutative. All varieties that appear in the proof of this fact are periodic again. So, a variety of degree > 2 is commutative whenever it is an upper-modular element in PER . In particular, the variety N is commutative.
Thus, N is a 0-reduced and commutative nilvariety. Therefore, N ⊆ ZM . This completes the proof of the implication c) -→ d), and hence the proof of Theorem 1.1 as a whole. /square
Theorem 1.1 immediately implies the following
Corollary 9.1. If an epigroup variety is a costandard element of the lattice EPI then it is a standard element of this lattice. /square
Comparing Theorem 1.3 in [27] with Theorem 1.1, we obtain the following
Corollary 9.2. For a periodic semigroup variety V , the following are equivalent:
- a) V is a costandard element of the lattice EPI ;
- b) V is a neutral element of the lattice EPI ;
- c) V is a costandard element of the lattice SEM ;
- d) V is a neutral element of the lattice SEM .
/square
## 10. Distributive and standard varieties
Here we prove Theorem 1.2. The implication b) -→ a) of this theorem is evident, whence it suffices to verify the implications a) -→ c), c) -→ a) and a) -→ b).
- a) -→ b) Suppose that an epigroup variety V is distributive. Hence it is lower-modular. Corollary 7.3 implies that V is modular. Now Lemma 2.17 implies with the conclusion that V is standard.
- a) -→ c) We denote by F m the free unary semigroup over the alphabet { x , x 1 2 , . . . , x m } and by S m the symmetric group on the set { 1 2 , , . . . , m } . If σ ∈ S m and u ∈ F m then σ u ( ) denotes the image of the word u under the automorphism of the unary semigroup F m induced by the action of the permutation σ on indexes of letters.The following statement easily follows from the proof of Lemma 5.1 of the paper [30].
Lemma 10.1. Let u and v be semigroup words from F m such that the identity u = v is non-balanced and none of the words u and v contains the image of other one over some endomorphism of the free unary semigroup, and let σ be a permutation from S m . If the variety X = var { u = v } satisfies a non-trivial identity of the form σ u ( ) = w where w is a semigroup word then w ≡ σ v ( ) . /square
Note that if the identity u = v is balanced then the class K { u = v } is not a variety by Lemma 2.4. Thus, the restriction to the identity u = v to be non-balanced in Lemma 10.1 is natural.
The following assertion is an 'epigroup analog' of Lemma 5.2 of the article [30].
Lemma 10.2. Let V be an epigroup variety that is a distributive element of the lattice EPI and let u , v be semigroup words such that c u ( ) = c v ( ) and one of the following holds:
- (i) none of the words u and v contains the image of other one over some endomorphism of the free unary semigroup;
- (ii) /lscript ( u ) = /lscript ( v ) .
If the variety V satisfies the identity u = 0 then it satisfies the identity v = 0 .
Proof. We may assume that u, v ∈ F m for some natural m . Suppose that V satisfies the identity u = 0.
(i) If m = 1 then u ≡ x q and v ≡ x r for some q and r . Then one of the words u and v contains an image of another one that contradicts the hypothesis. Therefore, m > 1, whence the group S m contains a non-trivial permutation α . Put w ≡ α u ( ). Then V satisfies the identity w = 0 that implies u = w . If the identity u = v is non-balanced then we may complete the proof repeating literally arguments from the corresponding part of the proof of item (i) in Lemma 5.2 of the paper [30]. Suppose now that the identity u = v is balanced. Let x be a letter with x / c u ∈ ( ). Then the identities xu = v and v = xw are non-balanced. According to Lemma 2.4, we may consider the varieties Y = var { xu = v } and Z = var { v = xw } . Since the variety V is distributive, V ∨ ( Y ∧ Z ) = ( V ∨ Y ) ∧ V ∨ Z ( ). Clearly, the variety V ∨ ( Y ∧ Z ) satisfies the identity xu = xw . Therefore, this identity holds in the variety ( V ∨ Y ) ∧ V ∨ Z ( ). Then there exists a deduction of the identity xu = xw from identities of the varieties V ∨ Y and V ∨ Z . In particular, one of these varieties satisfies a non-trivial identity of the form xu = w 1 .
Suppose at first that xu = w 1 holds in V ∨ Y . Then it holds in Y . Now Lemma 10.1 with the trivial permutation σ from S m applies, and we conclude that w 1 ≡ v . Since xu = w 1 in V , we have that V satisfies the identity v = xu that implies v = 0. It remains to consider the case when the identity xu = w 1 holds in V ∨ Z . In particular, it holds in Z . Let σ be a permutation of the alphabet such that the restriction of σ on the set c u ( ) coincides with α -1 and σ x ( ) ≡ x . Now we apply Lemma 10.1 with such permutation σ . Then we obtain that w 1 ≡ σ v ( ). But the identity xu = w 1 holds in V . Hence V satisfies the identity xu = σ v ( ) that implies σ v ( ) = 0 and v = 0.
(ii) Here we may repeat literally arguments from the proof of item (ii) in [30, 5.2]. /square
Now we are well prepared to complete the proof of the implication a) -→ c) of Theorem 1.2. Let V be a distributive epigroup variety. Then V is a lowermodular variety. In view of Theorem 1.3, V = M∨N where M is one of the varieties T or SL and N is a 0-reduced variety. Corollary 3.3 implies that the variety N is distributive. Being 0-reduced, this variety satisfies the identity u = 0 for some word u . We may assume that this identity fails in the class of all nil-semigroups. Lemma 2.3 implies that u is a semigroup word. We may assume that c u ( ) = { x, y } . Indeed, if u depends on one letter then we may substitute xy to this letter, and if u depends on two or more letters then we may substitute x to one of these letters and y to all other of them. Now we may repeat literally the corresponding part of the proof of Proposition 3.2 of the article [30] with referring to Lemmas 10.1 and 10.2 of the present article rather than Lemmas 5.1 and 5.2 of the paper [30] respectively. As a result, we obtain that N satisfies all identities of the form v = 0 with c v ( ) = c u ( ) and /lscript ( v ) ≥ 3. In particular, N satisfies the identities x y 2 = xyx = yx 2 = 0. We see that N is a 0-reduced subvariety of the variety Q . If N = Q then we are done. Let now N ⊂ Q . Then N is given in Q by some set of 0-reduced identities. By Lemma 2.12, N is given within Q either by the identity x 2 = 0 or by the identity x x 1 2 · · · x n = 0 for some n or by these two identities together. Thus N is one of the varieties Q Q , n , R or R n . The implication a) -→ c) is proved.
c) -→ a) Let V = M∨N where M is one of the varieties T or SL and N is one of the varieties Q Q , n , R or R n . In view of Corollary 3.3, it suffices to prove that the variety N is standard. In other words, we may assume that V is one of the varieties Q Q , n , R or R n . In particular, V is periodic, whence it may be considered as a semigroup variety.
Let Y and Z be arbitrary epigroup varieties. We need to verify that
$$\mathcal { V } \vee \left ( \mathcal { Y } \wedge \mathcal { Z } \right ) = \left ( \mathcal { V } \vee \mathcal { Y } \right ) \wedge \left ( \mathcal { V } \vee \mathcal { Z } \right ).$$
It suffices to check that ( V ∨ Y ) ∧ V ∨ Z ( ) ⊆ V ∨ ( Y ∧ Z ) because the contrary inclusion is evident. In other words, we have to prove that an arbitrary identity holds in ( V ∨ Y ) ∧ V ∨ Z ( ) whenever it holds in V ∨ ( Y ∧ Z ). In view of [30, Lemma 2.2] we may suppose that Y Z ⊇ SL , .
Let u = v be an identity that holds in V ∨ ( Y ∧ Z ). Then it holds in V and there exists a deduction of this identity from identities of the varieties Y and Z . Let
$$u \equiv w _ { 0 } \longrightarrow w _ { 1 } \longrightarrow \cdots \longrightarrow w _ { k } \equiv v$$
be the shortest such deduction. Since Y Z ⊇ SL , , Lemma 2.5(i) implies that c w ( 0 ) = c w ( 1 ) = · · · = ( c w k ).
We use an induction by k . The induction base is evident because if k = 1 then the identity u = v holds in one of the varieties V ∨ Y or V ∨ Z , whence it holds in the intersection of these varieties. Let now k > 1. The identity u = v holds in the variety V . Since V is 0-reduced, u = v = 0 holds in V . If w i = 0 in V for some 0 < i < k then our considerations may be completed by repeating literally arguments from the corresponding part of the proof of Proposition 3.3 of the paper [30]. In view of Lemma 2.12, we may assume that each of the words w , w , . . . , w 1 2 k -1 either is a linear word or coincides with the word x 2 .
Suppose that w i ≡ x 2 for some 0 < i < k . The choice of the deduction (10.1) guarantees that the words w , w , . . . , w 0 1 k are pairwise different. Since c w ( 0 ) = c w ( 1 ) = · · · = ( c w k ) = { x } , we have that each of the varieties Y and Z satisfies a non-trivial identity of the form x m = x n . Therefore, the varieties Y and Z are periodic. Hence they may be considered as semigroup varieties. This permits to complete our considerations by repeating literally arguments from the corresponding part of the proof of Proposition 3.3 of the paper [30].
It remains to consider the case when the words w , w , . . . , w 1 2 k -1 are linear. We may assume that the words u and v are non-linear (as usual, this claim may be verified by the same arguments as in the proof of [30, Proposition 3.3]). Besides that, we will assume without loss of generality that the identity u = w 1 holds in Y and therefore, w 1 = w 2 holds in Z . Further considerations are divided into three cases.
Case 1: k = 2. Here the deduction (10.1) has the form u → w 1 → v , u = w 1 in Y and w 1 = v in Z . Since the words u and v are non-linear, the varieties Y and Z are periodic. Hence they may be considered as semigroup varieties. This permits to complete a consideration of this case by repeating literally arguments from the Case 1 in the proof of Proposition 3.3 of the paper [30].
Case 2: k = 3. Here the deduction (10.1) has the form u → w 1 → w 2 → v and the identities u = w 1 and w 2 = v hold in Y . Since the words u and
v are non-linear, the variety Y is periodic. Hence it may be considered as a semigroup variety. This variety satisfies the identity x x 1 2 · · · x m = u and /lscript ( u ) > m . Therefore, Y is a variety of degree ≤ m (see [10, Lemma 1]). According to [25, Proposition 2.11], Y satisfies an identity of the form
$$x _ { 1 } x _ { 2 } \cdots x _ { m } = x _ { 1 } x _ { 2 } \cdots x _ { i - 1 } ( x _ { i } \cdots x _ { j } ) ^ { t } x _ { j + 1 } \cdots x _ { m }$$
for some t > 1 and 0 ≤ i ≤ j ≤ m . This permits to complete a consideration of this case by repeating literally arguments from the Case 2 in the proof of Proposition 3.3 of the paper [30].
Case 3: k > 3. This case may be considered by repeating literally arguments from the Case 3 in the proof of Proposition 3.3 of the paper [30].
We complete the proof of the implication c) -→ a) and of Theorem 1.2 as a whole. /square
Comparing Theorem 1.1 in [30] with Theorem 1.2, we obtain the following
Corollary 10.3. For a periodic semigroup variety V , the following are equivalent:
- a) V is a distributive element of the lattice EPI ;
- b) V is a standard element of the lattice EPI ;
- c) V is a distributive element of the lattice SEM ;
- d) V is a standard element of the lattice SEM .
/square
## 11. Open questions
It is verified in [27, Theorem 1.1] that if a proper semigroup variety V is codistributive in SEM then the square of any member of V is completely regular. We do not know, whether the epigroup analog of this fact is true.
Question 11.1. Is it true that if an epigroup variety V is a codistributive element of the lattice EPI then the square of any member of V is completely regular ?
As we have already mentioned in Section 4, it is proved in [25, Theorem 1.1] that if V is a proper upper-modular in SEM semigroup variety then every nilsubvariety of V is commutative and satisfies the identity (1.1). Proposition 4.2 gives a partial epigroup analog of this assertion. We do not know, whether the full analog is true.
Question 11.2. Suppose that an epigroup variety V is an upper-modular element of the lattice EPI and let N be a nilsubvariety of V . Is it true that the variety N
- a) is commutative;
- b) satisfies the identity (1.1)?
Proposition 4.2 shows that the affirmative answer to Question 11.2a) would immediately implies the same answer to Question 11.2b).
Further, it is verified in [26, Theorem 1] that every proper upper-modular in SEM variety is either commutative or has a degree ≤ 2. We do not know, whether the epigroup analog of this alternative is valid. We formulate the corresponding question together with its weaker version.
Question 11.3. Suppose that an epigroup variety V is an upper-modular element of the lattice EPI . Is it true that the variety V
- a) either is commutative or has a degree ≤ 2 ;
- b) either is permutative or has a finite degree ?
The affirmative answer to Question 11.3a) together with Theorem 1.5 would immediately imply a complete description of upper-modular epigroup varieties of degree > 2.
If V is a proper upper-modular in SEM variety then the lattice L ( V ) is modular [26, Corolary 2] and every subvariety of V is upper-modular in SEM [26, Corolary 3]. This makes a natural the following
Question 11.4. Suppose that an epigroup variety V is an upper-modular element of the lattice EPI . Is it true that
- a) the lattice L ( V ) is modular;
- b) every subvariety of V is an upper-modular element of the lattice EPI ?
We note that the affirmative answer to Question 11.4a) would immediately implies the same answer to Question 11.4b) (see [25, Lemma 2.1]).
## References
- [1] E. A. Golubov and M.V.Sapir, Residually small varieties of semigroups , Izv. VUZ. Matem., no. 11 (1982), 21-29 [Russian; Engl. translation: Soviet Math. Izv. VUZ, 26, no. 11 (1982), 25-36].
- [2] G.Gr¨ atzer, Lattice Theory: Foundation, Birkh¨user, Springer Basel AG, 2011. a
- [3] G.Gr¨ atzer and E. T. Schmidt, Standard ideals in lattices , Acta Math. Acad. Sci. Hungar., 12 (1961), 17-86.
- [4] S. V. Gusev and B. M. Vernikov, Endomorphisms of the lattice of epigroup varieties , Semigroup Forum, accepted; available at http://arxiv.org/abs/1404.0478v4.
- [5] T. J. Head, The lattice of varieties of commutative monoids , Nieuw Arch. Wiskunde, 16 (1968), 203-206.
- [6] J. Jeˇek, z The lattice of equational theories. Part I: modular elements , Czechosl. Math. J., 31 (1981), 127-152.
- [7] J. Jeˇek z and R.N.McKenzie, Definability in the lattice of equational theories of semigroups , Semigroup Forum, 46 (1993), 199-245.
- [8] M.Petrich, Inverse Semigroups, Wiley Interscience, New York, 1984.
- [9] M. Petrich and N. R.Reilly, Completely Regular Semigroups, John Wiley & Sons, Inc., New York, 1999.
- [10] M.V.Sapir and E. V. Sukhanov, On varieties of periodic semigroups , Izv. VUZ. Matem., no. 4 (1981), 48-55 [Russian; Engl. translation: Soviet Math. Izv. VUZ, 25, no. 4 (1981), 53-63].
- [11] V.Yu.Shaprynskiˇ ı, Distributive and neutral elements of the lattice of commutative semigroup varieties , Izv. VUZ. Matem., no. 7 (2011), 67-79 [Russian; Engl. translation: Russ. Math. Izv. VUZ, 55, no. 7 (2011), 56-67].
- [12] V. Yu. Shaprynskiˇ ı, Modular and lower-modular elements of lattices of semigroup varieties , Semigroup Forum, 85 (2012), 97-110.
- [13] V.Yu.Shaprynskiˇ ı, Periodicity of special elements of the lattice of semigroup varieties , Proc. Inst. of Math. and Mechan. of the Ural Branch of the Russ. Acad. Sci., 18, no. 3 (2012), 282-286 [Russian].
- [14] V.Yu.Shaprynskiˇ ı and B. M. Vernikov, Lower-modular elements of the lattice of semigroup varieties . III, Acta Sci. Math. (Szeged), 76 (2010), 371-382.
- [15] L. N. Shevrin, On theory of epigroups . I, II, Mat. Sbornik, 185, no. 8 (1994), 129-160; 185, no. 9 (1994), 153-176 [Russian; Engl. translation: Russ. Math. Sb., 82 (1995), 485-512; 83 (1995), 133-154].
- [16] L. N. Shevrin, Epigroups , V.B.Kudryavtsev and I.G. Rosenberg (eds.), Structural Theory of Automata, Semigroups, and Universal Algebra, Springer, Dordrecht (2005), 331-380.
- [17] L. N. Shevrin, B. M. Vernikov and M. V. Volkov, Lattices of semigroup varieties , Izv. VUZ. Matem., no. 3 (2009), 3-36 [Russian; Engl. translation: Russ. Math. Izv. VUZ, 53, no. 3 (2009), 1-28].
- [18] M.Stern, Semimodular Lattices. Theory and Applications, Cambridge: Cambridge Univ. Press, 1999.
- [19] E. V. Sukhanov, Almost linear semigroup varieties , Mat. Zametki, 32 (1982), 469-476 [Russian; Engl. translation: Math. Notes, 32 (1983), 714-717].
- [20] P. G. Trotter, Subdirect decompositions of the lattice of varieties of completely regular semigroups , Bull. Austral. Math. Soc., 39 (1989), 343-351.
- [21] B. M. Vernikov, On semigroup varieties whose subvariety lattice is decomposable into a direct product , L. N. Shevrin (ed.), Algebraic Systems and their Varieties, Sverdlovsk, Ural State Univ., 1988, 41-52 (Russian).
- [22] B. M. Vernikov, On modular elements of the lattice of semigroup varieties , Comment. Math. Univ. Carol., 48 (2007), 595-606.
- [23] B. M. Vernikov, Lower-modular elements of the lattice of semigroup varieties , Semigroup Forum, 75 (2007), 554-566.
- [24] B. M. Vernikov, Lower-modular elements of the lattice of semigroup varieties . II, Acta Sci. Math. (Szeged), 74 (2008), 539-556.
- [25] B. M. Vernikov, Upper-modular elements of the lattice of semigroup varieties , Algebra Universalis, 59 (2008), 405-428.
- [26] B. M. Vernikov, Upper-modular elements of the lattice of semigroup varieties . II, Fundam. and Appl. Math., 14, no. 3 (2008), 43-51 [Russian; Engl. translation: J. Math. Sci., 164 (2010), 182-187].
- [27] B. M. Vernikov, Codistributive elements of the lattice of semigroup varieties , Izv. VUZ. Matem., no. 7 (2011), 13-21 [Russian; Engl. translation: Russ. Math. Izv. VUZ, 55, no. 7 (2011), 9-16].
- [28] B.M.Vernikov, Proofs of definability of some varieties and sets of varieties of semigroups , Semigroup Forum, 84 (2012), 374-392.
- [29] B. M. Vernikov, Special elements in lattices of semigroup varieties , Acta Sci. Math. (Szeged), 81 (2015), 79-109.
- [30] B. M. Vernikov and V. Yu. Shaprynskiˇ ı, Distributive elements of the lattice of semigroup varieties , Algebra and Logic, 49 (2010), 303-330 [Russian; Engl. translation: Algebra and Logic, 49 (2010), 201-220].
- [31] B. M. Vernikov and D. V. Skokov, Semimodular and Arguesian epigroup varieties . I, Proc. of the Institute of Math. and Mechan. of the Ural Branch of the Russ. Acad. of Sci., 22, no. 3 (2016), 31-43 [Russian].
- [32] B. M. Vernikov and M. V.Volkov, Lattices of nilpotent semigroup varieties , L. N. Shevrin (ed.), Algebraic Systems and their Varieties, Sverdlovsk, Ural State University (1988), 53-65 [Russian].
- [33] B. M. Vernikov and M. V. Volkov, Modular elements of the lattice of semigroup varieties . II, Contrib. General Algebra, 17 (2006), 173-190.
- [34] B. M. Vernikov, M.V.Volkov and V.Yu.Shaprynskiˇ ı, Semimodular and Arguesian epigroup varieties . II [Russian], to appear.
- [35] M.V.Volkov, Semigroup varieties with modular subvariety lattices , Izv. VUZ. Matem., no. 6 (1989), 51-60 [Russian; Engl. translation: Soviet Math. Izv. VUZ, 33, no. 6 (1989), 48-58].
- [36] M.V.Volkov, Commutative semigroup varieties with distributive subvariety lattices , Contrib. General Algebra, 7 (1991), 351-359.
- [37] M.V.Volkov, Modular elements of the lattice of semigroup varieties , Contrib. General Algebra, 16 (2005), 275-288.
Ural Federal University, Institute of Mathematics and Computer Science, Lenina 51, 620000 Ekaterinburg, Russia
E-mail addresses : [email protected], [email protected], [email protected] | null | [
"V. Yu. Shaprynskii",
"D. V. Skokov",
"B. M. Vernikov"
] | 2014-08-02T10:16:06+00:00 | 2016-09-04T20:21:00+00:00 | [
"math.GR",
"20M07 (Primary), 08B15 (Secondary)"
] | Special elements of the lattice of epigroup varieties | We study special elements of eight types (namely, neutral, standard,
costandard, distributive, codistributive, modular, lower-modular and
upper-modular elements) in the lattice EPI of all epigroup varieties. Neutral,
standard, costandard, distributive and lower-modular elements are completely
determined. A strong necessary condition and a sufficient condition for modular
elements are found. Modular elements are completely classified within the class
of commutative varieties, while codistributive and upper-modular elements are
completely determined within the wider class of strongly permutative varieties.
It is verified that an element of EPI is costandard if and only if it is
neutral; is standard if and only if it is distributive; is modular whenever it
is lower-modular; is neutral if and only if it is lower-modular and
upper-modular simultaneously. We found also an application of results
concerning neutral and lower-modular elements of EPI for studying of definable
sets of epigroup varieties. |
1408.0357v1 | ## Solving the flow fields in conduits and networks using energy minimization principle with simulated annealing
## Taha Sochi
University College London, Department of Physics & Astronomy, Gower Street, London, WC1E 6BT
Email: [email protected].
## Abstract
In this paper, we propose and test an intuitive assumption that the pressure field in single conduits and networks of interconnected conduits adjusts itself to minimize the total energy consumption required for transporting a specific quantity of fluid. We test this assumption by using linear flow models of Newtonian fluids transported through rigid tubes and networks in conjunction with a simulated annealing (SA) protocol to minimize the total energy cost. All the results confirm our hypothesis as the SA algorithm produces very close results to those obtained from the traditional deterministic methods of identifying the flow fields by solving a set of simultaneous equations based on the conservation principles. The same results apply to electric ohmic conductors and networks of interconnected ohmic conductors. Computational experiments conducted in this regard confirm this extension. Further studies are required to test the energy minimization hypothesis for the non-linear flow systems.
Keywords: fluid dynamics; pressure field; tube; network; porous media; energy minimization; simulated annealing; stochastic method; electric conductor; electric network.
## 1 Introduction
One of the fundamental physical principles that regulate Nature's behavior is optimization which reflects a prejudice that leads to minimizing or maximizing selected physical quantities. For instance, Nature has a tendency to maximize the entropy of dynamic systems, as given by the second law of thermodynamics, but to minimize the passage time of light as summarized by Fermat's least time principle. Many physical laws have been deduced or derived from optimization arguments and hence it is one of the main pillars of modern science. As a result, extensive
branches of mathematical and computational disciplines have been developed to deal with modeling and quantifying optimization problems with widespread applications in physical and social sciences.
One of the powerful and widely used optimization methods is simulated annealing [1-3] which is a stochastic computational technique based on the physical principles of statistical mechanics. The essence of this method is to emulate the process of controlled and slow cooling of liquified substances so that they reach their minimal energy configuration in their solid state. The main advantages of simulated annealing are its simplicity and wide applicability to large classes of optimization problems as well as its high success rate of avoiding traps of local minima which other deterministic and stochastic methods are more likely to fall in. Furthermore, in many cases it is the only viable method as combinatorial enumeration and other analytical or conceptually-based methods are not viable or available in those circumstances.
The main disadvantage of simulated annealing is its generally high computational cost in terms of CPU time. Although this is true for commonplace problems where alternative methods are available, in some cases simulated annealing is more efficient even in terms of CPU time when the computational cost grows exponentially and hence the cost of traditional methods, assuming their viability, becomes much higher than the cost of SA. In fact this is one of the main reasons why simulated annealing and similar stochastic methods are invented and widely used in all sorts of optimization problems as can be inferred, for instance, from the number of citations of the SA founding papers [1-3]. Regardless of this, nowadays the computational cost is a trivial factor in many cases considering the huge advances over the last few decades in the hardware and software development and the availability of multi-processor platforms, even for personal use, with relative ease of parallelization.
In the present paper we suggest and examine a hypothesis that the driving field, like pressure and potential difference, in conducting elements and networks of interconnected elements will adjust itself to minimize the energy cost of transporting a given quantity of fluid through the transport device. We use simulated annealing with supporting arguments to achieve the minimization objective and establish the energy minimization hypothesis. We restrict our attention in the current investigation to the linear transport systems where the driving and induced fields are linearly correlated, such as the flow of Newtonian fluids in rigid tubes and networks of interconnected rigid tubes and the flow of electric current in ohmic components
and networks of interconnected ohmic components, although we will briefly discuss some issues related to the non-linear systems for the sake of completeness. The case of non-linear flow systems requires further investigation to reach definite conclusions although initial examination indicates that they are also subject to the energy minimization rules.
The model that is used in the current investigation to present the energy minimization principle and elaborate the SA optimizing algorithm is from fluid mechanics, namely the aforementioned example of the flow of Newtonian fluids in rigid tubes and networks. However, since this flow model is mathematically equivalent to the flow model of electric current in ohmic devices, our investigation and conclusions will naturally extend to this case as well.
## 2 Method
In this study we assume a laminar, incompressible, isothermal, time-independent, pressure-driven, fully-developed flow with minor entry and exit effects. Our plan for establishing the energy minimization principle through solving the flow fields by simulated annealing is summarized in the following points
- 1. We establish the existence and uniqueness of the flow solution in general.
- 2. We explain how to adapt simulated annealing to find a flow solution in single conduits and networks of interconnected conduits.
- 3. We demonstrate that the solution found by simulated annealing is a correct one, it minimizes the total energy of fluid transportation, and this minimum is a global rather than a local one.
As for the first point , the existence and uniqueness of flow solutions for single conduits is a thorny issue from the theoretical and mathematical viewpoint. However, it can be established by the forthcoming physical argument which we presented in the context of network discussion. Regardless of this, we can take this for granted by claiming it is an intuitive assumption. In fact most of the ongoing studies in this field should be based explicitly or implicitly on such an assumption especially the purely theoretical ones which are not supported by experimental or observational evidence.
With regard to networks, the existence and uniqueness conditions can be established by the following argument assuming the existence and uniqueness of the
solution on their individual conduits. For the linear case, to describe such flow networks we set a system of N simultaneous linear equations in N unknowns where N is the total number of nodes in the network which includes the boundary as well as the internal nodes. The equations of the inlet and outlet nodes are derived from the boundary conditions while the equations of the internal nodes are derived from the mass conservation principle in conjunction with the characteristic flow relation that correlates the driving and induced fields, like p and Q in Hagen-Poiseuille law. Since these equations are linearly independent, due to the fact that no two equations share the same sequence of conducting elements and hence they cannot be represented as scalar multiples of each other, we have a system of N linearly independent equations in N unknowns and hence a solution does exist and it is unique according to the rules of algebra.
With regard to the non-linear systems, there is no general condition that guarantees the existence or uniqueness of solution. However, physical rather than mathematical arguments can be proposed to establish the existence and uniqueness of solutions even for the non-linear systems in case such an extension is required. It can be argued that for both the linear and non-linear systems a solution should exist and it should be unique regardless of all these elaborate mathematical considerations because as long as our mathematical models reflect the essential features of the reality of these classical deterministic systems, the soundness and accuracy of these models will guarantee the existence and uniqueness conditions. Such line of reasoning should be sufficient for the purpose of establishing our energy minimization argument since we have no interest in those theoretical and mathematical subtleties.
As for the second point , the time rate of energy consumption, I , of fluid transport through a conducting device considering the type of flow systems that meet our stated assumptions, is given by
$$I = \Delta p \, Q$$
where ∆ p is the pressure drop across the conducting device and Q is the volumetric flow rate of the transported fluid. For a single conduit that is discretized into n sections indexed by i , the total energy consumption rate, I t , is given by
$$I _ { t } = \sum _ { i = 1 } ^ { n } \Delta p _ { i } Q _ { i }$$
A similar equation applies for a network of n interconnected conduits, with n standing for the number of conduits rather than sections, if only the nodal pressure values are required. If the axial pressure values at the midpoints of conduits are also required, a discretization scheme, similar to the one used for single conduits, can be used where it is needed. However, for the linear systems, only nodal pressures are necessary to compute since the midpoint values can be obtained by a simple linear interpolation scheme.
The role of the simulated annealing protocol then is to find the set of pressure values, p k where k indexes the discretization and nodal points, that minimizes the rate of total energy consumption, given by Equation 2. Accordingly, the p k values can be freely adjusted by the SA routine to satisfy the minimization requirement. The exception to this is the boundary values which are held constant in all SA iterations to satisfy the imposed boundary conditions. As for the Q i values, they are computed in each SA iteration from the analytical expression of the model that correlates the volumetric flow rate to the pressure drop, i.e. Hagen-Poiseuille law in the case of Newtonian flow systems, using the most recent values of p k . Unlike the traditional solution methods, such as the residual-based Newton-Raphson technique, no conservation principle, such as mass continuity, is needed in the proposed SA solution scheme.
Regarding our simulated annealing algorithm, we use a standard scheme as described by many papers and monographs written on simulated annealing and stochastic methods. Briefly, we start by initializing the pressure values, p k , randomly except the boundary ones which are set to the values required by the imposed boundary conditions. We then compute the cost function, which is initially set to a very high value, as given by Equation 2. The new pressure field solution, as defined by the set of p k values, is then accepted if the cost function obtained from the current iteration is less than the cost function of the most recently accepted solution. If the cost function of the new solution is higher, it may also be accepted but with a probability P given by
$$P = e ^ { \frac { I _ { t _ { m - 1 } } - I _ { t _ { m } } } { T _ { c } } }$$
where m is an index for the current iteration and T c is the current value of the annealing parameter (temperature) which we assume to have the same units as I . The p k values are then adjusted randomly using a random number generator and the cycle is repeated. The annealing control parameter, T , is decremented persistently as the annealing goes on. The algorithmic procedure will terminate
when the annealing control parameter reaches its lower limit which is normally set to virtually zero, and hence the final solution will be taken as the last accepted set of p k values according to the minimization criterion.
By finding the pressure field, the volumetric flow rate field will be easily obtained from the analytical expression of the flow in single conduits that links Q to ∆ . p For networks, the total inflow/outflow can be obtained by computing and summing the volumetric flow rates of the inlet/outlet boundary conduits.
As well as implementing this standard SA scheme, we experimented with a number of similar SA schemes which differ in elaboration and complexity, for instance by the way that T varies or by the accepted exit condition from the interim cycles, but since they all produce very similar results we decline to include these irrelevant details.
As for the third point , the verification of the single conduit solution is trivial since the axial pressure varies linearly with the conduit axial coordinate. For the networks, the obtained solution from the simulated annealing procedure can be checked for correctness by simultaneously satisfying the analytical flow relation that links Q to ∆ p on each conduit plus the mass conservation principle on each internal node. Detailed explanations about these verification conditions and other related issues are given in [4]. The simulated annealing solution can also be verified more easily by comparison to the solution found by other methods, mainly the deterministic ones such as the residual-based Newton-Raphson scheme which originates from the existence and uniqueness proof that we outlined earlier. Further details about this can also be found in [4].
Since simulated annealing is essentially a minimization algorithm, the minimal energy principle is established by finding the optimal solution so no further proof is required to establish the fact that the obtained solution is not a maximum or a stationary inflection point, unlike solutions obtained by other methods where such a proof may be needed. Since this minimizing solution is the only possible solution, as established by the uniqueness condition, it should be a global rather than a local minimum since no other solution, minimizing or non-minimizing, is possible.
## 3 Results
The simulated annealing algorithm, which is described in the last section, was implemented in a computer code and results were obtained and analyzed using various models of fluid transport devices including rigid tubes, one-, twoand
three-dimensional networks of interconnected rigid tubes. The multi-dimensional networks include fractal, cubic and orthorhombic lattice models; the detailed description of these networks is given, for instance, in [4, 5]. These model tubes and networks vary in their size, geometry, connectivity, number of nodes and segments, and statistical distributions. Various boundary conditions and discretization schemes with different model Newtonian fluids were also employed in these computations. A sample of these results is given in Figures 1 and 2 and Table 1. All the results represent averaged data over many similar runs where no run has failed to meet the rigorous criteria set for accepting the solutions.
As seen, the SA results agree very well with the verified deterministic solutions. One thing that should be highlighted is the general trend of increasing the size of the errors with increasing the size of the networks. The main reason is our tendency to reduce the required CPU time and hence the results can be improved if longer CPU time is allowed and more elaborate simulated annealing schemes are employed. The size of the errors in these results, nonetheless, does not affect the definite conclusion that can be drawn from this investigation about the validity of the energy minimization principle as a governing rule in determining the flow fields ( p and Q ) in the fluid transport systems. After all, such errors, and even larger ones, are expected to contaminate the results obtained by stochastic methods. In fact significant errors can occur even in the numerical deterministic methods although the size of the errors in the latter is usually less than that in the former.
These results do not only establish the energy minimization principle but also establish a novel way for solving the flow fields in conduits and networks by stochastic methods in general and by simulated annealing in particular. Most of the reported results are obtained within seconds or minutes of CPU time on a normal laptop computer where the length of the CPU time mainly depends on the number of discretized elements in the transport device assuming the employment of the same SA parameters and scheme. Although the proposed stochastic method is generally slower than the traditional deterministic methods, it may be possible to compete with the traditional deterministic methods even in speed for very large networks especially if more elaborate SA schemes are employed. A big advantage, however, of the proposed stochastic method is the trivial memory cost; a factor based on its serial nature and the redundancy of employing a numerical solver that is usually needed in the deterministic methods. Anyway, the proposed stochastic method is a tool that is advantageous to be available to scientists and engineers and it definitely can have useful applications in some exceptional circumstances at least.
Figure 1: Comparison between the pressure field solution, as a function of the axial coordinate for a single tube, as obtained from the simulated annealing method based on the energy minimization principle and the solution obtained by interpolation or from the classical deterministic residual-based Newton-Raphson method.
Figure 2: Comparison between the pressure field solution, as a function of the axial coordinate for a one-dimensional network of five serially-connected tubes with different lengths and radii, as obtained from the simulated annealing method based on the energy minimization principle and the solution obtained from the classical deterministic residual-based Newton-Raphson method.
Table 1: Statistical distribution parameters of the percentage relative difference of the nodal pressures between the Newton-Rapson deterministic solutions and the simulated annealing solutions for a number of 2D fractal (F) and 3D orthorhombic (O) networks with the given number of segments (NS) and number of nodes (NN). The results represent averaged data over multiple runs for each network. The meaning of the statistical abbreviations are: Min for minimum, Max for maximum, SD for standard deviation, and Avr for average.
| Index | Type | NS | NN | Min | Max | SD | Avr |
|---------|--------|------|------|-------|-------|------|-------|
| 1 | F | 15 | 16 | -0.42 | 0.35 | 0.15 | -0.06 |
| 2 | F | 31 | 32 | -0.7 | 0.67 | 0.22 | -0.16 |
| 3 | F | 63 | 64 | -0.79 | 0.75 | 0.24 | -0.05 |
| 4 | F | 127 | 128 | -0.77 | 0.85 | 0.27 | 0.19 |
| 5 | F | 255 | 256 | -1.29 | 1.15 | 0.25 | 0.07 |
| 6 | F | 511 | 512 | -1.87 | 1.74 | 0.35 | -0.06 |
| 7 | O | 72 | 45 | -0.51 | 0.58 | 0.2 | 0.1 |
| 8 | O | 99 | 60 | -0.89 | 0.78 | 0.22 | -0.11 |
| 9 | O | 136 | 80 | -1.11 | 1.12 | 0.26 | 0.23 |
| 10 | O | 176 | 96 | -1.42 | 1.37 | 0.32 | -0.17 |
| 11 | O | 216 | 112 | -1.65 | 1.57 | 0.37 | -0.29 |
| 12 | O | 275 | 140 | -1.74 | 1.68 | 0.44 | 0.35 |
## 4 Conclusions
We demonstrated, through the use of simulated annealing, the validity of the assumption of energy minimization principle as a governing rule for the flow systems in the context of obtaining the driving and induced fields in single conduits and networks of interconnected conduits where the driving and induced fields are linearly correlated, e.g. the flow of Newtonian fluids through rigid tubes and networks or the flow of electric current through ohmic devices. All the results support the proposed energy minimization principle.
There are two main outcomes of this investigation. First, a novel method for solving the flow fields that is based on a stochastic approach is proposed as an alternative to the traditional deterministic approaches such as the residual-based Newton-Raphson and finite element methods. Although this new numerical method may not be attractive in most cases where it is outperformed in speed by the traditional methods, it can surpass the other methods in other cases and may even
be the only viable option in some circumstances where the flow networks are very large.
The second outcome, which in our view is the most important one, is the theoretical conclusion that energy minimization principle is at the heart of the flow phenomena and hence it governs the behavior of flow systems; the linear ones at least. This is inline with our previous investigations [6, 7] which are based on minimizing the total stress in the flow conduits. The current investigation may be extended in the future to the non-linear case.
## Nomenclature
- I time rate of energy consumption for fluid transport
- I t time rate of total energy consumption for fluid transport
- P probability of accepting non-minimizing SA solution
- p pressure
- ∆ p pressure drop across flow conduit
- Q volumetric flow rate
- T annealing control parameter (temperature)
- T c current value of annealing control parameter
## References
- [1] N. Metropolis; A.W. Rosenbluth; M.N. Rosenbluth; A.H. Teller; E. Teller. Equation of State Calculations by Fast Computing Machines. Journal of Chemical Physics , 21:1087-1092, 1953.
- [2] S. Kirkpatrick; C.D. Gelatt Jr.; M.P. Vecchi. Optimization by Simulated Annealing. Science , 220(4598):671-680, 1983.
- [3] V. Cern´. Thermodynamical approach to the traveling salesman problem: An ˇ y efficient simulation algorithm. Journal of Optimization Theory and Applications , 45(1):41-51, 1985.
- [4] T. Sochi. Comparing Poiseuille with 1D Navier-Stokes Flow in Rigid and Distensible Tubes and Networks. Submitted , 2013. arXiv:1305.2546.
- [5] T. Sochi. Pore-Scale Modeling of Navier-Stokes Flow in Distensible Networks and Porous Media. Computer Modeling in Engineering & Sciences (Accepted) , 2014.
- [6] T. Sochi. Using the Euler-Lagrange variational principle to obtain flow relations for generalized Newtonian fluids. Rheologica Acta , 53(1):15-22, 2014.
- [7] T. Sochi. Further validation to the variational method to obtain flow relations for generalized Newtonian fluids. Submitted , 2014. arXiv:1407.1534. | null | [
"Taha Sochi"
] | 2014-08-02T10:19:49+00:00 | 2014-08-02T10:19:49+00:00 | [
"physics.flu-dyn",
"physics.comp-ph"
] | Solving the flow fields in conduits and networks using energy minimization principle with simulated annealing | In this paper, we propose and test an intuitive assumption that the pressure
field in single conduits and networks of interconnected conduits adjusts itself
to minimize the total energy consumption required for transporting a specific
quantity of fluid. We test this assumption by using linear flow models of
Newtonian fluids transported through rigid tubes and networks in conjunction
with a simulated annealing (SA) protocol to minimize the total energy cost. All
the results confirm our hypothesis as the SA algorithm produces very close
results to those obtained from the traditional deterministic methods of
identifying the flow fields by solving a set of simultaneous equations based on
the conservation principles. The same results apply to electric ohmic
conductors and networks of interconnected ohmic conductors. Computational
experiments conducted in this regard confirm this extension. Further studies
are required to test the energy minimization hypothesis for the non-linear flow
systems. |
1408.0358v1 | 
## OPEN
SUBJECT AREAS: RAMAN SPECTROSCOPY ELECTRONIC PROPERTIES AND DEVICES OPTICAL PROPERTIES AND
DEVICES
Received 24 April 2014
Accepted 30 June 2014
Published 21 July 2014
Correspondence and requests for materials should be addressed to Y.Y. ([email protected]. cn)
## Graphene, a material for high temperature devices - intrinsic carrier density, carrier drift velocity, and lattice energy
Yan Yin , Zengguang Cheng 2 1 , Li Wang , Kuijuan Jin 1 1 & Wenzhong Wang 3
1 The Institute of Physics, Chinese Academy of Sciences, Beijing 100190, P. R. China, 2 National Center for Nanoscience and Technology, Beijing 100190, P. R. China, 3 Minzu University of China, Beijing 100081, P. R. China.
Heat has always been a killing matter for traditional semiconductor machines. The underlining physical reason is that the intrinsic carrier density of a device made from a traditional semiconductor material increases very fast with a rising temperature. Once reaching a temperature, the density surpasses the chemical doping or gating effect, any p-n junction or transistor made from the semiconductor will fail to function. Here, we measure the intrinsic Fermi level ( j EF j 5 2.93 kBT) or intrinsic carrier density (n in 5 3.87 3 10 6 cm 2 2 K 2 2 ? T 2 ), carrier drift velocity, and G mode phonon energy of graphene devices and their temperature dependencies up to 2400 K. Our results show intrinsic carrier density of graphene is an order of magnitude less sensitive to temperature than those of Si or Ge, and reveal the great potentials of graphene as a material for high temperature devices. We also observe a linear decline of saturation drift velocity with increasing temperature, and identify the temperature coefficients of the intrinsic G mode phonon energy. Above knowledge is vital in understanding the physical phenomena of graphene under high power or high temperature.
S ince the discoveries of carbon nanotube and graphene 1,2 , a significant amount of attention has been directed toward the room temperature and low temperature measurements in pursuing a high carrier mobility 3-5 . As people seek device applications for graphene, high field and high current measurements start to appear 6-10 . The importance of electron-phonon coupling is well recognized 11-13 . However, little attention has been made to the role of intrinsic carriers in graphene, because the great ability of graphene to tolerate the impact of temperature masks the existence of its intrinsic carriers. Thus, the importance of graphene as a material for high temperature devices is not noticed. Here we use a Raman method to measure the level of intrinsic carriers in a voltage biased graphene device up to 2400 K. We find the level of intrinsic carriers in graphene is an order of magnitude less sensitive to the temperature than that of traditional semiconductors, silicon and Germanium. The result tells us that a properly fabricated graphene p-n junction or transistor can operate well above 1500 K and will not fail until other parts melt. Working together with silicon carbide (SiC), graphene might be one of the building materials for coming high temperature devices. We also observe a linear decline of saturation drift velocity in graphene with increasing temperature in contrast to a constant saturation drift velocity in traditional semiconductors, as the result of the degeneracy of carriers and the Pauli exclusion principle. At the same time, we identify the temperature coefficients of the intrinsic G mode lattice vibration energy up to 2400 K. Above information highlights the great application potentials for graphene in high temperature devices, suggests a new exploring direction for graphene research, and is critical in the physics of graphene under high field, high current or high temperature conditions.
The graphene devices were fabricated by mechanical exfoliation from graphite. Figure 1a shows a sketch of the layout of graphene devices and electrical measurement. Contacts were formed on top of the graphene after steps of E-beam lithography, metal evaporation and liftoff. The contacts are 100 nm Au with 5 nm Cr between Au and graphene. The substrate is doped Si with 300 nm SiO2 isolation layer on top. When an electrical bias is applied to a device, the drain electrode and Si substrate are grounded. A Kethley 2400 is used both as a voltage source and a current meter. During the measurements, the devices were kept in a flowing nitrogen gas environment in order to prevent the oxidization of the sample. A Horiba Raman microscope was used to measure the Raman spectra of the sample from the top. The excitation laser is a He-Ne 633 nm (1.96 eV) laser. The focusing objective is a long working distance 50 3 objective lens. Two graphene devices are measured; the optical pictures of these graphene devices are showing in Fig. 1b and 1c. Device A is 5 m minlength and 2.7 m minwidth, and device B is 4.85 m min
Figure 1 | Graphene device information. (a), The sketch of the stacking structure of the devices and the electrical measurement layout. (b), Optical picture of the graphene device A. (c), Optical picture of the graphene device B. The widths and lengths of the devices are marked in the pictures. The blue dots are the optical focus positions during the measurements of the devices. The letters D and S indicate the drain electrodes and source electrodes.
length and 17.1 m minwidth. For example, we measured the device A from Vsd 5 0 V to Vsd 5 21 V with an increasing step of 1 V, then a downward scan (with a step size of 2 1 V) back to Vsd 5 0. For each different Vsd , a Raman spectrum from the center of the device was recorded; the currents of the device before and after each Raman measurement were recorded as well.
Our measurement result reveals the dominating impact of intrinsic carrier due to increasing temperature in the devices under high bias conditions. In Figure 2, we plot the measurement results of Ramanspectra as a function of Vsd for device A. We can see two main Raman peaks, G mode around 1600 cm 2 1 and 2D mode 14 around 2600 cm 2 1 . In both subfigures, G mode intensity increases with a larger Vsd . At the same time, a dramatic increasing of the background signal appears with a high Vsd voltage and this background will drop back down with the decreasing Vsd during the downward scan. In an earlier report 15 , C.-F. Chen et al. presented that the G mode intensity has resonant behavior when the twice of the Femi level in graphene matches the excitation photon energy or the photon energy minus a G mode phonon energy. They also reported an increasing background in the spectra near the resonant condition, and they attributed this high background signal to hot luminescence in graphene. Our data's resonant behaviors in the G mode intensity and background are very similar to those reported behaviors, therefore we believe that the G mode intensity change is due to a changing Femi level in the graphene device. The Femi level ( EF ) in graphene is /C18 /C19
directly linked to the carrier density ( n ) with n ~ EF /C22 h u F 2 = p , where n F is the graphene Fermi velocity, /C22 h is the reduced Planck constant. A higher Fermi level will have a higher carrier density, and vice versa. There are four possible reasons may change the device carrier density. They are intrinsic carrier density increasing due to rising temperature, permanent defects in graphene, back gating effect due to increasing Vsd , and the desorbing of the surface molecules. In comparison with Fig. 3a in C.-F. Chen et al.'s paper 15 after shifting the resonant peak due to different excitation photon energies (785 nm to 633 nm, 1.58 eV to 1.96 eV), we use the relative intensity of the G mode at Vsd 5 21 V over base intensity at Vsd 5 0-5 V, and determine that j EF j < 0.8 eV or n < 4 3 10 13 cm 2 2 when Vsd 5 21 V. First of all, this high carrier density clearly is not the result of permanent defects in the device, because the Vsd downward scan shows this carrier density will go down with a lower voltage. For unintentional residual and surface adsorbate doping, the carrier density is normally at the scale of 10 11 -10 12 cm 2 2 , and this doping should only decrease with an increasing temperature. The back gating effect of the Vsd can be estimated after calculating the back gating capacitance. For a 300 nm SiO2 layer (permittivity e 5 4), the typical back gating capacitance Cbg 5 1.2 3 10 2 8 F cm ? 2 2 ; for Vsd 5 20 V, back gating carrier density at the middle of the device nbg < CbgVsd /2 5 7.5 3 10 11 cm 2 2 . This value is two orders of magnitude smaller than what we have ( n < 4 3 10 13 cm 2 2 ). The only reasonable explanation for this high rising carrier density is the increasing intrinsic carrier density due to rising temperature within the device. Also, the scales of other effects, back gating and desorption of surface adsorbate, are a couple orders of magnitude smaller than the intrinsic carrier density under
Figure 2 | Raman spectra of device A plotted in color maps for different source drain voltages (Vsd). (a), The spectra during the Vsd upward scan. (b), The spectra during the Vsd downward scan.
Figure 3 | Measurement and fitting results for G mode intensities as a function of the Vsd. (a), Device A during Vsd upward scan. (b), Device A during Vsd downward scan. (c), Device B during Vsd upward scan. The data points are fitted with the curve from Fig. 3a in C.-F. Chen et al.'s paper 15 after shifting 0.38 eV (the difference between excitation photon energies, 785 nm to 633 nm or 1.58 eV to 1.96 eV). We only use the Vsd $ 9 V data points for the fitting in order to give a better fitting weight for the high bias data points.
the high bias condition in our sample. We determine that the thermally generated intrinsic carrier is the dominating factor for those short samples under high bias voltage conditions.
First, we will use the G mode intensity to determine the Fermi level in the devices. The Fermi level in the graphene device is changed with Vsd, but not directly tuned by V sd . The Vsd drives the graphene device to generate heat; the heat generated carrier density rises in the device with an increasing temperature; with the help of external field from Vsd, heat generated carriers break balance between electron carriers and hole carriers, and the major carrier winning out raises the Fermi level in the graphene device. For graphene with Fermi level at the Dirac point and no external field, Tian Fang et al. calculated ther/C18 /C19 2
$$\text{mally generated carrier density} ^ { 1 6 }, \, n _ { t h } = \frac { \pi } { 6 } \left ( \frac { \bar { k } _ { B } T } { \hbar { v } _ { F } } \right ) ^ { 2 }, \, \text{where } k _ { B } \text{ is the} \quad \text{the } E _ { 0 } \, \text{va} \\ \text{n.i.d. } \text{and} s$$
Boltzmann constant, T is the temperature. Their result gives out a nth for both the electrons and the holes, the electrons and holes cancel each other so the Fermi level is still zero. Clearly, our graphene devices under high field don't fit into their calculation's pre-condition, and their calculation result probably is numerically invalid for our case as well. In a connected real device with a certain electric voltage level, the Fermi level can go well above zero. For our devices, weassume the Fermi level due to rising intrinsic carrier density at the measurement point of a device,
$$| E _ { F } | = \frac { \alpha } { 2 } V _ { s d } + E _ { 0 }, \quad \ \ ( 1 ) \begin{array} { c c } V ; \, t l \\ 0. 0 5 \\ \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
\quad \quad \quad$$
where a is a device and location dependent constant, E0 is a constant Fermi level when Vsd 5 0. The reasons behind this assumption are: it is reasonable to guess j EF j / kBT , and we notice that device temperature T has a linear relationship with Vsd. (This will be discussed later when we discuss transport data, and in supplementary information.) Equation (1) works much better for Vsd downward scan than Vsd upward scan. During Vsd upward scan, the graphene device goes through desorption of surface adsorbate and increasing of permanent defects due to rising temperature 17 , so the E0 is not a constant. During Vsd downward scan, the surface adsorbate was fully released at the upward scan step, and the permanent defects should be the same because the device is cooling, so the E0 is about a constant during downward scan. This argument is supported by Figure 4a, which shows that the downward scan data for G mode positions shows a much better smoothness and data continuity than the upward scan data. In addition, Figure 4a shows that the G mode peak position, V G 5 1590.44 cm 2 1 , for Vsd 5 0 after the downward scan. I. Calizo et al. reported that the graphene G mode energy linear coefficient with temperature is 2 0.016 cm 2 1 /K for a temperature range 83 K-373 K, and V G 5 1584 cm 2 1 for T 5 0 K 18 , so V G 5
1579.2 cm 2 1 for T 5 298.15 K (room temperature) if the Fermi level in graphene is zero. The blue shifted 11.24 cm 2 1 , in G mode energy at Vsd 5 0 after downward scan, is due to the Fermi level E0 in graphene. Because 15 D V G 5 j EF j 3 42 cm 2 1 eV 2 1 , we will have E0 5 0.2676 eV for device A during downward scan. Figure 3 shows our measured G mode intensities as a function of the Vsd . The fitting curves are the curve retrieved from C.-F. Chen et al.'s paper 15 Fig. 3a for I G vs. 2 j EF j , after blue shifted 0.38 eV. (They use a 785 nm-1.58 eV excitation laser and we use a 633 nm-1.96 eV excitation laser.) Our experimental data are fitted to the curve with two fitting parameters, a in equation (1) with constant E0 5 0.2676 eV and a linear scaling factor between our G mode intensity and I G in C.-F. Chen et al.'s paper. The fitting is most accurate for device A during downward scan, because the E0 value is measured for it. We do the same fitting for device A upward scan anyway, just to show a comparison. We only use the Vsd $ 9 V data points for the fitting in order to give a better fitting weight for the high bias data points, which matter the most in this fitting. The G mode intensities at the high bias voltages already rise significantly above the random error range, especially for the data from device A. Therefore, our data have enough signal-to-noise ratios to make this fitting method valid, although our data doesn't cover the full resonant peak. For device A, the upward scan data fitting gives a 5 0.0520 eV/V; the downward scan data fitting gives a 5 0.0513 eV/ V; these two values are close and we take the downward scan a 5 0.0513 eV/V as the value for device A. Data from device B has lower quality than the data from device A, however we use device B for the verification purpose only. Because device B broke down after the measurement at Vsd 5 17 V during the upward scan, no data is available for device B Vsd downward scan and we can't accurately determine the E0 . In Fig. 4a, the high bias ( Vsd $ 6 V) data points' behaviors are similar between the upward scan and downward scan for device A, this might suggest that the average E0 value for high bias conditions in upward scan is close to the E0 value in downward scan. We also consider device B has a similar length and goes through similar processes as device A, it is likely that device B has a similar defect level as device A. So, we use E0 5 0.2676 eV from device A, as an approximation value, for the fitting of device B high bias data in Fig. 3c. The fitting gives a < 0.0566 eV/V for device B.
Next, we will find out the temperatures at the measurement point for difference Vsd situations using G mode peak position information. The key information we used to determine temperature is: the value of room temperature, the intrinsic G mode peak position shift (after correcting the doping effect) as a result of temperature change, the reported G mode energy temperature coefficient for temperature range 83 K to 373 K from Ref. 18, and a 2 nd order polynomial fitting of our G mode peak position. The 2 nd order fitting gives the relation-
Figure 4 | The G mode peak positions as a function of the Vsd. (a), Device A. The blue square, red cross, and blue circle data points are the experimental data points for Vsd downward scan, Vsd upward scan, and the G mode position for downward scan after taking out the effect of carrier density, respectively. The blue curve is the fitting result for the corrected G mode position with a 2 nd order polynomial function. (b), Device B. The blue square and blue circle data points are the experimental data points for Vsd upward scan and the G mode position after taking out the effect of carrier density. The blue solid curve and dash curve are the fitting results for the corrected G mode position and the original data points, respectively, with a 2 nd order polynomial function. Only the data points of Vsd $ 6 V are used for the fittings of device B.
ship between the first and second temperature coefficients; our device temperature range overlaps with Ref. 18, so the reported the linear temperature coefficient from Ref. 18 sets another condition for the first and second temperature coefficients. Combining both conditions, we can have the values of the first and second temperature coefficients for G mode energy. With the value of room temperature and G mode Raman shift of each data point, we can have the device temperature of each data point. Figure 4 shows the G mode peak positions as a function of the Vsd for device A and device B. In comparison between upward scan and downward scan G mode positions for device A, the downward scan data shows a much better smoothness and data continuity. The surface adsorbate was fully released during upward scan, so no chemical doping changes during downward scan. The device goes through cooling during downward scan, the permanent defects have a constant amount, and only shift the downward scan data a constant amount in G mode energy. Therefore, we use the downward scan data for further data analyzing. During Vsd downward scan, the G mode peak position, V G , satisfies V G 5 V G\_T 1 D V G\_in 1 D V G\_defect , where V G\_T is the intrinsic G mode energy as a function of the temperature, D V G\_in is the G mode energy shift due to rising Fermi level (intrinsic carrier density) as a function of Vsd , D V G\_defect is a constant from the permanent defects level in the graphene. We know that D V G\_in 1 D V G\_defect 5 j EF j 3 42 cm 2 1 eV 2 1 and EF j j ~ 0 0513 : eV V N Vsd 2 z 0 2676 eV for device A, : then we can have a corrected G mode energy, V G\_T 5 V G 2 ( D V G\_in 1 D V G\_defect ), and we plot the new values in Fig. 4a as blue circle points. The new, after correction, G mode energies are fitted very well with a 2 nd order polynomial function,
$$\Omega _ { G _ { - T } } = a _ { 1 } V _ { s d } ^ { 2 } + a _ { 2 } V _ { s d } + a _ { 3 }, \quad \quad \ ( 2 ) \ \ \text{meas} \quad \ \text{$\varepsilon$}$$
where a1 , a2 and a3 are fitting parameters. For device A, we get a1 5 2 0.0474 cm 2 1 V 2 2 , a2 5 2 1.6901 cm 2 1 V 2 1 , and a3 5 1580 cm 2 1 . Wehave not found any report about the coefficients of the graphene G mode energy as a function of the temperature for a temperature range up to 2500 K. H. Herchen and M. Cappelli reported that a first order Raman mode energy of diamond has a 2 nd order polynomial dependence on temperature up to 2000 K 19 . So, it is reasonable to assume, for G mode in graphene, that
$$\Omega _ { G _ { L } T } = b _ { 1 } T ^ { 2 } + b _ { 2 } T + b _ { 3 },$$
where b1 , b2 and b3 are graphene dependent only constants. The temperature of the device appears a linear relationship with Vsd , and we define
$$T = \beta V _ { s d } + T _ { R T },$$
where b is a device and location dependent constant and TRT is the room temperature, 298.15 K. Combining equations (2), (3) and (4), we will have
$$a _ { 1 } = b _ { 1 } \beta ^ { 2 }, \, a _ { 2 } = 2 \beta T _ { R T } b _ { 1 } + b _ { 2 } \beta.$$
I. Calizo et al. reported that the graphene G mode energy linear coefficient with temperature is 2 0.016 cm 2 1 /K for a temperature range 83 K-373 K, and V G\_T 5 1584 cm 2 1 for T 5 0 K 18 . Because b1 , b2 and b3 are graphene dependent only constants, we have /C12
$$\varkappa _ { 1 }, \varkappa _ { 2 } \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \ dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \\ \varkappa _ { 1 } } & \frac { \Delta \Omega _ { G _ { T } } } { \Delta T } \Big | _ { T = 2 2 8 K } = 2 b _ { 1 } T | _ { T = 2 2 8 K } + b _ { 2 } = - 0. 0 1 6 \, c m ^ { - 1 } / K, \\ \intertext { a } b _ { 3 } = 1 5 8 4 \, c m ^ { - 1 },$$
in which 228 K is the average of 83 K and 373 K. Combining equations (5) and (6) with knowing values for a1 and a2 from fitting of device A, we can solve b 1 5 2 4.596 3 10 2 6 cm 2 1 K 2 2 , b 2 5 2 0.0139 cm 2 1 K 2 1 , b 3 5 1584 cm 2 1 , and b 5 101.5 K/V for the measurement point in device A. Finally, we combine equation (1) and (4), and have j EF j 5 c T to describe a graphene device intrinsic Fermi level as a function of the temperature, where c 5 a /2 b . From device A, our measurement gives out a graphene device intrinsic Fermi level, j EF j 5 2.53 3 10 2 4 eVK 2 1 ? T or j EF j 5 2.93 kBT . Using graphene /C22 h u F 5 1.16 3 10 2 28 J ? m , we have a graphene device intrinsic carrier density, nin 5 3.87 3 10 6 cm 2 2 K 2 2 ? T 2 .
In Fig. 4b, the square points are the measured G mode positions of device B as a function of Vsd . The circle points are the corrected G mode energies using the same correction method for device A, but with EF j j ~ 0 0565 : eV V N Vsd 2 z 0 2676 eV. For the : b value, it is invalid to process the limited data from device B with the same method for
Figure 5 | Device I-V curves, temperatures, and carrier drift velocities. (a), Device A, the defect level is 4 3 10 12 cm 2 2 . (b), Device B. Red solid square points and black solid round points are the I-V data during Vsd downward and upward scan, respectively, using left y axis. The current values are the averages of the currents before and after the Raman measurements, the differences of the two are plotted as error bars. Red hollow diamond points and black hollow star points are calculated carrier drift velocities during Vsd downward and upward scan, respectively, using right y axis. Only the data points with carrier density n $ 1.4 3 10 17 m 2 during the upward scan are plotted in order to control the error in drift velocities due to the 2 uncertainty in E0 . The top axis plots the device temperature, which is calculated from Vsd using equation (4).
device A. Luckily, a 1 is a fitting parameter independent from the values of a and E0 . In order to prove this point, we did the 2nd order polynomial fittings for both measured G mode positions and corrected G mode energies, and plotted the fitting curves as a dash line and a solid line, respectively, in Fig 4b. Both fittings give a 1 5 2 0.0603 cm 2 1 V 2 2 . Because b1 is a graphene dependent only constant and should be same for all graphene, we can use the b 1 5 2 4.596 3 10 2 6 cm 2 1 K 2 2 value from device A, for device B. Using equation (5), a 1 5 b 1 b 2 , we can get b 5 114.5 K/V for device B. Together with a < 0.0566 eV/V for device B, we get c 5 2.47 3 10 2 4 eV/K from device B. This c value of device B is consistent with the c value from device A. Such an agreement between device A and device B supports the validity of our method.
graphene transistor or a chemical doped graphene p-n junction will be still well functioning at a temperature above 1500 K. A lot of things melt at 1500 K, for example gold melting temperature is about 1340 K. So, such a graphene transistor or graphene p-n junction will never fail due to rising intrinsic carrier density before other parts start to melt. Above analyses reveal the great potential of graphene as the material for high temperature devices and point out a new research direction for graphene applications.
Graphene shows a great potential as a material for high temperature devices, compared to traditional semiconductors. High temperature devices are demanded for military, astronautic, and heavy industry applications, where high power density is required or high temperature environment can't be avoided. In order to make a nonlinear semiconductor device (p-n junction or transistor) functioning, the chemically doped carriers or the gating generated carriers have to be the controlling factor within the semiconductor material. However, when the device temperature increases, the semiconductor's intrinsic carrier density will increase. Once the temperature reaches a certain point that, the semiconductor's intrinsic carrier surpasses the chemical doping or gating effect and dominates the device's response, the nonlinear device will simply fail its function. For instance, a Si device's intrinsic carrier density will double if temperature increases 8 K from room temperature, the maximum operation temperature for a Si device is 250 u C; a Ge device's intrinsic carrier density will double if temperature increases 12 K from room temperature, the maximum operation temperature for a Ge device is 100 u C 20 . Normally for a semiconductor, a larger band gap gives a higher operation temperature; however, a larger band gap also increases the difficulty of carrier density tuning. Graphene is semimetallic, not a semiconductor. By a doping or field effect, graphene can be made into a p-n junction or a transistor as well. Running our numbers for graphene, we find that it will take 125 K for graphene intrinsic carriers to double from room temperature. At 1500 K, graphene intrinsic carrier density is 8.7 3 10 12 cm 2 2 . Multiple research reports 15,21,22 have demonstrated that a top gating method or a chemical doping can tune the graphene carrier density to a level above 6 3 3 10 13 cm 2 2 . Therefore, we foresee that a well-fabricated top gated
Finally, we will show and discuss carrier drift velocity in graphene up to 2400 K. Fig. 5 shows the device I-V curves, temperatures and carrier drift velocities of device A and device B. Carrier drift velocity, n , equates to I /( qWn ), where I is the current, q is the electron charge, W is the device width, n is the carrier density. The device B broke down after Vsd . 17 V and the device A was also on the brink of breakdown at Vsd 5 21 V and showed clear runaway effect, in which the current keeps increasing and feeds itself with a constant driving Vsd . The maximum temperature is about 2400 K for device A and 2200 K for device B, those temperatures are consistent with the theoretically calculated graphene breakdown temperature, 2230 6 700 K, in vaccum . The drift velocities from device A, 8 Vsd downward have the best accuracy compared to the values from upward scans, and map out a very nice three stage behavior under high field. In the first stage, the drift velocity increases linearly with the field, Ohm's law holds perfectly. Then with a further increased field, lattice scattering dominates, the drift velocity still increases but with a reducing rate, Ohm's law still works but with a reducing conductivity. In the last stage, after the field reaches a certain level, intrinsic carrier dominates, the drift velocity stops to increase. Traditionally, the drift velocity at this point is called the saturation drift velocity, and this saturation drift velocity stays the same for traditional semiconductors if the field increases further. For example, for Si, it is about 1 3 10 5 m/s. Our graphene sample shows a linear drop of ''saturation'' drift velocity with the increasing temperature. This reducing saturation drift velocity of graphene has been predicted by theoretical calculation considering the degeneracy of carriers and the Pauli exclusion principle 23 . The same calculation gives a saturation drift velocity value, 2 3 10 5 m/s, for an ideal graphene with a doping level 10 13 cm 2 2 at 300 K. The maximum drift velocity of our graphene with a defect level of 4 3 10 12 cm 2 2 is about 4 3 10 4 m/s at 1000 K. Our value is not far from the calculation value after considering the temperature difference and the scattering from the defects. The declining saturation drift velocity in graphene helps
delay the current runaway point, where the current keeps increasing on its own with a constant driving voltage, to an even higher temperature; but it also limits the ability of graphene in delivering large current. Another thing we want to point out is that Ohm's law fails totally in the intrinsic carrier dominated third stage. The electrons and the lattice are in a non-equilibrium state, in which the energy electrons taking from the field is not fully dumped to the lattice. This is why the device temperature has a linear relationship with V, not the IV. The difference in drift velocity for a same Vsd between downward and upward scan is due to different effective electron temperatures, despite a similar lattice temperature. During downward scan a graphene device appears a smaller contact resistant (or barrier) than the upward scan 24 , therefore, downward scan data has a larger current, electrons carrying more energy and moving faster, or a higher effective electron temperature than the upward scan data.
In summary, we use Raman method to quantitatively measure the graphene intrinsic carrier density, drift velocity and G mode phonon energy as functions of temperature up to 2500 K. We find that the intrinsic carrier density of graphene is an order of magnitude less sensitive to temperature than that of traditional semiconductor materials. We foresee graphene becomes the material for high temperature devices and a properly fabricated graphene transistor or p-n junction will not fail at all before other parts melt. We map out the three stage behavior of carrier drift velocity as a function of the field or temperature, and graphene shows a linear decline in ''saturation'' drift velocity in contrast to a constant saturation drift velocity in traditional semiconductor materials. We also find the G mode phonon intrinsic energy coefficients with temperature, V G 52 4.596 3 10 2 6 cm 2 1 K 2 2 T 2 1 2 0.0139 cm 2 1 K 2 1 T 1 1584 cm 2 1 . All of above reveal the great potential of graphene as a foundational material for high temperature devices, identify a new research direction for graphene applications, and are vital in understanding the physics of graphene under high field or high temperature.
- 1. Novoselov, K. S. et al . Two-dimensional gas of massless Dirac fermions in graphene. Nature 438 , 197-200; DOI:10.1038/Nature04233 (2005).
- 2. Zhang, Y. B., Tan, Y. W., Stormer, H. L. & Kim, P. Experimental observation of the quantum Hall effect and Berry's phase in graphene. Nature 438 , 201-204; DOI:10.1038/Nature04235 (2005).
- 3. Hwang, E. H., Adam, S. & Das Sarma, S. Carrier transport in two-dimensional graphene layers. Phys. Rev. Lett. 98 , 186806; DOI:10.1103/Physrevlett.98.186806 (2007).
- 4. Morozov, S. V. et al . Giant intrinsic carrier mobilities in graphene and its bilayer. Phys. Rev. Lett. 100 , 016602; DOI:10.1103/Physrevlett.100.016602 (2008).
- 5. Bolotin, K. I., Sikes, K. J., Hone, J., Stormer, H. L. & Kim, P. Temperaturedependent transport in suspended graphene. Phys. Rev. Lett. 101 , 096802; DOI:10.1103/Physrevlett.101.096802 (2008).
- 6. Moon, J. S. et al . Epitaxial-Graphene RF Field-Effect Transistors on Si-Face 6HSiC Substrates. IEEE Electr. Device L. 30 , 650-652; DOI:10.1109/ Led.2009.2020699 (2009).
- 7. Freitag, M. et al . Energy Dissipation in Graphene Field-Effect Transistors. Nano Lett. 9 , 1883-1888; DOI:10.1021/Nl803883h (2009).
- 8. Dorgan, V. E., Behnam, A., Conley, H. J., Bolotin, K. I. & Pop, E. High-Field Electrical and Thermal Transport in Suspended Graphene. Nano Lett. 13 , 4581-4586; DOI:10.1021/nl400197w (2013).
- 9. Chae, D. H., Krauss, B., von Klitzing, K. & Smet, J. H. Hot Phonons in an Electrically Biased Graphene Constriction. Nano Lett. 10 , 466-471; DOI:10.1021/ Nl903167f (2010).
- 10. Bae, M. H., Ong, Z. Y., Estrada, D. & Pop, E. Imaging, Simulation, and Electrostatic Control of Power Dissipation in Graphene Devices. Nano Lett. 10 , 4787-4793; DOI:10.1021/Nl1011596 (2010).
- 11. Yao, Z., Kane, C. L. & Dekker, C. High-field electrical transport in single-wall carbon nanotubes. Phys. Rev. Lett. 84 , 2941-2944; DOI:10.1103/ PhysRevLett.84.2941 (2000).
- 12. Yin, Y. et al . Optical determination of electron-phonon coupling in carbon nanotubes. Phys. Rev. Lett. 98 , 037404; DOI:10.1103/Physrevlett.98.037404 (2007).
- 13. Yin, Y. et al . Electron-phonon coupling of G mode and assignment of a combination mode in carbon nanotubes. Phys. Rev. B 84 , 075428; DOI:10.1103/ Physrevb.84.075428 (2011).
- 14. Ferrari, A. C. et al . Raman spectrum of graphene and graphene layers. Phys. Rev. Lett. 97 , 187401; DOI:10.1103/Physrevlett.97.187401 (2006).
- 15. Chen, C. F. et al . Controlling inelastic light scattering quantum pathways in graphene. Nature 471 , 617-620; DOI:10.1038/Nature09866 (2011).
- 16. Fang, T., Konar, A., Xing, H. L. & Jena, D. Carrier statistics and quantum capacitance of graphene sheets and ribbons. Appl. Phys. Lett. 91 , 092109; DOI:10.1063/1.2776887 (2007).
- 17. Cheng, Z. et al . Toward Intrinsic Graphene Surfaces: A Systematic Study on Thermal Annealing and Wet-Chemical Treatment of SiO2-Supported Graphene Devices. Nano Lett. 11 , 767-771; DOI:10.1021/nl103977d (2011).
- 18. Calizo, I., Miao, F., Bao, W., Lau, C. N. & Balandin, A. A. Variable temperature Raman microscopy as a nanometrology tool for graphene layers and graphenebased devices. Appl. Phys. Lett. 91 , 071913; DOI:10.1063/1.2771379 (2007).
- 19. Herchen, H. & Cappelli, M. A. 1st-Order Raman-Spectrum of Diamond at HighTemperatures. Phys. Rev. B 43 , 11740-11744; DOI:10.1103/PhysRevB.43.11740 (1991).
- 20. Liu, E., Zhu, D. & Luo, J. Semiconductor Physics (Publishing House of Electronics Industry, Beijing, 2008).
- 21. Das, A. et al . Monitoring dopants by Raman scattering in an electrochemically top-gated graphene transistor. Nature Nano. 3 , 210-215; DOI:10.1038/ nnano.2008.67 (2008).
- 22. Zhao, W. J., Tan, P. H., Liu, J. & Ferrari, A. C. Intercalation of Few-Layer Graphite Flakes with FeCl(3): Raman Determination of Fermi Level, Layer by Layer Decoupling, and Stability. J. Am. Chem. Soc. 133 , 5941-5946; DOI:10.1021/ Ja110939a (2011).
- 23. Fang, T., Konar, A., Xing, H. L. & Jena, D. High-field transport in two-dimensional graphene. Phys. Rev. B 84 , 125450; DOI:10.1103/Physrevb.84.125450 (2011).
- 24. Chiu, H. Y., Perebeinos, V., Lin, Y. M. & Avouris, P. Controllable p-n Junction Formation in Mono layer Graphene Using Electrostatic Substrate Engineering. Nano Lett. 10 , 4634-4639; DOI:10.1021/Nl102756r (2010).
## Acknowledgments
This work was supported by the National Basic Research Program of China (No. 2014CB921001 and No. 2014CB339800), by the Young Scientists Fund of the National Natural Science Foundation of China (Grant No. 11004231), by the National Natural Science Foundation of China (Grant No. 21161120321), and by the Scientific Research Start-up Fund for the Returned Overseas Chinese Scholars from Ministry of Education of China. Y.Y. also acknowledges Ying Fang's good suggestions for the experiments, Yuping Yang's help in arranging the Raman instrument, and helpful discussions with Wenxin Wang.
## Author Contributions
Y.Y. designed the experiments, carried out the measurements, performed the data analyses and drafted the initial manuscript; Z.C. contributed to sample fabrication; L.W. and K.J. provided the instruments for transport measurements and funding support, W.W. provided the Raman instrument. All authors discussed the results and wrote the paper together.
## Additional information
Supplementary information accompanies this paper at http://www.nature.com/ scientificreports
Competing financial interests : The authors declare no competing financial interests.
How to cite this article: Yin, Y., Cheng, Z., Wang, L., Jin, K. & Wang, W. Graphene, a material for high temperature devices - intrinsic carrier density, carrier drift velocity, and lattice energy. Sci. Rep. 4 , 5758; DOI:10.1038/srep05758 (2014).
This work is licensed under a Creative Commons Attribution-NonCommercialShareAlike 4.0 International License. The images or other third party material in this article are included in the article's Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder in order to reproduce the material. To view a copy of this license, visit http:// creativecommons.org/licenses/by-nc-sa/4.0/
 | 10.1038/srep05758 | [
"Yan Yin",
"Zengguang Cheng",
"Li Wang",
"Kuijuan Jin",
"Wenzhong Wang"
] | 2014-08-02T10:29:40+00:00 | 2014-08-02T10:29:40+00:00 | [
"cond-mat.mtrl-sci",
"physics.optics"
] | Graphene, a material for high temperature devices; intrinsic carrier density, carrier drift velocity, and lattice energy | Heat has always been a killing matter for traditional semiconductor machines.
The underlining physical reason is that the intrinsic carrier density of a
device made from a traditional semiconductor material increases very fast with
a rising temperature. Once reaching a temperature, the density surpasses the
chemical doping or gating effect, any p-n junction or transistor made from the
semiconductor will fail to function. Here, we measure the intrinsic Fermi level
(|E_F|=2.93k_B*T) or intrinsic carrier density (n_in=3.87*10^6 cm^-2 K^-2*T^2),
carrier drift velocity, and G mode phonon energy of graphene devices and their
temperature dependencies up to 2400 K. Our results show intrinsic carrier
density of graphene is an order of magnitude less sensitive to temperature than
those of Si or Ge, and reveal the great potentials of graphene as a material
for high temperature devices. We also observe a linear decline of saturation
drift velocity with increasing temperature, and identify the temperature
coefficients of the intrinsic G mode phonon energy. Above knowledge is vital in
understanding the physical phenomena of graphene under high power or high
temperature. |
1408.0359v1 | ## Spontaneous jamming and unjamming in a hopper with multiple exit orifices
Amit Kunte, Pankaj Doshi, ∗ and Ashish V. Orpe †
Chemical Engineering Division, National Chemical Laboratory, Pune 411008 India
(Dated: August 5, 2014)
We show that the flow of granular material inside a 2-dimensional flat bottomed hopper is altered significantly by having more than one exit orifice. For the hoppers with small orifice widths, intermittent flow through one orifice enables the resumption of flow through the adjacent jammed orifice, thus displaying a sequence of jamming and unjamming events. Using discrete element simulations, we show that the total amount of granular material (i.e. avalanche size) emanating from all the orifices combined can be enhanced by about an order of magnitude difference by simply adjusting the inter-orifice distance. The unjamming is driven primarily by fluctuations alone when the interorifice distance is large, but when the orifices are brought close enough, the fluctuations along with the mean flow cause the flow unjamming.
PACS numbers: 45.70.Mg,47.57.Gc
An assembly of discrete, non-cohesive particles, aka dry granular media, while trying to flow through a narrow opening can clog or jam, an occurrence widely observed in particle drainage through silos and hoppers which are used ubiquitously in several industrial applications. The jamming at the orifice is caused due to few particles forming a stable arch at the exit [1, 2] eventually causing the entire system to halt abruptly. The particles constituting the arch can be from anywhere in the system [2] and while occurrence of the arch is quite unpredictable, some information can be obtained through the spatial distribution of the velocity fluctuations in the system [3]. The shape of the arch, though, can be predicted quite well using a random walk model [1].
For a three dimensional hopper there exists a critical orifice width above which the system never jams [4] while no such limit exists for a two dimensional system which can jam for large enough orifice widths [5]. It has been shown that the output flow rate from a hopper can be increased as large as 10% by placing suitable inserts within at appropriate locations [6-8]. This also decreases the probability of jamming by about two orders of magnitude [9]. These studies have shown that the reduction in the tendency to jam is due to the reduction of the pressure in the region of arch formation. The motivation for placing inserts is derived from parallel interesting studies carried out to alter the flow behavior of pedestrians from a crowded room [10, 11].
Over here we report an interesting observation about the jamming behavior in a 2-dimensional hopper having two exit orifices of same width placed far away from each other. Whenever one of the orifice jams, which is expected given the orifice width incorporated, it unjams spontaneously if the flow is occurring through other orifice. Effectively, flow continues to occur through both orifices, alternately or together, for much longer duration than expected for a hopper with single orifice. The overall duration of flow, consequently the tendency of jamming, can be altered simply by changing the inter orifice distance. Such non-local interaction between regions exhibiting differing flow behavior has been observed previously, but in different configurations [12, 13] and under different flow conditions. In these studies it was shown that the origin of such non-local interaction can be attributed to a self activated process within which the stress fluctuations induced by a localized shear causes the material to yield and flow elsewhere [13].
In the present system, we conjecture that the observed behavior of spontaneous unjamming occurs due to rearrangement of the particles in the jammed region above one orifice. This unjamming behavior can be correlated with the fluctuations induced in the system due to flow from the nearby orifice. We believe that such non-local interaction between two widely separate regions can be used to systematically alter the jamming behavior in a hopper non-intrusively which could be quite significant for several industrial operations. We have explored this behavior in great detail through DEM simulations of soft particles. We measure the mean avalanche size for varying conditions and show their correlation with the fluctuations measured as root mean squared (r.m.s.) values.
The DEM simulations are carried out using the Large Atomic/Molecular Massively Parallel Simulator
FIG. 1. Sample snapshots of the flow occurrence in a twoorifice hopper for a specified w and D . (a) Flow occurring through the left orifice while the right orifice is jammed. (b) Spontaneous flow re-initialization through the right orifice at a later time.
(LAMMPS) developed at Sandia National Laboratories [14, 15]. The simulation employs Hookean force between two contacting particles described in detail elsewhere [16]. All the simulation parameters are the same as used in the systematic study of hopper flows carried out previously [16] except for a higher normal elastic constant ( k n = 2 × 10 6 mgd ) which corresponds to a more stiffer particle. The inter-particle friction coefficient ( µ ) is varied from 0 2 to 0 8 with no qualitative difference . . between the results. Over here, we report the results obtained for µ = 0 5. .
The simulation geometry consists of a 2-dimensional hopper of height about 1 2 times its width and thickness . 1d, where d is the mean particle diameter with a polydispersity of 15 %. The sidewalls are created out of the largest particles (rough wall) while the bottom surface is created from the smallest particles (relatively smooth wall) by freezing the particles so that their translational and angular velocities are kept zero throughout the simulation. The hopper has two orifices of width D placed at an inter-orifice distance w , both defined in terms of mean particle diameter d . The hopper width is large enough to prevent any confinement effects due to their proximity to either orifice. The fill height is maintained constant by re-positioning the particles exiting from the orifice just above the free surface. The hopper is filled using the sedimentation method as suggested previously [17] in which a dilute packing of non-overlapping particles is created in a simulation box and allowed to settle under the influence of gravity. The simulation is run for a significant time so that the kinetic energy per particle is less than 10 -8 mgd resulting into a quiescent packing of desired fill height H in the hopper.
The flow through the hopper is initiated by opening both the orifices simultaneously. The orifice width is chosen to be small enough to cause jamming after certain period of flow. After the flow is initialized, either of the orifice gets jammed but unjams again spontaneously. Note that this unjamming would not have been possible in the absence of second orifice through which the flow occurs for a very short duration before jamming itself. The jamming-unjamming sequence can, thus, flip from one orifice to other. Effectively the flow occurs through either one or both the orifices at any given time. After few of these jamming-unjamming sequences, the flowing orifice jams before it can unjam the other orifice and the overall flow stops. It is quite evident that the particles above the flowing orifice transmit some information to the jammed orifice causing it to unjam again. However, this information is available only for a short duration before the flowing orifice jams on its own. After both orifices jam, the flow is re-initiated by removing 2 -3 particles from each of the arch. This procedure is continued to get significant number of jamming events. The total number of particles flowing out from the hopper from the instant both orifices are opened until both are
FIG. 2. Jamming characteristics for varying distances ( w ) between orifices of different widths ( D ). (a) Mean avalanche size per unit orifice ( s m ) plotted against the inter-orifice distance ( w ). Dashed lines corresponds to the avalanche size ( s 0 ) for a hopper having single orifice of width D . (b) Scaled mean avalanche size ( s m /s 0 ) plotted against the inter-orifice distance ( w ).
jammed is defined as an avalanche size ( s ). The value of s is found to depend on the values of D and w which we discuss next. A sample snapshot of particles flowing through one orifice while other being jammed is shown in Fig. 1a. A small time later, the jammed orifice starts to flow spontaneously as shown in Fig. 1b.
The distribution of avalanche sizes ( s ) for a given D and w , normalized by the mean avalanche size 〈 s 〉 , exhibits an exponential tail for all cases which is typical of the random nature of the discrete avalanche events [9]. Fig. 2a shows the mean avalanche size per orifice ( s m = 〈 s / 〉 2) obtained for four different values of D and several inter-orifice distance ( w ). The behavior is qualitatively similar for all the orifice sizes. The avalanche size ( s m ) decreases monotonically with increasing w and at infinitely large w it asymptotically approaches a constant value, which corresponds to that for a single orifice hopper (see Fig. 3b). In this scenario, the two orifices will function unaware of existence of the other and no information is exchanged between the respective flowing regions. For w larger than those shown in Fig.2a (but not studied), an unjamming event can happen, but with much lesser probability. Now with decreasing w , the value of s m increases and it grows quite rapidly for small enough w . This is the consequence of the flow from one orifice aiding that through the other in some way, thus, effectively increasing the total time period over which flow occurs, hence larger s m . In the limit the two orifices are very close to each other ( w of the order of D or lower), the flow now occurs as if through a wider orifice
FIG. 3. (a) Schematic of a single orifice hopper of width L and height H . (b) Variation of mean avalanche size ( s 0 ) for a hopper with single orifice of width D . Variation of (c) mean velocity v , (d) r.m.s. velocity u and (e) r.m.s. normal stress T along the horizontal direction ( x ) obtained over a region 3 d in z -direction for four different orifice widths. Please refer to text for the method to obtain mean and rms quantities. Mean velocity and rms velocity are measured in units of d/τ , while rms stress is measured in units of mg/d 2 .
( ∼ 2 D ). Eventually it approaches the asymptotic limit for a single orifice of width 2 D . Within the entire range of w , the value of s m is observed to vary by almost an order of magnitude difference for the largest orifice width considered. The mean avalanche size ( s m ) normalized by the single orifice avalanche size ( s 0 ) is shown in Fig. 2b. The data shows reasonable collapse for w > 20 d , while increasing scatter is observed with decreasing w , which, perhaps, indicates a non-linear dependence on the orifice width ( D ) and interacting flow fields which cannot be scaled out by simple normalizing.
To elucidate the origin of the enhanced avalanche sizes, we perform simulations in a hopper fitted with single orifice for different D (see Fig. 3a). The width of the hopper is more than twice the maximum distance ( w ) used in two orifice system. For every orifice size, several avalanches are obtained by re-initiating the flow post jamming. The mean avalanche size shows an exponential-squared dependence on the orifice size ( e [0 26 . D 2 ] ) as shown in Fig. 3b and is in accordance with experimental results obtained previously for a 2-dimensional system [5] suggesting of the absence of a critical orifice width ( D ) separating the flowing and jamming conditions.
We next obtain the profiles of mean velocity, velocity fluctuations and normal stress fluctuations of the particles along the x -direction. To calculate these quantities, we save snapshots of particle positions at intervals of 0 0025 . τ , where τ = √ d/g , g is the gravitational acceleration and integration time step used in the simulations is δt = 2 5 . × 10 -5 τ . The mean velocity is defined as v x ( ) = √ 〈 c x 〉 2 + 〈 c z 〉 2 , while the average velocity fluctuations are measured in terms of root mean squared (rms) velocity defined as u x ( ) = √ [ 〈 c 2 x 〉 - 〈 c x 〉 2 ] + [ 〈 c 2 z 〉 - 〈 c z 〉 2 ]. Here, c x and c z are the instantaneous horizontal and vertical velocity components of every particle obtained from the displacements between two successive snapshots. The simulation algorithm also outputs the horizontal ( σ x ) and vertical ( σ z ) component of computed normal stress on each particle within each snapshot. The normal stress fluctuations are captured in terms of r.m.s. stress defined as T x ( ) = √ [ 〈 σ 2 x 〉 - 〈 σ x 〉 2 ] + [ 〈 σ 2 z 〉 - 〈 σ z 〉 2 ]. In all the above calculations, 〈〉 represents a temporal average over several time instants of flow and spatial average over a 3 d region in z -direction located about 4 d above the orifice. This spatial region is chosen so as to capture the essential dynamics in a simple manner. All the three quantities are shown in Fig. 3c,d,e for the four orifice widths. The profiles are symmetric about the orifice and hence only the right half of the profiles are shown in each case.
The magnitude of all the quantities increase with an increase in the orifice width which is expected given the faster flow and consequently, more collisions between particles. The profiles show similar behavior for different orifice sizes and collapse quite nicely (not shown) when normalized by the respective quantity at the x = 0. The mean velocity decays much rapidly with distance from the orifice while the r.m.s. velocity and stress show a much gradual decay. The ratio of r.m.s. to the mean velocity increases progressively away from the orifice and reaches a value of about 100 by x = 45 d . Almost similar numbers are obtained for all the orifice sizes studied.
Now consider the scenario of a two orifice hopper with one of the orifice located at x = 0 which is in unjammed (flowing) state. Provided the other orifice located at some distance ( w or x ) is in a jammed state, the flow profile due to this orifice will the same as shown in Fig. 3ce. The occurrence of flow through the unjammed orifice over a time period (which corresponds to the time interval between jamming and unjamming instances of second orifice), causes particle rearrangements in the system and breaks the arch above second orifice leading to flow. For an inter-orifice distance lesser than 10 d , with the mean and r.m.s. velocity magnitudes being of same order it is difficult to clearly isolate the effect of each on the spontaneous unjamming behavior. However, for larger interorifice distances ( > 20 d ), the mean velocity is over an order of magnitude lower than the rms velocity and the fluctuations are expected to dominate the spontaneous unjamming behavior. It is expected that the fluctuations too will decay at infinitely large distances and will not able to re-initiate flow in the jammed orifice. The flow and jamming behavior of the two orifices in that case resemble that from the two isolated orifices unaware of the other's existence. The mean avalanche size then approaches that for a single orifice (dashed lines in Fig. 2b).
Given the observed dependence of the mean avalanche
FIG. 4. Correlation between mean avalanche size per unit orifice width ( s m ) and (a) the r.m.s. velocity and (b) the r.m.s normal stress. All quantities obtained at distance ( w ). Dashed lines show an exponential fit. (c, d) Data in (a) and (b) respectively after normalizing s m with mean avalanche size ( s 0 ) for a hopper with single orifice. The solid line in (c) and (d) shows an exponential fit.
size (see Fig. 2a) and that of the flow variables (see Fig. 3c,d,e) on the inter-orifice distance, we now proceed to determine the relation between these quantities. We consider data only for inter-orifice distance greater than 10 d beyond which the mean flow decreases rapidly compared to fluctuations. Fig. 4a and b shows the avalanche sizes ( s m ) for different inter-orifice distances ( w ) plotted against the rms velocity u w ( ) and rms normal stress T w ( ), respectively. All quantities are evaluated at the same distance ( w or x ) in all the cases. An exponential relation (shown as dashed lines) is observed for all the cases where with the y-intercept approximately equal to the avalanche size ( s 0 ) for a single orifice hopper (i.e. two orifices infinitely far away from each other). The normalized mean avalanche size ( s m /s 0 ) plotted against the normalized r.m.s velocity and normalized r.m.s. normal stress values is shown in Fig. 4c and d, respectively. The rms velocity and normal stress are normalized with their corresponding values at x = 0. The collapse is much better than that observed in Fig. 2b which indicates that the fluctuations do play a role and w is not the only parameter influencing the flow behavior. The solid line shows an exponential fit. For smaller u w /u ( ) (0) and T w /T ( ) (0) ( w > 20 d ), where the primary quantity responsible for unjamming is fluctuations, the data collapse is quite good. However, the scaling shows increasing scatter at increasing values of u w /u ( ) (0) and T w /T ( ) (0) (or smaller w ) which perhaps is indicative of the more complex dependence on the mean flow in addition to fluctuations and the inter-orifice width. Similar scaling behavior for both, r.m.s. velocity and r.m.s. normal stress data, is not quite surprising as both represent average fluctuations in the system and are quite inter-related to each other. Fluctuation of velocity is expected to generate collisions between particles leading to fluctuations in stresses and vice-versa.
In conclusion, our study suggest that the jamming occurrences within a hopper can be altered non-intrusively using multiple orifices by varying the inter-orifice distance. The hopper can be made to flow for prolonged duration for a small enough orifice size by having another orifice of same size at different distances. We observe that the fluctuations arising locally cause rearrangements of particles in the region located far away which leads to spontaneous unjamming. We would also like to note that while the results have been reported for a strictly 2-dimensional system, similar qualitative behavior is observed in simulations of 3-dimensional or quasi3d systems. This mechanism of spontaneous jammingunjamming can be of immense importance for (a) more detailed exploration of jamming characteristics for granular systems studied previously [18-21] and (b) investigating other disordered systems which exhibit jamming, viz. bubbles escaping from an orifice [22] or colloidal hard spheres flowing through a constriction [23]. As a system of practical utility, the multi-orifice hopper can be operated as an efficient mixing device [24] wherein the mixing can be initialized through interaction between different zones of an hopper through either random or controlled closing and opening of orifice.
We thank Ken Kamrin and Chris Rycroft for several stimulating discussions, Mayuresh Kulkarni and Sandesh Kamath for help with some simulations and preliminary experiments and the funding from Department of Science and Technology, India, Grants No. SR/S3/CE/037/2009 and SR/S3/CE/0044/2010.
- ∗ [email protected] † [email protected]
- [1] K. To, P. Y. Lai, and H. K. Pak, Phys. Rev. Lett. 86 , 71 (2001).
- [2] J. Tang and R. P. Behringer, Chaos 21 , 041107 (2011).
- [3] S. Tewari, M. Dichter, and B. Chakraborty, Soft Matter 9 , 5016 (2013).
- [4] I. Zuriguel, L. A. Pugnaloni, A. Garcimartı'n, and D. Maza, Phys. Rev. E 68 , 030301(R) (2003).
- [5] A. Janda, I. Zuriguel, A. Garcimart´ ın, A. Pugnaloni, and D. Maza, Eur. Phys. Lett. 84 , 44002 (2008).
- [6] S. C. Yang and S. S. Hsiau, Powder Technol. 120 , 244 (2001).
- [7] B. Zelinski, E. Goles, and M. Markus, Phys. Fluids 21 , 031701 (2009).
- [8] F. Alonso-Marroquin, S. I. Azeezullah, S. A. GalindoTorres, and L. M. Olsen-Kettle, Phys. Rev. E 85 , 020301(R) (2012).
- [9] I. Zuriguel, A. Janda, A. Garcimart´ ın, C. Lozano,
- R. Ar´ evalo, and D. Maza, Phys. Rev. Lett. 107 , 278001 (2011).
- [10] G. A. Frank and C. O. Dorso, Physica A 390 , 2135 (2011).
- [11] D. Yanagisawa, A. Kimura, A. Tomoeda, R. Nishi, Y. Suma, K. Ohtsuka, and K. Nishinari, Phys. Rev. E 80 , 036110 (2009).
- [12] K. Nichol, A. Zanin, R. Bastien, E. Wandersman, and M. van Hecke, Phys. Rev. Lett. 104 , 078302 (2010).
- [13] K. A. Reddy, Y. Forterre, and O. Pouliquen, Phys. Rev. Lett. 106 , 108301 (2011).
- [14] S. J. Plimpton, J. Comp. Phys. 117 , 1 (1995).
- [15] http://lammps.sandia.gov/.
- [16] C. H. Rycroft, A. V. Orpe, and A. Kudrolli, Phys. Rev. E 80 , 031305 (2009).
- [17] J. W. Landry, G. S. Grest, L. E. Silbert, and S. J. Plimpton, Phys. Rev. E 67 , 041303 (2003).
- [18] D. Bi, J. Zhang, B. Chakraborty, and R. P. Behringer, Nature 480 , 355 (2011).
- [19] A. J. Liu and S. R. Nagel, Soft Mat. 6 , 2869 (2010).
- [20] E. I. Corwin, H. M. Jaeger, and S. R. Nagel, Nature 435 , 1075 (2005).
- [21] T. S. Majmudar and R. P. Behringer, Nature 435 , 1079 (2005).
- [22] Y. Bertho, C. Becco, and N. Vandewalle, Phys. Rev. E 73 , 056309 (2006).
- [23] L. Isa, R. Besseling, E. R. Weeks, and W. C. K. Poon, J. Phys.: Conf. Ser. 40 , 124 (2006).
- [24] S. Kamath, A. Kunte, P. Doshi, and A. V. Orpe, (2014), in preparation. | 10.1103/PhysRevE.90.020201 | [
"Amit Kunte",
"Pankaj Doshi",
"Ashish V. Orpe"
] | 2014-08-02T10:30:49+00:00 | 2014-08-02T10:30:49+00:00 | [
"cond-mat.soft"
] | Spontaneous jamming and unjamming in a hopper with multiple exit orifices | We show that the flow of granular material inside a 2-dimensional flat
bottomed hopper is altered significantly by having more than one exit orifice.
For the hoppers with small orifice widths, inter-mittent flow through one
orifice enables the resumption of flow through the adjacent jammed orifice,
thus displaying a sequence of jamming and unjamming events. Using discrete
element simulations, we show that the total amount of granular material (i.e.
avalanche size) emanating from all the orifices combined can be enhanced by
about an order of magnitude difference by simply adjusting the inter-orifice
distance. The unjamming is driven primarily by fluctuations alone when the
inter-orifice distance is large, but when the orifices are brought close
enough, the fluctuations along with the mean flow cause the flow unjamming. |
1408.0360v1 | ## High-spin torus isomers and their precession motions
T. Ichikawa, 1 K. Matsuyanagi, 1,2 J. A. Maruhn, 3 and N. Itagaki 1
1
Yukawa Institute for Theoretical Physics, Kyoto University, Kyoto 606-8502, Japan 2
RIKEN Nishina Center, Wako 351-0198, Japan
3 Institut fuer Theoretische Physik, Universitaet Frankfurt, D-60438 Frankfurt, Germany
(Dated: August 5, 2014)
Background: In our previous study, we found that an exotic isomer with a torus shape may exist in the high-spin, highly excited states of 40 Ca. The z component of the total angular momentum, Jz = 60 /planckover2pi1 , of this torus isomer is constructed by aligning totally twelve single-particle angular momenta in the direction of the symmetry axis of the density distribution. The torus isomer executes precession motion with the rigid-body moments of inertia about an axis perpendicular to the symmetry axis. The investigation, however, has been focused only on 40 Ca.
Purpose: Wesystematically investigate the existence of exotic torus isomers and their precession motions for a series of N = Z even-even nuclei from 28 Si to 56 Ni. We analyze the microscopic shell structure of the torus isomer and discuss why the torus shape is generated beyond the limit of large oblate deformation.
Method: We use the cranked three-dimensional Hartree-Fock (HF) method with various Skyrme interactions in a systematic search for high-spin torus isomers. We use the three-dimensional time-dependent Hartree-Fock (TDHF) method for describing the precession motion of the torus isomer.
Results: Weobtain high-spin torus isomers in 36 Ar, 40 Ca, 44 Ti, 48 Cr, and 52 Fe. The emergence of the torus isomers is associated with the alignments of single-particle angular momenta, which is the same mechanism as found in 40 Ca. It is found that all the obtained torus isomers execute the precession motion at least two rotational periods. The moment of inertia about a perpendicular axis, which characterizes the precession motion, is found to be close to the classical rigid-body value.
Conclusions: The high-spin torus isomer of 40 Ca is not an exceptional case. Similar torus isomers exist widely in nuclei from 36 Ar to 52 Fe and they execute the precession motion. The torus shape is generated beyond the limit of large oblate deformation by eliminating the 0 s components from all the deformed single-particle wave functions to maximize their mutual overlaps.
PACS numbers: 21.60.Jz, 21.60.Ev, 27.40. + z
## I. INTRODUCTION
Nuclear rotation is a key phenomenon to study the fundamental properties of finite many-body quantum systems. In particular, the rotation about the symmetry ( z ) axis produces a unique quantum object with its density distribution of a torus shape, as shown in our previous studies for 40 Ca [1, 2]. In a classical picture for such rotation the oblate deformation develops with increasing rotational frequency due to the strong centrifugal force [3]. However, such a collective rotation about the symmetry axis is quantum-mechanically forbidden. Instead, it is possible to construct extremely high-spin states by aligning individual angular momenta of single-particle motion in the direction of the symmetry axis [4, 5].
A drastic example is a high-spin torus isomer in 40 Ca [1], where totally twelve single particles with the orbital angular momenta Λ = + 4, + 5, and + 6 align in the direction of the symmetry axis and construct a z component of the total angular momentum of Jz = 60 /planckover2pi1 . Thus, a 'macroscopic' amount of circulating current emerges in the torus isomer state, which may be regarded as a fascinating new form of the nuclear matter suggested by Bohr and Mottelson [6].
Another important kind of rotation is a collective motion that restores the symmetry spontaneously broken in the selfconsistent mean field. The density distribution of the torus isomer largely breaks the symmetry about an ( x or y ) axis perpendicular to the symmetry axis [2]. Below, we call this axis a perpendicular axis. Thus, the torus isomer can rotate about a perpendicular axis, although the collective rotation about the symmetry axis is quantum-mechanically forbidden. This rotational degree of freedom causes the precession motion of the system as a whole. Then, an interesting question arises how such a 'femto-scale magnet' rotates collectively to restore the broken symmetry about a perpendicular axis.
A physical quantity characterizing such a collective rotation is the moment of inertia about a perpendicular axis. It has been theoretically recognized that an independent-particle configuration in a deformed harmonic-oscillator potential rotates with the rigid-body moment of inertia when the selfconsistency between the mean-field potential and the density is fulfilled [7]. However, measured moments of inertia for the case of the precession motion of prolately deformed nuclei are often much smaller than the rigid-body values even when pairing correlations are negligible [8, 9]. This is because of shell e ff ects in highK prolate isomers [8]. Although precession modes of highK oblate isomers have not been observed yet, their moments of inertia would be much reduced from the rigid-body values due to oblate shell structures at small deformations [10].
From these observations, it might be conjectured that the moment of inertia about a perpendicular axis for the torus isomer also significantly deviates from the classical rigidbody value, because the torus isomer is a unique quantum object characterized by the alignment of angular momenta of independent-particlemotions. It is thus surprising that the moment of inertia about a perpendicular axis, evaluated with the time-dependent Hatree-Fock (TDHF) method, from the rotational period of the precession motion of the torus isomer in
40 Ca takes a value close to the classical rigid-body value [2]. We analyzed the microscopic structure of the precession motion by using the random-phase approximation (RPA) method. In the RPA calculation, the precession motion of the torus isomer is generated by a coherent superposition of many oneparticle-one-hole excitations across the sloping Fermi surface. We found that the precession motion obtained by the TDHF calculation is a pure collective motion well decoupled from other collective modes. In our previous studies, however, we focused only on the torus isomer of 40 Ca. It is thus important to investigate whether torus isomers exist also in other nuclei and the properties of the precession motion found there are universal or not.
In this paper, we first perform a systematic investigation of the high-spin torus isomers for a series of N = Z even-even nuclei from 28 Si to 56 Ni. We show that the high-spin torus isomer of 40 Ca is not an exceptional case. About forty years ago, Wong suggested, using a macroscopic-microscopic method, the possible existence of torus isomers at highly excited states of a wide region of nuclei [12]. Quite recently, Staszczak and Wong systematically explored the existence of torus isomers using the constrained cranked Hartree-Fock (HF) method and found some torus isomers at highly excited states in several nuclei [11]. However, they use the harmonic-oscillator basis expansion method, which is insu ffi cient to treat unbound states. It is therefore di ffi cult to examine the stability of the torus isomers against the nucleon emission in their calculation, although some of them would contain single particles in unbound states.
We then perform a systematic TDHF calculation to investigate the properties of the precession motion. For all the highspin torus isomers obtained by the cranked HF calculation, we find the periodic solutions of the TDHF equation of motion, which describe the precession motions. Among them, the precession motion of the 60 /planckover2pi1 torus isomer of 40 Ca is particularly stable and continues for many periods.
To understand the microscopic origin of appearance of the torus isomers, we analyze the process during which the shell structure of the large oblate shape and that of the torus shape grow up from that of the spherical shape. Using the radially displaced harmonic-oscillator (RDHO) model [12] and the oblately deformed harmonic-oscillator potential, we finally discuss why the lowest 0 s components disappear from all the single-particle wave functions of the occupied states and how a large 'hole' region is created in the center of the nucleus to generate the torus shape.
This paper is organized as follows: In Section II, we describe the theoretical framework and parameters of the numerical calculation. In Section III, we present results of the systematic calculation for static and dynamical properties of the high-spin torus isomers including their precession motions. In Section IV, we analyze microscopic shell structures of the torus isomers and discuss the reason why the torus shape emerges beyond the limit of large oblate deformation. Finally, we summarize our studies in Section V.
## II. THEORETICAL FRAMEWORK
## A. Cranked HF calculation
To investigate systematically the existence of high-spin torus isomer states in a wide range of nuclei, we use the cranked three-dimensional Skyrme HF method. To build high-spin states rotating about the symmetry axis of the density distribution ( z axis), we add a Lagrange multiplier, ω , to the HF Hamiltonian, ˆ . H Then, the e ff ective HF Hamiltonian, ˆ H ′ , is written as ˆ H ′ = ˆ H -ω ˆ , Jz where ˆ Jz denotes the z component of the total angular momentum. We minimize this e ff ective HF Hamiltonian with a given Lagrange multiplier, which is equivalent to the cranked HF equation given by δ 〈 ˆ H -ω ˆ Jz 〉 = 0.
For this purpose, we slightly modify the code Sky3d . The details of the code are given in Ref. [13]. In the code, the single-particle wave functions are described on a Cartesian grid with a grid spacing of 1.0 fm, which is a good approximation for not only bound states but also unbound states in contrast to the harmonic-oscillator basis expansion. We take 32 × 32 × 24 grid points for the x , y , and z directions, respectively. This is su ffi ciently accurate to provide converged configurations. The damped-gradient iteration method [14] is used, and all derivatives are calculated with the Fourier transformation method.
In the calculation, we use the SLy6, SkI3, and SkM ∗ Skyrme forces to check the interaction dependence of the calculated results. These e ff ective interactions were well constructed based on nuclear bulk properties but di ff er in details; SLy6 as a fit which includes information on isotopic trends and neutron matter [15], SkI3 as a fit taking into account the relativistic isovector properties of the spin-orbit force [16], and SkM ∗ as a widely used traditional standard [17]. However, except for the e ff ective mass, the bulk properties (equilibrium energy and density, incompressibility, and symmetry energy) are comparable to each other. In the energy density functional, we omit terms depending on the spin density, because it may be necessary to extend the standard form of the Skyrme interaction in order to properly take into account the spin-density dependent e ff ects [18] (see also a review [19]), but such e ff ects are inessential to the torus isomers.
## B. Setting of initial configurations
In the cranked HF calculations, we first search for stable torus configurations in a series of N = Z even-even nuclei from 28 Si to 56 Ni. We use, as an initial configuration of the HF calculation, an α -cluster ring configuration placed on the x y -plane, as shown in Fig. 1 of Ref. [1]. The α -cluster wave function is described by a Gaussian function with the width of 1.8 fm. The center positions of the Gaussian functions are placed equiangularly along a circle with a radius of 6.5 fm on the ( x y , ) plane. Only for the calculations of 52 Fe with the SkM ∗ interaction, we use a radius of 7.55 fm and a width of 1.63 fm. Using these initial configurations, we perform 15,000 HF
iterations. We search for stable torus solutions varying ω from 0.5 to 2.5 MeV / /planckover2pi1 with a step of 0.1 MeV / /planckover2pi1 . After these calculations, we check the convergence of the total energies, the density distributions, and the total angular momenta. In the calculations of the excitation energies, we subtract the expectation value of the center-of-mass motion in both the ground and the torus isomer states.
We next calculate all single-particle states including those above the Fermi energy. To calculate those, we use, as initial wave functions of the HF calculation, the single-particle wave functions of the RDHO model [12]. This model is a good approximation to the mean-field of torus-shaped isomers. In this model, the single-particle potential is given by
$$V ( r, z ) = \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } z ^ { 2 }, \quad \ \ ( 1 ) \quad \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
V ( r, z ) = \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 3 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 3 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 3 } \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega _ { 0 } ^ { 2 } ( r - R _ { 0 } ) ) ^ { 2 } + \frac { 1 } { 2 } m \omega \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \. \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \$$
where m denotes the nucleon mass, ω 0 the oscillator frequency, r and z the radial and the z components of the cylindrical coordinate system, and R 0 the radius parameter of the torus shape. Since the radial wave function of the lowest energy in the RDHO model is described by a shifted Gaussian function with the width d = √ /planckover2pi1 / m ω 0 , we determine ω 0 from the radius of a cross section of a torus ring. The optimal values of R 0 and d are determined through the global investigation mentioned above. Using this initial condition and a value of ω 0 obtained by the global investigation, we perform the HF iteration over 20000 times and calculate the single-particle states up to the 40th for both protons and neutrons.
## C. Sloping Fermi surface
It is important to note that the cranking term -ω ˆ Jz does not change the single-particle wave functions for rotation about the symmetry axis. Thus, it is useful to introduce the concept of 'sloping' Fermi surface. As usual, the single-particle Hamiltonian is given by ˆ H ′ = ∑ i ( ˆ hi -ω ˆ j z ( ) i ), where ˆ hi and ˆ j z ( ) i denote the mean-field Hamiltonian and the z component of the total angular momentum for each single particle, respectively. The eigenvalue of ˆ H ′ is written as E ′ = ∑ i [( ei -λ ) -/planckover2pi1 ω Ω i ], where λ denotes the Fermi energy at ω = 0. The symbols ei and Ω i denote the single-particle energy and the eigenvalue of ˆ j z ( ) i , respectively. By introducing the sloping Fermi surface defined by λ ′ ( Ω ) = λ + /planckover2pi1 ω Ω , we can rewrite E ′ as E ′ = ∑ i { ei -λ ′ ( Ω i ) . } Therefore, aligned configurations can be easily constructed by plotting the single-particle energies as a function of Ω and tilting the Fermi surface in the ( e , Ω ) plane. It is important to note that the value of ω to specify an aligned configuration is not unique. As we can immediately see in Figs. 3-7 below, individual configurations do not change for a finite range of ω .
## D. Optimally aligned torus configurations
Let us focus on optimally aligned torus configurations where all the single-particle states below the sloping Fermi surface are occupied. They are expected to be more stable than other aligned configurations involving particle-hole excitations across the sloping Fermi surface. Before carrying out the cranked HF calculations, we can easily presume candidates of optimally aligned torus configurations. Since the e ff ects of the spin-orbit potential are negligibly weak in the torus configurations, not only Ω but also the z component of the orbital angular momentum, Λ , are good quantum numbers ( Ω = Λ + Σ , where Σ denotes the z component of the spin, ± 1 2) [1]. Single-particle states having the same / Λ value with di ff erent spin directions are approximately degenerated and simultaneously occupied . Thus, the lowest-energy configurations for the torus shapes at ω = 0 are Λ = 0, ± 1, ± 2, and ± 3 for 28 Si, Λ = 0, ± 1, ± 2, ± 3, and + 4 or -4 for 32 S, Λ = 0, ± 1, · · · , ± 4 for 36 Ar, Λ = 0, ± 1, · · · , ± 4, and + 5 or -5 for 40 Ca, Λ = 0, ± 1, · · · , ± 5 for 44 Ti, Λ = 0, ± 1, · · · , ± 5, and + 6 or -6 for 48 Cr, Λ = 0, ± 1, · · · , ± 6 for 52 Fe, and Λ = 0, ± 1, · · · , ± 6, and + 7 or -7 for 56 Ni.
For instance, in 40 Ca, possible aligned configurations at ω /nequal 0 are (i) Λ = 0, ± 1, · · · , ± 4, and + 5 for Jz = 20 /planckover2pi1 [ = 5 /planckover2pi1 × 2 (spin degeneracy) × 2 (isospin degeneracy)], (ii) Λ = 0, ± 1, ± 2, ± 3, + 4, and + 5 for Jz = 60 /planckover2pi1 [ = 15 /planckover2pi1 × 2 × 2], and (iii) Λ = 0, ± 1, ± 2, + 3, + 4, + 5, + 6, and + 7 for Jz = 100 /planckover2pi1 [ = 25 /planckover2pi1 × 2 × 2]. However, we could not obtain stable HF solutions for the configurations (i) and (iii): the centrifugal force is insu ffi cient for stabilizing the configuration (i), while the last occupied single-particle state with Λ = 7 is unbound for the configuration (iii). Indeed, we confirmed that the torus isomer configuration (iii) with Jz = 100 /planckover2pi1 slowly decays. In the systematic calculations, it is often di ffi cult to discuss the stability of torus isomers when such unbound states are included. To avoid this di ffi culty, in this paper, we focus on torus configurations without involving unbound single-particle states.
## E. TDHF calculation for the precession motion
For the stable torus isomers obtained above, we performed TDHF calculations to investigate their precession motions. The time evolution of the density distribution is determined by solving the TDHF equation of motion i /planckover2pi1 ˙ ρ = [ ˆ H , ρ ]. When an impulsive force is provided in a direction perpendicular to the symmetry axis at t = 0, the torus isomer starts to execute the precession motion. This precession motion is associated with a rotation about a perpendicular axis, i.e., an axis perpendicular to the symmetry axis. In Ref. [2], we already showed that this precession motion is a pure collective motion to restore the broken symmetry and well described as coherent superpositions of many 1p-1h excitations across the sloping Fermi surface. We investigate whether other torus isomers also execute the precession motion well decoupled from other collective modes and whether their moments of inertia are close to the rigid-body values or not. In this way, we can also check the stability of the obtained torus isomers against given impulsive forces.
Figure 1 illustrates the schematic picture of the precession motion. At t = 0, the torus isomer is placed on the x y -plane with the angular momentum K ( = Jz ) along the z axis in the
FIG. 1. Schematic picture for the precession motion of torus isomers taken from Ref. [2]. The bold solid line denotes the symmetry axis of the density distribution. The dashed line denotes the precession axis. The symbols θ and φ denote the tilting and the rotational angles, respectively.
laboratory frame. When an impulsive force is provided in the negative x direction (the dotted line) at t = 0, the total angular momentum becomes /vector I (the dashed line). We call this vector the precession axis. After that, the symmetry axis of the density distribution in the body-fixed frame (the bold solid line) starts to rotate about the precession axis with the rotational angle φ . In the precession motion, the value K is conserved and its direction is identical to the bold solid line. The tilting angle θ is defined as the angle between the bold solid and the dashed lines. Then, the moment of inertia for the rotation about a perpendicular axis, T ⊥ , can be estimated by T ⊥ = I /ω prec , where ω prec denotes the rotational frequency of the precession motion. To build the first excited state of the precession motion, we provide an impulsive force such that the total angular momentum becomes I = K + 1.
To solve the TDHF equation, we use the code Sky3d and take the Taylor expansion of the time-development operator up to the 12th order. The setups of spatial grid points and interactions are the same as those of the cranked HF calculations described above. We start to perform calculations from the initial density distribution obtained by the cranked HF calculations. The time step of the TDHF calculations is 0.2 fm / c . We calculate the time evolution until 3000 fm / c . To excite the precession motion, we provide an impulsive force at t = 0 by the external potential given by V ext ( r , ϕ, z ) = V z 0 cos ϕ exp[ -( r -R 0 ) 2 / d 2 ]. This impulsive force gives an angular momentum in the negative x direction at t = 0. The parameter V 0 is chosen such that the total angular momentum becomes I = K + 1.
## III. RESULTS OF CALCULATION
## A. Static properties
We have carried out a systematic search for stable torus isomers for the N = Z even-even nuclei from 28 Si to 56 Ni. The result of the calculation is summarized in Table I. We obtain the stable torus isomers in 36 Ar for Jz = 36 /planckover2pi1 , 40 Ca for Jz = 60
TABLE I. Stable torus isomers obtained in the cranked HF calculation with various Skyrme interactions. The excitation energy, E ex , is measured from the ground state. The calculated density distributions are fitted to the Gaussian function ρ ( r z , ) = ρ 0 e -[ ( r -R 0 ) 2 + z 2 ] / d 2 and the resulting values of the parameters, ρ 0 , R 0 , and d , are listed. The symbols, T rid ⊥ and T rid ‖ , denote the rigid-body moments of inertia for the rotations about a perpendicular and the symmetry axes, respectively.
| System | J z ( /planckover2pi1 ) | E ex (MeV) | ρ 0 (fm - 3 ) | R 0 (fm) | d (fm) | T rid ⊥ ( /planckover2pi1 2 / MeV) | T rid ‖ ( /planckover2pi1 2 / MeV) |
|----------|---------------------------|--------------|-----------------|------------|----------|--------------------------------------|--------------------------------------|
| (SLy6) | | | | | | | |
| 36 Ar | 36 | 123.89 | 0.137 | 5.12 | 1.62 | 14.3 | 26.4 |
| 40 Ca | 60 | 169.71 | 0.129 | 6.07 | 1.61 | 21.0 | 39.6 |
| 44 Ti | 44 | 151.57 | 0.137 | 6.30 | 1.61 | 24.6 | 46.5 |
| 48 Cr | 72 | 191.25 | 0.132 | 7.19 | 1.60 | 33.8 | 64.7 |
| 52 Fe | 52 | 183.70 | 0.138 | 7.47 | 1.60 | 39.1 | 75.1 |
| (SkI3) | | | | | | | |
| 36 Ar | 36 | 125.15 | 0.146 | 5.01 | 1.58 | 13.7 | 25.3 |
| 40 Ca | 60 | 173.52 | 0.138 | 5.90 | 1.58 | 19.9 | 37.5 |
| 44 Ti | 44 | 153.02 | 0.146 | 6.17 | 1.58 | 23.6 | 44.6 |
| 48 Cr | 72 | 193.66 | 0.141 | 7.00 | 1.57 | 32.0 | 61.3 |
| 52 Fe | 52 | 183.70 | 0.147 | 7.31 | 1.57 | 37.5 | 71.9 |
| (SkM ∗ ) | | | | | | | |
| 36 Ar | 36 | 124.80 | 0.131 | 5.16 | 1.65 | 14.6 | 26.9 |
| 40 Ca | 60 | 167.84 | 0.122 | 6.17 | 1.64 | 21.8 | 41.0 |
| 44 Ti | 44 | 152.20 | 0.131 | 6.36 | 1.64 | 25.1 | 47.5 |
| 48 Cr | 72 | 192.40 | 0.125 | 7.30 | 1.63 | 34.9 | 66.7 |
| 52 Fe | 52 | 187.08 | 0.132 | 7.55 | 1.63 | 40.0 | 76.7 |
/planckover2pi1 , 44 Ti for Jz = 44 /planckover2pi1 , 48 Cr for Jz = 72 /planckover2pi1 , and 52 Fe for Jz = 52 /planckover2pi1 with all the three Skyrme interactions used in this study. On the other hand, we have not found any stable torus isomer in 28 Si, 32 S, and 56 Ni. In Fig. 3, we plot the total density distributions of the torus isomers obtained in the cranked HF calculation with the SLy6 interaction.
The total density distribution of each of the torus isomers obtained in the cranked HF calculation is well fitted by the Gaussian function ρ ( r z , ) = ρ 0 e -[ ( r -R 0 ) 2 + z 2 ] / d 2 , where ρ 0 , R 0 , and d denote the maximum value of the nucleon density, the radius of the torus ring, and the width of a cross section of the torus ring, respectively. The resulting values of the parameters, ρ 0 , R 0 , and d , are tabulated in the middle part of Table I. We see that the values ρ 0 and d are almost constant for all the torus isomers. The interaction dependence of these values is weak. It is interesting that, in all the results, ρ 0 is smaller than the saturation nuclear density ( ρ sat ∼ 0 17 fm . -3 ) and d is close to the width of an alpha particle used in Brink's α -cluster model ( d α ∼ 1.46 fm) [20].
Using the total density distribution, we also calculate the rigid-body moments of inertia for rotation about a perpendicular axis, T rid ⊥ , and the symmetry axis, T rid ‖ . The results are also shown in Table I. Later, we shall compare these values for T rid ⊥ with those obtained by an analysis of the precession motions in the TDHF time evolution.
To investigate microscopic structures of the torus isomers,
FIG. 2. (Color online) Density distributions on the z = 0 plane of the obtained stable torus isomers. The contours correspond to multiple steps of 0.015 fm -3 . The color is normalized by the largest density in each plot.
we plot in Figs. 3-7 neutron single-particle energies versus Ω for each torus isomer calculated with the SLy6 interaction. In the figures, the solid and the open circles denote the positiveand negative-parity states, respectively. The gray area in each plot denotes the occupied states. In each plot, we see that the single-particle energies with the same Λ are almost degenerate. This indicates that the e ff ects of the spin-orbit force are negligibly small and Λ is approximately a good quantum number in all the torus isomers. One may also notice that the Kramer's degeneracy for a pair of single-particle states with ± Λ is lifted. This is due to the time-odd components (dependent on the current density) of the cranked HF mean fields associated with the macroscopic currents, which are produced by the alignment of the single-particle angular momenta with large values of Λ .
Because of the negligible spin-orbit splittings, the spinorbit partners are always occupied simultaneously. Therefore, the Jz values of the optimally aligned configurations are easily determined by summing up the Λ values of the occupied single-particle states: they are Λ = 0 , ± 1 , ± 2 , ± 3 , + 4 , + 5[ Jz = 9 /planckover2pi1 × 2 (spin degeneracy) × 2 (isospin degeneracy) = 36 /planckover2pi1 ] for
FIG. 3. Single-particle energies versus the z component of the total angular momentum, Ω , for 36 Ar. Solid and open circles denote the single-particle energies of the positive- and negative-parity states, respectively. To illustrate the degeneracy of positive- and negativeparity states, some negative-parity states are shown by double open circles.
FIG. 4. Single-particle energies versus Ω for 40 Ca. All symbols are the same as in Fig. 3.
36 Ar, Λ = 0 , ± 1 , ± 2 , ± 3 , + 4 , + 5 , + 6 [ Jz = 15 /planckover2pi1 × 2 × 2 = 60 /planckover2pi1 ] for 40 Ca, Λ = 0 , ± 1 , · · · , ± 4 , + 5 , + 6 [ Jz = 11 /planckover2pi1 × 2 × 2 = 44 /planckover2pi1 ] for 44 Ti, Λ = 0 , ± 1 , · · · , ± 4 , + 5 , + 6 , + 7 [ Jz = 18 /planckover2pi1 × × 2 2 = 72 /planckover2pi1 ] for 48 Cr, and Λ = 0 , ± 1 , · · · , ± 5 , + 6 , + 7 [ Jz = 13 /planckover2pi1 × × 2 2 = 52 /planckover2pi1 ] for 52 Fe.
We can estimate from the figures a region of ω for which each of the torus isomers stably exists. This is done by determining the steepest and the most gradual slopes of the Fermi surface for which the occupied single-particle configuration remains the same. The results are plotted in Fig. 8. The solid, dashed, and dotted lines denote the regions of ω for each of the stable torus isomers obtained with the SLy6, SkI3, and SkM ∗ interactions, respectively. We see that the result does not strongly depend on the Skyrme interaction employed, although the width is weakly dependent on it.
FIG. 5. Single-particle energies versus Ω for 44 Ti. All symbols are the same as in Fig. 3.
FIG. 6. Single-particle energies versus Ω for 48 Cr. All symbols are the same as in Fig. 3.
## B. Dynamic properties
We carried out a systematic TDHF calculation for each of the torus isomers and found that that the TDHF time evolution of the density distribution is quite similar to that displayed in Fig. 2 of [2]. Figure 9 shows the calculated time evolution of the precession motion for each of the torus isomers obtained with the SLy6 interaction. In each plot in Fig. 9, panels (a), (b), and (c) denote the total angular momentum, I , the tilting angle, θ , and the rotational angle, φ , respectively. In panel (a) in each plot, we can see that the total angular momentum is conserved very well. This indicates that the TDHF calculations are su ffi ciently accurate. We find that the precession motion of the 40 Ca torus isomer is especially stable [see panel (b) in each plot], where the rotational angle φ lineally increases with time, indicating that the rotation of the symmetry axis about the precession axis keeps a constant velocity through all the periods. This indicates that the strong shell e ff ects responsible for the appearance of the torus isomer in 40 Ca also
FIG. 7. Single-particle energies iversus Ω for 52 Fe. All symbols are the same as in Fig. 3.
FIG. 8. Regions of ω for which each of the torus isomers stably exists. The solid, dashed, and dotted lines denote the results calculated with the SLy6, SkI3, and SkM ∗ interactions, respectively
stabilize the precession motion. We find that the precession motions emerge also for other torus isomers and they are stable at least for two periods. After that, however, the tilting angle gradually starts to fluctuate. Correspondingly, the rotational angle φ also starts to deviate from the linear time evolution [see panel (c) in each plot]. We have also carried out similar TDHF calculations with the use of the SkI3 and SkM ∗ interactions. The results are similar to those shown above for the SLy6 interaction, which implies that the properties of the precession motion are robust and depend on the choice of the Skyrme interaction only weakly.
To evaluate the moment of inertia for the rotation about a perpendicular axis, we take the average of the two periods starting from t = 0 during which the precession motion is especially stable. The results are tabulated in the third column of Table II. Using these values, we calculate the frequency of the precession motion by ω prec = 2 π/ T prec and the moment of inertia for the rotation about a perpendicular axis by T TDHF ⊥ = I /ω prec . The results are tabulated in the forth and fifth columns
FIG. 9. Time evolution of the precession motion for each of the torus isomers from 36 Ar to 52 Fe calculated by solving the TDHF equation of motion for the SLy6 interaction. In each plot, panels (a), (b), and (c) denote the total angular momentum I , the tilting angle θ , and the rotational angule φ , respectively.
of Table II. The obtained moments of inertia are very close to the rigid-body values tabulated in Table I for all the Skyrme interactions employed. As discussed in Ref. [2], these results indicate that the precession motions under consideration are pure collective motions generated by coherent superpositions of many 1p-1h excitations across the sloping Fermi surface.
## IV. DISCUSSION
## A. Radial density distributions of individual single-particle states
Let us examine the radial density distributions of individual single-particle wave functions on the z = 0 plane for the torus isomer of 40 Ca. For this purpose, we interpolate the density distributions described with a Cartesian coordinate in the cranked HF calculations by means of a third-order B-spline function. After that, we transform those to a cylindrical coordinate representation [ ρ i ( x y , ) → ρ i ( r , ϕ )]. We then integrate
TABLE II. Results of the TDHF calculation for the precession motions of the torus isomers from 36 Ar to 52 Fe. The symbol I denotes the resulting total angular momentum after an impulsive force is provided. The symbol T prec denotes the average over the two periods from t = 0 for the precession motion. The symbol ω prec denotes the precession frequency estimated by ω prec = 2 π/ T prec . The symbol T TDHF ⊥ denotes the moment of inertia for the rotation about a perpendicular axis estimated by T TDHF ⊥ = I /ω prec .
| System | I ( /planckover2pi1 ) | T prec (MeV / /planckover2pi1 ) | ω prec (MeV) | T TDHF ⊥ ( /planckover2pi1 2 / MeV) |
|----------|-------------------------|-----------------------------------|----------------|---------------------------------------|
| (SLy6) | | | | |
| 36 Ar | 37 | 450.1 | 2.75 | 13.5 |
| 40 Ca | 61 | 402.5 | 3.08 | 19.8 |
| 44 Ti | 45 | 651.0 | 1.90 | 23.7 |
| 48 Cr | 73 | 554.5 | 2.24 | 32.6 |
| 52 Fe | 53 | 872.8 | 1.42 | 37.3 |
| (SkI3) | | | | |
| 36 Ar | 37 | 427.9 | 2.90 | 12.8 |
| 40 Ca | 61 | 378.6 | 3.28 | 18.6 |
| 44 Ti | 45 | 624.4 | 1.99 | 22.7 |
| 48 Cr | 73 | 524.5 | 2.36 | 30.9 |
| 52 Fe | 53 | 839.0 | 1.48 | 35.9 |
| (SkM ∗ ) | | | | |
| 36 Ar | 37 | 464.2 | 2.67 | 13.9 |
| 40 Ca | 61 | 418.2 | 2.96 | 20.6 |
| 44 Ti | 45 | 666.1 | 1.86 | 24.2 |
| 48 Cr | 73 | 572.8 | 2.16 | 33.7 |
| 52 Fe | 53 | 894.8 | 1.39 | 38.3 |
ρ i ( r , ϕ ) in the ϕ direction and obtain ρ i ( r ). The calculated results are plotted in the upper panel of Fig. 10.
As shown in Refs. [1, 2], the RDHO model can describe well the microscopic structures of torus isomers. To illustrate this, we solve the Schr¨ odinger equation with the RDHO potential, Eq. (1), by means of the deformed harmonic-oscillator basis expansion and calculate ρ i ( r ). In the calculation, we take R 0 = 6 07 fm and . d = 1 61 fm for the RDHO model and . the same aligned single-particle configuration as that obtained by the cranked HF calculation for 40 Ca. The obtained radial density distributions of the individual single-particle states are plotted in the lower panel of Fig. 10. Using these density distributions, we calculate the rigid-body moments of inertia about a perpendicular axis and the symmetry axis: they are T RDHO ⊥ = 21 3 . /planckover2pi1 2 / MeV and T RDHO ‖ = 40 2 . /planckover2pi1 2 / MeV, respectively. These values are in good agreement with those obtained by the cranked HF calculation.
In Fig. 10, it is clearly seen that the radial density distributions of the individual single-particle states in the RDHO model are quite similar to those obtained by the cranked HF calculations. In particular, the peak positions of each radial density distribution are in good agreement between the two calculations. As a matter of fact, the peak position of each density distribution shifts to a larger r with increasing orbital angular momentum. Looking into details of the density distributions obtained by the cranked HF calculations, one may
FIG. 10. Radial density distributions of individual single-particle states on the z = 0 plane for 40 Ca obtained by the cranked HF calculations (upper panel) and the RDHO model (lower pannel). The densities for the φ direction in cylindrical coordinates are integrated.
notice that some radial density distributions with high angular momentum slightly shift due to the spin-orbit potential. In Fig. 4, the degeneracy of single-particle energies with the same high Λ is indeed slightly broken for the spin-orbit partner with Ω π = 9 2 / -and 11 2 / -( Λ = 5) and that with Ω π = 11 2 / + and 13 2 / + ( Λ = 6). These spin-orbit e ff ects are absent in the RDHO model.
## B. Shell structure of torus nucleus
Using the RDHO model, we next investigate shell structures of a torus isomer and examine how single-particle configurations change from spherical to torus shapes. Figure 11 shows a Nilsson diagram versus the parameter η = R 0 / d for 40 Ca. At η = 0, the nuclear shape is spherical. At η = 4, a torus shape is well developed, which is a size similar to that obtained by the cranked HF calculations. Note that we take into account volume conservation inside an equi-potential surface of a torus isomer (see Ref. [12] and Appendix for the volume conservation in 0 ≤ η ≤ 1). To eliminate the volume e ff ect, we plot the single-particle energies in unit of /planckover2pi1 ω 0 ( η ). In this figure, we slightly shift the single-particle energies with higher Ω in order to illustrate the degeneracy of the states.
FIG. 11. Nilsson diagram versus η = R 0 / d of the RDHO model for 40 Ca. The solid and dashed lines denote the single-particle states with positive- and negative-parities, respectively. The single-particle energies are plotted in units of /planckover2pi1 ω 0 ( η ). To illustrate the degeneracy of the levels, the single-particle energies with higher Ω are slightly shifted.
In Fig. 11, we see the spherical major shell with E = /planckover2pi1 ω 0 ( N sh + 3 2) at / η = 0, where N sh denotes the total number of oscillator quanta. With increasing η , the single-particle energies with Ω = 1 2 approach the asymptotic value given / by E = /planckover2pi1 ω 0 ( N ′ sh + 1), where N ′ sh = nr + nz , nz and nr denote the quantum number for oscillations in the z and the radial directions, respectively. The energies of other single-particle states with larger Ω in the same N sh shell steeply decrease as a function of η . At η = 4, the 10th and 11th (from the bottom) single-particle states with Ω = p 11 2 / -and 13 / 2 -( Λ = 6) become lower than the 14th level with N ′ sh = 1 (the 1 /planckover2pi1 ω 0 state). These two single-particle states originate from those with a spherical harmonic-oscillator quantum number of N sh = 5 (the 5 /planckover2pi1 ω 0 state).
It is easy to understand these behaviors. As Wong showed in Ref. [12], the single-particle energies for large R 0 are approximately given by E ∼ /planckover2pi1 ω 0 ( N ′ sh + 1) + /planckover2pi1 2 Λ 2 / 2 mR 2 0 . Thus, the single-particle energies belonging to the same N ′ sh shell are proportional to Λ 2 at lager R 0 .
Figure 12 shows the single-particle energies in the RDHO model versus Ω from η = 0 to 4. At η = 0 [Fig. 12 (a)], the familiar shell structure of the spherical harmonic-oscillator is seen. With increasing η [Figs. 12 (b)-(d)], single-particle energies with high Ω rapidly decrease. Then, the single-particle energies start to form parabolic structures. At η = 4 [Fig. 12 (e)], two important properties emerge: (i) the curvature of the parabolic structure becomes large, and (ii) the single-particle energies within the same N ′ sh shell are proportional to Λ 2 . These two properties play an essential role in stabilizing the torus isomers when single particles are aligned in the direction of the symmetry axis.
It is surprising that the single-particle shell structure of the RDHOmodelat η = 4 is very similar to that of Fig. 5 obtained by the cranked HF calculation. The RDHO model is therefore a good approximation for describing the microscopic structures of the torus isomers.
## C. Emergence of the torus shape beyond the limit of oblate deformation
Lastly, let us discuss the reason why the torus nucleus emerges beyond the limit of large oblate deformation. In the spherical harmonic-oscillator potential, the radial wave function of the lowest single-particle state is given by a Gaussian function peaked at the center (the 0 s state): accordingly, the central part of the total density distribution is quite stable. Then, a question arises why such a stable and robust wave function vanishes and how the torus shape emerges.
To investigate why the 0 s state disappears, we calculate the single-particle energies for the deformed harmonic-oscillator potential as a function of oblate deformation. Figure 13 shows the obtained Nilsson diagram versus the aspect ratio of the short (the z direction) to the long (the radial direction) axes for an ellipsoidal nuclear surface (oblate deformation). The aspect ratio 1:1 corresponds to the spherical shape. The aspect ratio 1:5 corresponds to an oblate shape with the same aspect ratio as that of the torus isomer of 40 Ca obtained by the cranked HF calculation. The single-particle energies are plotted in unit of /planckover2pi1 ω 0 ( /epsilon1 ), where ω 0 ( /epsilon1 ) denotes the frequency of the harmonic-oscillator potential depending on the Nilsson perturbed-spheroid parameter /epsilon1 to describe ellipsoidal nuclear shapes. In ω 0 ( /epsilon1 ), the volume conservation inside an equipotential surface is taken into account [21]. In the figure, we see that some single-particle energies associated with high N sh spherical major shells rapidly decrease with increasing oblate deformation. At the aspect ratio 1:5, the last occupied state for 40 Ca ( N = 20) originates from that with a spherical harmonicoscillator quantum number of N sh = 3 (the 3 /planckover2pi1 ω 0 state).
In Figs. 14, the single-particle energies are plotted versus Ω at each aspect ratio. We see that the shell gaps of the singleparticle energies decrease with increasing oblate deformation. However, the basic pattern of deformed shell structure does not change, in contrast to that of the RDHO model shown in Fig. 12. In Fig. 14 (e), the dashed line denotes the Fermi level for N = 20 at ω = 0. The neutron density distribution, ρ ( r z , ), for the occupied configuration is shown in Fig. 15 (a). The densities in the ϕ direction are integrated. Two prominent peaks are seen in the density distribution. We next consider an aligned single-particle configuration at ω = 1 6 MeV . / /planckover2pi1 . This ω corresponds to a value for the torus isomer of 40 Ca. The occupied states at this ω are shown by the gray area in Fig. 14 (e). By the alignments, totally five single-particle states with 11 2 / -[505] and 9 / 2 [505] ( -Λ = 5), 9 2 [404] / + and 7 / 2 [404] ( + Λ = 4), and 5 / 2 [402] ( + Λ = 2) are occupied
FIG. 12. Single-particle energies of the RDHO model versus Ω at various values of η = R 0 / d . The solid and open circles denote the singleparticle states with positive and negative parities, respectively.
(the asymptotic Nilsson label Ω π [ Nnz Λ ] is used here). On the other hand, the single particle states with -7 2 [303] and / --5 2 / -[303] ( Λ = 3), -5 2 [202] ( / + Λ = 2), and -3 2 / -[301]
and -1 2 [301] ( / -Λ = 1) become unoccupied. Summing up the aligned single-particle angular momenta, we obtain the neutron contribution to the z component of the total angular momentum Jz = 31 /planckover2pi1 . Taking into account the proton contribution as well, we finally obtain the total angular momentum Jz = 62 /planckover2pi1 for this oblate configuration, which is close to that of the torus isomer for 40 Ca obtained by the cranked HF calculation. The neutron density distribution at ω = 1 6 MeV . / /planckover2pi1 is shown in Fig. 15 (b). The two peaks seen in Fig. 15 (a) vanish and densities in the central region become flat and stretch to radial direction, as the single-particle states with high Ω are occupied.
FIG. 13. Nilsson diagram versus the aspect ratio of the short (the z direction) to long (the radial direction) axes for an ellipsoidal nuclear surface (oblate deformations). The solid and dashed lines denote the single-particle states with positive and negative parities, respectively. The aspect ratio 1:1 corresponds to the spherical shape. The aspect ratio 1:5 is close to that of a torus isomer obtained by the cranked HF calculation.
Figure 16 shows the density distributions of individual single-particle states of special interest at aspect ratio 1:5. The densities for the ϕ direction in the cylindrical coordinate are integrated. The dashed line shows the density distribution of the lowest Λ = 0 state. On the other hand, the solid lines depict those of the aligned Λ = 2, 4, and 5 states mentioned above that are occupied at ω = 1 6 MeV . / /planckover2pi1 . The single-particle density distributions of these aligned states peak around r = 6 fm. Apparently, the overlap between the aligned nucleons and the nucleons in the lowest Λ = 0 state is very small. Namely, the lowest Λ = 0 state largely containing the spherical 0 s component is rather isolated from the others. To gain the attractive interactions between nucleons, the total system tends to maximize the overlaps between the density distributions of individual single particles. Thus, it would be energetically favorable to concentrate the densities of individual nucleons around r = 6 fm. In this way, the nucleus with extremely large oblate deformation may start to generate the torus shape. This seems to be the basic reason why a large 'hole' is created in the central region of the nucleus by eliminating the spherical 0 s wave function.
FIG. 14. Single-particle energies of the deformed harmonic-oscillator versus Ω at various oblate deformations. In panel (e), the dashed line denotes the Fermi surface for 40 Ca ( N = 20) at ω = 0. The gray area denotes the occupied states in 40 Ca at ω = 1 6 MeV . / /planckover2pi1 for rotation about the symmetry axis.
FIG. 15. (a) Density distributions of neutrons in 40 Ca calculated for the deformed harmonic-oscillator model at an oblate deformation of the aspect ratio 1:5 with ω = 0. The contours correspond to multiple steps of 0.05 fm -2 . The densities for the φ direction in the cylindrical coordinate are integrated. (b) The same as (a) but with ω = 1.6 MeV / /planckover2pi1 . The colors are normalized by the largest density of panel (a).
## V. SUMMARY
We have systematically investigated the existence of highspin torus isomers for a series of N = Z even-even nuclei from 28 Si to 56 Ni using the cranked HF method. We found the stable torus isomers from 36 Ar to 52 Fe for all the Skyrme interactions used in this study. In the obtained torus isomers, the z components of the total angular momentum are Jz = 36 /planckover2pi1 for 36 Ar, 60 /planckover2pi1 for 40 Ca, 44 /planckover2pi1 for 44 Ti, 72 /planckover2pi1 for 48 Cr, and 52 /planckover2pi1 for 52 Fe. We fitted the density distribution of each of the obtained torus isomers with the Gaussian function. We
FIG. 16. Density distributions of the single-particle states of special interest for 40 Ca on the z = 0 plane in an oblate deformation of the aspect ratio 1:5 calculated with the deformed harmonic-oscillator model. The densities in the φ direction in the cylindrical coordinate are integrated. The solid lines show the density distributions of the aligned single-particle states with Λ = 2, 4, and 5 that are occupied at ω = 1 6 MeV . / /planckover2pi1 . The dashed line shows the density distribution of the lowest Λ = 0 state.
also analyzed the microscopic structure of the obtained torus isomers by plotting the single-particle energies versus Ω and using the concept of sloping Fermi surface. We determined the regions of ω for which the obtained torus isomers can stably exist in each Skyrme interaction. The dependence of the obtained results on the Skyrme interactions employed is found to be weak.
We have also performed TDHF calculations to explore the
properties of the precession motion rotating with angular momentum I = K + 1, which is built on each of the obtained torus isomer with I = K . For all the obtained torus isomers, the precession motion emerges and the symmetry axis rotates about the precession axis for at least two periods. It was found that the precession motion of the 60 /planckover2pi1 isomer in 40 Ca is especially robust and stably rotates for many periods, We obtained similar results for all the Skyrme interactions used, We also estimated the moment of inertia for the rotation about a perpendicular axis from the calculated rotational periods of the precession motion. The obtained moments of inertia are close to the rigid-body values for all the obtained torus isomers.
We have discussed the radial density distribution of each single-particle wave function in the high-spin torus isomer of 40 Ca. We showed that the density distributions are well approximated by those of the RDHO model. We then discussed how the shell structure develops from spherical to torus shapes. There are two important mechanisms for stabilizing torus isomers: (i) the development of the major shells consisting of single-particle states whose energies are given by E = ( N ′ sh + 1) + /planckover2pi1 2 Λ 2 / 2 mR 2 0 , where N ′ sh = nr + nz and Λ is the z component of orbital angular momentum. (ii) a large value of R 0 that reduces the energies of high Ω single-particle states. We finally discussed why the 0 s components of all the single-particle wave functions vanish and generate a torus shape. We showed that in an aligned single-particle configuration with extremely large oblate deformation, the overlaps between the density distributions of the lowest Λ = 0 singleparticle state and the aligned highΩ single-particle states become very small due to the strong centrifugal force. To gain the attractive interaction energy as much as possible, nucleons tend to maximize the overlaps of their wave functions. An optimal configuration beyond the limit of large oblate de-
- [1] T. Ichikawa, J. A. Maruhn, N. Itagaki, K. Matsuyanagi, P.-G. Reinhard, and S. Ohkubo, Phys. Rev. Lett. 109 , 232503 (2012).
- [2] T. Ichikawa, K. Matsuyanagi, J. A. Maruhn, and N. Itagaki, Phys. Rev. C 89 , 011305(R), (2014).
- [3] S. Cohen, F. Plasil and W.J. Swia ¸tecki, ´ Ann, Phys. 82 , 557 (1974).
- [4] A. Bohr, in Elementary Modes of Excitation in Nuclei, Proceedings of the International School of Physics 'Enrico Fermi', Course LXIX , edited by A. Bohr and R. A. Broglia (NorthHolland, Amsterdam, 1977), p. 3.
- [5] A.V. Afanasjev, D.B. Fossan, G.J. Lane, and I. Ragnarsson, Phys. Rep. 322 , 1 (1999).
- [6] A. Bohr and B.R. Mottelson, Nucl. Phys. A354 , 303c (1981).
- [7] A. Bohr and B.R. Mottelson, Nuclear Structure , Vol. II (World Scientific, Singapore, 1998).
- [8] M. A. Deleplanque , S. Frauendorf, V. V. Pashkevich, S. Y . Chu, and A. Unzhakova, Phys. Rev. C 69 , 044309 (2004).
- [9] Y. R. Shimizu, M. Matsuzaki, and K. Matsuyanagi, Phys. Rev. C 72 , 014306 (2005).
- [10] C. G. Andersson, J. Krumlinde, G. Leander, Z. Szyma´ nski, Nucl. Phys. A 361 , 147 (1981).
formation is the one creating the localization of single-particle density distributions around a torus ring. This seems to be a basic mechanism of the emergence of high-spin torus isomers.
## ACKNOWLEDGMENTS
TI was supported in part by MEXT SPIRE and JICFuS. This work was undertaken as part by the Yukawa International Project for Quark-Hadron Sciences (YIPQS). J.A. M. was supported by BMBF under contract number 06FY9086 and 05P12RFFTG, respectively.
## Appendix: PARAMETERS AND VOLUME CONSERVATION IN THE RDHO MODEL
In the RDHO model, we take the oscillator frequency, ω 0 , to conserve the inner volume of an equi-potential energy surface. It is given by
$$\Lambda \text{ is } \begin{pmatrix} \dots \cdots \cdots \cdots \cdots \cdots \cdots \\ \arg \\ \text{title} \\ \text{of all} \\ \text{orus} \\ \text{ifgu} \\ \text{large} \\ \text{is} \\ \end{pmatrix} ^ { 3 } = \begin{cases} ( 1 + \frac { \eta ^ { 2 } } { 2 } ) \sqrt { 1 - \eta ^ { 2 } } \\ ( 1 + \frac { 3 } { 4 } \pi \eta ( 1 + \frac { 2 } { \pi } \arctan \frac { \eta } { \sqrt { 1 - \eta ^ { 2 } } } ) \\ ( 1 - \eta ^ { 2 } ) \\ \end{cases} \\ \text{ifgu} \\ \text{like} \\ \end{cases} \begin{pmatrix} ( 0 \leq \eta < 1 ) \\ ( 1 - \eta ^ { 2 } ) \\ ( 1 - \eta ^ { 2 } ) \\ \end{pmatrix} ^ { 3 } = ( \eta \geq 1 ), \\ \end{cases}$$
where η = R 0 / d and ◦ ω 0 denotes the oscillator frequency in the spherical limit. Here, we take /planckover2pi1 ◦ ω 0 = 41 A -1 3 / ρ torus /ρ gr MeV, where A is the number of nucleons, and ρ torus and ρ gr denote the average densities of a torus isomer and the ground state, respectively [11]. In the calculations, we use ρ torus = (2 / 3) ρ gr .
- [11] A. Staszczak and C. Y. Wong, arXive:1312.3413.
- [12] C.Y. Wong, Ann. Phys. 77 , 279 (1973).
- [13] J.A. Maruhn, P.-G. Reinhard, P.D. Stevenson, A.S. Umar, Comp. Phys. Comm. 185 , 2195 (2014).
- [14] P.-G. Reinhard, and R. Y. Cusson, Nucl. Phys. A 378 , 418 (1982).
- [15] E. Chabanat, P. Bonche, P. Haensel, J. Meyer, and R. Schae ff er, Nucl. Phys. A 627 , 710 (1997).
- [16] P.-G. Reinhard and H. Flocard, Nucl. Phys. A 584 , 467 (1995).
- [17] J. Bartel, P. Quentin, M. Brack, C. Guet, and H.-B. Hakansson, Nucl. Phys. A386 , 79 (1982).
- [18] P. Wen, L-G. Cao, J. Margueron and H. Sagawa, Phys. Rev. C 89 , 044311 (2014).
- [19] J.R. Stone and P.-G. Reinhard, Prog. Part. Nucl. Phys. 58 , 587 (2007).
- [20] D. Brink, Proceedings of the International School of Physics 'Enrico Fermi', Course 36, Varenna, 1965, edited by C. Bloch (Academic Press, London, 1966), pp. 247-277.
- [21] S. G. Nilsson and I. Ragnarsson, Shapes and Shells in Nuclear Structure , (Cambridge University Press, Cambridge, 1995). | 10.1103/PhysRevC.90.034314 | [
"T. Ichikawa",
"K. Matsuyanagi",
"J. A. Maruhn",
"N. Itagaki"
] | 2014-08-02T10:39:04+00:00 | 2014-08-02T10:39:04+00:00 | [
"nucl-th"
] | High-spin torus isomers and their precession motions | We systematically investigate the existence of exotic torus isomers and their
precession motions for a series of $N=Z$ even-even nuclei from $^{28}$Si to
$^{56}$Ni. We analyze the microscopic shell structure of the torus isomer and
discuss why the torus shape is generated beyond the limit of large oblate
deformation. We use the cranked three-dimensional Hartree-Fock (HF) method with
various Skyrme interactions in a systematic search for high-spin torus isomers.
We use the three-dimensional time-dependent Hartree-Fock (TDHF) method for
describing the precession motion of the torus isomer. We obtain high-spin torus
isomers in $^{36}$Ar, $^{40}$Ca, $^{44}$Ti, $^{48}$Cr, and $^{52}$Fe. The
emergence of the torus isomers is associated with the alignments of
single-particle angular momenta, which is the same mechanism as found in
$^{40}$Ca. It is found that all the obtained torus isomers execute the
precession motion at least two rotational periods. The moment of inertia about
a perpendicular axis, which characterizes the precession motion, is found to be
close to the classical rigid-body value. The high-spin torus isomer of
$^{40}$Ca is not an exceptional case. Similar torus isomers exist widely in
nuclei from $^{36}$Ar to $^{52}$Fe and they execute the precession motion. The
torus shape is generated beyond the limit of large oblate deformation by
eliminating the $0s$ components from all the deformed single-particle wave
functions to maximize their mutual overlaps. |
1408.0362v1 | ## Radiation friction vs ponderomotive effect
A.M. Fedotov, 1 N.V. Elkina, 2 E.G. Gelfer, 1 N.B. Narozhny, 1 and H. Ruhl 2
1 National Research Nuclear University 'MEPhI' (Moscow Engineering Physics Institute), 115409 Moscow, Russia 2 Ludwig-Maximilians Universit¨t, a M¨ unchen, 80539, Germany
The concept of ponderomotive potential is upgraded to a regime in which radiation friction becomes dominant. The radiation friction manifests itself in novel features of long-term capturing of the particles released at the focus and impenetrability of the focus from the exterior. We apply time scales separation to the Landau-Lifshitz equation splitting the particle motion into quivering and slow drift of a guiding center. The drift equation is deduced by averaging over fast motion.
PACS numbers: 41.75.Ht, 52.20.Dq, 52.25.Os, 52.65.Vv
A concept of ponderomotive potential [1] had been proven to be an extremely useful heuristic tool for interpreting electron scattering in experiments with focused laser fields, see e.g. [2-5]. Generalizations to relativistic intensities are straightforward but technically more tedious [6-8]. However, none of the papers [1-8] took into account effect of radiation friction because of relatively low laser intensity available at that time. The intensity of recent laser sources has been increased up to 2 × 10 22 W/cm 2 [9], thus enabling entering into ultrarelativistic regime for which the effects of radiation friction could become highly significant (see [10-12] for an overview). The unusual behavior of electrons in ultrastrong fields was recently revealed in numerical simulations [13-15]. However, none of these examples have yet convincing physical explanation. The purpose of this Letter is to analyze general properties of electron motion under the dominance of radiation friction as well as to outline a unifying framework of time scale separation leading to guiding center equations of motion for radiative electrons. As a result, we develop a new analytical approach which allows to formulate, explore and explain long-time radiative capturing of particles inside the focus and impenetrability of the focus for particles from the exterior.
A classical particle motion with radiation friction is governed by the Landau-Lifshitz equation [10, 16], where we retain just the leading term of radiation friction force F R proportional to the energy squared [26]:
$$\frac { d p } { d t } = F _ { L } + F _ { R }, \ \ F _ { L } = e \left ( E + v \times B \right ), \quad \left ($$
Then, substituting these expansions into Eq. (1) and equating the terms of the same powers of µ , we obtain where p , ε ( p ) = √ p 2 + m 2 and v = p /ε are the momentum, energy and velocity of a particle, respectively. Note that radiation friction dominates over the Lorentz force if the dimensionless field strength a 0 = e m √ 〈 A 2 〉 exceeds the critical value a 0 ∗ = ( 3 m/ e ω 2 2 ) 1 / 3 /similarequal 470, where the numerical value is given for optical frequency /planckover2pi1 ω /similarequal 1eV [12, 17]. This roughly corresponds to laser intensity I /greaterorsimilar 5 × 10 23 W/cm . In what follows we assume 2 that the field is that strong, a 0 /greaterorsimilar a 0 ∗ .
We also assume that the field is weakly inhomogeneous of the form A ( µ , t r ), where µ /lessmuch 1 is a small parameter. A 2D field configuration of this kind can be realized in a focal plane ( z = 0) of a couple of counter-propagating identically shaped focused monochromatic laser beams. In this context µ is the angular aperture of the pulses, w 0 = ( ωµ ) -1 is their waist and z R = (2 ωµ 2 ) -1 is the Rayleigh length. Note that at the focal plane the electric field E = -∂ A /∂t = O (1), while the magnetic field B = ∇× A is O µ ( ).
We split the particle motion r ( ) = t µ -1 R ( µ t 2 ) + ξ ( ) t into a slow drift of a guiding center and rapidly oscillating motion with small amplitude around it. Both functions can be expanded in powers of µ , R ≈ R 0 + µ R 1 , ξ ≈ ξ 0 + µ ξ 1 . Accordingly, r ≈ µ -1 r 0 + r 1 , v ≈ v 0 + µ v 1 and p ≈ p 0 + µ p 1 , where r 0 = R 0 ( µ t 2 ), v 0 = ξ ′ 0 ( ) and t
$$r _ { 1 } = \xi _ { 0 } ( t ) + R _ { 1 } ( \mu ^ { 2 } t ), \ \ v _ { 1 } = \xi _ { 1 } ^ { \prime } ( t ) + R _ { 0 } ^ { \prime } ( \mu ^ { 2 } t ). \ \ ( 2 )$$
$$& \frac { d p } { d t } = F _ { L } + F _ { R }, \quad F _ { L } = e \left ( E + v \times B \right ), \quad \left ( 1 \right ) \\ & F _ { R } \approx - \frac { 2 } { 3 } \frac { e ^ { 4 } } { m ^ { 4 } } \left [ \left ( \varepsilon E + p \times B \right ) ^ { 2 } - \left ( p \cdot E \right ) ^ { 2 } \right ] v, \quad \text{$\text{$\text{$\text{$equa}$} }$}$$
$$\dot { p } _ { 0 } = & e E _ { 0 } ( t ) - \frac { 2 } { 3 } \frac { e ^ { 4 } } { m ^ { 4 } } \left [ \varepsilon _ { 0 } ^ { 2 } E _ { 0 } ^ { 2 } - ( p _ { 0 } \cdot E _ { 0 } ( t ) ) ^ { 2 } \right ] v _ { 0 }, \\ \dot { p } _ { 1 } = & \, e ( E _ { 1 } + v _ { 0 } \times B _ { 1 } ) - \frac { 2 } { 3 } \frac { e ^ { 4 } } { m ^ { 4 } } \left \{ \left [ \varepsilon _ { 0 } ^ { 2 } E _ { 0 } ^ { 2 } - ( p _ { 0 } \cdot E _ { 0 } ) ^ { 2 } \right ] v _ { 1 } + 2 \varepsilon _ { 0 } E _ { 0 } ^ { 2 } ( v _ { 0 } \cdot p _ { 1 } ) v _ { 0 } - 2 ( p _ { 0 } \cdot E _ { 0 } ) ( E _ { 0 } \cdot p _ { 1 } ) v _ { 0 } + \\ & + 2 \varepsilon _ { 0 } ^ { 2 } ( E _ { 0 } \cdot E _ { 1 } ) v _ { 0 } - 2 \varepsilon _ { 0 } \left [ ( E _ { 0 } \times B _ { 1 } ) \cdot p _ { 0 } \right ] v _ { 0 } - 2 ( p _ { 0 } \cdot E _ { 0 } ) ( p _ { 0 } \cdot E _ { 1 } ) v _ { 0 } \right \}, \\ v _ { 1 } = & \frac { 1 } { \varepsilon _ { 0 } } \left [ p _ { 1 } - ( v _ { 0 } p _ { 1 } ) v _ { 0 } \right ],$$
$$v _ { 1 } = & \frac { 1 } { \varepsilon _ { 0 } } \left [ p _ { 1 } - \left ( v _ { 0 } p _ { 1 } \right ) v _ { 0 } \right ],$$
where v 0 = p 0 /ε 0 , ε 0 = ε ( p 0 ), E 0 = E r ( 0 , t ), E 1 = ( r 1 · ∇ ) E r ( 0 , t ) and B 1 = B r ( 0 , t ).
Now suppose that the vector E 0 ( ) = t E r ( 0 , t ) is lying in the plane z = 0 and at each fixed r 0 is rotating uniformly around the z -axis with angular velocity ω , while the magnetic field is oscillating along z -axis (this implies circular polarization of the colliding beams). In such a case the zeroth order equation (3) admits an exact solution [18]. Namely, after transition into the reference frame rotating along with the field this equation becomes autonomous, with an attractor, for which the momentum p 0 is lagging behind the field by an angle δ . This angle, together with the magnitude of the momentum, are determined by the equations
$$\cos \delta = ( a _ { 0 } / a _ { 0 * } ) ^ { 3 } \sin ^ { 4 } \delta, \ \ p _ { 0 } = m a _ { 0 } \sin \delta. \quad ( 6 ) \quad \underset { \text{domi} } { \ s i e t }$$
If a 0 /lessmuch a 0 ∗ , then radiation friction is negligible and δ ≈ π/ 2. In the opposite regime of dominance of radiation friction ( a 0 /greaterorsimilar a 0 ∗ ) δ is small and (6) can be reduced to δ ≈ ( a 0 ∗ /a 0 ) 3 / 4 /lessmuch 1, p 0 ≈ m a ( 3 0 ∗ a 0 ) 1 / 4 . The settling time τ s ∼ δ/ω is estimated from linearized near the attractor Eq. (3). For δ /lessmuch 1 it is much smaller than the rotation period and this is one of the typical features of the regime with dominating radiation friction. This essentially means that strong radiation friction reduces the effective number of degrees of freedom at time scales larger than rotation period.
Next we turn to consideration of the first order equations (4), (5). As follows from (2), the corrections r 1 , v 1 and p 1 contain both slow and fast motions, p 1 = p ( s ) 1 ( µ t 2 ) + p ( f ) 1 ( ). t In our case, the first order correction ξ 1 ( ) to fast motion represents rotation with doubled fret quency 2 ω . To separate slow and fast motions, we consider balance of the second Fourier harmonics in Eq. (4) to find p ( f ) 1 , then substitute it back and after averaging over a rotation period obtain p ( s ) 1 . Finally, by substituting the results into Eq. (5) and averaging, we obtain the equation
$$\frac { 1 } { \mu } \frac { d R _ { 0 } } { d t } \approx - \frac { \mu } { 2 \omega a _ { 0 } ( R _ { 0 } ) } \left [ n _ { \omega } \times \nabla a _ { 0 } ( R _ { 0 } ) + \frac { 5 } { 4 } \delta \nabla a _ { 0 } ( R _ { 0 } ) \right ], \quad \text{$\text{ is con}$} \\ ( 7 ) \quad \text{ing tc} \\ \text{can in}$$
where n ω is a unit vector directed along ω (i.e., the focal axis). For the sake of clarity, we present here Eq. (7) in a simplified form, with the accuracy O δ ( 2 ) [27].
In a sharp contrast to the ponderomotive effect regime for moderate a 0 /lessorsimilar a 0 ∗ , the slow drift motion is governed by a first order equation instead of the ponderomotive potential (i.e. second order equation). This is a principle feature of the regime of dominance of radiation friction and is attributed to the aforementioned reduction of the degrees of freedom due to presence of a strong friction. The same degeneration takes place e.g. in an inhomogeneous magnetic field, although due to a different physical reason of energy conservation in magnetic field [19].
Interestingly, in virtue of Eq. (7), drift is directed almost along the equipotentials a 0 = const, while ponderomotive force usually pulls the particles across the
FIG. 1: (Color online) Balance of forces acting on a single loop path of fast motion in inhomogeneous field: a) case of relativistic ponderomotive effect ( a 0 /lessmuch a 0 ∗ ): both the Lorentz force v 0 × B 1 and the resultant of electric forces acting on top and bottom of a loop are directed upwards, i.e. opposite to the gradient of the field amplitude ∇ a 0 ; b) case of dominance of radiation friction ( a 0 /greatermuch a 0 ∗ ): the resultant of radiation friction forces on top and bottom of the loop is directed almost to the left, i.e. opposite to n ω ×∇ a 0 .
equipotentials of the ponderomotive potential U pond = m √ 1 + a 2 0 . Thus, the actions of ponderomotive force in the regimes a 0 /greaterorsimilar a 0 ∗ and a 0 /lessorsimilar a 0 ∗ are almost orthogonal to each other. This feature is illustrated at a qualitative level in Fig. 1b), which explains why the resultant force acting on a single loop part of the trajectory r = ξ 0 ( ) t in an inhomogeneous field is directed orthogonally to the gradient of the field amplitude. This can be compared to the force balance in Fig. 1a), which illustrates just the usual ponderomotive effect. However, for finite values of a 0 , due to a smaller term ∝ δ originating from the Lorentz force, the drift still possesses a small component directed outwards. Only in the formal limit a 0 →∞ this component vanishes and particle drifts strictly along the equipotential.
This is demonstrated in Fig. 2 for a Gaussian profile of the beam. In order to emphasize the effect, we have chosen very high a 0 = 10 4 and µ = 0 1. . The numerical solution of the Landau-Lifshitz equation (curve 2), is compared to the solution without radiation friction (curve 1) and the untwisting spiral 3, which corresponding to the solution of the drift equation Eq. (7). One can immediately observe a drastically different behavior of curves 1 and 2. At the same time, curve 3 approximates correctly the drift of a guiding center in the time range t /lessorsimilar µ -2 ω -1 , for which r ∼ µ -1 R 0 /lessorsimilar w 0 . Further divergence of the curves 2 and 3 is explained by the corrections O µ ( 2 ) not considered here. Right after intersecting the level a 0 = a 0 ∗ the curve 2 stops circling and continues like the curve 1, as expected. For smaller values of a 0 /greaterorsimilar a 0 ∗ the number of windings of the trajectory around the focus is of course smaller, but the whole picture looks much the same.
Because of strong Lorentz time contraction in the lab frame, the time scales of quivering and drift for the curve 1 for our choice of parameters look commensurable. But with radiation friction taken into account, the drift remains non-relativistic in the time range t /lessorsimilar µ -2 ω -1 , no matter how strong is the field, just because the drift ve-
locity in Eq. (7) is O µ ( ). This point is illustrated at the inset in Fig. 2, where distance from the origin r t ( ) is depicted with and without account for radiation reaction. As a result, the trajectory 2 indeed demonstrates a nice separation of time scales thus proving applicability of time scales separation technique. But it also means that straightforward solution of the Landau-Lifshitz equation becomes a challenge for numerical methods in use. In order to gain the desired resolution and accuracy at an acceptable expense, we apply the approach suggested in Ref. [15], which is based on application of implicit schemes to the covariant equations of motion formulated in a proper time.
FIG. 2: (Color online) The numerical solutions of the NewtonLorentz equation (with radiation friction neglected, curve 1), the Landau-Lifshitz equation (curve 2) and the solution of the drift equation Eq. (7) (curve 3) combined together with the levels of a 0 . The employed field model is E x = ( mω/e a ) 0 ( r ) cos ωt , E y = ( mω/e a ) 0 ( r ) sin ωt , B z = ( m/e a ) ′ 0 ( r ) cos( ωt -φ ) (other field components vanish in the xy plane), where a 0 ( r ) = a 0 m exp( -µ ω r / 2 2 2 2), a 0 m = 10 4 and µ = 0 1. . Particle starts from the origin. Inset: dependence r t ( ) (all in units of ω -1 ) for the same solutions.
Another, and even more general, feature of the radiation friction dominated regime is impenetrability of the core of a localized field for high energy particles incoming from the exterior. This effect admits a rather simple explanation in terms of Pomeranchuk bound for radiative cooling [16, 20]. This bound was introduced in order to explain energy losses by cosmic rays in the earth magnetosphere. The energy of a particle coming from the exterior with initial gamma-factor γ i /greatermuch ( a 3 0 ∗ /a 0 ) 1 / 2 after a long enough path (in our case ∆ t ∼ w / 0 2 /greatermuch ω -1 ) in the field is reduced due to radiative cooling down to the value bounded by
$$\gamma _ { f } < \gamma _ { f } ^ { ( m a x ) } = \frac { 3 m ^ { 3 } / 2 e ^ { 2 } } { \int _ { t _ { i } } F _ { L \perp } ^ { 2 } ( r ( t ), t ) \, d t } \sim \frac { 4 a _ { 0 * } ^ { 3 } \mu } { a _ { 0 } ^ { 2 } } \ll \gamma _ { i }, \quad ( 8 ) \quad \text{tic, w}$$
FIG. 3: (Color online) Trajectories of particles scattered from the same Gaussian-shaped field as in Fig. 2 for different values of impact parameter with (solid lines) and without (dashed lines) account for radiation friction ( γ i = 2 × 10 , 3 a 0 m = 10 , 3 µ = 0 1). . Numbers mark the levels of a 0 ( r ).
where F L ⊥ is a component of the Lorentz force transverse to the velocity. After cooling is completed, as a qualitative estimate, we can apply a criterion γ f > a 0 for penetrability of a ponderomotive potential hill (thus neglecting radiation friction at this stage). But it is evident that such a criterion, being combined with Eq. (8), means impenetrability of the core of the field with respect to arbitrarily high energy particles if a 0 /greaterorsimilar a 0 ∗ (4 µ ) 1 / 3 . For example, for µ = 0 1 the threshold value of . a 0 would be of the order of 350. These estimates are in perfect agreement with numerical calculations. Indeed, all the scattering particles are reflected outwards just around the pointed out equipotential level, see Fig. 3. The 2D consideration in the focal cross section can be readily extended into 3D, where field satisfying Maxwell equations must possess a node-antinode structure along the focal axis. Then the high-energy particle crossing the field may partially penetrate through the nodes. However, the core regions surrounding antinodes still remain impenetrable for high-energy particles even in a 3D setup.
The results obtained within classical treatment based on the Landau-Lifshitz equation (1) have to be taken with certain care since at the specified extreme values of parameters quantum nature of radiation could be also important. However, quantum simulations based on the Monte-Carlo algorithm described in Ref. [21] reveal qualitatively the same behavior. The only important difference is that, due to random quantum recoils from photon emissions, the sampled trajectories become stochastic, with a noticeably wide spread around the classical trajectories Ref. [22]. In order to explain the origin of qualitative accordance of classical and quantum simulations, we estimate the dynamical quantum parameter
(proper acceleration in Compton units) χ /similarequal F L ⊥ γ/m 2 for parameters under consideration. For a particle released at the origin and a 0 = 10 4 we have δ ≈ µ = 0 1, . so that the components Eδ , B ∼ µE of the electric and magnetic fields transverse to velocity are of the same order. Then, taking into account Eq. (6), we can estimate χ ∼ (2 ω/m a δ ) 2 0 2 ∼ 4. Since it exceeds unity but not too much, quantum effects such as the spread of sampled trajectories must be indeed noticeable though not changing the whole picture completely. On the other hand, in a scattering setup the parameter χ remains even smaller than unity, attaining its maximum during radiative cooling, as it goes faster than particle approaches the core of the field. After cooling, the particles in fact behave classically. Nevertheless, quantum spread shows up in this setup as well because quantum fluctuations are amplified by the billiard instability of scattering dynamics.
To conclude, by analysis of the motion of charged particles in a 2D weakly inhomogeneous circularly polarized fields in radiation friction dominated regime, we reveal three rather basic features which discriminate this regime against the regime of usual motion in ponderomotive potential: (i) The trajectories of particles set in an equilibrium regime after a short time less than a period. In this regime the motion is well separated into a fast quiver motion and a drift of its guiding center. This drift is directed almost along the levels of the field amplitude, constituting with them a small angle δ . This is a principle difference with ponderomotive pulling across the levels of ponderomotive potential; (ii) Even more importantly, the drift velocity (7) is proportional to µ , making drift nonrelativistic [the component of the drift velocity across the levels of a 0 is even more suppressed by a factor δ because of (i)]. Combined together, (i) and (ii) mean that a particle placed initially inside the core of the field, would be long-term captured inside. This is in great contrast to the case of weaker fields, for which it would be rapidly pushed away with almost the speed of light from the top of a ponderomotive potential hill. (iii) Finally, we predict existence of a threshold field strength, above which any focused field must become completely impenetrable for high-energy particles. Our predictions (i)-(iii) could be tested with the forthcoming ELI and/or XCELS projects [23, 24]. More sophisticated experiments could be proposed for probing the internal structure of focused fields, since more elaborate spatial envelope profiles of the laser pulses with localized wells may allow to capture radiatively cooled fractions of particles and trap them inside over laser pulse duration. This sort of plasma trapping can have potentially a number of useful applications, including design of new γ -ray sources [25] and stimulating production of self-sustained QED cascades [21].
This research was supported by the Russian Fund for Basic Research (grant 13-02-00372), the RF President program for support of young Russian scientists and lead- ing research schools (grant NSh-4829.2014.2), the Sonderforschungsbereiche TransRegio-18 (Project B13), the Cluster-of-Excellence Munich-Centre for Advanced Photonics (MAP), and Arnold Sommerfeld Zentrum (ASZ).
- [1] A. V. Gaponov and M. A. Miller, ZhETF 34 , 252 (1958).
- [2] P. H. Bucksbaum, M. Bashkansky, and T. J. McIlrath, PRL 58 , 349 (1987).
- [3] J. L. Chaloupka and D. D. Meyerhofer, PRL 83 , 4538 (1999).
- [4] F. V. Hartemann, S. N. Fochs, G. P. Le Sage, N. C. Luhmann, Jr., J. G. Woodworth, M. D. Perry, Y. J. Chen, and A. K. Kerman, Phys. Rev. E 51 , 4833 (1995).
- [5] Y. I. Salamin and F. H. M. Faisal, Phys. Rev. A 55 , 3678 (1997).
- [6] T. W. Kibble, Phys. Rev. 150 , 1060 (1966).
- [7] D. Bauer, P. Mulser and W.-H. Steeb, PRL 75 , 4622 (1995).
- [8] N. B. Narozhny and M. S. Fofanov, JETP 90 , 753 (2000).
[9]
V. Yanovsky, V. Chvykov, G. Kalinchenko, et
al
,
Opt.
Express
16
,
2109 (2008).
- [10] F. Rohrlich, Classical Charged Particles (World Scientific, Singapore, 2007).
- [11] A. Macchi, A Superintense Laser-Plasma Interaction Theory Primer (Springer, Heidelberg, 2013).
- [12] A. Di Piazza, C. M¨ller, K. Z. Hatsagortsyan, and C. H. u Keitel, Rev. Mod. Phys. 84 , 1177 (2012).
- [13] A. Gonoskov, A. Bashinov, I. Gonoskov, C. Harvey, A. Ilderton, A. Kim, M. Marklund, G. Mourou and A. Sergeev, PRL 113 , 014801 (2014).
- [14] L. L. Ji, A. Pukhov, I. Yu. Kostyukov, B. F. Shen and K. Akli, PRL 112 , 145003 (2014).
- [15] N. V. Elkina, A. M. Fedotov, C. Herzing, H. Ruhl, Phys. Rev. E 89 , 053315 (2014).
- [16] L. D. Landau and E. M. Lifshitz, The classical theory of fields (Pergamon Press, Oxford, 1975).
- [17] S. V. Bulanov, T. Zh. Esirkepov, J. Koga and T. Tajima, Plasma Phys. Rep. 30 , 196 (2004).
- [18] Ya. B. Zel'dovich, Sov. Phys. Usp. 18 , 79 (1975).
- [19] N. G. van Kampen, Phys. Rep. 124 , 69 (1985).
- [20] I. Ya Pomeranchuk, JETP 9 , 915 (1939).
- [21] N. V. Elkina, A. M. Fedotov, I. Yu. Kostyukov, M. V. Legkov, N. B. Narozhny, E. N. Nerush, and H. Ruhl, Phys. Rev. STAB 14 , 054401 (2011).
- [22] N. Neitz and A. Di Piazza, PRL 111 , 054802 (2013).
- [23] E. Gerstner, Nature 446 , 16 (2007); G. Mourou, G. Korn, W. Sandner and J. K. Collier (eds) ELI - Extreme Light Infrastructure Science and Technology with Ultra-Intense Lasers, Whitebook (Berlin: THOSS Media GmbH, 2011).
- [24] Exawatt Center for Extreme Light Studies (XCELS), http://www.xcels.iapras.ru/ .
- [25] A. Pipahl, E. A. Anashkina, M. Toncian, T. Toncian, S. A. Skobelev, A. V. Bashinov, A. A. Gonoskov, O. Willi, and A. V. Kim, Phys. Rev. E 87 , 033104 (2013).
- [26] We use units /planckover2pi1 = c = 1 throughout the Letter.
- [27] Note that 60% contribution to the value of the numerical coefficient of the second term comes from coupling with second harmonic. It can be shown that higher even order harmonics are also present, but are much smaller and in our case slow motion is decoupled from them. | 10.1103/PhysRevA.90.053847 | [
"A. M. Fedotov",
"N. V. Elkina",
"E. G. Gelfer",
"N. B. Narozhny",
"H. Ruhl"
] | 2014-08-02T12:07:05+00:00 | 2014-08-02T12:07:05+00:00 | [
"physics.plasm-ph",
"physics.comp-ph",
"physics.optics"
] | Radiation friction vs ponderomotive effect | The concept of ponderomotive potential is upgraded to a regime in which
radiation friction becomes dominant. The radiation friction manifests itself in
novel features of long-term capturing of the particles released at the focus
and impenetrability of the focus from the exterior. We apply time scales
separation to the Landau-Lifshitz equation splitting the particle motion into
quivering and slow drift of a guiding center. The drift equation is deduced by
averaging over fast motion. |
Subsets and Splits
No saved queries yet
Save your SQL queries to embed, download, and access them later. Queries will appear here once saved.